The Temperature Prediction of Hydro-generating Units based on Temporal Convolutional Network and Recurrent Neural Network

Abstract

Temperature prediction is an important means to ensure the safety of hydro-generating unit (HGU) due to the fact that a lot of faults of the HGU are reflected in the form of temperature rise. Because of huge monitoring data of the HGU, it is difficult to accurately predict the trend of temperature by the traditional recurrent neural network (RNN). In this paper, a time series prediction model based on temporal convolutional network (TCN) and RNN is proposed. First of all, a TCN including a temporal convolutional layer and a maximum pooling layer is built. By introducing a sliding window mechanism and a maximum pooling structure, the effective information is extracted from large-scale time series data, thus to reduce data size. Subsequently, a RNN model is established to predict the trend of the temperature. The TCN-RNN model avoids the problem of ineffective use of hidden layer information in traditional models, improves accuracy and speed of large-scale time series data prediction. The experimental results show that compared with RNN, CNN and CNN-RNN, the prediction accuracy of this model is significant improved, which can better meet the temperature prediction requirements of the HGU.