Research on Line Loss Prediction Method Based on Improved DBN Model

Abstract

The line loss rate is an important indicator for measuring the technical and non-technical losses in the distribution process. This paper proposes a prediction method for power distribution networks based on an improved Deep Belief Network (DBN) model and deep learning for line loss prediction in medium and low-voltage courts. Considering that the line loss data is time series data, this paper proposes the Cycle_DBN_A model. In the model’s training process, the greedy algorithm performs unsupervised pre-training layer by layer on the network layer in the model. Then the Adam optimizer is used to perform supervised global fine-tuning on the Cycle_DBN_A model. Test sets are to verify the algorithm. The Cycle_DBN_A model is superior to other models as it has a Mean Relative Error (MRE) of 2.0662% and is also the best in multi-data verification.