Research On Non-intrusive Load Identification Algorithm Based On Smart Meter

Smart micro-grids play an important role in the development of smart grids,they have drawn much attention due to the development and research of a new generation of power grids based on the details of the users’ electricity consumption.As a non-invasive device,the smart meter can obtain the power information of each electricity load in the user’s home by analyzing the power data,and can give the result to the user so that the user can learn more detailed electricity consumption details information from the smart meter.A smart meter is designed in this thesis,and the data is processed based on the smart meter.The smart meter is able to achieve the user’s home online load decomposition in order to achieve the purpose of saving electricity.The system in this thesis includes three aspects: smart meter data acquisition system,connection module between smart meter and PC,smart meter data analysis algorithm.The Smart meter is used for power data collection,data storage,transmission and multi-function menu display.Upper computer software is to achieve two-way communication and set the smart meter system parameters between PC-side and smart meters.The data analysis algorithm based on smart meters includes two types of algorithms: current-based transient event detection and load identification based on transient current characteristics.In the transient event detection,the maximum current signal sequence is obtained first,and the sliding window residual model is used to monitor the switching events of the load.In order to reduce the range of load identification,the load corresponding to transient events is divided into non-resistance type(non-R type)and resistance type(R type).Two load identification algorithms for most non-R loads in life are proposed is this thesis.The first method is to extract the multi-dimensional waveform characteristics of transient current and the harmonic amplitude characteristics of S-transform,and use Canonical Correlation Analysis to combine these two types of features.The bi-directional 2DPCA is used to compress the transient current S-transform matrix in the second method.Finally,six categories from BLUED dataset are classified by SVM.Experimental results show that both algorithms achieve better classification results.Both algorithms have some advantages especially for on-line decomposition of similar electrical loads.