Research On Non-intrusive Load Identification Method For Home Load

In order to build a new type of power system and realize the two-way interactive service between customers and the power grid,it is a very important work to classify and measure the electricity consumption of customer-side equipment,among which load monitoring technology is a typical technology to classify and measure the electricity consumption of customer-side equipment.The non-intrusive load identification technology can monitor the operation and energy consumption of the customer’s home appliance load only by collecting the voltage and current data from the customer’s incoming line.In this paper,the feature extraction and identification algorithm of non-intrusive load identification process are thoroughly studied and experimentally tested,and the main work and research contents are as follows:Transient steady-state analysis is performed on the data collected at the user inlet end,and gaussian white noise is filtered and denoised,then the time period of load throwing and cutting is detected by the sliding bilateral window CUSUM algorithm to obtain data of transient and steady-state operation,and the transient steady-state features of nine common household appliances are extracted and analyzed.To address the problems of high feature redundancy and low distinguishability,a load identification method based on VMD and RF is proposed.Firstly,the steady-state current waveform data are decomposed using VMD to obtain the modal components,and then the features are extracted from the modal components to form the original feature set;then RF is used for feature selection to reduce the redundancy of the feature set.Finally,the SSA algorithm is improved and the DASSA algorithm is proposed to optimize the SVM to improve the accuracy of identification.The simulation example analysis shows that the used method not only improves the recognition rate,but also reduces the time used for calculation.To address the shortcomings of the above methods: the randomness of the random forest algorithm feature scoring,which leads to a slight decrease in the discrimination of some appliances caused by the low scoring of a few features with good results and the influence of the optimization of SVM classifier parameters on the discrimination accuracy.In this paper,we image the numerical features and propose a discrimination method of V-I trajectory images incorporating power and harmonic features.Based on the original V-I trajectory features,the power and harmonic features are incorporated into the V-I trajectory in an encoded form to form a three-channel color image feature to compensate for the loss of power and harmonic features after high pixelation.The Lenet-5 neural network,which does not require optimized parameters,is then used to train and test the images for feature extraction and deep learning to verify the superiority of the method.The experimental results verify that the method is more accurate.For the two load identification methods proposed in this paper,a real-time non-intrusive load identification experimental platform is designed for detecting appliance start-stop and identification in real-time homes,and the proposed two methods are compared and experimented to conclude that the VMD and RF based load identification methods are more favorable to quickly identify household appliance categories of each load,while the V-I based on fused power and harmonic features trajectory images are more suitable for accurate prediction of household appliance categories.