ConvOPF-DOP: A Data-Driven Method for Solving AC-OPF Based on CNN Considering Different Operation Patterns

AC optimal power flow (AC-OPF) is a significant problem in the economic operation of power systems. Traditional AC-OPF calculation methods only consider a specific operation pattern, which has limitations. We propose ConvOPF-DOP, a novel data-driven approach based on Convolutional Neural Network (CNN), to solve the AC-OPF problem in different operation patterns. This study uses the cluster analysis method to obtain the network structure and operating condition information from historical data. In this way, even when the network structure is unknown, the corresponding network topology label can also be identified from the collected data. Then we use the network topology label and load data as input data, optimize hyperparameters of the CNN model by the Bayesian optimization method, and train the CNN model to learn the relationship between the input data and the voltage output. Finally, the Power Flow (PF) calculation method is used to get the remaining optimal solutions. The effectiveness and superiority of ConvOPF-DOP are verified through 30-bus systems in four different operation patterns. ConvOPF-DOP brings 350x speed increase compared with the traditional method while ensuring high accuracy of generated optimal solutions.

Automated summary

AC optimal power flow (AC-OPF) is a significant problem in the economic operation of power systems. Traditional AC-OPF calculation methods only consider a specific operation pattern, which has limitations. We propose ConvOPF-DOP, a novel data-driven approach based on Convolutional Neural Network (CNN), to solve the AC-OPF problem in different operation patterns. The effectiveness and superiority of ConvOPF-DOP are verified through 30-bus systems in four different operation patterns. ConvOPF-DOP brings 350x speed increase compared with the traditional method while ensuring high accuracy of generated optimal solutions.