Lasso-based Non-invasive Home Power Load Decomposition Feature Selection Algorithm

Non-intrusive family load decomposition only needs to install monitoringdevices on the electric power entrance to the user, through the collection andanalysis of power use total current, voltage and power can identify the load classand operation of the individual load with the measured data. The utility can hasthe great significance on the safety operation of the electric power companypower grid stability, the rationality of the planning, scientific and normal andorderly management. And also for the normal production and life of the society,user frugal consumption, reduce the cost, and improve the economic benefit andsocial benefit.To counter the load power curve expression problem of the familyappliance load, according to different devices with different power curves, weestablish the autoregressive time series model to each electrical load in thefamily, with a Lasso and related feature selection method for parameterestimation and the order determined. The experimental results indicate that thismethod can effectively compress model coefficient, can accurately show powerconsumption of every electrical load. In this paper, we use the day before thepower data to establish model then to predict after a day of power load data,The experimental results indicate that the accuracy of prediction were above90%.Paper use the established time series model as recognition characteristics，using integer programming algorithm to the total power to decompose directly.This method can avoid the running status of equipment complex estimates; canbe used to measure the total power of access to information within the family ofevery detail of electrical equipment. Through the simulation analysis using thismethod decomposes the average accuracy can reach above85%. At the sametime, the author applied a family of forecast data of time series modelestablished by Lasso method, is to the same equipment in the other families. Theexperimental results indicate that the time series model established of this paperis only applicable to power load forecasting and decomposition of the same family.