[00:00:02]
WE'LL HAVE AN OPEN MEETING.[1. Call General Session to Order]
MORNING EVERYONE.WELCOME TO THE TECHNOLOGY AND SECURITY COMMITTEE OF ERCOT.
UM, THE, WE HAVE A QUORUM PRESENT.
ONE OF OUR MEMBERS, CARLOS AGUILAR, IS PRESENT BY TELEPHONE.
BEFORE WE BEGIN, I'D LIKE TO ASK, UH, PC CH PUC CHAIR TOM GLEASON, IF HE WOULD LIKE TO CALL AN OPEN MEETING OF THE PUBLIC UTILITY COMMISSION OF TEXAS TO ORDER.
THIS MEETING OF THE PUBLIC UTILITY COMMISSION OF TEXAS WILL COME TO ORDER TO CONSIDER MATTERS THAT HAVE BEEN DULY POSTED WITH THE SECRETARY OF STATE FOR TODAY, FEBRUARY 3RD, 2025.
BEFORE WE GET STARTED, I, I LIKE TO DRAW YOUR ATTENTION TO THE ANTITRUST ADMONITION AND THE SECURITY MAP THAT ARE SHOWN IN FRONT OF YOU.
ALWAYS IMPORTANT TO KNOW HOW TO GET OUT.
UM, FIRST, FIRST ITEM OF BUSINESS IS NOTICE OF PUBLIC COMMENT.
UM, THE, UH, TODAY'S MEETING AGENDA WAS POSTED PUBLICLY ON THE 24TH OF JANUARY, AND WE PROVIDE INSTRUCTIONS TO THE PUBLIC.
UH, AS OF THIS DATE, I DO NOT HAVE ANY, UM, COMMENTS.
UM, AND SO IF PEOPLE WANT TO MAKE COMMENTS, THEY SHOULD GO THROUGH THAT WEBSITE.
[3. December 2, 2024 General Session Meeting Minutes]
NEXT ITEM IS AGENDA ITEM THREE, WHICH IS THE MINUTES FROM THE DECEMBER 2ND GENERAL SESSION MEETING.UM, THERE IS A DRAFT IN THE MEETING MATERIALS.
I ASSUME EVERYONE'S LOOKED AT THEM.
ARE THERE ANY COMMENTS OR QUESTIONS? HEARING NONE, UH, CAN I HAVE A MOTION AND A SECONDER PLEASE? MR. THANK YOU.
UM, UH, ANY, UH, ANY OPPOSED? ANY, EVERYONE? FOUR.
I ASSUME THE MOTION IS CARRIED.
[4. Annual Charter Review and Approval]
ITEM IS THE REVIEW OF OUR ANNUAL CHARTER.UM, THE ANNUAL CHARTER IS UNCHANGED FROM LAST YEAR, AND, UH, THERE ARE NO CURRENT REVISIONS PLANNED.
IT'S IN, IT'S IN YOUR, UH, BOARD BOOKS.
UM, IS THERE, ARE THERE ANY QUESTIONS ABOUT THE CHARTER? UM, JULIE, I KNOW WE HAD SOME CLARIFICATION QUESTIONS ABOUT THE ROLE OF THE TWO DIFFERENT COMMITTEES, WHICH I THINK HAVE BEEN RESOLVED.
I THINK RTCS A GOOD EXAMPLE WHERE IT'S MOVING FROM TNS TO R AND M.
UH, I LOOK FORWARD TO HOW THE, YOU KNOW, EXECUTIVE TEAM WANTS TO SPLIT THAT UP BETWEEN THE COMMITTEES AND MAYBE WE'LL LEARN FROM THAT.
THE POINT OF DISCUSSION WAS THE PURPOSE OF THIS COMMITTEE IS OBVIOUSLY TO LOOK AT TECHNOLOGY, IT TECHNOLOGY IN PARTICULAR, UM, AS IT PROGRESSES THROUGH THE SYSTEM, BUT ALSO TO LOOK AT TECHNOLOGY IN GENERAL AS IT MIGHT AFFECT ERCOT IN THE FUTURE.
AND IN TODAY'S GUEST SPEAKER IS GONNA BE A GOOD EXAMPLE OF THAT.
UM, AND AS THINGS TRANSITION FROM FUTURE TO REALITY, AND RTC IS A GOOD EXAMPLE OF THAT, UM, IT SWITCHES OVER TO THE DOMAIN OF THE R AND M COMMITTEE.
UM, SO WITH NO FURTHER DISCUSSION, I DON'T THINK WE NEED A, I DON'T THINK THERE'S A MOTION NEEDED FOR THAT ONE, MR. SHANNON? NO.
[5. Emerging Technologies Guest Presentation “AI for Power Systems: Concepts and Transformative Applications”]
NEXT UP IS, I THINK A FASCINATING TOPIC.ONE THAT I'M PERSONALLY INTERESTED IN, UM, AND THAT IN, IN OUR EMERGING TECHNOLOGY SERIES IS A GUEST PRESENTATION ON AI FOR POWER SYSTEMS. UM, I'D LIKE TO THANK PASCAL NON HENDRICK, UH, FOR JOINING US TODAY.
PASCAL IS THE DIRECTOR OF THE AI HUB AT THE GEORGIA TECH UNI, GEORGIA TECH UNIVERSITY, AND IS THE DIRECTOR OF THE NATIONAL SCIENCE FOUNDATION AI INSTITUTE FOR ADVANCES AND OPTIMIZATION.
UM, HE'S A PROFESSOR OF COMPUTER SCIENCE AT BROWN UNIVERSITY FOR 20 YEARS.
HE LED OPTIMIZATION RESEARCH GROUPS AT NATIONAL ICT RESEARCH IN AUSTRALIA.
AND HIS CURRENT RESEARCH FOCUSES ON AI FOR ENGINEERING WITH APPLICATIONS AND ENERGY SYSTEMS, SUPPLY CHAINS, MANUFACTURING, HEALTHCARE, AND MOBILITY.
WE CARE A LOT ABOUT THE FIRST OF THOSE.
UM, COMMITTEE MEMBERS CAN RAISE QUESTIONS REGARDING RISK MANAGEMENT.
IN THE SECOND PART, WE'RE GONNA GO INTO EXECUTIVE SESSION, AND PASCAL'S GOING TO JOIN US FOR THAT AS WELL.
SO IF THERE ARE QUESTIONS THAT YOU WANT TO ASK IN A PRIVATE SESSION, PLEASE RESERVE THEM FOR THAT TIME.
SO I'M GONNA TALK ABOUT, UH, AI FOR POWER SYSTEM.
AS, UH, AS YOU JUST HEARD, I'M THE DIRECTOR OF, UH, ONE OF THE NATIONALS, UH, THE ONE OF THE NATIONAL SCIENCE FOUNDATION AI INSTITUTE, UH, FOR ADVANCING OPTIMIZATION.
AND I'M ALSO, UM, THE DIRECTOR OF THE AI HYPER GEORGIA TECH.
WE PRODUCE THE MOST GRADUATE, UH, THE MOST, UH, THE MOST STUDENTS, UH, IN AI ACROSS THE COUNTRY WITH ABOUT 3.2%.
UH, THIS IS MY IVORY TOWER, UH, THAT YOU CAN SEE THERE.
UM, WE LIVE IN THE, THE 12TH FLOOR
[00:05:01]
OF THE, OF, OF THIS CODA BUILDING.AND ONE FUN FACT IS THAT WE, WE USED TO HAVE THE LARGEST SPIRAL STAIRCASE, UH, IN THE WORLD.
NOW, UH, THIS BEING SAID, MOST OF WHAT WE DO IS USE INSPIRE RESEARCH.
SO WE ARE REALLY GUIDED, UM, BY CHALLENGES, UH, OF THE INDUSTRY, UH, WHETHER IT'S IN POWER SYSTEMS OR IN MANUFACTURING AND SUPPLY CHAINS.
AND WE APPLY THE FUNDAMENTAL RESEARCH THAT WE DO.
UH, WE HAVE A VERY RICH, UH, INDUSTRIAL, UH, PARTNER PROGRAM.
AND THIS IS REALLY WHAT IS DRIVING THE RESEARCH THAT WE DO.
UH, AS I TOLD YOU, WE ARE A NATIONAL SCIENCE FOUNDATION AI INSTITUTE, BUT FOR OPTIMIZATIONS, AND OBVIOUSLY WE LOVE OPTIMIZATION, AND THIS IS A TECHNOLOGY WHEN IT WORKS, WHICH IS MAGIC, AS YOU KNOW, UH, IN POWER SYSTEMS. UH, SO IF YOU HAVE AN OPTIMIZATION MODEL, UH, YOU GIVE IT AN INPUT, YOU GET AN OUTPUT, WHICH IS AN OPTIMAL SOLUTIONS, UH, BY USING AN OPTIMIZATION SOLVER.
AND SO, UH, IN GENERAL, WHEN IT WORKS, IT'S AMAZING, UH, BUT IT HAS ALSO SOME LIMITATIONS.
AND I'M GONNA TALK ABOUT THREE OF THEM, AND HOPEFULLY THE REST OF THE PRESENTATION WILL ADDRESS THEM LATER ON IN THE CONTEXT OF ENERGY SYSTEM.
BUT ONE OF THE THINGS THAT OPTIMIZ OPTIMIZATION TECHNOLOGIES HAVE, HAVE CHALLENGES IS WHEN YOU ARE IN THE VERY STRICT REAL TIME CONSTRAINTS, WHICH IS OBVIOUSLY A CASE IN IN POWER SYSTEM.
IT'S ALSO THE CASE WHERE YOU HAVE HUMAN IN THE LOOPS.
HUMAN TYPICALLY DON'T LIKE TO WAIT FOR 40 MINUTES TO GET AN ANSWER.
AND FINALLY, AND THIS IS GONNA BE RELEVANT, AND I WILL SHOW YOU A CASE STUDY WHEN YOU HAVE TO RUN VERY LARGE, UH, SIMULATION, FOR INSTANCE, FOR RISK ASSESSMENT.
UH, IF YOU DON'T HAVE THE RIGHT TECHNOLOGY, IT MAY TAKE TOO LONG FOR ACTUALLY GETTING THE RIGHT ANSWER AT THE RIGHT TIME.
AND HOPEFULLY WHAT I CAN DO TODAY IS TO SHOW YOU THAT, UH, WHEN YOU ARE USING AI, SOME OF THESE CHALLENGES THAT OPTIMIZATIONS ARE GONNA DISAPPEAR.
UH, ONE OF THE THINGS THAT WE, UH, WE LEVERAGE IS THE FACT THAT WE LIVE IN THE REAL WORLD.
AND IN THE REAL WORLD WE HAVE, WE ARE OPTIMIZING OVER VERY COMPLEX INFRASTRUCTURES.
AND THESE INFRASTRUCTURES ARE NOT CHANGING VERY QUICKLY.
YOU DON'T BUILD A TRANSMISSION LINE EVERY DAY, OR YOU DON'T PUT A NEW KEY CRANES, UH, KEY CRANE IN A PORT EVERY DAY.
SO WHAT WE ARE DOING, REALLY MANY OF THE PROBLEMS THAT WE FACE, UH, IN, IN, IN PRACTICE ARE, ARE THE, ARE, ARE SOLVING THE SAME PROBLEM REPEATEDLY OVER VERY SIMILAR INSTANCES.
AND THEREFORE WE CAN, WE CAN LEVERAGE THAT.
AND WE TYPICALLY HAVE A LOT OF HISTORICAL DATA, OR WE HAVE VERY GOOD FORECASTING METHODS.
AND SO WE ARE IN A SITUATION WHERE, YES, THE PROBLEMS ARE EXTREMELY COMPLEX, BUT THEY ARE THE SAME EVERY DAY OR EVERY FIVE MINUTES OR EVERY TWO MINUTES.
UH, AND WE HAVE A LOT OF DATA ABOUT THEM, WHICH IS TYPICALLY VERY GOOD FOR AI.
SO TO GET SO, SO TO, TO GET THE SPEED THAT WE NEED TO ACTUALLY, UH, UH, SPEED UP THIS OPTIMIZATION ALGORITHM, ONE OF THE THINGS THAT WE CAN DO IS, IS START IN THINKING ABOUT USING MACHINE LEARNING.
AND WHY CAN WE DO THAT? BECAUSE ESSENTIALLY WHEN YOU LOOK AT OPTIMIZATION, OPTIMIZATION IS A FUNCTION.
YOU GET AN INPUT, LET'S SAY, YOU KNOW, THE LOAD IN YOUR SYSTEMS AND, AND, AND, AND THE NET LOAD IN YOUR SYSTEM.
AND THE OUTPUT IS GONNA BE YOUR SET POINTS FOR THE GENERATOR.
IT'S A FUNCTION FROM INPUT TO THE OPTIMAL DISPATCH.
AND THIS IS WHAT MACHINE LEARNING AI IN GENERAL IS VERY GOOD AT.
UM, MACHINE LEARNING IS A UNIVERSAL FUNCTION, APPROXIMATOR.
SO WE BASICALLY HAVE A DEEP LEARNING SYSTEM THAT WE, WE'VE, WE'VE, YOU KNOW, UH, MILLIONS OR TRILLIONS OF PARAMETERS THAT YOU WANT TO TUNE SUCH THAT IT APPROXIMATES THIS OPTIMIZATION AND THIS OPTIMIZATION MODELS.
NOW, UH, SO IF I, IF I CAN SUMMARIZE WHAT, WHAT I'M TRYING TO PITCH YOU HERE IS THAT WHAT WE ARE TRYING TO DO IS SOLVING WHAT IS CALLED A MULTIPARAMETRIC OPTIMIZATIONS.
WE HAVE A DISTRIBUTION OF INPUT, AND WE WANT TO SOLVE THOSE PROBLEMS. WE PUT ALL THE COMPUTATIONAL BACKGROUND OFFLINE, LEARNING THE OPTIM, THE, THE MACHINE LEARNING ALGORITHM, THE AI SYSTEM, AND THEN WHEN WE APPLY THE MODEL IN REAL TIME, IT TAKES MILLISECONDS.
YOU PUSH EVERYTHING OFFLINE, AND THEN IN REAL TIME, YOU ARE ORDERS OF MAGNITUDE FASTER.
