부산대학교 도서관

Electricity plays a vital role in industrial production, residential life, and transportation. With the increasing power load demands, it is of great importance to accurately forecast power loads for power system planning and generation operation scheduling. To this end, this paper proposes an optimization model for short-term power load forecasting, utilizing Sand Cat Swarm Optimization (SCSO) in conjunction with a Bidirectional Long Short Term Memory (LSTM) Network. To address the issue of variable initial population quality in the SCSO algorithm, which can lead to local optima, a Modified-SCSO approach is introduced and applied to optimize the hyperparameters of the BiLSTM. Enhancing the accuracy of short-term power load forecasting can significantly improve the efficiency of power generation, dispatching, and maintenance, thereby ensuring the safe and efficient operation of power systems. Traditional LSTM models are hindered by high computational costs associated with hyperparameter tuning, and the tuning outcomes are often reliant on human expertise. The sand cat swarm algorithm inherently excels in hyperparameter tuning, achieving global optimal results through information exchange among individuals. This approach notably enhances neural network performance by discovering the most suitable hyperparameters. Experimental results indicate that, when comparing LSTM with SCSO-LSTM, MSCSO-BiLSTM consistently outperforms in all evaluation criteria. Moreover, with larger training data sets, there is a higher likelihood of achieving more accurate model predictions.