Equipment switch state detection method based on deep learning

Abstract

The invention belongs to the technical field of non-intrusive load identification, and particularly relates to an equipment switch state detection method based on deep learning. According to the method, a sequence-point processing mode is improved, that is, prediction of a single point is changed into prediction of a certain target sequence length, the target sequence is closer to a midpoint and a value near the midpoint, and the prediction efficiency is improved on the premise that good prediction performance is guaranteed; meanwhile, a network module used for modeling an original audio signal is introduced, a receiving domain of the network is expanded by adopting stacked cavity convolution, and efficient modeling and prediction output of a long data sequence are realized; the network module comprises a regression network used for performing load decomposition on the aggregated load data to obtain load data of different devices; and the network is classified to obtain the load state of the corresponding equipment, and the information of the two parts is integrally output and integrated to obtain a more accurate equipment on-off state prediction result. The method has remarkable superiority in the aspect of non-intrusive equipment switch state recognition.

Description

technical field [0001] The invention belongs to the technical field of non-intrusive load identification, and specifically relates to a deep learning-based device switch state detection method. Background technique [0002] Non-intrusive load identification is to try to identify the current operating state of the equipment by decomposing the power of the corresponding equipment from the total power of the meter. Therefore, how to infer the power load of a specific equipment, and how to decompose and identify the power load is the current stage. The main research objective. At present, deep learning has penetrated into all walks of life. Neural networks are used to solve various problems. How to use deep learning-related methods to improve the accuracy of load identification is a hot topic in current research. [0003] The total power data is a sequence of data. At present, there are two main processing methods for sequence data: sequence-sequence and sequence-point. Sequenc...

AI Technical Summary

Problems solved by technology

Traditional neural networks are divided into convolutional neural networks and cyclic neural networks. When the input sequence is long, the receptive field of the network increases, and the computational complexity of convolutional neural networks increases in a flat manner. Although the computational complexity of cyclic neural networks It increases linearly, but it can only be calculated sequentially, and cannot perform effective parallel computing processing. Therefore, long sequence modeling has always been a challenge for traditional neural networks.

Images