Optimizing LSTM for medium and long-term electricity load forecasting based on the improved mayfly algorithm

Medium and long-term electricity load forecasting is crucial in power system planning and system dispatching. To improve the accuracy of load forecasting, this paper proposes an improved mayfly algorithm (MA) to iteratively optimize three parameters of the long short-term memory (LSTM) model: training times, learning rate, and the number of hidden layers. The MA algorithm is improved from three aspects: First, the introduction of the Tent chaotic map and reverse learning initializes the population to enhance global search capability. Second, the adaptation of the incomplete gamma function and Beta cumulative distribution for adaptive inertia weight balancing. Third, the introduction of Gaussian mutation and Lévy flight strategies to alleviate the issue of local optimal solutions during the optimization process. In load forecasting based on historical data, the LSTM model based on the improved MA algorithm can fit the actual data well, and the prediction results are accurate and reliable.