Parameters prediction of EV cluster aggregation model for low-voltage stations based on IGWO-BiLSTM

• Proposes IGWO-BiLSTM for accurate EV cluster power/capacity boundary prediction. • Uses hybrid convergence factor & fitness-distance weighting to enhance IGWO search. • Aggregates EVs into privacy-preserving GES via Minkowski summation theory. • Reduces RMSE > 16.67 % (capacity) & >25 % (power) vs BiLSTM; meets 35 s runtime need. The electric vehicle (EV) represents a critical new source of variable load. Accurately estimating the aggregate power and total energy storage capacity of an EV cluster is essential for effective grid regulation. To improve the accuracy of parameters prediction of the electric vehicle cluster aggregation model, a prediction method based on the improved gray wolf algorithm (IGWO) optimized bidirectional long-short-term memory network (BiLSTM) is proposed. First, the EV cluster generalized energy storage aggregation model is established based on Minkowski summation theory, then the feasible domains of its power and capacity boundaries are constructed. Secondly, a fitness-distance dynamic weighting strategy and a hybrid convergence factor, which combines the Sigmoid function and linear regulation, are designed to improve the global search and local development capability of IGWO. Finally, the high-accuracy prediction of power and capacity boundaries for the EV cluster aggregation model is achieved by optimizing the hyperparameters configuration of BiLSTM. The simulation results show that the prediction effect of the proposed method is significantly better than that of the traditional BP neural network, LSTM and BiLSTM.