Short and medium term power load forecasting based on bidirectional long-short-term memory network models

Realizing the dynamic balance between power supply and power load is a prerequisite for the stable and economic operation of the power system. This paper proposes a short- and medium-term load forecasting method based on the Bidirectional Long-Short-Term Memory (Bi-LSTM) network to fully exploit the nonlinear relationship between power load and multidimensional characteristic factors. First, the proposed Bi-LSTM network improves the forecasting performance by capturing complex temporal patterns and interdependencies. The forecasting model absorbs and utilizes information from previous and future time steps, significantly enhancing the forecasting capability. Secondly, the two-way LSTM network strengths and weaknesses are tested by simulation analysis using load data considering the effects of holidays. Finally, the performance of the proposed method is measured against the traditional forecasting methods utilizing root mean square error and average absolute error. The results show that the proposed method has high prediction accuracy in different validation periods of days, weeks, and months, and its accuracy reaches 98.847%, 98.520%, and 98.480%, respectively, which can provide a reference for accurate prediction of short-term power loads.