SO THE QUESTION I HAVE FOR YOU IS, DOES IT WORK? AND THE ANSWER TO THAT QUESTION IS PRESENTED THIS WAY, IT DOESN'T WORK, RIGHT? SO BECAUSE WE HAVE TO SATISFY ALL THE PHYSICAL ENGINEERING AND BUSINESS CONSTRAINTS THAT THIS ENGINEERING SYSTEM, THIS OPTIMIZATION MODEL HAVE, SO WE HAVE TO MAKE SURE THAT THE, YOU KNOW, THE SYSTEM IS, AND THE MACHINE LEARNING SYSTEM, THE AI SYSTEM IS GONNA BE RELIABLE, SUCH THAT IT CAN BE DEPLOYED IN PRACTICE.
IT HAS ALSO TO BE HIGH QUALITY.
YOU WANNA MAKE SURE THAT THE SOLUTIONS THAT IT ARE PRO, THAT'S PRODUCING ARE ACTUALLY VERY CLOSE OR, YOU KNOW, OPTIMAL.
AND THEN FINALLY, YOU WANT IT TO SCALE.
THE PROBLEMS THAT WE FACE IN THE POWER SYSTEM INDUSTRY HAVE MILLIONS OF VARIABLES AS INPUT AND OUTPUT, WHICH ARE, YOU KNOW, OF THE OTHER, OF HUNDREDS OF THOUSANDS.
NOW, THIS IS VERY DIFFERENT FROM A CLASSIFICATION PROBLEM IN MACHINE LEARNING, TYPICALLY, UH, WHERE YOU ARE, YOU ARE BASICALLY TRYING TO FIND THE CATEGORY OF SOME IMAGES OR SOMETHING LIKE THAT.
VERY DIFFERENT NATURE OF THESE PROBLEMS. SO WHAT ARE WE GONNA DO? WHAT I WANT TO AS A MESSAGE FOR YOU IS THAT DESPITE ALL THESE THINGS THAT, YOU KNOW, A COUPLE OF YEARS WE COULDN'T DO, I'M ABOUT TO TELL YOU THAT AI IS ACTUALLY READY FOR CRITICAL POWER SYSTEM APPLICATIONS.
AND I'M ACTUALLY GONNA TELL YOU A SECOND MESSAGE, WHICH IS THAT THIS IS THE KEY ENABLER TECHNOLOGY TO DO THINGS THAT WE ARE NOT CAPABLE OF, THAT WE WERE NOT CAPABLE OF DOING
[00:10:01]
A FEW YEARS AGO.SO THESE ARE THE TWO MESSAGE THAT I WANT TO, YOU KNOW, BASICALLY CONVEY TODAY.
AND I'M GONNA SHOW YOU TWO CASE STUDIES FOR ILLUSTRATING THIS.
AND THEY ARE BASED ON, YOU KNOW, REAL CASE STUDY, UH, WITH REAL PARTNERS IN THE POWER SYSTEM INDUSTRY.
SO, UH, ONE OF THE BIG THINGS THAT WE HAVE TO DO IS TO BASICALLY CAPTURE ALL THE PHYSICS AND THE, AND THE, AND THE ENGINEERING CONSTRAINTS OF THE APPLICATION.
SO LET ME SHOW YOU AN EXAMPLE.
UH, THIS IS, UH, BASED ON A COLLABORATION THAT WE HAD WITH MISO.
AND WHAT YOU SEE THERE IS THE TYPICAL, UH, THE TYPICAL PIPELINE OF AN ISO IN THE UNITED STATES.
NOW, THEY ARE ALL DIFFERENT, A LITTLE BIT FOR, FOR EVERY ONE OF THE ISO, BUT ROUGHLY SPEAKING, THERE ARE TWO THINGS THAT YOU ARE DEALING WITH, THE COMMITMENT OF THE GENERATORS AND THEN THE DISPATCHING.
AND TYPICALLY YOU HAVE, YOU KNOW, UM, A UNIQUE COMMITMENT THAT DAY BEFORE, AND THEN A NUMBER OF RELIABILITY COMMITMENTS, YOU KNOW, A COUPLE OF HOURS BEFORE, AND THEN MAYBE EVERY 15 MINUTES.
AND THEN YOU HAVE THE REAL TIME DISPATCH.
SO ONE OF THE THINGS THAT, UM, UM, SO I'M GONNA SHOW YOU A LITTLE BIT, YOU KNOW, THE CONSTRAINTS THAT YOU HAVE, BECAUSE THIS, THEY'RE GONNA BE RELEVANT WHEN WE TALK ABOUT THE AI SYSTEM.
THE FIRST THING THAT YOU WANNA MAKE SURE IS THAT AT EVERY POINT IN TIME YOU SATISFY THE POWER BALANCE.
SO WHAT I'M SHOWING YOU HERE IS KIND OF A STYLIZED VERSION OF WHAT, YOU KNOW, MISO IS ACTUALLY SOLVING.
SO AT EVERY POINT THE LOAD AND THE GENERATIONS HAVE TO MATCH, THEN YOU HAVE TO SATISFY ALL THE RELIABILITY CONSTRAINTS.
AND THE WAY MANY OF THE ISOS ARE DOING THIS IS UP BY HAVING RESERVES, AND YOU HAVE TO HAVE IN, IN OF GLOBAL RESERVE, TYPICALLY LOCAL RESERVE AS WELL.
AND THEN YOU HAVE ALL THE ENGINEERING CONSTRAINTS.
YOU DON'T WANT TO DESTROY YOUR GENERATORS, YOUR TRANSFORMERS, AND SO ON.
THE TRANS LINE ARE TYPICALLY, UM, SOFT CONSTRAINTS WITH A VERY HIGH PENALTY IN THE OBJECTIVE.
NOW, NORMALLY I HAVE A BIG DEMO THAT SHOWS YOU EVERYTHING MOVING, UH, MOVING ON THE FRENCH SYSTEM.
OBVIOUSLY I CAN'T SHOW YOU, YOU KNOW, US SYSTEM IN GENERAL.
UH, BUT WHAT YOU SEE THERE IS, UH, UH, ESSENTIALLY HOW THE, HOW THE SYSTEM WOULD WORK IF YOU WOULD APPLY, YOU KNOW, THE, THE WAY ISOS IN THE UNITED STATES ARE RUNNING THE SYSTEM, UH, IF YOU WOULD APPLY THAT TO THE FRENCH SYSTEM, YOU SEE THE ENERGY ON THE LEFT, YOU SEE THE RESERVE ON THE RIGHT.
AND TYPICALLY, WHEN, WHEN I HAVE A DEMO OF THIS, I SHOW YOU HOW FAST THESE THINGS ARE CHANGING EVERY FIVE MINUTES, UH, THE GENERATORS ARE CHANGING VERY SIGNIFICANTLY.
THE RESERVE, UH, ARE CHANGING VERY, VERY FAST.
NOW, THIS WORKS, BUT ONE OF THE THINGS THAT WE HAVE IS THAT, UH, UM, MOST OF THE ISOS ARE FACING INCREASED VOLATILITY, UM, BECAUSE OF, YOU KNOW, THE DIFFERENT TYPES OF GENERATIONS THAT WE HAVE, BECAUSE, BECAUSE OF THE VOLATILITY IN THE GAS PRICES THAT WE HAVE, BUT ALSO BECAUSE OF THE ELECTRIFICATION OF EVERYTHING DISTRIBUTED, ENERGY, RE RESOURCES, DATA CENTERS, EVERYTHING, UH, THE VOLATILITY IS SUBSTANTIALLY LARGER THAN IT USED TO BE.
AND THIS IS RAISING FUNDAMENTAL CHALLENGES.
SO A LOT OF THE, THE PEOPLE WE, THE, THE INDUSTRIES WE ARE WORKING WITH, WHAT THEY WOULD LIKE TO HAVE IS A REAL TIME RISK ASSESSMENT.
SOMETHING THAT THEY CAN RUN EVERY TWO MINUTES OR EVERY MINUTE, SUCH THAT THEY CAN SEE WHAT IS THE RISK IN ON OF THIS SYSTEM IN THE NEXT 24 HOURS.
AND SO IF YOU SORT OF PIPELINE THAT I'VE SHOWN YOU BEFORE, WE WANNA DO EXACTLY THAT.
BUT INSTEAD OF HAVING ONE SCENARIO, WE ARE GONNA HAVE PLENTY OF SCENARIOS OF WHAT THE FUTURE CAN BE.
AND FOR EVERY ONE OF THEM, WE CAN EVALUATE WHAT THE RISK OF THE SYSTEM WOULD BE.
NOW, IF YOU TRY TO DO THAT IN PRACTICE, AND YOU WANT TO DO THAT, LIKE IN, YOU KNOW, LESS THAN A MINUTE, YOU WOULD'VE TO SOLVE ABOUT 300 OPTIMIZATION PROBLEMS FOR ONE SCENARIO.
AND THAT WOULD TAKE 15 MINUTES.
SO 15 MINUTES FOR JUST ONE SCENARIO, EVEN IF YOU HAVE A GOOD PLATFORM.
SO MY GOAL IS TO SHOW YOU THAT WE CAN DO THAT EXTREMELY FAST WITH AI TECHNOLOGIES.
AND SO HOW ARE WE GONNA DO THAT? THE FIRST THING WE NEED TO DO IS TO HAVE A CRYSTAL BALL.
UH, THE CRYSTAL BALL IS GONNA TELL YOU WHAT THE FUTURE CAN BE.
AND SO ONE OF THE THINGS THAT HAS HAPPENED IN AI AS WELL IS THAT THE PROGRESS IN PROBABILISTIC FORECASTING HAS BEEN VERY SUBSTANTIAL.
WE CAN PR, YOU KNOW, WE CAN PREDICT JOINTLY THOUSANDS OF TIME SERIES AT THE SAME TIME, WE CAN GENERATE CORRELATED SCENARIOS FOR EVERY ONE OF THEM AND QUANTIFY UNCERTAINTY.
SO WHAT YOU SEE THERE, AGAIN, NORMALLY, YOU KNOW, YOU, YOU WOULD SEE A, A, AN ANIMATIONS OF THIS, BUT YOU CAN SEE THE PREDICTION FOR THE WIN FOR THE NEXT 24 HOURS.
AND YOU CAN ALSO SEE ON THE RIGHT, YOU KNOW, WHETHER YOU ARE OVERESTIMATING OR UNDERESTIMATING, UH, THE WIND PRODUCTION.
UH, THIS IS ON THE, ON THE NORTH OF THE MISO SYSTEM USING A, A PUBLIC DATA SET HERE.
SO MOST OF THESE MODELS ARE BOTH DATA DRIVEN, WHICH AI LIKES AND ALSO WEATHER DRIVEN.
UH, YOU GET SOME OF THE PHYSICS AS WELL BECAUSE THE COMBINATIONS OF THE TWO GIVE YOU MUCH BETTER MODEL.
SO JUST TO GIVE YOU A SENSE, YOU KNOW, THE LATEST TECHNOLOGY, WHICH ARE CALLED TEMPER INFUSION TRANSFORMERS, THIS IS BASED ON, YOU KNOW, SOME OF THE, THE TYPE, THE TRANSFORMERS TECHNOLOGY THAT, UH, UH, IS DRIVING SOME OF THE LARGE LANGUAGE MODELS THAT YOU ALL KNOW ABOUT, UH, BUT ALSO OTHER TECHNOLOGIES INSIDE, UH, THIS IS THE KIND OF RESOURCE THAT THEY GET ABOUT 1% FOR THE LOAD, WITHIN 1% OF ACCURACY FOR THE LOAD, WITHIN 8% FOR THE WIND, UH, WHICH IS REMARKABLE.
AND WE WITH ABOUT 3% FOR SOLAR.
SO NOW WE HAVE OUR CRYSTAL BALL, WE KNOW WHAT THE FUTURE COULD BE, WE CAN GENERATE MANY SCENARIOS OF THE FUTURE.
AND WHAT WE WANT TO DO NOW IS TO TRY TO PREDICT, YOU KNOW, WHAT THE, WHAT'S GONNA HAPPEN TO THE SYSTEM.
[00:15:01]
WE ALSO WANT TO CAPTURE WHEN WE DO THAT, ALL THE PHYSICS.AND SO THIS IS WHAT, THIS IS PROBABLY THE MOST IMPORTANT SLIDE.
THIS IS WHAT THE AI SYSTEM FOR ENGINEERING APPLICATIONS ARE ALL ABOUT.
IT'S A COMBINATIONS OF A DEEP LEARNING MODEL WITH A REPAIR LAYER.
AND SO, AS I TOLD YOU, THE MACHINE LEARNING SYSTEM ON ITS OWN WILL NEVER SATISFY ANY CONSTRAINTS.
AND SO WHAT WE ADD IS ANOTHER LAYER, WHICH IS A REPAIR LAYER, WHICH TAKES THAT, THAT APPROXIMATION AND MAP IT INTO A FEASIBLE, RELIABLE SOLUTION.
THAT'S THE, THE AI SYSTEM THAT I'M GONNA SHOW YOU ARE ALL BASED ON THESE OPTIMIZATION PROXIES.
AND SO THIS IS IN TWO DIMENSION THAT I WANT TO ILLUSTRATE THAT FOR YOU.
OBVIOUSLY, YOU KNOW, ON A REAL SYSTEM, YOU DO THAT IN 60,000 DIMENSION.
AND WHAT YOU CAN SEE HERE, WHAT YOU CAN SEE HERE IS THAT THIS IS, THIS IS THE TWO GENERATORS THAT YOU SEE THERE AND THERE, AND YOU SEE THE MACHINE LEARNING PREDICTION HERE, WHICH IS OUTSIDE THE FEASIBLE REGION, WHICH IS HERE.
AND SO THE ROLE OF THE REPAIR LAYER IS TO TAKE THIS APPROXIMATION AND PROJECT IT BACK IN THE FEASIBLE REGION WHERE YOU KNOW THAT YOUR SYSTEM CAN OPERATE RELIABLY.
AND SO ONE OF THE THINGS THAT YOU HAVE TO KNOW IS THIS IS HOW YOU TRAIN THOSE MODELS.
