Short-Term Power Load Forecasting Based on CNN-BiLSTM-Attention Neural Network with ISSA Optimized

Accurate short-term electricity load forecasting is essential to ensure the safe, reliable, and economically efficient operation of power systems. However, traditional forecasting methods are increasingly inadequate to meet the demand for ever more precise load predictions. In response to this issue, a CNN-BILSTM-Attention neural network method optimized based on an improved Sparrow Search Algorithm is proposed for short-term electricity load forecasting. Firstly, the Sin chaotic mapping strategy is utilized to initialize the Sparrow population, enhancing the diversity of population positions. Additionally, the cosine optimization algorithm is introduced into the formula for updating the joiner's position. This aims to address the decrease in global search capability in the later iterations of the Sparrow algorithm, thereby enhancing the accuracy of the Sparrow search algorithm; Next, the ISSA algorithm is employed to optimize the CNN-BILSTM-Attention neural network. This helps avoid overly random initialization of the CNN-BILSTM-Attention neural network's initial weight thresholds, thus enhancing the stability of prediction results; Finally, load forecasting is conducted and compared with other algorithmic models for analysis. Additionally, for the WOA-CNN-BILSTM-Attention model, RMSE decreases by 1.832%, with a corresponding MAE reduction of 1.073%. Similarly, for the PSO-CNN-BILSTM-Attention model, RMSE decreases by 1.804% and MAE by 1.026%. The results of the case study demonstrate that the proposed model can effectively enhance prediction accuracy and is valid for short-term load forecasting. By accurately predicting the power demand, the reasonable distribution and optimal dispatch of power can be realized, and the power grid fluctuations caused by the load mutation can be reduced, so as to ensure the quality of power supply.