An adaptive hybrid forecasting framework for community net load considering distributed generation and nonlinear demand dynamics

Abstract The increasing penetration of distributed power generation (DPG) introduces significant uncertainty and nonlinear characteristics into community-level electricity demand forecasting. Conventional forecasting approaches mainly rely on historical consumption patterns and often fail to capture the interactions between consumer demand, local generation, and rapidly changing load behaviors. This paper proposes a Hybrid Demand–Generation Adaptive Forecasting Model (HDGMA-FM) for community net load prediction under high distributed generation penetration. The proposed framework integrates a linear ARIMAX-based forecasting branch with a nonlinear temporal kernel learning branch, enabling simultaneous modeling of interpretable temporal trends and complex nonlinear demand variations. An adaptive fusion mechanism is introduced to dynamically balance the contributions of linear and nonlinear representations according to changing load characteristics. In addition, a distributed generation-aware net demand formulation is developed to explicitly describe the impact of local renewable generation on grid-side demand. Extensive experiments under different distributed generation penetration scenarios demonstrate that HDGMA-FM consistently outperforms ARIMAX, LSTM, and CNN-LSTM benchmarks in terms of MSE, RMSE, MAE, and MAPE. Ablation analysis verifies the effectiveness of the nonlinear feature learning and adaptive fusion components, while statistical significance testing confirms the reliability of the performance improvements. The proposed framework provides an accurate and interpretable forecasting solution for community-level smart grids and supports future demand-side management applications in renewable-rich energy systems.