THIS IS, THIS IS, OR ANY MODEL THAT YOU HAVE SEEN, UH, IN, IN THAT YOU KNOW ABOUT IN A, YOU KNOW, IN AI ARE BASICALLY TRAINED.
THIS IS CALLED, UH, GRADIENT, YOU KNOW, UH, STOCHASTIC GRADIENT DESCENT.
WHAT YOU HAVE TO DO IS THAT YOU HAVE TO SEE, YOU KNOW, WHAT IS THE OUTPUT OF YOUR, OF YOUR, OF YOUR MACHINE OR YOUR AI SYSTEM, AND YOU HAVE TO MAKE SURE THAT IT'S ACTUALLY AS CLOSE AS POSSIBLE TO THE BEST SOLUTION OF THE OPTIMIZATION MODEL.
AND SO WHAT YOU DO IS YOU DO A GRADIENT STEP FOR ACTUALLY, UM, ADJUSTING ALL THE PARAMETERS, THE WEIGHT OF YOUR, OF YOUR MACHINE LEARNING SYSTEM.
SO, YOU KNOW, WHEN YOU TALK ABOUT, WHEN YOU THINK ABOUT WHAT OPENAI AND OTHER COMPANIES ARE DOING, THEY HAVE TRILLIONS OF THESE PARAMETERS.
THAT'S WHAT WE ARE TALKING ABOUT, THE PARAMETER HERE, WHICH ARE ADJUSTED HERE TO TRY TO FIND THE BEST AND AND FEASIBLE SOLUTION.
SO HOW DO WE DO THAT IN PRACTICE? THIS IS WHAT WE HAVE TO DO FOR THE MISO SYSTEM, FOR INSTANCE.
SO WE HAVE THE PREDICTION ON THE LEFT, WE HAVE A SIGMOID LAYERS FOR MAKING SURE THAT YOU NEVER BREAKS ANY OF YOUR GENERATORS, AND THEN YOU HAVE THE REPAIR LAYERS FOR THE POWER BALANCE AND FOR THE, UH, AND FOR THE RESERVE LAYERS.
NOW, UH, WHAT DOES, WHAT DO THESE DO? I WILL JUST GIVE YOU THE INTUITION AGAIN IN TWO DIMENSION, BUT WE HAVE TO DO THAT IN 60,000 DIMENSION IN GENERAL, RIGHT? AND SO, UH, THE WAY WE DO THIS IS THAT IF WE HAVE NOT ENOUGH POWER, WE'RE GONNA SCALE THE GENERATORS UP, YOU KNOW, USING ID FORM CONTROL.
AND THEN IF WE HAVE TOO MUCH POWER, WE SCALE THEM DOWN.
AND THE BEAUTIFUL THING HERE IS WHAT YOU SEE HERE ON THE RIGHT, THIS IS THE FORMULA THAT YOU HAVE TO USE, AND THEY ARE BEAUTIFUL BECAUSE THEY'RE DIFFERENTIABLE ALMOST EVERYWHERE.
SO THEY ARE PART OF THIS MACHINE LEARNING SYSTEM, AND YOU CAN DIFFERENTIATE OVER AND YOU CAN TRAIN YOUR MACHINE LEARNING MODEL THAT WAY.
SO, UH, I'M GONNA SKIP THIS ONE, BUT THE BOTTOM LINE IS THAT INSTEAD OF TAKING 15 MINUTES, NOW, WE ARE ABLE TO DO THIS REAL TIME RISK ASSESSMENT FOR ALL THE SCENARIOS FOR A THOUSAND SCENARIO IN ABOUT FIVE SECONDS.
AND THE KEY POINT IS THAT EVERY OF THESE OPTIMIZATION NOW IS REPLACED BY THESE OPTIMIZATION PROXIES.
SO YOU HAVE THESE PROXIES THAT ARE TAKING MILLISECONDS, AND THEY ARE REPLACING THESE OPTIMIZATION MODELS THAT ARE TAKING TWO OR THREE ORDERS OF SUBMITTED MORE, AND YOU DECREASE THE TIME TREMENDOUSLY.
SO WHAT IS AVAILABLE NOW IS ESSENTIALLY A SYSTEM THAT CAN EVALUATE RISK IN REAL TIME FOR YOUR OPERATORS.
OKAY? SO LET ME MOVE TO THE SECOND PART, WHICH IS CONTINGENCIES AND SECURITY CONSTRAINTS.
OPF, THIS IS PROBABLY ONE OF THE BIGGEST CHALLENGE THAT WE FACE AS A, AS AN INDUSTRY.
IT'S LIKE, HOW CAN YOU ACTUALLY DEAL WITH MAJOR CONSTITUENCY? YOU LOSE ALL THE WIND, YOU LOSE MAJOR GENERATORS, YOU HAVE MASSIVE CONGESTION IN YOUR SYSTEM.
HOW CAN YOU DO THAT? UH, AGAIN, I'M GONNA ILLUSTRATE THAT ON THE FRENCH SYSTEM.
I'M NOT FRENCH BY THE WAY, BUT YOU KNOW, UH, WE HAVE A DEEP COLLABORATION WITH THE FRENCH SYSTEM.
SO IF YOU LOSE A NUCLEAR GENERATOR THAT YOU SEE THERE, SO THE TWO, YEAH, THERE ARE TWO THINGS HAPPENING.
WE HAVE PRIMARY CONTROL, AND SO THAT'S BASICALLY STABILIZING THE FREQUENCY AND THEN SECONDARY CONTROL, THAT'S A GC, WHICH IS BRINGING THE FREQUENCY BACK.
NOW, I'M GONNA SHOW YOU A SLIDE IS GONNA BE SCARY, IS SCARY FOR ME TOO.
THIS IS THE WAY YOU WOULD SOLVE THIS PARTICULAR MODEL.
SO THIS IS CALLED A SECURITY CONSTRAINED OPF WITH PROPORTIONAL RESPONSE OF THE GENERATOR.
THAT'S WHAT YOU SHOULD ACTUALLY DO IN PRACTICE.
THAT'S WHAT THE CONTROL SYSTEM IS DOING.
MODELING THIS IS VERY DIFFICULT, BUT YOU CAN RECOGNIZE SOME OF THE THINGS THAT WE HAVE TALKED ABOUT.
YOU HAVE THE POWER BALANCE THERE, YOU HAVE THE, THE THERMAL LIMITS THERE.
THIS PART THAT I'M HIDING, THERE ARE ALL THE CONTINGENCIES.
THIS IS EXTREMELY DIFFICULT TO DO BECAUSE YOU HAVE A VERY LARGE NUMBER OF THEM, AND FOR EVERY ONE OF THEM, YOU ARE DUPLICATING ALL THESE CONSTRAINTS IN THERE, AND YOU'RE ALSO HAVING TO MODEL THE PROPORTIONAL RESPONSE OF THE GENERATORS.
NOW, I CAN TELL YOU THAT NOBODY CAN DO THIS IN REAL TIME.
UM, AND SO WHAT I'M GONNA SH WHAT I'M GONNA, WHAT I'M GONNA TELL YOU IS THAT WITH AI TECHNOLOGIES, YOU CAN REDUCE THE TIME FOR 2.5 HOURS.
YOU HAVE TO DO THAT EVERY FIVE MINUTES, RIGHT? UH, WE ARE REDUCING THAT TO 10 MILLISECONDS.
SO THIS IS ON THE FRENCH SYSTEM THERE, YOU SEE THAT JUST RUNNING THE BEST ALGORITHM, WHICH IS AVAILABLE IS TAKING 2.5 HOURS.
[00:20:01]
KNOW WE CAN DO THAT IN, IN ABOUT 10 MILLISECONDS USING AI TECHNOLOGIES.AGAIN, WHAT IS HAPPENING IS THAT WE ARE LEARNING THIS OFFLINE, AND THEN WE ARE APPLYING THIS IN REAL TIME.
AND THAT TAKES ME 10 MILLISECONDS ON THE GPU, YOU KNOW, FROM NVIDIA, FOR INSTANCE.
SO LET ME JUST TELL YOU A LITTLE BIT ABOUT THE MATH BEHIND THIS.
SO WHAT WE ARE DOING IS LEARNING TWO THINGS.
WE ARE LEARNING HOW TO PRODUCE FEASIBLE SOLUTION, BUT WE ARE ALSO LEARNING AUTOMATICALLY THE PENALTY THAT WE HAVE TO PUT ON THE CONSTRAINTS TO MAKE SURE THAT, UH, WE GET A FEASIBLE SOLUTION AT THE END.
SO THIS IS PROBABLY THE, THE MOST SOPHISTICATED ARCHITECTURE THAT WE HAVE IS CALLED PRIMAL DUAL LEARNING.
THE ONE THING THAT I WANT TO TELL YOU ABOUT, WHICH IS REALLY IMPORTANT, IS THIS SELF SUPERVISED NATURE.
WE DON'T NEED LABELED DATA, WE JUST NEED A VERY GOOD INPUT DISTRIBUTION OR FORECAST OF THE FUTURE.
AND THEN THE SYSTEM IS BASICALLY TRAINED AUTOMATICALLY USING THAT.
SO WE DON'T NEED SOMEBODY TELLING US, HEY, THIS IS HOW YOU SOLVE THESE OPTIMIZATION PROBLEMS, BECAUSE THAT WOULD TAKE TOO MUCH TIME.
SO, UH, THIS IS THE PRIMAL NETWORK INSIDE.
SO IT'S, IT'S WHAT I'VE SHOWN YOU BEFORE.
NO, BECAUSE WE DON'T ONLY HAVE, YOU KNOW, THE MACHINE, THE DEEP LEARNING SYSTEM, THEN THE, THE, THE, THE REPAIR LAYERS THAT I'VE SHOWN YOU.
BUT NOW WE HAVE TO DO THESE REPAIR LAYERS FOR EVERY ONE OF THE CONTINGENCIES AS WELL.
SO THIS IS CALLED DIFFERENTIABLE PROGRAMMING.
THAT'S THE WAY ACTUALLY DEEP AI SYSTEMS ARE ACTUALLY WORKING.
SO, SO LET ME CONCLUDE BEFORE MY, YOU KNOW, I, I, I ANSWER SOME OF YOUR QUESTIONS IF YOU HAVE SOME.
SO WHAT I WANTED TO TELL YOU IS THAT AI IS READY FOR CRITICAL POWER SYSTEM APPLICATION.
WE CAN APPLY THE TECHNOLOGY AND IT'S GONNA BE RELIABLE BECAUSE WE CAN TAKE INTO ACCOUNT THE PHYSICAL AND ENGINEERING CONSTRAINTS INSIDE THIS AI SYSTEM, AT LEAST FOR SOME, YOU KNOW, UH, CRITICAL POWER SYSTEM APPLICATION.
THE OTHER THINGS THAT I WANTED TO TELL YOU IS THAT THIS IS A KEY TECHNOLOGY ENABLERS, BECAUSE THE TWO APPLICATIONS THAT I'VE SHOWN YOU ARE APPLICATIONS THAT WE COULDN'T DO FIVE YEARS AGO.
AND AT THIS POINT, WE CAN ACTUALLY APPLY THE TECHNOLOGY, YOU KNOW, IN PRACTICE, AND THEY MAKE, THEY MAKE A, A SIGNIFICANT DIFFERENCE IN THE SPEED AT WHICH YOU CAN ACTUALLY SOLVE THESE OPTIMIZATION PROBLEMS. WE'RE TALKING ABOUT FOUR ORDERS OF MAGNITUDE IN SPEED.
SO YOU HAVE THE FUNCTIONALITY THAT YOU REALLY, REALLY WANT, BUT YOU HAVE IT, YOU KNOW, AT YOUR FINGER, AT YOUR FINGERTIPS, YOU KNOW, WITH FOUR ORDERS OF MAGNITUDE IMPROVEMENT IN SPEED.
SO THE KEY OBVIOUSLY HAS BEEN THIS CONCEPT OF OPTIMIZATION PRO, WHICH IS ACTUALLY PUTTING INSIDE THE AI SYSTEM A REASONING SYSTEM THAT RESTORE THE FEASIBILITY OF YOUR CONSTRAINTS.
SO THIS IS LIKE, THIS IS THE FUTURE OF AI.
THIS IS THIS COMBINATIONS OF, OF PREDICTION AND THEN REASONING SYSTEM.
WHEN YOU PUT THE TWO TOGETHER, YOU HAVE SOMETHING WHICH IS RATHER UNIQUE AND THAT ALLOWS US TO BE A RELIABLE BY DESIGN AND ALSO GETTING PERFORMANCE GUARANTEES.
I DIDN'T TALK ABOUT THAT, BUT YOU SAW SOME OF THE RESULTS ALSO ON THE SCALABILITY AND THE SIZES OF PROBLEMS THAT WE CAN SOLVE.
SO THAT'S WHAT I WANTED TO TELL YOU.
UH, I'LL BE, UH, VERY HAPPY TO ANSWER ANY, ANY, ANY QUESTIONS.
I THINK AI IS GONNA TRANSFORM MANY INDUSTRIES, AND I BELIEVE THAT IT HAS A SIGNIFICANT ROLE TO DO IN THE POWER SYSTEM INDUSTRY.
QUESTIONS FROM THE COMMITTEE? JOHN, SIR CARLOS? UH, WELL, AS YOU, AS YOU SAID AT THE START OF THIS, THIS IS, UH, ABSOLUTELY INCREDIBLE TO SEE HOW THESE THINGS CAN BE DONE, UH, WITH AI NOWADAYS.
WHEN IT TOOK LITERALLY HOURS TO DO BEFORE.
ACTUALLY MY, MY TRAINING, MY GRADUATE TRAINING WAS IN OPTIMIZATION, DYNAMIC PROGRAMMING, BELLMAN EQUATIONS, ALL THESE KINDS OF THINGS THAT ARE, I GUESS THE, THE SUBSTANCE BEHIND THESE KINDS OF OPTIMIZATIONS.
BUT MY QUESTION IS MORE ON THE, HOW DOES THE SYSTEM REACT OR OPERATE WHEN THE MARKET PARTICIPANTS ARE ACTUALLY ALSO USING AI IN TERMS OF THEIR PLANNING AND BIDDING INTO THE MARKET.
UH, BEING THAT THIS IS AN ENERGY ONLY MARKET AND ONE THAT'S VERY DYNAMIC, HOW DO WE, UH, CONTROL FOR THAT? YEAH, I THINK THAT'S A, THAT'S A VERY GOOD QUESTION.
