Integrating NeuralProphet (NP) forecasting with MATLAB/simulink for energy-efficient PV-grid water pumping

This study proposes an integrated forecasting–simulation framework to improve the energy efficiency and operational resilience of municipal water pumping systems. To overcome the interpretability and long-horizon planning limitations of conventional black-box deep learning approaches, an interpretable NeuralProphet (NP) forecasting model is coupled with a physics-based MATLAB/Simulink hybrid PV–grid digital twin. Using three years of SCADA irradiance and operational data (2021–2023) from a 22-kW municipal pumping station, the NP model was trained and benchmarked against LSTM and standard Prophet models. NeuralProphet demonstrated superior forecasting performance (RMSE = 92.7136 W/m², R² = 0.8998), statistically outperforming LSTM (RMSE = 100.26 W/m², R² = 0.8828; p < 0.05) while providing interpretable trend and seasonality decomposition essential for infrastructure decision-making. The validated forecasts were dynamically integrated into the Simulink hybrid PV–grid model for year-ahead operational simulation, indicating an annual photovoltaic generation potential of approximately 35 MWh , resulting in reduced grid dependency while maintaining stable electromechanical performance (∼143 Nm torque stability). The proposed framework establishes a transferable digital-twin methodology for planning, sizing, and risk-aware deployment of renewable-assisted municipal water-pumping infrastructure. • Proposed a unified framework coupling NP forecasting with MATLAB/Simulink dynamic simulation. • Framework overcomes "black box" limitations of Deep Learning, with interpretable decomposition for infrastructure planning. • Simulated hybrid PV-grid system achieves ∼35 MWh annual renewable generation, reducing grid dependence. • Demonstrates a transferable "digital twin" methodology for sizing and de-risking municipal PV retrofits. • Statistically validated forecasting superiority over LSTM and Prophet using RMSE, R², Diebold–Mariano significance testing.