The large-scale integration of electric vehicles (EVs) has caused significant spatial and temporal fluctuations in distribution network loads, posing challenges to power system operation security and charging infrastructure planning. Traditional time-series and statistical models fail to capture the influence of traffic behavior on the formation of charging loads. To address this issue, a hybrid EV charging load forecasting method combining traffic equilibrium theory and deep neural networks (DNN) is proposed. First, a traffic equilibrium model considering energy constraints is established to simulate traffic flow and charging behavior under different travel demand scenarios, thereby generating network operation data. Then, correlation analysis and SHAP methods are applied to identify key influencing features and determine the primary factors affecting charging loads. Finally, a multi-layer fully connected neural network is constructed to predict charging loads at multiple charging station nodes. It is demonstrated that the proposed model exhibits good convergence and stability, and that charging loads can be accurately forecast while the coupling relationship between traffic behavior and power demand is effectively revealed. Technical support is provided for charging infrastructure planning and the optimization of modern distribution systems.