Research on traffic energy scheduling based on photovoltaic forecasting using stacked ensemble learning

In the process of traffic energy scheduling, the proportion of renewable energy utilization is gradually increasing, and accurate photovoltaic (PV) power generation forecasting is a prerequisite for the safe and stable integration of high proportions of PV into the power grid. However, existing machine learning models for PV power forecasting generally suffer from low prediction accuracy and weak generalization capability. To address these issues, this paper proposes a Stacking ensemble forecasting model to improve PV prediction accuracy. An improved stacked ensemble algorithm is adopted, integrating three base models – Artificial Neural Network (ANN), Long Short-Term Memory network (LSTM), and Random Forest (RF) – to combine time-series features, nonlinear relationships, and robustness, with the outputs of the base models used as inputs to a Linear Regression (LR) meta-model to effectively avoid overfitting and enhance prediction stability. Data from two different locations were collected, with missing and abnormal values processed, and the base models were trained using five-fold cross-validation to ensure data diversity. The model performance was evaluated using RMSE, MAE, MAPE, R 2 , and a performance comparison was conducted between the proposed model and individual baseline models. Furthermore, the SHapley Additive exPlanations (SHAP) method was applied to analyze the importance of nine input features, quantifying their contributions to the prediction results. Based on the forecast results of photovoltaic output and load demand, an economically-oriented energy dispatch scheme has been formulated. The improved whale optimization algorithm is employed to maximize the economic benefits of the grid system while meeting load demands, followed by an analysis of the system’s self-sufficiency rate during this period.