Research on Load Forecasting Based on Deep Learning Framework Optimized by Sparrow Search Algorithm

Precise power load forecasting is crucial for maintaining the stable and efficient operation of power systems. To enhance predicted accuracy, this study introduces a hybrid forecasting approach that integrates Sparrow Search Algorithm (SSA) with deep learning framework. Convolutional neural network (CNN) is employed to capture nonlinear relationships among factors influencing power load, generating high-quality feature representations. Then the Sparrow Search Algorithm (SSA) optimizes the parameters of the Long Short-Term Memory (LSTM) to enhance its fitting ability to handle raw load data. Subsequently, the LSTM models the time-dependence of power loads and accurately captures the short-term fluctuation characteristics. Finally, the Attention mechanism is incorporated to assign weights to the input features, emphasizing critical features and improving the prediction accuracy. The experimental findings indicate that, compared with alternative models, the proposed CNN-SSA-LSTM-Attention model attains an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}$</tex> value of 0.983, with the MAPE significantly decreased to 2.85%. In addition, the RMSE and MAE are reduced to 2.94kW and 2.12kW, respectively, which are better than other models.