Integration of Combined Cycle Units into Economic Dispatch

Integration of Combined Cycle Units into Economic Dispatch Computation May 04 -11 Brent Miller Mun-Hong Chong Jason Mardorf Zobair Molla

Presentation Outline I. Introductory materials A. Problem statement B. Operating environment C. Intended use(s) and user(s) D. Assumptions and limitation E. End product and other deliverables Project Activity Description A. Previous accomplishments B. Present accomplishments C. Approaches considered and used D. Project definition activities E. Research activities

Presentation Outline III. F. Design activities G. Implementation activities H. Testing, results and modification I. Other important activities Resources and Schedules A. Resources and schedules 1. Personal effort requirements 2. Other resource requirements 3. Financial requirements B. Schedules

Problem Statement • Problem statement –Combined cycle units have non-monotonically increasing curves –Economic dispatch: meet demand at lowest cost –Can’t use standard optimization techniques

Presentation Outline IV. Closing materials A. Project evaluation B. Commercialization C. Recommendation for additional work D. Lessons learned E. Risk and risk management F. Closing summary

Problem statement Combined cycle plant -Two simple cycle combustion turbines -One heat recovery unit -Together comprise a combined cycle plant -Heat rate curve is not a typical -Acts as turbocharger

Problem statement • Solution approach statement – Separate the linear units and the combined cycle units – Determine a way to combine them to get lowest cost

Operating Environment • Windows based PC • Normal computer operating environment

Intended user(s) • Introductory knowledge of economic dispatch • Understanding of power system analysis • Understanding of elementary differential calculus

Intended use(s) • Be able to input generation parameters • For a given demand produce lowest cost solution • Provide proof of concept for client

Assumptions and Justifications Assumptions Justification Given real world representative data Needed to perform reasonable test Given genetic algorithm code to study Needed to fully understand the algorithm Given data in MW range Need to know proper range of values Client will receive program code Part of original agreement End product shall be on a windows PC Team members have the most experience on PC To be used in USA Countries have different power systems Input data from client The output is only as good as the input

Limitations and Justifications Limitations Justification Max and min number of units Design requirements Max solution time Client requirement Required accuracy Client requirement

End product and other deliverables 1. Program code – Do file I/O – Determine lowest cost solution to meet electric demand – Output each unit’s power output – Output each unit’s cost for a specified power output 2. Test Results – Give client results of test data – Give optimal parameters of code

Resources and Schedules • Resource Requirements - • Personal effort requirements Other resources requirements Financial requirements Schedules - Tasks and Subtasks vs. Calendar

Project Activity Description

Previous Accomplishments • Learning genetic algorithm concepts • Did a conventional dispatch of generators with segmented operating areas • Project Plan • Poster • Design Plan

Present Accomplishments • Finalized the end product design • Developed a flow chart of this design • Wrote a large part of the code

Approaches considered • Standard La. Grangian techniques • Convex optimization • Genetic algorithms • C/C++ • Matlab

Advantages/Disadvantages • Classical techniques – Advantage • Easy • Standard • no issues – Disadvantage • Not accurate • phony data

Advantage/Disadvantage • Convex optimization – Advantage • Mathematically grounded • Apparently easier • implement equations – Disadvantage • Too mathematical based • Didn’t feel comfortable with it

Advantage/Disadvantage • Genetic Algorithms – Advantage • No solution space problem • Will work on any weird function – Disadvantage • A high learning curve • Takes computing power

Advantage/Disadvantage • C/C++ – Advantage • Fast • More easily integrated into client’s system – Disadvantage • Comfort level • Not as easy to test

Advantage/Disadvantage • Matlab – Advantage • Members’ familiarity level • Ease of testing • Natural use of matrices – Disadvantage • Programs don’t run as fast • Global variables can cause problems

Selected Approach and Why • Genetic algorithm/Classical approach – Faculty advisor has extensive knowledge of genetic algorithms – Made best use of each technique • Matlab – Very easy to test changes – We are more familiar with it than any other

GA/La. Grangian • Main Idea: Use each technique at it’s strong point • La. Grangian techniques excel at optimizing monotonically increasing functions • Genetic algorithms excel at optimizing any type of function • Result: Split the problem into two parts – Linear units – Combined cycle units

