Research On Non-Intrusive Load Decomposition Method Based On Artificial Intelligence

With the rapid development of the energy Internet,the demand for intelligence on both the power distribution side and the power consumption side is also increasing,and intelligence is the theme of today’s social development.As a key technology in smart electricity consumption,load monitoring can help users grasp the electricity consumption information of home load in real time,and provide the power grid company with detailed information about users’ electricity consumption behavior.At this time,it is time to make changes.Nowadays,the load monitoring in society mainly adopts intrusive methods.The sensor devices used to collect electricity information and transmit information are directly installed in the user’s electrical equipment.This not only requires a lot of economic costs,but also There is a certain violation of the user’s security and privacy.In contrast,non-intrusive load decomposition only needs to install data collection and information transmission equipment at the total electricity meter,and then the total current or voltage data collected can be analyzed to obtain the operation of each electrical equipment in the home.And energy usage.Therefore,non-intrusive load decomposition has broad development prospects.This paper mainly studies the related methods of non-intrusive load decomposition in the steady-state process.First,based on the related principles of non-intrusive load decomposition,the basic framework of non-intrusive load decomposition is given,and the modules in the framework are briefly introduced.The load feature extraction methods are detailed in terms of transient feature extraction and steady-state feature extraction,and load decomposition algorithms are detailed in terms of mathematical optimization algorithms and pattern recognition algorithms.Secondly,a non-invasive load decomposition method based on convolutional recurrent neural network is proposed.While constructing the convolutional recurrent neural network,a stochastic gradient descent algorithm is added to optimize the convergence of the loss function,and a large amount of load data is generated by generating a confrontation network to ensure the huge demand for data by the neural network.Experimental results show that under the conditions of historical data and generated data,the convolutional recurrent neural network has high accuracy.Then,a non-invasive load decomposition method based on improved adaptive differential evolution algorithm is proposed.Construct the load state probability matrix,obtain the load state probability factor,improve the differential evolution algorithm by optimizing the control parameters of the differential evolution algorithm,and add probability and time dimensions to the objective function to construct a multi-characteristic objective function.Experimental results show that the improved differential evolution algorithm has a higher decomposition accuracy.