| Smart community system supported by the Internet of Things and digital technology has greatly improved the intelligence of power grid on the distribution side.In the modern smart community system,large numbers of smart devices are connected to the Internet realizing the interconnection of devices by installing smart meters and smart appliances.On this basis,the power company is able to obtain detailed residential electricity consumption to carry out load forecasting,establish electric load simulation models and develop reasonable demand response strategies.Non-intrusive load monitoring(NILM)technology achieving detailed power consumption information and user consumption habits by applying intelligent algorithm to aggregated power signal,providing a low cost and strong expandability method,becomes a research hotspot.The mathematical model of non-intrusive load monitoring is presented,and key technologies of four constituent modules,which is data acquisition,feature extraction,algorithm foundation and evaluation,that define a general NILM framework are analyzed.This thesis summarizes the most commonly used load features and chooses active power as the only feature of load signatures considering the convenience of data acquisition and system scalability.On the basis of defining NILM as a time series problem,the advantages of replacing LSTM by deep convolutional neural network in tackling time series problems are analyzed.An encoder-decoder structured deep convolutional neural network is proposed as well as removing fully connected layers.The encoder part is stacked by hybrid dilated convolutional blocks.The performance of different convolutional kernel size as well as the influence of the dilate rate of dilated convolution on the receptive field is analyzed.The decoder part is stacked by multi-branch up-sampling blocks.Both nearest neighbor up-sampling and deconvolutional up-sampling are applied to recover power sequence.The architecture and technical route of the smart community system is designed by analyzing user demands.Ionic framework was chosen to build smart community software which allows cross-platform development of both Android and i OS.The terminal software is logically designed by modules.The cloud server was designed by sorting out its business logic,realizing HTTP based data communication as well as Web Socket based video streaming.The proposed energy disaggregation model is tested on UK-DALE dataset and proves its validity.The smart community terminal software turns out to be a human-centered design. |
