A Survey of Electric Load Forecasting Algorithm Models

Abstract

With the increasing economic and social development, the load forecasting of the power system plays an increasingly important role in many energy-related fields such as energy dispatching, market-based price forecasting, and capacity allocation for the government and power companies. Based on the long-term research of many scholars, this paper aims to provide a general overview of some specific categories of load forecasting methods and models. Linear models can meet the real needs of forecasting in the early days, represented by autoregressive moving average (ARIMA) , but the lack of modeling ability requires researchers to develop more novel models based on machine learning and deep learning. Algorithms are widely used in the field of load forecasting, including human neural network (ANN) , support vector machine (SVM) , various variants of convolutional neural network (CNN) , memory network (RNN , LSTM...) , etc., which enhance the prediction performance and each have their own unique advantages. While the depth of the network continues to deepen, residual networks are proposed to optimize the model and solve a series of intractable problems. Different models are integrated and integrated together to form a new integrated network to give full play to the advantages of each model. However, these models do not solve the problem once and for all and have their own limitations. Finally, three different experiments are used to compare the three types of models mentioned in this article, and some typical network models are used as examples for performance analysis and demonstration. This review paper summarizes the valuable information of the prediction network, provides valuable information for subsequent research, and provides perspectives and entry points for subsequent research work.