OBVIOUSLY THE ONLY THING THAT I HAVE FOCUSED ON IS THE OPERATIONS.
I COULD HAVE FOCUSED ALSO ON PLANNING AND THINGS LIKE THIS, BUT, YOU KNOW, UH, THERE IS LIMITED TIME AND I DIDN'T TALK ABOUT AT ALL ABOUT THE MARKET, UH, AND THE MARKET PARTICIPANT AND HOW YOU BID ON THE MARKETS, AND YOU ARE COMPLETELY RIGHT.
IS THAT MORE AND MORE, YOU KNOW, UH, MACHINE LEARNING AND AI SYSTEMS ARE BEING USED ON THE MARKET? I THINK THE WAY I WOULD, I WOULD PHRASE THIS IS THAT THE MARKET, BY THE WAY IT IS STRUCTURED AND THE COMPLEXITY OF WHAT WE ARE DEALING WITH, IT'S A VERY COMPLEX ENGINEERING SYSTEM.
THE MARKET ITSELF IS NOT PERFECT.
AND WHAT THE BIDDING, THE, THE, THE MARKET PARTICIPANTS ARE DOING IS USING AI TO TRY TO REDUCE AS MANY EFFICIENCIES AS POSSIBLE.
AND SO, IN A SENSE, WE HAVE BEEN WORKING ON THIS AS WELL WITH SOME OF OUR PARTNERS IN TRYING TO SEE HOW WE CAN BID, HOW WE CAN USE AI FOR BIDDING.
AND, AND I DON'T VIEW THAT NEGATIVELY IN A SENSE, BECAUSE WHAT THEY ARE TRYING TO DO IS REALLY EXPLOITING THE INEFFICIENCY
[00:25:01]
AND CONTRACTING THESE INEFFICIENCIES TO MAKE THE MARKET MORE EFFICIENT IN A SENSE.SO, UH, I, I DON'T WORRY TOO MUCH ABOUT THIS.
I THINK, UH, ONE OF THE THINGS THAT THEY DON'T HAVE IS ALL THE, YOU KNOW, THEY DON'T REALLY KNOW HOW THE SYSTEM IS OPERATING, SO THEY ARE SOMEWHAT LIMITED IN WHAT THEY CAN DO UNLESS WE WOULD ACTUALLY PUBLISH THE ENTIRE, YOU KNOW, THE GRID ITSELF, UH, AND THE, AND THE VARIOUS PARAMETERS OF THE GRID.
SO I THINK IT'S, I, I THINK I FIND IT, I FIND IT NORMAL THAT THEY ARE TRYING TO USE MORE AND MORE AI TECHNOLOGY, AND WHAT I BELIEVE THEY ARE TRYING TO DO IS REMOVE THIS INEFFICIENCIES OF THE MARKET.
UH, SO I, I, I THINK ALSO, YEAH, THERE IS ANOTHER POINT THAT I WANTED TO MAKE, WHICH IS VERY IMPORTANT.
SO CONTRARILY TO WHAT HAPPENED IN THE FINANCIAL MARKET WHERE PEOPLE WERE REALLY PLAYING WITH TRENDS.
THIS IS NOT WHAT IS HAPPENING REALLY INSIDE THE, INSIDE THE POWER SYSTEM, INSIDE OF, YOU KNOW, THE MARKET FOR A, FOR AN ENERGY SYSTEM BECAUSE IT'S VERY, VERY STRUCTURED.
AND SO YOU CAN BID IN THE REAL TIME MARKET, BUT IT'S NOT LIKE YOU ARE GOING TO, YOU KNOW, SEE TRENDS, MICRO TRENDS AND EXPLOIT THEM.
SO I THINK WE ARE IN A MUCH BETTER POSITION INSIDE THE ENERGY SYSTEM FOR MAKING SURE THAT, YOU KNOW, PEOPLE DON'T EXPLOIT, YOU KNOW, THINGS THAT, YOU KNOW, IN MY OPINION, FOR INSTANCE, SHOULD NOT BE EXPLOITED.
JUST A QUESTION ABOUT WHAT KIND OF SYSTEM DO YOU NEED TO ACTUALLY RUN THESE OPTIMIZATIONS AND WHAT SYSTEM WERE YOU RUNNING THESE TESTS ON? YEAH, SO THERE ARE TWO THINGS, AND, AND THIS IS A GREAT QUESTION.
THERE ARE TWO THINGS THAT I, I WOULD LIKE TO, TO POINT OUT, UH, ABOUT THE TECHNOLOGIES THAT WE ARE USING, THE KIND OF COMPUTER ARCHITECTURE, THE GRIDS THAT WE'RE USING.
SO YOU HAVE TO DISTINGUISH BETWEEN TRAINING AND INFERENCE AND, YOU KNOW, UH, UH, REAL TIME OPERATION.
AND YOU MAY KNOW THAT THERE ARE COMPANIES NOW THAT ARE ACTUALLY DOING CHIPS, JUST, YOU KNOW, FOR INFERENCE AND NOT ONLY FOR TRAINING, BECAUSE THESE TWO THINGS ARE VERY DIFFERENT.
SO WHEN YOU RUN, WHEN YOU WANNA RUN A MACHINE LEARNING MODEL THAT HAS BEEN TRAINED, YOU DON'T NEED A VERY, YOU KNOW, DON'T, YOU DON'T NEED A VERY LARGE CLUSTER WITH THOUSANDS AND THOUSANDS OF, YOU KNOW, NVIDIA PROCESSOR.
YOU NEED ONE MACHINE THAT CAN RUN THIS VERY, VERY FAST OR A COUPLE OF MACHINE FOR RELIABILITY PURPOSES.
BUT THIS IS A VERY DIFFERENT PROPOSITION, IS WHAT WE HAVE TO DO FOR TRAINING.
WHEN YOU TRAIN, OBVIOUSLY YOU WANT THE BEST GRID AS POSSIBLE AS POSSIBLE.
WE HAVE A VERY GOOD GRID THAT WE HAVE, THAT WE HAVE AT GEORGIA TECH, FOR INSTANCE, TO DO ALL THIS TRAINING.
BUT WHEN WE RUN, THIS IS NOT A MAJOR ISSUE.
SO YOU HAVE TO ACTUALLY DISTINGUISH BETWEEN, YOU KNOW, WHAT YOU WANT TO DO IN REAL TIME AND WHAT YOU DO FOR TRAINING.
UH, FOR INSTANCE, ONE OF THE POSSIBILITIES THAT YOU MAY HAVE IS THAT YOU CAN TRAIN USING SECURE CLOUD ARCHITECTURE AND YOU CAN RUN ON MACHINE LOCALLY WHEN YOU ARE ACTUALLY RUNNING THE MODEL.
WHEN YOU THINK ABOUT THESE MODELS, THE RUNNING THE MODELS IS VERY DIFFERENT FROM TRAINING THE MODEL, TRAINING THE MODEL IS COMPUTATIONALLY INTENSIVE, ALTHOUGH SOME OF THE MODELS THAT I'VE SHOWN YOU, WE CAN ACTUALLY TRAIN THEM IN A COUPLE OF HOURS OR 30 MINUTES FOR SOME OF THEM.
SO THE ONE THAT I'VE SHOWN YOU FOR THE REAL TIME RISK ASSESSMENT, WE CAN TRAIN THEM IN ABOUT 30 MINUTES FOR A SYSTEM WITH ABOUT 10,000 BUSES.
SO PRESUMABLY WE COULD DO THAT ON A PUBLIC CLOUD PROVIDER AND NOT BUILD OUR OWN, YOU KNOW, YOU CAN DO ALL THE TRAINING, TRAINING, YES, ABSOLUTELY, YOU CAN DO THE TRAINING.
AND THEN WHAT I PERSONALLY WOULD RECOMMEND, BUT OBVIOUSLY I'M NOT IN YOUR SHOES, IS THAT YOU COULD RUN IT, UH, INSIDE YOUR OPERATION FOR, YOU KNOW, WITH A MUCH LOWER COST BECAUSE YOU DON'T HAVE THIS LATENCY TO COME FROM THE CLOUD TO ACTUALLY THE, UH, TO, TO YOUR OPERATION.
AND YOU DON'T HAVE, HAVE ALL THE GPUS AND, AND, AND THEY WILL BE ALSO DIFFERENT KIND OF TECHNOLOGIES THAT ARE COMING AT THAT WILL BE THE DEDICATE ON RUNNING, BECAUSE RUNNING IS VERY DIFFERENT FROM TRAINING.
JULIE, THE, THE INPUTS TO THIS CALCULATION ARE VERY IMPORTANT, AND AS WE RELY MORE ON SOLAR AND WIND WEATHER, PREDICTIONS BECOME MORE AND MORE IMPORTANT.
COULD YOU COMMENT ABOUT YOUR WORK WITH WEATHER MODELS? YES.
SO, UH, FIRST I THINK THE PROGRESS IN FORECASTING HAS BEEN VERY SIGNIFICANT IN THE LAST 10 YEARS.
THE MODELS THAT YOU HAVE SEEN THERE ARE PREDICTING JOINTLY, RIGHT? SO JOINTLY IT'S NOT INDIVIDUALLY YOU ARE, YOU ARE PREDICTING JOINTLY ABOUT A THOUSAND DIFFERENT GENERATION SOURCES, INCLUDING WIND, SOLAR.
AND SO THAT'S THE FIRST THING.
THE TECHNOLOGY HAS BEEN, UH, PROGRESSING SUBSTANTIALLY.
AT THE CORE, YOU HAVE A TRANSFORMER MODEL WITH ATTENTION MECHANISM, BUT YOU HAVE ALSO OTHER COM, YOU KNOW, OTHER COMPONENTS THAT ARE REALLY CAPTURING THE, THE NATURE OF A POWER SYSTEM BECAUSE YOU KNOW, SOMETHING.
SO FOR INSTANCE, YOU KNOW, IN ALABAMA ON FRIDAY NIGHT, YOU HAVE ALL THE FOOTBALL GAME OF THE HIGH SCHOOLS, YOU KNOW THAT THE LOAD IS GONNA BE, IS GONNA BE MUCH HIGHER.
AND SO YOUR MODELS, YOU KNOW, YOU HAVE A STATIC COVA, FUTURE COVA THAT IS ACTUALLY CAPTURING THAT.
SO WE HAVE A, A KIND OF A TRANSFORMER ARCHITECTURE, BUT THEN OTHER LAYERS FOR CAPTURING WHAT VARIABLES ARE IMPORTANT AT DIFFERENT POINT IN TIME.
THE SECOND THING THAT YOU WANT IS ACTUALLY THE ABILITY OF GENERATING CORRELATED SCENARIOS.
AND SO THIS IS AGAIN, THE TECHNOLOGY THAT IT'S NOT ONLY PREDICTING KIND OF A, YOU KNOW,
[00:30:01]
A SINGLE POINT PREDICTION, BUT IT'S ALSO GIVING YOU, UH, A, A WIDE VARIETY OF SCENARIOS THAT WILL CAPTURE WHAT COULD HAPPEN IN THE FUTURE.AND THIS IS REALLY IMPORTANT BECAUSE SOMETIMES THERE ARE THESE VERY OUTLIER SCENARIO SUPPLIERS THAT YOU NEED TO ACTUALLY TAKE INTO ACCOUNT SUCH THAT YOU POSITION YOUR GENERATORS CORRECTLY.
SO FOR INSTANCE, YOU CAN LOSE THE WIND IN THREE HOURS IN, YOU KNOW, IN NORTH OF, YOU KNOW, IN MINNESOTA.
AND SO YOU HAVE TO BE, HAVE A SCENARIO THAT WILL CAPTURE THAT SUCH THAT YOU RECOGNIZE THAT.
THE, THE ACCURACY OF HAVING A GOOD WEATHER FORECAST WILL IMPROVE YOUR, YOUR, YOUR, YOUR, THE, THE, THE FIDELITY OF YOUR MODEL BY ORDERS OF MAGNITUDE.
SO WHAT I WOULD RECOMMEND IN GENERAL IS TO GET THE AS BEST FOR, YOU KNOW, WEATHER FORECAST AND COMBINE THAT WITH THE DATA-DRIVEN CAPABILITY OF AI, AND YOU GET THE BEST OF BOTH WORLD AND YOU HAVE THE MOST PRECISE, YOU KNOW, THE MOST ACCURATE, YOU KNOW, UH, FORECAST THAT WE CAN GET.
SO IN A SENSE, I THINK INVESTING IN GETTING VERY GOOD WEATHER FORECAST IS REALLY IMPORTANT, UH, FOR, FOR FEEDING THIS MODEL.
SO THIS COMBINATIONS OF PHYSICS-BASED MODELING AND DATA DRIVEN MODELING IS REALLY IMPORTANT.
ON THAT, ON THAT POINT, I NOTICED IN ONE OF YOUR SLIDES, IT LOOKED LIKE YOU HAD FOUR OR FIVE DIFFERENT WEATHER MODELS THAT WERE INPUT INTO, INTO THAT PARTICULAR MODEL YOU WERE SHOWING.
IS THAT, IS THAT THE WAY YOU WOULD TEND TO DEAL WITH IT? SO ABSOLUTELY.
IF YOU HAVE MORE MODELS, YOU YEAH, PLEASE, YEAH, BY ALL MEANS, USE AS MUCH.
SO MY ANSWER TO YOUR QUESTION WILL BE ALWAYS, THE MORE DATA THAT YOU HAVE, THE MORE HAPPY YOU MAKE ME.
IF YOU HAVE MULTIPLE, MULTIPLE MODELS, I WOULD USE ALL OF THEM.
AND YOU LET, AND YOU LET THE TRAINING ACTUALLY TAKE THEM AND SEE WHEN THEY ARE THE BEST.
AND YOU CAN DO VARIOUS THINGS LIKE VARIABLE SELECTION, SELECT MODEL SELECTION AS WELL INSIDE YOUR, INSIDE YOUR ARCHITECTURE.
THE MORE MODEL YOU HAVE, THE BETTER YOU, THE BETTER WE WILL BE.
