Charging decision modelling and load collaborative simulation of electric vehicles based on multi-source fusion: Adaptive Huff-LSTM method

This paper proposes a modelling framework for electric vehicle (EV) charging decision-making and load coordination simulation based on multi-source heterogeneous data fusion and an adaptive Huff-LSTM model. The goal is to address stability issues in the power system that arise from large-scale EV charging, as well as to overcome the limitations found in existing studies related to the dynamic interplay between the traffic-grid system and uncertainties in user behavior. A dynamic road model that couples time, speed, and flow is initially established, which integrates the road network with the power grid systems. This model ensures the accurate assignment of vehicle origin-destination (OD) pairs and optimizes path planning using an enhanced Dijkstra algorithm. Thereafter, energy consumption per mile is modeled by incorporating factors such as ambient temperature and driving speed. Subsequently, the Huff-Long Short-Term Memory (Huff-LSTM) model is introduced to characterize user behavior in selecting a charging station. This innovative approach combines the LSTM neural network with the Huff model, accommodating multi-dimensional constraints such as charging station availability and queuing time. The comparison of multiple schemes using real charging station data in Shenzhen demonstrates that the coefficient of determination for this decision model is 0.8912. Finally, Monte Carlo simulation, in conjunction with co-simulation using the IEEE 33-node distribution network, demonstrates that the proposed model effectively captures the spatial and temporal divergence characteristics of temperature-sensitive charging behavior under typical scenarios, thereby enhancing the accuracy of EV charging load simulation.