Power Load Decomposition Of Multi-appliance Operation Based On Deep Learning

After decades of research and development investment and construction,the State Grid has basically met the requirements of the ubiquitous Internet of Things on the grid side.However,in the user terminals of the power grid,the data generated by nearly 400 million smart meters has not been fully and effectively utilized.How to use these data for power energy management value service is the key to the ubiquitous construction of power Internet of Things.Non-intrusive power load decomposition can estimate the usage time and power consumption of various appliances in a household from a smart meter that measures household electricity consumption.This technology can not only meet the demand for smart electricity in the new era,but also promote the development of smart grids.It can also effectively reduce electrical energy loss and play an important role in solving energy problems.In recent years,with the continuous increase of energy data,many data-driven deep learning frameworks have been applied to the field of non-intrusive power load decomposition.However,most of the proposed methods can only predict the use status or power consumption of a single appliance.In this situation,LSTM neural network is used to perform non-intrusive power load decomposition on multiple loads simultaneously.This article uses the UK-DALE data set for training and testing.First,the data is extracted,and then a state detection network is built.The extracted data is used to train the network.The trained network is used to detect the switching states of multiple appliances.Then build a load decomposition network to identify the power consumption of multiple appliances.In this work,the multi-appliance load recognition neural network architecture proposed in this paper avoids repeated modeling of a single appliance,reduces model training costs,and shortens model training time.This article compares the experimental results with existing deep learning methods.The load detection network comparison index is the recognition accuracy rate.The load decomposition network comparison index includes four items:accuracy,recall rate,F1 score,and average absolute error.This proves that the model in this article has Good decomposition effect.