OKAY, SO I, I'M JUST REFERRING TO THE GRAPH AND DAVID DATA DRIVEN VERSUS WEATHER INFORMED, JUST BECAUSE I WANT TO TALK A LITTLE BIT.
SO YOU TAKE THE MODEL, YOU HAVE YOUR INPUT DATA, YOU HAVE YOUR HISTORIC TREND ANALYSIS, YOU HAVE YOUR ASSUMPTIONS THAT GIVE YOU YOUR FUTURE TREND ANALYSIS.
AND SO YOU DEVELOP YOUR MODEL.
SO NOW YOU'RE RUNNING IT AGAINST REAL.
HOW DO YOU CROSSCHECK THAT YOU'RE DRIFTING TOO FAR, THAT REALITY IS DRIFTING AWAY FROM YOUR MODEL? LIKE, AND WHEN DO YOU JUST SAY STOP THE MODEL ISN'T NO LONGER REPRESENTATIVE? YEAH, YEAH.
SO, UH, UH, SO ONE OF THE THINGS THAT WE ARE PUSHING VERY MUCH IS WHAT WE CALL JUST IN TIME LEARNING, WHICH BASICALLY MEANS THAT WE HAVE THE CAPABILITY OF TRAINING SOME OF THESE MODEL VERY QUICKLY.
AND SO IF WE SEE CONDITION CHANGING AND YOU HAVE, YOU KNOW, TECHNOLOGY WHICH ARE CALLED, YOU KNOW, UH, CHANGE PREDICTION MODE, YOU KNOW, STEP CHANGE PREDICTIONS AND THINGS LIKE THIS, YOU CAN RETRAIN THEM VERY QUICKLY.
SO THIS IS ONE OF THE THINGS THAT WE HAVE BEEN PUSHING, FOR INSTANCE, WHEN WE DO, UH, THE ECONOMIC DISPATCH THAT YOU HAVE SEEN, WE ACTUALLY TRAIN A DIFFERENT MODEL FOR EVERY HOURS OF THE DAY BECAUSE THEY HAVE DIFFERENT COMMITMENT, AND THEN DURING THE DAY IT TAKES ABOUT 30 MINUTES TO RETRAIN THEM.
SO IF YOU SEE A, BASICALLY A CHANGE IN THE WEATHER FORECAST, WE CAN RETRAIN THEM SUCH THAT WE DON'T GET OUT OF DISTRIBUTION.
SO IT'S REALLY THAT, IT'S REALLY THAT FAST.
I MEAN, I'M IMPRESSED 'CAUSE YES, IN EIA WE DO ITERATIVE MODELS, WHICH TAKE FOREVER BECAUSE NO, SO THE MODEL, THE MODEL THAT WE HAD FOR THE ECONOMIC DISPATCH, WE TRAIN THEM IN ABOUT 30 MINUTES FOR SOMETHING WHICH HAS ABOUT SIX, 6,000 BUSES, 10,000 BUSES.
WE TRAIN THOSE MODELS IN ABOUT 30, 45 MINUTES FOR 30,000 BUSES.
ONCE AGAIN, I MEAN, WE HAVE THIS DISTRIBUTION, WHICH IS REASONABLY WELL BEHAVED IN A SENSE, YES, YOU HAVE SCENARIOS THAT ARE DIFFERENT, BUT IT'S NOT LIKE IT'S GONNA BE COMPLETELY, IS GONNA BE COMPLETELY WILD, RIGHT? SO IT'S NOT THAT THE LOAD CAN COMPLETELY, YOU KNOW, QUADRUPLE IN, IN FIVE MINUTES.
SO I THINK WE ARE IN A SITUATION WHERE WE LIVE IN THE REAL WORLD, AND BECAUSE OF THAT, WE CAN PREDICT REASONABLY WELL, AT LEAST IN THE SHORT TERM, WITH GOOD VISION OF THE MEDIUM TERM AND THE LONG TERM.
AND IF YOU NEED TO RETRAIN THE MODEL VERY QUICKLY, BUT THAT, THAT THEN IMPLIES THAT YOU NEED TO HAVE ACCESS TO THE TRAINING SYSTEM AS WELL AS THE, THE REAL TIME SYSTEM.
BUT AGAIN, THE LATENCY IN THAT PARTICULAR, YEAH, IT'S A GREAT QUESTION AGAIN, BUT THE LATENCY IN THAT CASE IS NOT THAT IMPORTANT BECAUSE YOU'RE GONNA TRAIN THEM FOR 30 MINUTES, YOU GET THE RESULTS BACK, AND THEN YOU CAN AGAIN RUN YOUR SYSTEM WITH THIS.
YOU SEE WHAT I MEAN? SO YOU, SO THE ONLY THING THAT YOU REALLY NEED AT THAT PARTICULAR POINT IS THE NEW WEIGHTS, YOU KNOW, FOR YOUR, UH, MACHINE LEARNING SYSTEM, YOUR AI SYSTEM.
WHAT ARE THE LIMITATIONS THAT YOU SEE IN THE SYSTEM? SO LIMITATION OF THE SYSTEM.
SO, UH, UH, SO LET ME TELL YOU ONE THING THAT I'M, THAT WE ARE WORKING ON THAT, UH, I THINK WE NEED TO IMPROVE, UH, ON THIS SYSTEM.
ALL THE RESULTS THAT I'VE SHOWN YOU, THE SPEED IS, IS WHERE WE WANT IT TO BE.
IT'S FOR US, ORDERS OF MANY IS FASTER THAN WHAT WE CAN DO.
UH, OTHERWISE THE, THE RESULTS IN TERM OF OPTIMALITY ARE AMAZING IN THE EXPECTED CASE.
SO WE ARE TYPICALLY, YOU KNOW, 0.5% OF OPTIMALITY.
AND WE CAN PROVE THAT BECAUSE WE HAVE ALSO ANOTHER CONCEPT THAT I DIDN'T TALK ABOUT, UH, WHICH IS KIND OF THE DUAL PROXY.
WE CAN APPROXIMATE, YOU KNOW, WE CAN GET LOWER BOND AS
[00:35:01]
WELL.THAT BEING SAID, SOMETIMES WE HAVE OUTLIERS THAT ARE OUTSIDE THIS, THIS 0.5 IS 1%, AND THESE CLIENTS ARE VERY DIFFICULT TO UNDERSTAND.
NOW, WE CAN ACTUALLY SPOT THEM.
WE, WE KNOW THAT THEY ARE THERE, UH, WHEN WE SEE THEM, SO WE CAN DETECT THEM, BUT THAT MEANS THAT IN THOSE PARTICULAR CASES, WE HAVE TO RUN THE OPTIMIZATION AT THAT POINT, OR WE HAVE TO RUN SOMETHING THAT, YOU KNOW, UH, IMPROVE THE SOLUTION.
SO WE NEED ANOTHER STEP THERE.
SO THIS IS ONE OF THE THINGS THAT MY TEAM IS ACTUALLY LOOKING AT.
REALLY, YOU HAVE ONE OF THESE OUTLIERS THAT IS NOT, UH, AS GOOD IN QUALITY AS, AS YOU EXPECT.
AND, AND WE DON'T REALLY UNDERSTAND WHY IT IS, THAT'S WHY I HAVE SOME STUDENTS WORKING ON THIS, BUT IT'S VERY RARE IT HAPPENS.
YOU ALWAYS HAVE A BACKUP, RIGHT? SO, BUT IT DOESN'T RUN AS FAST, OBVIOUSLY.
AND SO THIS IS ONE OF THE THINGS THAT WE ARE TRYING TO LOOK, WHY DO WE HAVE THESE OUTLIERS FROM TIME TO TIME AND CAN WE ACTUALLY GET RID OF THAT? UH, SO, UH, AND, AND THE THE GOOD POINT IS THAT WE KNOW THAT THEY ARE THERE.
WE KNOW THAT AT SOME POINT THE QUALITY IS NOT EXACTLY WHAT WE SHOULD HAVE, BUT IT'S VERY RARE, RIGHT? THESE ARE LIKE THESE OUTLIERS, YOU KNOW, UH, UH, ONE OUT OF A THOUSAND CASES YOU'RE GONNA HAVE ONE OF THESE OUTLIERS.
SO THAT'S ONE OF THE THINGS THAT WE NEED TO IMPROVE.
A QUESTION HERE FROM PEGGY, COULD YOU SHARE WITH US WHERE, WHERE THE OTHER MARKETS ARE IN UTILIZING THIS INFORMATION? YOU HAD, YOU TALKED ABOUT MISO.
WHAT EXACTLY ARE THEY, WHAT PHASE ARE THEY? WHAT DATA WE, WE HAVE NO, WHAT PHASE ARE TO IMPLEMENTATION? HOW FAR ALONG? SO, SO I, I CAN'T REALLY TALK ABOUT THIS IN PUBLIC, RIGHT? SO, UH, BECAUSE WE HAVE, WE, WE HAVE, UM, UH, WE HAVE AGREEMENTS, UH, THAT WE CAN, WE CAN SAVE IT.
WHAT WE, WHAT I CAN TELL YOU IS THAT WE ARE IN THE PROCESS OF, OF, UH, DEPLOYING SOME OF THIS TECHNOLOGY IN PARALLEL WITH EXISTING SYSTEM TO VALIDATE THEM.
OBVIOUSLY YOU WORK IN A VERY, UM, SENSITIVE INDUSTRY, I WOULD SAY.
UH, YOU HAVE TO MAKE SURE THAT EVERYTHING IS RELIABLE.
UH, YOU DON'T HAVE TO TRUST THE NUMBERS THAT I GIVE YOU.
AND SO WHAT WE ARE TRYING TO DO AS A, AS A, AS A FIRST STEP FOR DEPLOYMENT IS ACTUALLY, UH, DEPLOYING THAT IN PARALLEL TO AN EXISTING SYSTEM SUCH THAT THEY CAN ACTUALLY VALIDATE THOSE THINGS.
THAT'S WHAT WE ARE TRYING TO DO.
THIS IS A VERY IMPORTANT STEP, UH, BUT THIS IS WHAT NEEDS TO BE DONE FOR BEFORE ACTUALLY DEPLOYING THIS SYSTEM, LIKE ANY KIND OF ENGINEERING SYSTEM.
THAT'S WHY WE, WE LOVE TO WORK WITH INDUSTRIAL PARTNERS BECAUSE IF, IF IT OTHERWISE, IT'S GONNA STAY AT THE ACADEMIC LEVEL, BUT YOU REALLY NEED TO GO TO THE NEXT STEP.
UH, WE HAVE AN ENGINEERING TEAM THAT IS ACTUALLY DOING THINGS LIKE THIS SO THAT WE CAN INTEGRATE INSIDE THE OPERATIONS OF A PARTNER AND RUN THE SYSTEM IN PARALLEL SO THAT WE CAN VALIDATE EVERYTHING THAT I JUST TOLD YOU.
IT DOESN'T MEAN THAT, YOU KNOW, WE, WE CAN DO THAT ALL, ALL IN ALL KINDS OF HISTORICAL DATA, BUT AGAIN, YOU KNOW, WHAT YOU WANT TO SEE IS THAT IN REAL TIME WITH THE REAL DATA THAT IS HAPPENING AT THAT TIME, YOU HAVE THOSE RESULTS AND YOU HAVE THE SPEED THAT WE ARE THAT, THAT WE NEED TO HAVE.
SO THAT'S THE STAGE WHERE WE ARE AT THIS PARTICULAR POINT.
ALRIGHT? REMEMBER, NONE OF THE THINGS THAT I TOOK THAT I TALKED ABOUT COULD BE DONE ABOUT TWO OR THREE YEARS AGO, RIGHT? SO ARE OTHER ISOS OTHER THAN MISO, UH, DOING SIMILAR EFFORTS WITH YOU? SO WE, WE ARE ALSO WORKING WITH SOUTHERN COMPANY IS ONE OF OUR BIGGEST PARTNER.
UH, WE ARE WORKING CLOSELY WITH THE, THE FRENCH OPERATOR NOW, UH, AS WELL.
UH, WE ARE DEPLOYING, SO ONE OF THE THINGS THAT WE DO WITH THEM NOW IS ACTUALLY TRYING TO GET THE MOST, UH, COMPREHENSIVE TEST CASE, UH, LONGITUDINAL TEST CASES.
UH, WE HAVE THE ENTIRE GRID, UH, ALL THE SNAPSHOT WITH DIFFERENT TOPOLOGIES AT DIFFERENT TIME.
UH, WHAT WE NEED TO PRODUCE NOW IS ALL THE TIME SERIES SUCH THAT YOU HAVE THE MOST, YOU KNOW, UH, SOPHISTICATED TEST CASES THAT WILL BE EVER AVAILABLE BECAUSE THIS IS, THIS IS ONE OF THE THINGS THAT WE HAVE BEEN PUSHING, GETTING MORE AND MORE TEST CASES TO THE COMMUNITY.
WE VOTE GOING INTO THE SENSITIVITIES OF EVERY ONE OF THOSE GRADES.
SO FOR A FRENCH SYSTEM FOR INSTANCE, WE CAN'T, WE DON'T HAVE THE TIME SERIES BECAUSE THIS IS VERY SENSITIVE AND WE ARE GONNA, WE ARE GONNA GENERATE THEM SUCH THAT THEY'RE REPRESENTATIVE OF THE SYSTEM, BUT THEY'RE NOT GONNA BE THE REAL ONES, BUT THEY'RE GONNA BE SO GOOD THAT PEOPLE WILL DO, YOU KNOW, REALLY ADVANCED RESEARCH ON THAT AS WELL.
SO THAT'S THE KIND OF COLLABORATIONS THAT WE HAVE.
ANY OTHER QUESTIONS FOR PASCAL? AGAIN, REMINDER, HE'LL BE COMING BACK IN EXECUTIVE SESSION.
WE CAN TALK ABOUT SOME OF THESE OTHER THINGS AS WELL.
THAT WAS EXTREMELY HELPFUL, THANK YOU.
