TS2VLGNN-based prediction of water temperature at the outlet valve of a valve cooling system

Abstract

During the operation of a power grid system, devices such as converter valves generate much heat, which leads to a constant increase in the device’s operating temperature, thus significantly reducing the stability and safety of the system. The temperature of the valve cooling system is critical to the stability of the system. In this paper, according to the limitations of traditional multivariate time series prediction methods, such as harrowing feature extraction and poor prediction performance, we propose a model TS2VLGNN to predict the valve cooling system’s outlet water temperature. TS2VLGNN first extracts the features of system variables through the time series-representation learning method TS2Vec and then combines the advantages of LSTM and GNN in multivariable time series prediction. Meanwhile, the GNN module is improved to predict the water temperature at the outlet valve. During the comparative experiment, the model was compared with several models, and the experimental results verified that the TS2VLGNN model has higher prediction performance. Finally, ablation experiments demonstrated each component of the model’s effectiveness.