Adaptive Physics-Informed Neural Network for Energy-Efficient Consumer Systems via Reliable Wind Power Forecasting Under Cold-Wave Conditions

A reliable electricity supply under cold weather underpins consumer electronics, smart homes, and electric vehicles. When renewable output fails, it can lead voltage sags that damage sensitive electronics, unpredictable demand-response events that disrupt smart home automation, and volatile electricity prices that increase the cost of charging electric vehicles. Achieving accurate wind-power forecasting during cold waves is crucial for ensuring energy-efficient and reliable consumer systems. However, state-of-the-art deep learning models that rely purely on data-driven learning struggle to handle the abrupt, nonstationary fluctuations caused by cold waves, resulting in significant prediction deviations. To address this issue, an adaptive physics-informed neural network (PI-LSTM) is proposed to integrate turbine aerodynamic energy-conversion principles into an LSTM structure, injecting physical priors and constraints into training to align the model with both physical laws and observational data. In addition, aCp-PINN subnetwork is constructed to adaptively model the power coefficientCp, enabling dynamic recalibration and improved physical interpretability. Meanwhile, cold-wave sequences are identified via a sliding-window procedure, and key meteorological features are refined using Pearson-correlation– based correction. Experiments on real cold-wave datasets demonstrate that the method enhances prediction accuracy and physical consistency. Our method provides a more stable and predictable electricity supply, which is fundamental for the reliable and cost-effective operation of smart homes, electric vehicles, and energy-aware consumer technologies.