4th China International Youth Conference on Electrical Engineering (CIYCEE 2023)

In order to solve the problem of weak generalization ability of existing temperature prediction models when adapted to multiple devices in substation distribution cabinets, this paper proposes a CNN-BiGRU temperature prediction model with an attention mechanism. First, data preprocessing is carried out using normalization and outlier removal. Second, the BiGRU layer with attention mechanism is introduced in the hidden layer to filter the non-important information, and the residual-connected CNN architecture layer is utilized to avoid the problem of gradient vanishing. Finally, the effectiveness of the proposed model is verified by combining the actual temperature data of four distribution cabinets in a substation. The results show that the temperature prediction model proposed in this paper exhibits higher temperature prediction accuracy compared to LSTM, GRU, CNN-LSTM, CNN-GRU models.
 

Substation distribution cabinet; Temperature prediction; CNN-BiGRU model; Attention mechanism; Residual network