Non-intrusive load decomposition method based on Informer model coding structure

Abstract

The invention relates to a non-intrusive load decomposition technology, and aims to provide a non-intrusive load decomposition method based on an Informer model coding structure. The method comprises the following steps: preprocessing power data in electrical loads, and forming a sample pair according to a total power and a time power sequence of a single electrical load; building a training model by referring to an Informer model, wherein the model comprises a feature extraction part, a feature processing part and a feature mapping part which are arranged in sequence; initializing each parameter of the training model, and selecting a proper activation function and a loss function; training a training model by using the preprocessed sample data; inputting the total power curve into the trained model, and decomposing to obtain a power curve of a single load. According to the method, while long-term dependence of model input and output is improved, the occupancy rate of the model on a memory space and the calculation time complexity are reduced, and the decomposition precision of non-intrusive load decomposition is improved; and the calculation complexity and the space storage complexity are lower.

Description

technical field [0001] The invention relates to the field of non-intrusive load decomposition, in particular to a non-intrusive load decomposition method based on an Informer model coding structure. [0002] technical background [0003] In recent years, building an energy Internet has become an increasingly popular topic, and it has also attracted the attention of many researchers. To realize energy interconnection, energy management must first be carried out, and the most important foundation is to monitor and control the load and optimize energy reasonably. The configuration can achieve the purpose of saving energy and reducing energy consumption. The traditional load monitoring method mainly adopts intrusive means. The advantage of this method is that the data is accurate, but the disadvantage is that the actual operability is very poor, the implementation cost is high, and the user's acceptability is low. Therefore, non-invasive load decomposition methods are favored by...

AI Technical Summary

Problems solved by technology

However, methods based on deep learning still have certain problems, such as the generalization ability of the model, the imbalance of training data, and the lack of comprehensive evaluation indicators, etc., which require a lot of research to solve

Images