Use Machine Learning to Assist Generation Scheduling Optimization Problem

Topic: Use Machine Learning to Assist Generation Scheduling Optimization Problem

Student: Teoh Jing Yang

Supervisor: Dr. Tan Wen Shan

Teoh Jing Yang, who is in his 2nd year 2nd semester of Robotics and Mechatronics engineering, had an opportunity to work on a project “Use Machine Learning to Assist Generation Scheduling Optimization Problem” under the supervision of Dr. Tan Wen Shan.

The project discusses the accuracy and the computational time of implementing a machine learning model in optimizing stochastic unit commitment problems. By formulating a 118-bus unit commitment problem, we can solve the mixed-integer programming problem by finding the minimum total operation cost, but the time taken to achieve that is very long. Hence, convolutional neural network and long short-term memory are introduced to obtain the on-off status of the generators before feeding them into the solver. The computational time is reduced from 352.72 seconds to around 20 seconds in a 4 scenario problem, while the objective value rises by about 0.18%.

Figure 1. Accuracy and Computational time of CPLEX, CNN and LSTM model

Figure 2. Visualization of the percentage of error and difference to optimal value

What Jing Yang said about his UROP experience

“This opportunity allowed me to work on a research project under supervision and it was a valuable experience. This project expanded my knowledge on Mixed Integer Linear Programming (MILP) based optimization and machine learning algorithm which is essential in my future academia and career.”