Load forecasting of hybrid deep learning model considering accumulated temperature effect

Abstract

Considering the non-linear, multi-dimensionality and time-series of load data, a short-term power load forecasting method based on TCN-DNN hybrid deep learning model is proposed. Firstly, considering the common influencing factors of short-term load, and analysing the correlation, the temperature variable quantitative model considering the accumulated temperature effect is established. Secondly, based on the deep learning, the load influencing factors are classified and processed; In order to study the internal dynamics of the data, the Temporal Convolutional Network (TCN) is used to extract and construct time-series feature vector; The outputs integrate with non-time-series data factors (date type) into new input features. Finally, the internal relationship between the characteristics and load changes is analysed as a whole by the Deep Neural Network (DNN), and the load forecasting is finally completed. Taking the actual load data of the power grid in a certain area of East China as practical example, the forecasting accuracy of the proposed method is 97.92%. The proposed method is compared with Long Short-Term Memory (LSTM), TCN, DNN and the hybrid TCN-DNN model without temperature correction, the experimental results show that the proposed method has higher optimization efficiency and forecasting accuracy.

Keywords

• Short-term load forecasting
• Temperature correction
• Temporal Convolutional Network
• Deep Neural Network
• TCN-DNN