Short-Term Electricity Load Forecasting Model Based on Bidirectional Long Short-Term Memory Network with Adaptive Boosting

Short-term load forecasting plays a vital role in the operation and management of power systems. While traditional forecasting models are capable of predicting general load trends, their predictions often exhibit significant deviations from actual values in specific regions, particularly in cases of rapid load fluctuations. To address this issue and minimize prediction errors, this paper introduces a novel short-term load forecasting model that integrates the Bidirectional Long Short- Terrm Memory (BiLSTM) network with the Adaptive Boosting (AdaBoost) algorithm, collectively referred to as the BiLSTM-AdaBoost model. The AdaBoost algorithm iteratively refines the weights and updates the weak predictors by focusing on error localization, while the BiLSTM network, functioning as the weak predictor, enhances the model's ability to learn both forward and backward dependencies in the load time series. A case study based on a real-world load dataset was conducted to compare the short-term forecasting results of this model with other models. The experimental results demonstrate that the proposed BiLSTM-AdaBoost model significantly improves prediction accuracy, particularly in managing complex load fluctuations, and effectively reduces overall prediction errors.