Research on dynamic flexibility of charging load in electric vehicle charging station based on time dynamic adaptive convolutional neural network

The uneven spatiotemporal distribution of electric vehicle (EV) charging demand causes load concentration during peaks and specific areas, leading to grid overload and poor dynamic flexibility analysis. This study proposes a method based on a Time Dynamic Adaptive Convolutional Neural Network (TDACNN) to address this. Key charging parameters (start time, duration, and power) are input to the TDACNN model to extract spatiotemporal correlations and predict EV charging loads. Based on these forecasts, a dynamic EV charging load adjustment model is constructed, with minimization of total load variance and peak-valley difference as the objective function under constraints. A Discrete Particle Swarm Optimization algorithm solves this model to output a dynamic adjustment scheme. Test results show that the method effectively analyzes EV charging characteristics, aligning with reality. Load forecasting proves highly reliable, with continuous ranked probability score and prediction interval average width values below 0.056 and 0.044, respectively. Voltage fluctuation remains below 5.66%, mitigating the dynamic impact of charging. Thus, the method effectively solves grid overload from uneven charging demand, significantly improves forecast accuracy/reliability, reduces voltage fluctuations, and provides vital support for grid stability and charging load optimization.