Non-intrusive load decomposition method based on seq2point network

Abstract

The invention discloses a non-intrusive load decomposition method based on a seq2point network, and the method comprises the steps: constructing a seq2point non-intrusive load decomposition model, and carrying out the training of the model; reading the total load power time sequence by a sliding window to generate an input sequence; inputting the input sequence into a one-dimensional convolutional layer so as to improve a one-dimensional convolutional network to automatically extract the characteristics of the input sequence and obtain the distributed characteristics of power data; and storing the extracted distributed power features in a fixed-length full-connection layer, and outputting the distributed features of the power data integrated into a sample space through an activation function to obtain a decomposed power sequence, thereby realizing Seq2point load decomposition. According to the method, data feature extraction and time sequence-based data features are fully considered, feature self-extraction is performed on the data through Conv1D, and identification errors under low-frequency sampling are reduced. The method has good generalization ability, and can identify a plurality of electric appliances.

Description

technical field [0001] The invention relates to a non-intrusive load identification technology, in particular to a non-intrusive load decomposition method based on a seq2point network. Background technique [0002] After the "carbon peak, carbon neutrality" policy was put forward, the State Grid took carbon neutrality as the ultimate goal, accelerated the construction of a clean energy Internet optimization configuration platform, and strengthened the use of big data, cloud computing, Internet of Things, artificial intelligence and other technologies in power generation. Integrate innovation and application with energy and other aspects. In this context, the research on the non-intrusive monitoring technology of consumer electrical appliances has become far-reaching. On the one hand, non-intrusive research on user electrical appliances can reduce the energy consumption of users, such as monitoring the usage of electrical appliances, and turning off unnecessary no-load elect...

AI Technical Summary

Problems solved by technology

[0005] There are a series of problems in the current non-intrusive load identification, such as poor data quality and low identification accuracy

First, sensor failures lead to abnormal data collection, and the number of samples in which appliances are in operation is low, resulting in poor data quality

For machine learning models, abnormal data will cause the model to learn wrong input-output correspondence, reducing the recognition accuracy of the model

And fewer running samples make the model tend to learn all the states of the appliance as not running, making it difficult to train an excellent learning model

Secondly, traditional machine learning methods rely on a priori parameter selection methods, which often lead to the selection of inappropriate parameters and the inability of the algorithm to fit real data. As a result, the accuracy of non-intrusive load identification is low, and it is difficult to meet the high requirements of actual scenarios. accuracy requirements

Images