A data-driven approach to urban charging facility expansion based on bi-level optimization: A case study in a Chinese city

Abstract

This paper addresses the optimization of urban charging facilities expansion in response to the growing charging demand and proposes a data-driven bi-level optimization framework. The proposed framework combines a mixed-integer programming model for siting and capacity planning, an improved sparrow search algorithm (ISSA) for hyperparameter optimization of the CNN-BiLSTM model, and the application of Benders decomposition with Lagrangian relaxation using Gurobi for solving the model. The proposed approach is empirically evaluated through a case study conducted in an administrative area of a Chinese city, using comprehensive charging records and other relevant information. The results demonstrate that the ISSA consistently outperforms other algorithms, showing an average improvement of 23.97 % in root mean square error, 19.25 % in average absolute percentage error, and 1.50 % in the determination coefficient. The modified Diebold-Mariano test confirms the significant advantage of ISSA. Based on the model analysis, the study recommends constructing 20 new charging stations in the case city, with more than half equipped with fast chargers. This strategy ensures that 95 % of the area can find a charging station within 5 min while also ensuring the operator's profitability. The research findings provide effective support for investment decisions related to charging facilities.