Research On Non-intrusive Load Disaggregation Algorithm

Non-intrusive load disaggregation can provide detailed power consumption data from the aggregate power consumption data,which is the basis for the system operator to perform user management.In this paper,the non-intrusive load disaggregation related technology is comprehensively analyzed,and the non-intrusive load disaggregation algorithm is further studied.This paper mainly studies the following aspects:Firstly,aiming at the low accuracy of the event detection algorithms for non-intrusive load disaggregation,three event detection algorithms are improved based on steady-state active power characteristics: improved sliding CUSUM bilateral cumulative sum algorithm,improved sliding chi-square-GOF algorithm and improved sliding Cepstrum algorithm.Compared with the literature results,the three event detection algorithms have better accuracy.Compared with the traditional sliding chi-square-GOF algorithm,the improved sliding chi-square-GOF algorithm can effectively detect high-reference power events and avoid false detection in such cases.At the same time,based on the improved sliding CUSUM bilateral cumulative sum algorithm for event detection,power feature extraction is carried out by using the difference feature extraction algorithm.The comparison between the extracted feature sequence and the template feature sequence shows that the proposed method has a good feature extraction effect.Then K-means clustering is used to obtain the template feature sequence and the extracted feature sequence clustering centers.At the same time,the template feature sequence clustering centers and the extracted feature sequence clustering centers are matched to obtain matching distance ranking.Furthermore,the DTW algorithm is used to calculate the global matching distance between the template feature sequence and the extracted feature sequence to obtain the global similarity ranking.The most advanced equipment in the two ranking results is the identified device.Experiments show that this method can effectively identify the target device in certain scenarios.Secondly,aiming at the low accuracy of traditional non-intrusive load disaggregation 0-1 programming model,a non-intrusive load disaggregation 0-1 programming model based on the load power characteristic sequence is established.Four algorithms based on dimension-by-dimension computation are improved: quantum genetic algorithm(QGA),quantum cuckoo algorithm(QCS),modified quantum cuckoo algorithm(MQCS)and improved binary cuckoo algorithm(IBCS).Experiments show that these four algorithms can effectively solve the non-intrusive load disaggregation 0-1 programming model.Finally,the open dataset is used to construct the experimental data,and the above four algorithms are used to solve the non-intrusive load disaggregation 0-1 programming model.Experiments show that the above four algorithms for solving non-intrusive load disaggregation 0-1 programming model can achieve certain results with and without noise data,and MQCS algorithm has better disaggregation performance.Therefore,the above four algorithms can effectively solve the non-intrusive load disaggregation 0-1 programming model,and have a wide range of applicability.