A Two-step BP Neural Network Based Machine Learning Algorithm for Solving Unit Commitment

• Published in: 2022 IEEE/IAS Industrial and Commercial Power System Asia (I&CPS Asia) pp. 380 - 385
• Main Authors: Lu, Zhiyuan, Shi, Xiaohan, He, Cunzhe, Wang, Xiaolei
• Format: Conference Proceeding
• Language: English
• Published: IEEE 08.07.2022
• Subjects: BP neural network; electric system; Machine learning; Machine learning algorithms; Neural networks; Prediction algorithms; Predictive models; Simulation; Training data; unit commitment
• DOI: 10.1109/ICPSAsia55496.2022.9949710

In order to speed up the solution of the unit commitment problem, a two-step BP(Back-Propagation) neural network-based machine learning algorithm is proposed in this paper for performing fast prediction of feasible solutions to the unit commitment. Firstly, this two-step BP neural network is constructed and trained with the training data generated by traditional unit commitment solution method. After that, prediction results of the two-step BP neural network are compared with those of the one-step BP neural network and simulation results of traditional unit commitment solution method in a case study to demonstrate this method's feasibility and better accuracy. The case study shows that it is feasible to use a two-step BP neural network for fast prediction of unit commitment solutions to support electrical power system optimization, with significantly shorter time compared to the traditional solution method and higher accuracy compared to a one-step BP neural network.