Research On Non-intrusive Load Monitoring Algorithms

Energy is an important material basis for the survival and development of human society,and it is also a key guarantee for the progress of the whole society.However,the energy shortage problem is manifested vividly in all walks of life,which has become a key problem restricting social development,and is gradually evolving into a global problem.In the proportion of energy consumption,electric energy has become the main consumption form of social production,and electric energy has also become the most widely used energy in people’s lives.Power energy consumption monitoring is the basis of energy saving.Strengthening power energy consumption monitoring has important practical significance in improving energy efficiency,realizing sustainable energy development,building a conservation-oriented society and alleviating energy pressure.Load monitoring can be divided into intrusive and non-intrusive types.Traditional Intrusive Load Monitoring(ILM)needs to install sensors at each load to monitor the operation of each load.Non-Intrusive Load Monitoring(NILM)was first proposed by George William Hart of MIT in the 1980 s.A sensor was installed only at the entrance of the total power supply to collect and analyze the total current and terminal voltage of the user’s electricity and monitor the working status of the electrical appliances in the whole system.The latter has become the main research direction in the field of load monitoring.For power users,it is possible to understand the operation status of each load in different periods in real time.For power networks,it is important to analyze and process load monitoring data,make scientific planning,and play an important role in reliability,safety,economy and environmental protection of power networks.For the whole society,it provides a basis for realizing energy saving,emission reduction and sustainable development of the whole society.On the basis of studying the basic concepts,principles and methods of NILM,this paper makes a thorough study on event detection and load decomposition,and carries out the following work:(1)Research on probabilistic model detection algorithm: deduce the decision functions of three detection algorithms,namely Cumulative Sum(CUSUM),Bayesian Information Criterion(BIC)and Generalized Likelihood Ratio Test(GLRT).Three probabilistic model detection algorithms are studied to detect multi-dimensional signals.The results show that the three algorithms can improve the accuracy of signal detection.(2)Research on matched filtering detection algorithm: Mathematical Morphology Gradient(MMG)algorithm is applied to NILM event detection.The simulation experiments are carried out using REDD data sets.The three probability model detection algorithms are compared with the matched filtering detection algorithm.The results show that MMG algorithm has the highest detection accuracy and can provide direction information.(3)Gauss Mixture Model Clustering: Based on the Gauss Mixture Model(GMM),the number of working states of electrical appliances is clustered.The GMM clustering algorithm assumes that the data obeys the Mixture Gauss distribution.It is suitable for NILM load data with multiple different distributions or the same type of distribution but different parameters,and GMM can provide information such as classification probability and positive-Pacific distribution parameters.(4)A Hidden Markov Model(HMM)is established for a single electrical appliance,and a Factorial Hidden Markov Model(FHMM)is established for all electrical appliances,which considering the load duration and external conditions.A load decomposition simulation experiment is carried out using AMPds data set.