A load decomposition method based on deep learning

Abstract

The invention discloses a load decomposition method based on deep learning, which involves using a GRU-CNN fusion network to decompose the total household electrical load to obtain load data of specific electrical appliances. The method includes: using a sliding window to capture the time series data of the total household electricity load, constructing a GRU-CNN fusion network, and using the network to extract long-distance features and short-distance features of the input sequence in parallel, and fuse the two features to obtain a specific electrical appliance electricity load. The invention gives full play to the feature extraction capability of deep learning, effectively combines the long-distance feature capture capability of the GRU network for the input sequence and the CNN network's short-range feature capture capability for the input sequence, and avoids the limitation of extracting the input sequence features by a single network. , so that the method of the present invention has stronger adaptability, wider application range and higher accuracy.

Description

technical field [0001] The invention belongs to the technical field of smart grids, and relates to a load decomposition method for the total load of household electricity by a deep learning-based GRU-CNN fusion network. Background technique [0002] As an important branch of the energy field, the load decomposition problem is a single-channel blind source problem. The purpose of load decomposition is to decompose the total load of multiple electrical appliances into the electrical load of a single electrical appliance, so that people can know the details of their own electricity consumption, rather than just relying on monthly electricity bills to obtain electricity consumption feedback. Studies have shown that load splitting can help a home save 5% to 15% of its electricity consumption. In addition, the abnormal operation of electrical appliances can also be detected through load decomposition, so as to achieve the effect of timely detection of power failures. [0003] Tr...

AI Technical Summary

Problems solved by technology

[0003] Traditional load decomposition methods such as combinatorial optimization algorithm, hidden Markov model, support vector machine and other algorithms are more dependent on the setting of hyperparameters, the ability to learn features independently is poor, and there are certain limitations in load decomposition

Images