PSO-LSTM-Based Ultra-Short-Term Load Forecasting Study for Solar Heating System

To address issues such as unstable heating loads, uneven heat consumption, and precise heating in solar heating systems, efficient and accurate heating load forecasting is essential. A suitable solar heating system model was established using the TRNSYS18 thermodynamic simulation platform. Taking a building in Alar City, Xinjiang, as the research subject, ultra-short-term prediction data parameters for the area were obtained. Using the acquired data parameters and historical heating load data as inputs, the particle swarm optimization (PSO) algorithm was employed to optimize the LSTM neural network, establishing a prediction model based on the PSO-LSTM neural network. For load forecasting in 7 min ultra-short-term time series, both the LSTM neural network model and the PSO-LSTM neural network prediction model underwent optimization. Through simulation experiments verifying indoor temperature, heat collection, and energy consumption, two model error evaluation metrics were used as results. Comparative analysis revealed that the PSO-LSTM model achieved a 3.3–86.7% increase in R2 compared to the LSTM model, a 38.2–84.8% reduction in RMSE, a 57.8–91.1% decrease in MAE, and a 58–90.3% reduction in MAPE. The research results demonstrate the PSO-LSTM model’s effectiveness in southern Xinjiang, confirming its superiority as a forecasting model. This provides data support for operational adjustments and load forecasting in solar heating systems.