[6.1 Projects and Technology Update]
WE ARE NOW GONNA PROCEED ON TO THE, UH, TO JPS SECTION, AND WE'RE GOING TO, UM, HAVE OUR PRESENTATION ON COMMITTEE BRIEFS SECTION SIX WITH, UH, PROJECT AND TECHNOLOGY UPDATE.[00:40:02]
SO IF YOU RECALL, WE MET IN DECEMBER AND THIS MEETING, I HAD TO TURN MY MATERIALS IN MID-JANUARY.SO MY SLIDES ARE SHORTER THAN USUAL, SO I'M SURE, BUT IT ALSO, IT'S A GREAT OPPORTUNITY TO LOOK AT THE SLIDES IN A DIFFERENT WAY THAT, UH, WE CAN LOOK AT HOW WE FARED OVER THE LAST YEAR COMPARED TO THE NORMAL SLIDES I TALKED ABOUT, UH, LOOKING FORWARD.
SO, UM, LOOKING AT THIS SLIDE, YOU KNOW, WE HAVE, UM, IF YOU SEE AN AVERAGE OF 60 CONCURRENT PROJECTS THAT WE WERE RUNNING THROUGHOUT THE YEAR, OKAY? AND, UM, AND YOU SEE THE BLUE BARS ARE PROJECTS THAT WE STARTED AND THE GRAY BARS ARE PROJECT THAT WE CLOSED.
AND, AND THAT'S AN IMPORTANT THING BECAUSE IF YOU'RE NOT ABLE TO CLOSE PROJECTS, WE ARE NOT ABLE TO START NEW PROJECTS.
SO THIS HEALTHY VOLUME OF BEING ABLE TO CLOSE PROJECTS GETS US THIS, UM, ABILITY TO START NEW PROJECTS.
SO IT, IT SOUNDS STRAIGHTFORWARD, BUT, BUT SOMETIMES YOU CAN BE STUCK ON PROJECTS.
AND YOU CAN SEE FROM OUR DAYS LAST YEAR THAT WE WERE GENERALLY MAKING PROGRESS ON ALL PROJECTS, NOT STUCK ON ANYTHING.
WE, WE DID HAVE NORMAL UPS AND DOWNS OF PROJECTS, BUT YOU CAN SEE THE NET RESERVE, UM, THAT, UM, WE HAVE A GOOD, UH, THROUGHPUT OF THINGS.
UM, ONE THING YOU MAY NOTICE ON THE SLIDE IS THAT THINGS HAVE SLOWED DOWN IN TERMS OF NEW INTAKES, UH, SINCE SEPTEMBER.
AND THAT'S BECAUSE IN SEPTEMBER WE LOCKED DOWN RTC.
AND, UM, AND SINCE THEN WE ARE CAREFUL ABOUT STARTING NEW PROJECTS THAT WILL IMMEDIATELY RUN INTO RTC AND THEN WE'LL HAVE TO PUT THAT PROJECT ON HOLD.
SO IF YOU KNOW THAT UPFRONT, WE DON'T START THOSE KIND OF PROJECTS.
OKAY? SO, SO YOU SEE SOME KIND OF A SLOWDOWN, UH, POST SEPTEMBER, BUT NEVERTHELESS, UH, PROJECTS WHERE WE DON'T HAVE CONFLICT WITH RTC, WE ARE STILL STARTING.
AND, AND THOSE ARE WHAT YOU SEE.
THE FLIP SIDE IS POST RTC, THERE WILL BE ALL THIS PENT UP DEMAND THAT WE HAVE TO HANDLE.
SO SOMETHING TO WATCH OUT FOR.
AND, UM, UH, EVERY OPPORTUNITY THAT WE GET, AND I'LL TALK ABOUT THIS IN RTC TOO, THAT WHEN YOU PASS OUR DEVELOPMENT CYCLE, WHATEVER WE CAN GET STARTED, WE WILL GET STARTED SO THAT WE ARE NOT WAITING FOR THIS BIG DAM TO BURST AT THE END OF THE RTC DELIVERABLE.
UM, SO THIS IS, UM, ANOTHER VIEW OF THE, UH, LABOR HOURS AND THIS IS A THREE YEAR VIEW.
OKAY? UM, AND, UM, AND AT THIS POINT I DON'T HAVE ANY KIND OF, UH, LABOR PROBLEMS ACROSS THE PROJECTS THAT, UM, UH, WE ARE LOOKING AT.
UM, ONE THING I WOULD POINT OUT IS, YOU KNOW, IN 23 WE HAD THIS PEAK HERE, AND THAT'S DRIVEN MORE BY THE EMS UPGRADE.
OKAY? AND WE DON'T HAVE A SIMILAR, UH, IN THE BLUE AS TECH HEALTH.
THE EMS PROJECT WAS A TECH HEALTH PROJECT.
UM, SO WE HAD THAT SPIKE WITH THAT EMS UPGRADE GOING LIVE.
UM, AND, AND THIS, UH, REGULATORY PROJECT IS RTC, WHICH IS THE MOST OF THAT EFFORT THAT YOU SEE THAT, AND, AND YOU'LL JUST SEE THIS SPIKE UP EVEN MORE TOWARDS THE END OF THE YEAR.
SO, UM, UM, SO OUR, OUR LABOR TENDS TO, UM, UM, TAKE INTO ACCOUNT PROJECTS THAT ARE STARTING AND THEN WE LOAD THAT IN, YOU KNOW, THE, THE LABOR HOURS BUMP UP, BUT YOU'LL SEE, YOU SHOULD SEE A SPIKE IN DECEMBER.
BUT NEVERTHELESS, I'M NOT REPORTING ANY KIND OF, UM, LABOR PROBLEMS. AND, AND LASTLY, UH, USUALLY I HAVE A SLIDE THAT SAYS WHAT DID WE RELEASE? BUT, UH, WITH THE HOLIDAYS AND THE TIMING OF THE SLIDES THAT HAD TO SUBMIT, AND I'LL FOCUS ON THINGS THAT ARE COMING UP.
AND, UM, THESE ARE SOME OF THE MARKET FACING OR MARKET IMPACTING, UM, DELIVERABLES THAT WE'LL BE, UH, DOING IN BETWEEN JANUARY AND UH, MARCH.
SO NOTHING PARTICULARLY TO CALL OUT THERE IS, UM, THE AMERICAN DISABILITY ACT THAT WE ARE, UH, IMPLEMENTING ON OUR, UM, ERCOT MOBILE APP.
SO THAT'LL ACCOUNT FOR SOME GAPS THAT WE SEE THERE.
THAT'S A, UM, BIG EXTERNAL FACING ONE.
UM, OTHERWISE, UH, THAT'S PRETTY MUCH MY PRESENTATION FOR THE, UH, PROJECTS UPDATES.
FOR JP ON PROJECTS, HOW DO YOU GET, UH, REQUESTS? WHAT IS THE MECHANISM TO GET REQUESTS FROM STAKEHOLDERS TO, TO UPDATE OUR INTERFACES TO THE, SO THERE ARE TWO MECHANISMS. ONE IS A REGULAR NPRR PROCESS ITSELF, WHERE YOU, YOU APPROVE THAT, RIGHT? I MEAN THE, THE STAKEHOLDER PROCESS WILL REQUEST, UM, UH, CHANGES.
SOME OF THOSE CHANGES HAVE NO IMPACT TO OUR SYSTEMS. THERE MAY BE PROCESS CHANGES, THINGS LIKE THAT.
BUT SOME OF THOSE CHANGES, UM, DOES HAVE IMPACT, LIKE IN THIS CASE NPR 1 1 4 5 WITH THIS, UH,
[00:45:01]
DLFS TRANSMISSION LOAD FACTORS, LOSS FACTORS.UM, SO THAT IS AN IMPACT THAT WE HAD TO IMPLEMENT.
UM, SO, SO IN THIS CASE, THE MARKET PARTICIPANTS IS SAYING THAT, UH, THE WAY WE ARE USING SEASONAL INFORMATION OR HISTORIC INFORMATION FOR THIS TRANSMISSION LOSS FACTORS IS NOT AS GOOD AS IF YOU USE REAL TIME INFORMATION FROM THE EMS. SO, SO THAT'S A REQUEST THAT THE MARKET PARTICIPANT HAS MADE, AND THEN WE USE STATE ESTIMATE TO CALCULATE THAT TTLF.
SO, SO THAT'S AN EXAMPLE OF SOMETHING THAT THE MARKET PARTICIPANT OR SOMETIMES WE MAY PROPOSE IT AS WELL.
SO THAT'S IF IT'S MARKET IMPACTING AND REQUIRES A RULE CHANGE.
AND THAT ALSO, UH, SCRS, WHICH ALSO COME TO YOUR WAY THAT A SYSTEM CHANGE REQUEST.
AND, AND THOSE ARE ALL VERY WELL DEBATED IN THE TECH AND OTHER, UM, COMMITTEES.
OKAY? SO IN ADDITION TO THAT, WE MAY HAVE OUR OWN INTERNAL PROJECT.
LIKE FOR EXAMPLE, THE EMS UPGRADE, WHICH IS USUALLY A TECH HEALTH PROJECT.
I DON'T NEED TO GO TO THE MARKET PARTICIPANT TO GET AN NPR ARE APPROVED OR WHATEVER.
SO IF YOU HAVE A TECH HEALTH REASON TO UPGRADE IT OR OTHER REASONS, LIKE MAYBE IT'S A COMMERCIAL REASON, A VENDOR, WE THINK WE WANT TO SWITCH THE VENDOR, THOSE KIND OF PROJECTS COME AS TECH HEALTH.
SO THOSE ARE VARIOUS MECHANISMS THAT, UH, WE HAVE TO GET THINGS GOING, BUT WE HAVE A VERY HEALTHY PROCESS OF GETTING THE REQUIREMENTS UP THERE, UM, ANALYZED.
AND SO IT TAKES A LONG BIT OF TIME TO START UP, BUT IT'S WELL WORTH IT BECAUSE THEN YOU KNOW, WHAT KIND OF PROJECT ARE YOU GOING TO RUN? DOES THAT ANSWER YOUR QUESTION? SORRY, LINDA HAS A QUESTION.
GO BACK TO THE LAST, GO BACK TO THE LAST SLIDE.
SO, YOU KNOW, THEY'RE IN CONTROL.
THEY KNOW, THEY KNOW WHAT TO PREDICT, IT'S TRACKING.
SO FOR US THOUGH, SINCE WE'RE TRYING TO LOOK AT TRENDS, MAYBE WE SHOULD BE DOING A THREE MONTH ROLLING AVERAGE OR SOMETHING THAT'LL GET THE PEAKS AND VALLEYS LOWER SO WE CAN, IT'S EASIER FOR US TO SEE TRENDS.
SO, SO USUALLY WHEN I PRESENT IT, I GIVE THIS, UM, UH, FORECAST WINDOW AND THAT'S WHAT THESE GRAY, VARIOUS SHADES OF GRAY ARE, RIGHT? BUT THIS TIME, BECAUSE WE DID AN END OF THE YEAR VIEW, I JUST SAID, OKAY, I'M NOT GONNA DO THAT THIS TIME.
I'M GOING TO LOOK AT MORE BACKWARD LOOKING.
UM, YEAH, SO MY QUESTION IS MORE ABOUT TRENDS.
OKAY? LIKE YOU, YOU'RE, YOU'RE OKAY ON THE DETAILS.
SO THIS ISN'T, BUT I SIT HERE AND LIKE, YOU KNOW, ON MY BRAIN I'M TRYING TO AVERAGE LIKE, YOU KNOW, IS IT GOING THIS WAY OR THAT WAY? AND THE TROUBLE IS, WITH THIS MUCH JUMPING AROUND, IT'S HARD TO SEE THAT IF YOU, SINCE YOU'RE IN CONTROL AND YOU KNOW WHAT YOU'RE DOING, IF YOU, IF WE DID A ROLLING AVERAGE, IT WOULD BE EASIER FOR US TO SEE THE TRENDS OR EVEN IF YOU DID A THREE MONTH ROLLING AVERAGE, I DON'T KNOW, LINE OR SOMETHING, JUST SOMETHING FOR US TO DO A QUICK SNAP.
'CAUSE I MEAN, I'M GOOD AT LOOKING AT CHARTS.
I MEAN, I'D MAKE A LIVING DOING THAT, BUT THIS IS HARD.
UH, I'LL COME BACK WITH SOMETHING AND I'LL RUN WITH YOU OFFLINE AS WELL, LINDA, IF THAT MAKES SENSE.
I, I, I WOULD WARN YOU TO TEST YOUR PROTOTYPE WITH LINDA FIRST.
I'M NOT TRYING TO MAKE ANYTHING COMPLICATED, IT'S JUST IF IT COULD BE A SIMPLE ARITHMETIC PROBLEM, YES.
BUT, BUT I WOULD LIKE TO GET MORE DATA, WHICHEVER WAY TO ANALYZE IT.
AND WE'LL BE TRYING TO DO THIS MANY DIFFERENT WAYS.
THE OTHER THING I THINK THAT WOULD BE INTERESTING HERE IS THE EXTENT TO WHICH SOME OF, SOME OF THIS LABOR IS INTERNAL IT LABOR, SOME OF IT IS INTERNAL TO ERCOT ITSELF, BUT NOT NECESSARILY IN IT.
SO THAT, THAT WOULD BE ANOTHER SLICE THAT WE COULD OKAY.
AND, AND WE HAVE DONE THAT BEFORE, BUT WE WILL DO THAT.
AND I ALSO SEPARATE OUT RTC IN THIS SO THAT, YOU KNOW, SINCE THAT'S A BIG THING, I MIGHT R SUCH A BIG CHUNK OF IT.
I WILL HAVE THAT FOR NEXT TIME.
SPEAKING OF RTC, WHY DON'T WE ALL KEEP ROLLING?
[6.2 Real-Time Co-Optimization Implementation Update]
SO HERE I'M AN RTC.UH, USUALLY I HAVE THIS INTRODUCTORY SLIDE OF SHOWING WHAT RTC IS AND I SAID, OKAY, I'VE DONE THIS ENOUGH AND I TOOK IT OUT AND, AND RIGHT AWAY I HAVE TWO NEW BOARD MEMBERS.
SO TUTORIAL NEEDED FOR THE NEW BOARD MEMBERS.
SO, UM, YOU KNOW, BUT, BUT I CAN STILL USE THIS SLIDE TO, UM, GIVE A HIGH LEVEL VIEW.
SO, SO THIS WHOLE PROJECT IS BROKEN DOWN INTO FOUR MAJOR PROJECTS.
