Enhanced Sampling And Improved HMM Based Non-intrusive Load Monitoring

With the deteriorating energy shortage problem and the need of smart grid development,Non-intrusive Load Monitoring(NILM)technology has increasingly become the key to energy monitoring,and the detailed load usage information obtained is beneficial to residents’ life science Electricity consumption and provide theoretical support for the power sector to formulate flexible electricity consumption policies.In the research of model-based methods,the Hidden Markov Model(HMM)has received extensive attention because it is suitable for describing the load state sequence.This paper studies the load training and load modeling disaggregation.A Non-intrusive load monitoring method based on enhanced sampling and improved HMM is proposed.Compared with general-purpose and model-based methods,the improved method adapts to lower data frequency and is suitable for family-level disaggregation requirements.Aiming at the problem of inaccurate load modeling,an intensive training algorithm is proposed.The method of delayed sampling and short-term logic learning reduces the influence of voltage instability and power quality degradation,and extracts the load curves of single and multi-modal loads more accurately.The experimental results show that the method overcomes the above problems,and the improvement in the training phase disaggregation accuracy.In order to solve the problem of loss of effective information in load decomposition,a model based on time information and an improved Viterbi algorithm are proposed.The duration of the electrical operating state is added to the load modeling,and the amount of calculation is reduced by compressing the matrix.The experimental results show that the method in this paper can not only retain the time information,but also improve the accuracy and robustness of disaggregation and identification.Design and complete simulation experiments for load clustering and load disaggregation.The improved method is compared in terms of state value and overall disaggregation performance;load disaggregation experiments are carried out to achieve modeling and identification in different scenarios,and the maximum number of simultaneous loads in the whole day reaches 15,of which multi-modal loads include washing machines and dish washers.The result analysis shows that the correct rate of adaptive clustering is over 95%,the average correct recognition rate of load events is better than that of the reference method in complex electrical appliances,and the standardized energy error is reduced to 0.22,which shows that the proposed method is effective in load disaggregation.Better performance in both state recognition and energy estimation.