Short-Term Temperature Forecasting of Cable Joint Based on Temporal Convolutional Neural Network

Abstract

Precise forecasting of cable joint temperature is vital for guaranteeing the safe operation of power supply systems. Nonetheless, the existing predictions are primarily ultra-short-term predictions with brief time steps. This study presents a short-term temperature prediction method based on a work index and an improved Temporal Convolutional Neural Network (TCN). Firstly, the effects of the variables gathered in the monitoring system on the cable joint temperature were examined. To integrate and reconstruct these influencing factors, such as date and human activities, a variable called “work index” was introduced. Subsequently, considering the impact of multidimensional data features, a TCN was used to process long-term historical data effectively and extract potential time series features. Finally, short-term forecasting of cable joint temperature was achieved through multi-feature fusion mapping by using a Back Propagation (BP) neural network. The accuracy of the proposed method was validated using real data from cable joints. The experimental results demonstrated that the proposed method has a mean absolute error (MAE) and a mean squared error (MSE) of 0.929 and 1.529, respectively. It exhibited a 13.3% reduction in MAE and a 12.1% reduction in MSE when compared to state-of-the-art models.

Keywords

• Temperature forecasting of cable joint
• short-term time series forecasting
• work index
• temporal convolutional network (TCN)
• back propagation neural network