Research On Event-based HMM For Non-intrusive Electrical Load Monitoring Algorithm

With the continuous progress of society,energy demand is exponential growth.Energy saving becomes a challenging problem.As the saying goes,sustainable development begins with the family.Research shows that by providing consumers with more detailed information about electricity consumption,consumers can manage their energy consumption patterns more reasonably and develop better energy efficiency plans.And so that residents can reduce living expenses effectively.One of the most important aspects of achieving this goal is energy monitoring.Non-intrusive load monitoring(NILM)has attracted much attention because of its low cost.This article chooses the power consuming situation of the residents as the study object.First of all,we analyze the type of the electrical appliances and the type of energy meter in the house.In our research,we consider ON state and OFF state of the electrical appliances and exploit the low frequency sampling power data to conduct experiments.Then,we point that three kinds of window-based event detection algorithm have the same shortcomings that their window size is difficult to set.So we proposed an event detection algorithm based on the extreme points of which the parameter setting is particularly simple.The algorithm also performs some processingon the noise.And it applies to the power data of various sampling rates.This paper adopts an expansion model of the hidden Markov model(HMM)as the load monitoring model and combines it with the event detection.The steady-state average power and the steady-state differential power obtained by the event detection are taken as the observed values.The total power information of all devices and the power transition information caused by the state transitions of some devices are fully utilized.The use of steady-state average power can greatly reduce the data inputs and capably smooth the noise.In this paper,we propose a Viterbi algorithm based on state simplicity to obtain the implicit state sequence of the model.Different from the traditional Viterbi algorithm,on the one hand,the new algorithm allows that two appliances can change their states at the same time.This can be concluded through the analysis of the actual data,so it is more realistic.On the other hand,this new algorithm sets two thresholds to simplify the state space,largely reduces the computational complexity,so this algorithm can quickly achieve load decomposition.Finally,this paper introduces the experiment with using MATLAB,including the performance comparison experiments of event detection algorithms and the performance comparison experiments of the load disaggregation algorithms.We also expounded several evaluation criteria.The simulation results show that the result of event detection algorithm based on extreme points is more accurate when compared to the other three window-based algorithms.The speed of our load disaggregation approach is nearly 30 times higher than that of the traditional algorithm.For some high power devices,the decomposition accuracy of our approach could reach 90%.And for most low power appliances,the decomposition accuracy could reach more than 60%.