Abstract: In order to improve the accuracy of load forecasting for the four key periods of power grid operation, namely low valley load, noon peak load, waist load load, and evening peak load, a short-term load forecasting method based on segmented forecasting and weather similar day selection is proposed. Firstly, the paper analyzes the impact of different factors, including meteorological and economic factors, on the load of the regional power grid at different time periods, and select relevant features as the training set for construction.Secondly, the paper adopts a long short term memory neural network model to achieve load forecasting for different time periods. Using mutual information and Euclidean distance, the paper selects similar days with weather conditions close to the day to be predicted, and uses the load curve of that day as a reference,combining with the segmented load forecasting results as the load forecasting result for the day to be predicted. The experimental results show that the proposed short-term load forecasting method can ef‍fectively improve the accuracy of short-term load forecasting, especially for low valley, noon peak, waist load, and evening peak periods, with a significant improvement in prediction accuracy.

Key words: short-term load forecasting, similar day selection, long short term memory(LSTM), neural network, segmented prediction