Electricity Consumption Forecasting Based on PC-CNN-BiLSTM Combined with Layered Transfer Learning Strategy

Abstract

Electricity consumption forecasting is a key participant in the smart grid system. Owing to the current electricity consumption data collected by smart meters having the characteristics of small data size and strong volatility, a parallel-channel convolutional neural network (CNN) and Bi-direction Long Short-Term Memory (BiLSTM) model (PC-CNN-BiLSTM) is proposed. Furthermore, an improved layered transfer learning strategy is proposed to extract the similar characteristics of the source domain data and the target domain data to improve the electricity consumption prediction accuracy as the available data sample is insufficient. We compare our model performance with that of deep learning to verify our approach, experimental results show that the proposed parallel-channels model can improve the prediction accuracy. And the improved layered transfer learning strategy can effectively reduce the prediction error compared with traditional transfer learning as the data samples are insufficient.