EE 303 Review • La. Grangian function • Form of cost curve --- • Two conditions for optimality • This gives n+1 unknowns and n+1 equations

LG • Key advantage – Incremental costs of all units are equal – A linear equation-(Incremental cost curves) – Can develop a system chart to treat the system as one unit. (Graphical method)

GA Part of solution • Genetic algorithms are for optimization • No proof as to how they work…they just do • Model nature…survival of the fittest 1. 2. 3. 4. 5. 6. Represent solution as a binary chromosome Determine the “fitness” of the encoded solution Crossover: The fittest solutions exchange their “DNA” The results of this crossover form a new generation Mutation: Random bit flipping to avoid local minima Stop after X number of generations.

GA-Chromosome 1. Encode a solution in binary chromosome 1 1 0 1 0 2. 0 1 0 1 1 Make a population of these chromosomes 1 1 0 1 0 0 1 1 1 0 0 0 0 1

GA-Fitness 3. Evaluate the fitness of each chromosome, for each member of the population 1 1 0 1 0 0 1 0 1 1 Fitness Function unit i

GA-Fitness 4. Determine each chromosome’s relative fitness to the whole population

GA -Crossover 6. Crossover (DNA swapping) -Randomly select site to do crossover (swap) Crossover sites 1 1 0 0 1 1 0 1 0 1 1 0 0 0 0 1 0 0 1 1 0 Parents - Generation “n” Children - Generation “n+1” 7. This process completes one generation

Design 1. Input the demand to be dispatched 2. GA selects operating point for CC units 3. Do table lookup of linear units to dispatch the demand minus Pcc 4. Evaluate the total cost of using a particular chromosome as a solution 5. Use this cost as the fitness function to determine chromosomes for next generation

Project Design 1. Use GA to randomly chose operating points for the combined cycle units 2. Subtract the power generated by the CCs at these settings from the total demand 3. Do a table lookup on the linear units to dispatch the remainder 4. Evaluate the total cost of dispatching the entire demand. Minimize this cost.

Implementation Activities • Code that: – – – Reads in the units’ IHR data and system data Generates the first generation of the GA Decodes the chromosome Determines the amount for the linear units to dispatch Does table lookups determining the operating point for each unit – Does cost calculation for operating each generator at a particular point – Assigns a fitness value to a chromosome – Randomly selects chromosomes for mating based on normalized fitness value

Resources and Schedules • Resource Requirements - • Personal effort requirements Other resources requirements Financial requirements Schedules - Tasks and Subtasks vs. Calendar

Personal effort requirements

Other resources requirements • Reference book: – Genetic Algorithms in Search, Optimization and Machine learning

Financial requirements Item W/0 Labor With Labor Poster $50 Printing and Binding $10 $31. 50 $91. 50 Parts and materials: Book Subtotal Labor at $10. 00 per hour a. Molla, Zobair $1, 920 b. Miller, Brent $2, 000 c. Chong, Mun Hong $1, 880 d. Mardof, Jason $1, 930 Subtotal Total $7, 730 $ 91. 50 $7, 821. 50

Tasks and Subtasks

Closure Materials

Project Evaluation Understanding problem 100% Developing linear model 95% Learning SGA theory 100% Develop Code 50% Testing 15%

Commercialization • NO Commercialization

Lessons Learned • Things that went well – Team/FA meetings – Design – Linear Dispatch • Things that did not go well – Understanding previous code – Determining which design to use – Understanding all of client’s data

Lessons Learned • Technical knowledge gained – La. Grangian optimization techniques – Simple genetic algorithms – Matlab programming • Non-technical knowledge gained – Communication – Documentation

Lessons Learned • Things to be done differently if done again – Develop linear system data earlier – Understand the workings of a GA sooner – Decided on a design sooner

Risk and Risk Management • Risk 1: Loss of team member • Management: Make sure of working knowledge of the design • Risk 2: Future users not able to understand our code • Management: Provide Simple comments and documentation of theory

Closing Summary • Problem: Normal optimization techniques don’t work on non-monotonically increasing curves • Our solution minimizes computation by splitting up the linear and non-linear units • This project is important because it involves saving money. This is nearly always a motivating factor.

Questions ?