Research On Non-Intrusive Residential Electricity Load Disaggregation Method

Non-intrusive load monitoring technology has always been one of the hot technologies in smart grid and home energy management.It can get the detailed information of the user’s household power consumption by analyzing the total power consumption data.It is beneficial for power companies to formulate power supply strategies,guide users’ electricity usage behavior,and also help users to check equipment failures,so as to achieve safe and economical electricity usage.This paper makes a detailed analysis of the technical framework of non-intrusive load monitoring,and focuses on the non-intrusive load disaggregation method.This paper studies the load disaggregation method based on the heuristic optimization algorithms,and improves the population initialization method and the construction method of fitness function of the optimization algorithm.Firstly,treating the characteristics of electricity data as time series data,the population initialization method for the heuristic optimization algorithm is improved,and the randomization of the previous steady-state result is added based on the traditional initialization method.By reducing the number of populations,the execution time of the algorithm is reduced.At the same time,experiments have shown that the improvement of initialization method can also improve the accuracy of Load disaggregation.Secondly,when constructing the fitness function of the heuristic optimization algorithm,this paper chooses to use multiple powers as features and uses the entropy method to calculate the weights of different powers.At the same time,it takes into account the predicted power and the constraints of operating equipment.Power parameters and connection parameters are introduced into the fitness function,and experiments show that the improved fitness function has a higher accuracy rate of load disaggregation.Finally,by comparing with the effectiveness of the improved particle swarm algorithm,genetic algorithm,and differential evolution algorithm,it is found that the improved particle swarm algorithm has a relatively good effect,and the particle swarm algorithm is selected for further research.This article refers to the idea of ensemble learning,using multi-model fusion method to solve the load disaggregation problem.First of all,based on the previous improvements,the particle swarm optimization algorithm was further improved by using the time probability table and the idea of optimizing the iterative process.Then based on the secondary improved particle swarm algorithm,difference feature model,and mixed integer programming model,the algorithm-parallel fusion strategy is used to construct a multi-model fusion load disaggregation method.In order to verify the effectiveness of the improved algorithm and the multi-model fusion method,this paper uses the foreign public data set to design six groups of experiments.The control variable method is used to investigate the effect of the single-model algorithm on the multi-model fusion method.The experiments show that the second improvement of the particle swarm algorithm has significantly improved the accuracy of the load disaggregation.At the same time,the single model algorithm has a positive contribution of different magnitudes to the load disaggregation under most samples.Finally,compared with other literatures,it is concluded that the load disaggregation method based on multi-model fusion is a relatively good method.