Electricity Load Forecasting Method Based on Improved PSO-LSTM Modeling

The swift advancement of artificial intelligence and intelligent algorithms provides powerful technical support for various data analysis and forecasting tasks. With the continuous development of power system load forecasting technology, it becomes a key research direction to improve the model's prediction ability in complex time series data. Therefore, the study raises a power load forecasting approach that integrates the improved particle swarm optimization algorithm with the long and short-term memory network model. Initially, a reverse learning strategy is introduced into the particle swarm algorithm. Second, the hyperparameters of the long and short-term memory network are optimized by the improved particle swarm algorithm, and combined with the empirical modal decomposition to decompose the electric power load data at multiple levels, fully mining the temporal features in the data. In the experiment of predicting the future 24-hour power load, in the segment at 16-17 o'clock, the error of the proposed model is studied to be 24 kW, and the prediction curve is most consistent with the real value curve. When the sample size is 2000, the memory occupation of the proposed model is <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{3 1 8 9 ~ M B}$</tex>, and the prediction speed is <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1 7}$</tex> FPS. The outcomes indicate that the raised method improves the prediction accuracy, and provides new optimization ideas and application references in the field of power load prediction.