Recurrent Neural Networks Application to Forecasting with Two Cases: Load and Pollution

Abstract

Forecasting problems exist widely in our life. Its purpose is to enable decision makers to make effective responses to future changes. The traditional prediction methods based on probability and statistics cannot guarantee the accuracy of multivariable dynamic prediction under the background of high randomness and big data. In recent years, with the improvement of hardware computing ability and the large-scale increase of training data, deep learning has been widely applied in the field of forecasting. This paper focuses on the analysis of the application of recurrent neural networks (RNN), an advanced algorithm in deep learning, in the forecasting task. The forecasting models based on long short-term memory (LSTM) and gated recurrent unit (GRU) were established respectively, and the real data of power load and air pollution were verified. Compared with traditional machine learning algorithms, the simulation proves the superiority of the forecasting model based on RNN.