THE ONE FIRST PROJECT, WHICH IS THE NUMBER TWO REALLY IS FOCUSING ON THE GRID AND MARKET SYSTEMS. AND THAT'S CARRIES THE BULK OF THE WORK.
THAT'S WHERE THE MOST OF THE RISK IS 'CAUSE THAT'S WHERE THE CORE OPTIMIZATION ENGINE IS CHANGING.
THE NEXT ONE IS COMMERCIAL APPS, WHICH IS A SETTLEMENT SYSTEM.
UM, AND THE THIRD ONE IS TYPICALLY ON AN IT PROJECT.
THAT'S WHERE THE LONG PULL IS.
IT'S AN INTEGRATION AND REPORTS KIND OF PROJECT.
AND OVERSEEING ALL THESE THREE PROJECTS IS A PROGRAM CONTROL.
AND THAT'S WHERE BUDGET WISE WE ARE MANAGING HARDWARE AND SOFTWARE, AND THAT'S WHAT A LOT OF THOSE NUMBERS ARE.
AND ALSO A LOT OF, UM, OLD PROJECT MANAGEMENT OVERHEAD AND THINGS LIKE THAT.
SO, SO THAT'S HIGH LEVEL HOW THE PROJECT IS SET UP.
[00:50:01]
WHAT I WOULD SAY IS THAT THE FIRST COLUMN, AND JUST AGAIN WALKING BACK SOME OF THESE NUMBERS IS SINCE YOU DON'T HAVE ANY BACKGROUND, THE, THE FIRST NUMBER COLUMN, THE IA BUDGET IMPACT ANALYSIS BUDGET IS THE, UM, UM, NUMBERS WE USE TO FUND THIS PROJECT.SO IT'S OUR BEST GUESS AS AT WHAT THIS PROJECT WOULD COST, HOW LONG IT WOULD TAKE, UM, AND THINGS LIKE THAT.
AND BASED ON THAT, WE HAD A $50 MILLION ESTIMATE WITH A PROJECTED DELIVERY IN 2026 JUNE.
SINCE THEN, AND WE BROUGHT THIS TO THE BOARD ALL THE TIME, WE'VE MADE GOOD PROGRESS, PRIMARILY BECAUSE OF GOOD REQUIREMENTS, GOOD INPUT FROM THE RTCB TASK FORCE.
SO ALL THOSE THINGS HELP BRING US, BRING THE SCHEDULE IN AND, AND IF YOU SCHEDULE IF YOU DROP YOUR SCHEDULE ALMOST SIX, SEVEN MONTHS, OBVIOUSLY THERE IS COST SAVING.
AND YOU CAN SEE THAT, AND THAT'S ONE OF THE PART OF THE COST SAVING.
THERE ARE OTHER FACTORS FOR COST SAVINGS IN TERMS OF HARDWARE AND SOFTWARE AND DIFFERENT WAYS OF MANAGING LICENSE.
BUT THE NET RESULT IS FROM A $50 MILLION TARGETED PROJECT, WE AT GATE TO EXECUTION.
GATE TO EXECUTION MEANS WE HAVE FINISHED OUR PLANNING AND, AND WE ARE INTO EXECUTION.
SO WE CAME UP WITH A BUDGET OF ABOUT 35 MILLION.
SO THAT'S WHAT THE SECOND COLUMN IS SHOWING AND, AND ALSO SHOWING THE BREAKDOWN ON A PER PROJECT BASIS.
THE THIRD COLUMN, WHICH WILL BE THE, SO THESE FIRST TWO COLUMNS ARE GOING TO BE STATIC EVERY TIME I COME HERE BECAUSE THEY'RE LOCKED DOWN.
THE THIRD COLUMN IS THE ONE THAT EVERY TIME I COME HERE, YOU'LL SEE NEW NUMBERS.
AND AT THIS POINT THEY'RE ALL GREEN.
AND I KNOW IN, UH, EVERY TIME I SAY THERE ARE NORMALS UPS AND DOWNS, BUT IN THIS PARTICULAR CASE TODAY, IT'S ALL UPS.
BUT THERE WOULD BE A TIME WHEN THERE'LL BE DOWNS.
I'M JUST GOING TO REMIND YOU THAT, WAIT, YOU DIDN'T TELL ME, ASK ME THAT BY, WERE THEY ALL UP SOME YEARS OR SOMETIMES.
BUT, BUT THERE IS NORMAL UPS AND DOWNS, BUT AT THIS POINT THERE ARE, UH, IT'S ALL UPS THERE.
I'M, WE ARE COMING UNDER OUR, UH, UH, GATE TO EXECUTION BUDGET, BUT WE'RE STILL EARLY DAYS.
OKAY? SO, SO, UM, SO THAT'S HIGH LEVEL ON THIS SHEET, AND THIS IS WHAT I WILL COME BACK AND SHOW EVERY TIME.
UM, BY THE WAY, UM, I, I THINK I MISSED SOMETHING THAT, UM, WE HAD AN ACTION ITEM, JULIE, THAT YOU SAID THAT WE SHOULD COME BACK AND GIVE A TRAINING ON RTC.
UM, AND WHAT WE DID IS WE INCLUDED THE APPENDIX, THE MATERIALS THAT WE HAVE SO FAR, AND WE ALSO HAVE, UM, A NEW VIDEO THAT WE PRODUCE, TRAINING VIDEO, UM, UM, AND, AND YOU DON'T HAVE TO GO TO THE APPENDIX TO FIND IT.
IF YOU JUST DO A SIMPLE GOOGLE LINK SEARCH RTCB TRAINING VIDEOS, IT'LL TAKE YOU STRAIGHT TO THE VIDEO.
AND THIS, FOR EXAMPLE, A GREAT VIDEO BY DAVE MAGGIO, UH, INTRO 35 MINUTES VIDEO THAT YOU CAN WATCH THAT INTRODUCES RTC.
OKAY? AND, UM, AT LEAST FOR THE NEW BOARD MEMBERS, I RECOMMEND THAT THAT WILL GIVE YOU GOOD, GREAT START ON YOUR OWN TIME.
UM, UM, AND IF YOU HAVE FOLLOW UP QUESTIONS, PLEASE ASK US.
OKAY? SO, SO THIS IS ANOTHER VIEW, UM, MORE FORWARD LOOKING.
UM, AND THE THING I WANT TO SHOW YOU HERE IS THAT I, I KNOW I'VE SAID OUR BUDGET IS 35 MILLION, BUT IF I LOOK AT THE WAY THE PROJECT IS LAID OUT, I AM GOING TO EXCEED THE 35 MILLION BETWEEN THE MONTHS OF, UM, I DUNNO, FEBRUARY AND, UM, AUGUST NEXT YEAR.
OKAY? AND I SAY IS IT'S BY DESIGN.
OKAY? WHAT IS HAPPENING IS THAT WHILE WE EXCEED IT AT THE END OF THE PROJECT, WHEN THE PROJECT RETURNS HARDWARE BACK TO IT, WE GIVE CREDIT TO THE PROJECT BACK FOR RETURNING THE HARDWARE BECAUSE THEY DON'T KNOW THE PROJECT NO LONGER NEEDS THE HARDWARE AND IT'S, IT'S PURCHASES OF THE PROJECT MADE.
SO WHEN THEY RETURN THE HARDWARE, WE GIVE THAT CREDIT BACK AND THAT CREDIT BRINGS IT BACK 2 35.
SO, SO THERE WILL BE A PERIOD WHERE IT LOOKS LIKE IT'S OVER BUDGET, BUT BY DESIGN.
OKAY? SO, BUT JUST THE WAY WE RUN IT HERE, OKAY? SO, BUT THIS IS THE FIRST TIME I'M SHOWING THIS PICTURE.
UM, AND I KNOW WE HAVE TALKED ABOUT THE 35 MILLION NEW NUMBER AND THAT NEW NUMBER IS COMPUTED AT THE END OF THE PROJECT, RIGHT? THAT'S, THIS IS WHERE WE THINK WILL END UP PROJECT WISE, BUT THE PATH TO IT DOES TAKE US OUTSIDE THAT RED LINE AND COMES BACK, OKAY? SO THIS IS A DIFFERENT VIEW.
I MEAN, AND I'VE SHOWN THIS VIEW BEFORE A FEW TIMES.
IT TALKS ABOUT THE FUNCTIONAL PROGRESS OF OUR PROJECT.
AND, UM, AND I MEAN, THE KEY THING I WANT TO POINT OUT IS, IS THAT PERHAPS THE NEXT TIME I MEET HERE ON APRIL SEVEN, I WOULD LIKE TO SAY THAT WE ARE DEVELOPMENT COMPLETE ON OUR CORE SYSTEMS THAT ARE NEEDED FOR MARKET TRIALS.
OKAY? THAT, UM, WE HAVE SYSTEMS READY TO GO.
WE ARE IN OUR INTERNAL INTEGRATED TESTING, AND THIS IS THE CORE SYSTEM.
THERE ARE OTHER SYSTEMS THAT WE'LL STILL BE WORKING ON.
[00:55:01]
THOSE ARE THOSE, UH, ARE RELATED TO OUR INTEGRATION OR THE REPORTS OR OUR OTS SYSTEM.THEY'RE NOT SO MUCH MARKET FACING, BUT WE JUST PRIORITIZE IT SO THAT WE CAN START MARKET TRIALS AS EARLY AS POSSIBLE.
AND THOSE SYSTEMS, THE DEVELOPMENT WILL BE READY AND WE, WE INTEND TO HIT MARKETER.
SO HOPEFULLY NEXT TIME THAT I'M HERE, I CAN REPORT THAT APRIL 7TH.
SO THIS ALSO, UM, I, I DO EVERY TIME, AND THIS KIND OF PRESENTATION HAS HELPED ME WHERE I TRY TO GIVE UPFRONT WHAT ARE THE UPCOMING MILESTONES AND, UM, AND INDICATE TO YOU HOW MUCH ON TRACK WE ARE.
AND, AND IF YOU REMEMBER FROM LAST YEAR, UM, I THINK WE MET ALL OUR GOALS AND WE ONLY MISSED ONE STRETCH GOAL, AND, BUT WE STILL MET THE GOALS AND, AND THOSE KIND OF VISIBILITY THAT I BRING HERE REALLY HAS HELPED.
UM, THE FIRST LINE HERE, IT SAYS ON TRACK, I HAD TO SUBMIT THE SLIDES BEFORE THAT DATE, BUT IT, IT IS DONE.
I JUST WANT TO REPORT ON THAT.
THE NEXT ONE, WE HAVE THE VENDOR DELIVERY OF SC, WHICH IS WHERE MOST OF THE WORK IS, SECURITY CONSTRAINT, ECONOMIC DISPATCH.
UM, AND UH, BASED ON THE DATA RELEASE THAT WAS GIVEN TO US, WE ARE VERY OPTIMISTIC THAT WE'LL MEET THOSE STATES.
AND THIS ALSO BASED ON, UH, REQUEST, UM, I THINK IT WAS JULIE HAS, OR I CAN'T REMEMBER, UM, THAT SAY TALK ABOUT THE RISK FOR THE PROJECT.
NOW I THINK WE HAVE TALKED ABOUT SOME OF THOSE RISKS, UM, BUT UM, BUT THIS KIND OF, UM, THERE OBVIOUSLY MORE RISK THAN THIS, BUT IS THE RISK THAT ARE HIGHLIGHTED AT THIS LEVEL, RIGHT? UM, AND SO BEFORE I GO THERE, I JUST THOUGHT I'LL EXPLAIN THIS SHEET, RIGHT? SO WHAT WE HAVE IS ON THE Y AXIS, YOU HAVE ALL THE IMPACTS COMING FROM THE RES, THE X AXIS IS THE LIKELIHOOD OF THAT, AND THE COLOR IS KIND OF SHOWING HOW IMPACTFUL WE FEEL IT IS.
AND UM, AND THE LINE IS LIKE, OKAY, IF YOU CROSS IT, YOU'RE IN A BIT OF TROUBLE.
OKAY? UM, SO FAIRLY STRAIGHTFORWARD, UM, HOPE AND, AND HOPEFULLY WITH THIS KIND OF A FORMAT, I CAN ALSO BRING BACK THE DYNAMICS SO I CAN SHOW NEXT TIME AND WE COME HERE, HOW HAVE THE WRIST MOVED TO THE LEFT, TO THE RIGHT TOP LINE? SO I THOUGHT IT'S A GOOD WAY TO SHOW THAT.
UM, SO YOU'VE QUESTION, SO YOU'VE OPENED YOURSELF UP TO WHO ARE THE PEOPLE, I, WHO'S THE, WHO'S THE GROUP THAT'S DECIDING ON THE RISK? IS IT ALL INTERNAL PEOPLE OR DO YOU BRING IN YES.
SO, SO RIGHT NOW WE HAVE A PROJECT MANAGEMENT STEERING COMMITTEE THAT TALKS ABOUT THIS.
SO, SO DEFINITELY, UM, WE TALK ABOUT THIS INTERNALLY.
UH, LIKE FOR EXAMPLE, UM, THE, UM, SO I'M NOT TRYING TO PUT YOU ON THE SPOT.
IS LIKE WHEN YOU, WHEN YOU SAY THAT YOU'VE ASSESSED RISKS, SO THE BIG QUESTION IS WHO ARE THE, YOU KNOW, WHAT ARE THE POSITIONS THAT IS NOT THE INDIVIDUAL NAMES, BUT IS IT ONLY INDIVID, YOU KNOW, IS IT ONLY THE TECH TEAM THAT DOES IT? OR DO YOU BRING IN SOMEONE ELSE TO TAKE A LOOK? LIKE HOW FAR OF A SPAN DO YOU, AS, DO YOU GET INPUT AS TO THE PERCENT RISK? SO I THINK THE, WHAT I WOULD SAY IS THAT THE PROJECT PRIMARILY RUNNING WHAT IS HAPPENING AT THE PROJECT.
THEN WE HAVE PEOPLE LIKE MATT WHO GO TO THE RTCB UH, MEETINGS.
WE COMMUNICATE THEM FROM AND FROM MARKET PARTICIPANTS.
WHAT SORT INPUT? SO, SO YOU'LL SEE SOME OF THE RISK ARE EXTERNAL RISK.
SO A LOT OF THOSE INPUTS COME BACK IN SOME OF THE RISKS THAT WE ELEVATE TO THE EXECUTIVE TEAM.
FOR EXAMPLE, UM, YOU REMEMBER THE DISCUSSION ON THE A SD SEASON ANSWER IS SERVICE GOALS, RIGHT? THAT OKAY, WE THOUGHT THERE COULD BE A RISK TO THE PROJECT.
SO, SO, SO SOMETIMES THERE'S AN ESCALATION THERE.
SOMETIMES WE BRING, SO ON A MATURITY ONE TO FIVE, DO YOU THINK YOUR ASSESSMENT IS A GOOD, IS UP AT FOUR AND FIVE OR YOU THINK IT'S DOWN AT, AT THIS POINT? I MEAN, I'M ONLY ASKING 'CAUSE YOU'RE PUTTING IT UP AND THIS IS EXACTLY IT.
AND I'M NOT TRYING TO PUT YOU ON THE SPOT, BUT I AM, UM, YOU'RE MATURING THIS PROCESS.
YOU PUT THIS UP AND NOW THE QUESTION IS WHAT'S YOUR ROUTINE INPUT TO GET THIS, UH, YOU KNOW, UH, TO GET THESE NUMBERS? JUST WHAT POSITIONS PARTICIPATE? THEY DON'T NEED A LONG EXPLANATION ABOUT IT.
JUST IS IT BROAD ENOUGH FOR, FOR THE SITUATIONS YOU'RE ASSESSING? I FEEL IT IS BROAD ENOUGH.
UM, BECAUSE, BECAUSE OF THE, UM, SUBSTANTIAL CONVERSATIONS WE'VE HAD IN SOME OF THOSE ISSUES, LIKE THE, THE MARKET READINESS IS ONE TOPIC, THE, THE RISK WE SAW WITH PEOPLE TRYING TO CHANGE OUR REQUIREMENTS, RIGHT? I MEAN, WE WERE ABLE TO PROACTIVELY ADDRESS THAT VERSUS BEING MORE REACTIVE.
SO I'M THINKING THAT BECAUSE JUST WRITE IT DOWN.
YOU DON'T HAVE TO ANSWER IT ON YOUR FEET.
UM, SO JUST GOING TO THE DETAILS OF WHAT THOSE WERE.
THE FIRST TWO, UM, WE HAVE COVERED IN THIS
[01:00:01]
FORUM ACTUALLY MULTIPLE TIMES.UM, AND, AND THAT IS THE RISK WHERE, UM, WHERE THERE IS A REQUIREMENTS CHANGE TO THE PROJECT, RIGHT? AND, AND AS YOU KNOW, OUR PROJECT THAT'S SO CLOSE TO COMPLETING MY DEVELOPMENT AND AT THIS POINT IF SOMEBODY SAYS, WELL, NEW REQUIREMENTS THAT THROWS THINGS FOR A LOOP, ESPECIALLY WHEN I HAVE A VENDOR TO WORK WITH AND I'LL START REQUIREMENTS ALL OVER WITH THE VENDOR.
SO THAT HAS BEEN OUR BIGGEST RISK.
AND WE HAVE ALWAYS SAID, LOOK, IF WE, THERE ARE ALWAYS GOOD IDEAS, MORE IDEAS, AND WE CAN ALWAYS DO THAT POST RTC, BUT LET'S GO DOWN, GO LIVE, AND THEN WE CAN ALWAYS IMPLEMENT INSTEAD OF, UM, YOU KNOW, MESSING WITH THE PROJECT.
THE OTHER RACE WE TALKED ABOUT IS ALSO THE MARKET READINESS.
UM, WHERE, UM, UNLIKE A TYPICAL AIRCRAFT PROJECT WHERE WE COULD JUST GO LIVE BY OURSELF, IN THIS CASE WITH RTC, WE NEED TO GO LIVE SIMULTANEOUSLY WITH OUR MARKET PARTICIPANTS AND THEY NEED TO BE READY AS WELL.
AND UM, AND, AND THAT'S ONE OF THE REASONS THAT WE ALSO POINTED OUT BEFORE THIS FORUM, AND WE ARE MANAGING IT BY BRINGING AWARENESS TO THE MARKET PARTICIPANTS AS WELL AS GETTING READINESS SCORES FROM THEM BACK TO THE, UM, BACK TO US.
SO AT THIS POINT WE THINK WE ARE DOING OKAY.
THE, CAN I ASK YOU A QUESTION ABOUT THAT? YES.
UM, IN PAST MEETINGS WE'VE DISCUSSED HOW THIS CHANGED TO RTC ON DECEMBER 5TH IS VERY SIMILAR TO HOW WE TRANSITIONED TO THE NODAL MARKET DESIGN.
WELL, PROBABLY A LITTLE EASIER THAN THAT, BUT HERE OKAY, WERE THERE, WELL THAT'S RELATED TO MY QUESTION.
IS THERE, ARE THERE ADDITIONAL LESSONS LEARNED ABOUT THE MARKET READINESS SIDE THAT WE COULD OR SHOULD APPLY TO OUR GO LIVE? YES.
I MEAN, BECAUSE IT'S SO COMPLEX AND, AND I THINK COUPLE OF THINGS THAT I HELPED THIS, UH, FOR EXAMPLE, THE DEPLOYMENT OF ECRS, SO, YOU KNOW, SO, SO WE ARE DOING A LOT OF, UM, MARKET CHANGING THINGS AND WE ARE TRYING TO INCORPORATE LESSONS THAT ARE FAIRLY RECENT AND RECENT WOUNDS.
YOU KNOW, WE TEND TO REMEMBER MORE THAN, UH, WOUNDS FROM 12 YEARS AGO AND MANY OF US WERE NOT HERE, RIGHT? UM, SO I THINK WE HAVE INCORPORATED THAT, BUT DEFINITELY FROM SOMETHING AS BIG AS, UM, NORDAL GO LIVE, UH, THERE ARE SOME OTHER WOUNDS WILL NEVER GO AWAY, RIGHT? SO WE DO WANT TO INTERPRET THAT.
YES, I I JUST ADD THAT WE'RE PROBABLY THE LAST MAJOR ISO TO DO RTC.
SO THIS IS SOMETHING THAT HAS BEEN DONE AND NOW NOT ALL GENERATORS IN AN IS ALL ISOS, BUT THIS IS NOT SOMETHING THAT'S NEW TO THE INDUSTRY.
AND, AND JP, MY, MY MAIN QUESTION IS THAT ONE ABOUT COORDINATION WITH MARKET PARTICIPANTS, IMM AND EVERYBODY ELSE THAT'S INVOLVED TO ENSURE ALIGNMENT YES.
ON THE ONE HAND AND ALSO COORDINATION AND, UH, THE ACTUAL PRACTICE OF EXECUTING THIS, UH, JOINTLY.
SO THERE'S CONCEPT, WHAT IS WHAT? AND, AND AGAIN, THAT, THAT'S MY MAIN CONCERN ABOUT THE SCHEDULE AND KEEPING IT AND ENSURING THAT WE HAVE THE RIGHT PRODUCT GOING OUT.
AND, AND THAT'S WHY, YOU KNOW, IT'S, IT'S A RISK TOPIC, BUT, BUT CERTAINLY WHAT WE ARE DOING ON OUR END IS CONSTANT COMMUNICATION TO THE MARKET PARTICIPANTS IN VARIOUS FORUMS, UM, INCLUDING THE TECHNOLOGY WORKING GROUP WHERE WE NOT REALLY TALK TO THE MARKET PARTICIPANTS, BUT TO THEIR VENDORS AS WELL.
UM, AND UM, AND WE ARE WORKING ON WHAT WILL THAT CUT OVER LOOK LIKE, SO WE CAN GO OVER THAT IN MORE DETAILS WITH THEM.
UM, AND JP MAYBE IT'D BE HELPFUL TO, UH, REFERENCE THE VERY RECENT VERY BROAD MARKET CHANGE THAT WAS IMPLEMENTED AROUND THE TEXAS AT FIVE, WHICH WAS A MULTI-YEAR, UH, TRANSMISSION RETAILER, UH, VERY BROAD MARKET CHANGE THAT HAD THIS KIND OF, YOU KNOW, PLANNING TECHNOLOGY CHANGES CUT OVER PREPARATIONS.
AND THAT WAS JUST COMPLETED VERY WELL IN, UH, IN DECEMBER OF THIS LAST YEAR.
DIFFERENT SET OF MARKET PARTICIPANTS, RETAIL FOLKS, BUT, BUT STILL IT'S A GOOD I GOOD WAY OF SAYING HOW WE PREPARE FOR SUCH THING WHERE, UM, WHERE A LOT OF COMMUNICATIONS UPFRONT IS NEEDED.
UH, WE NEED TO SAY, THIS IS WHAT YOU'RE GOING TO DO AT THIS MINUTE, THIS IS HOW WE PLAN TO CUT OVER.
THIS IS WHAT THE IMPACT YOU'RE GONNA SEE.
SO ALL THAT COMMUNICATION WILL COME THROUGH AND, AND AT SOME POINT I'LL BRING THAT TO THE BOARD AS WELL SO WE CAN USE THIS FORUM TO UM, ASSURE YOU HOW WE ARE GOING THROUGH THAT.
SO JUST GOING DOWN THAT RISK, THE, THERE'S SOME NEW ONES THAT I'VE NOT REALLY TALKED ABOUT HERE, BUT THERE'S STILL RISK.
UH, THE, THE, THE THIRD ONE THERE IS THE BATTERY DATA QUALITY.
SO THIS PROJECT IS NOT JUST RTC, BUT IT'S RTC PLUS B.
THE PLUS B IS THE BATTERIES PART.
AND UM, AND, AND, AND SO THIS GOES A LITTLE BIT BEYOND CO-OP OPTIMIZATION, WHERE IN OUR SYSTEMS WE HAVE, UM, BATTERIES MODEL AS BOTH THE GENERATOR AND LOAD WILL BE COMBINING THAT INTO A SINGLE MODEL, WHICH IS MORE REPRESENTATIVE OF THE, UM, UM, EQUIPMENT OUT THERE AND INCORPORATING OPTIMIZATIONS RELATED
[01:05:01]
TO BATTERIES.AND THAT BRINGS IN ITS OWN, UM, UH, COMPLEXITIES.
UM, SO, SO THIS IS NEWER AND, AND WE WANT TO WORK WITH THE, UH, MARKET PARTICIPANTS, UM, AS THEY SUBMIT THE DATA AND, AND WE CAN COMMUNICATE BACK TO THEM WHAT WE ARE SEEING SO THEY CAN IMPROVE ON, ON THINGS.
UM, SO ONE, THE NEXT ONE IS KIND OF OBVIOUS.
YEAH, WE COULD HAVE LOSS OF RESOURCES, KEY RESOURCES, WE HAVE, UM, SPECIALIZED, UM, UM, SUBJECT MATTER EXPERTS.
AND IF YOU LOSE, UM, FEW KEY PEOPLE, UH, YOU COULD HAVE IMPACT AT LEAST SCHEDULE WISE TO THE PROJECT AS WE TRY TO BACKFILL THEM OR, OR SUBSTITUTE THEM WITH OTHER RESOURCES THAT MAY TAKE MORE TIME OR, UM, OR HAS A LEARNING CURVE.
SO, UM, SO THAT'S, THAT'S ALWAYS THE RISK FOR ANY PROJECT.
IT'S NOT AS SPECIFIC TO RTC, BUT, BUT HERE BECAUSE OF THE COMPLEXITY AND SPECIALIZATION, UM, JUST, UM, MORE OF THAT.
UM, NOW ON MY SLIDE, THE LAST ONE IS HIGH, HIGH IMPACT, BUT LOWER, UM, PROBABILITY IS A POSSIBILITY OF EXTERNAL EVENT DISRUPTION, RIGHT? AND, AND AN EXAMPLE IS IF I HAVE A WEATHER EVENT FOR EXAMPLE, RIGHT? THAT, THAT IT'LL IMMEDIATELY SUCK OUT ALL THOSE SMES AND IT'S THE SAME THING AS LOSING THOSE SMES BECAUSE NOW I'LL NEED THOSE SMES TO WORK ON, UH, REPORTS OR, UM, DEPOSITIONS OR WHATEVER THAT NEEDS TO HAPPEN NOW AND THAT MAY HAVE HIGHER PRIORITY THAN THE RTC PROJECT.
AND AS FAR AS THE PROJECT IS CONCERNED ARE THE SAME THING AS NOT HAVING THOSE RESOURCES AND, AND NOT JUST ONE OR TWO.
I'LL USE A WHOLE BUNCH OF RESOURCES AT THE SAME TIME.
SO, SO OBVIOUSLY THAT'S MORE IMPACTFUL, BUT SO FROM A RISK PERSPECTIVE, IT'S THERE.
SO HOPEFULLY THIS WAS MORE HELPFUL FOR ME AS A FORMAT TO COMMUNICATE THE RISK AND I'LL USE THIS IF YOU GUYS ARE OKAY WITH THIS.
AND, UM, THAT'S MY PRESENTATION.
LET ME KNOW IF I HAVE ANY QUESTION.
I DO THINK THAT'S GOOD 'CAUSE IT'S HELPING US UNDERSTAND THE RISK IN THE MARK FROM THE MARKET PARTICIPANTS VERSUS THE INTERNAL TEAM.
AND I'LL BRING BACK THE, SOME OF THE SUGGESTIONS AND, UM, BASED ON WHAT PABLO ALSO MENTIONED.
LAST ITEM RIGHT UP, UP TO THE JP.
UM, WE HAVE A COUPLE OF, UM, TOPICS FOR OUTSIDE SPEAKERS THAT, UH, JP AND I ARE EXPLORING AND SOME OF THAT'S SUBJECT TO AVAILABILITY OF THOSE PEOPLE, BUT, UH, ANY OTHER ITEMS THAT THE COMMITTEE WOULD LIKE US TO ADD TO THE LIST? NOPE.
WITH THAT, I THINK WE'RE AT THE END OF THE GENERAL SESSION AGENDA.
UM, IF THERE'S NO OTHER BUSINESS, I'LL CALL THE MEETING, UM, ADJOURNED AND WE WILL GO INTO EXECUTIVE SESSION IN ABOUT FIVE MINUTES.