Research On Non-intrusive Household Load Monitoring Metho

With the construction of smart grid and information technology,the power industry has accumulated massive data,which has the characteristics of big data in terms of data volume,diversity,speed and value,while distribution and power consumption are important parts of smart grid,and the acquisition of information on electrical equipment inside the user’s home is the basis of building smart grid,and nonintrusive load monitoring(NILM)can achieve this task.NILM provides a low-cost,durable and easy to install solution for energy consumption and equipment monitoring.It only needs to obtain the load data from the power inlet,and the power consumption of the internal load on the user side can be obtained through intelligent algorithm analysis.However,the traditional intrusive load monitoring needs to install sensors on each household load to detect the use of the load,which has the disadvantages of high cost,difficult maintenance and easy to affect the normal life of users.Compared with the traditional load monitoring method,NILM can monitor the internal load of the entire system only by installing the monitoring equipment at the entrance of the power supply,which has the advantages of simple operation and maintenance,low investment cost,strong information security,and noninfringement of user privacy.This paper studies non-intrusive load monitoring,and the specific work is as follows:(1)For the power signal sampled at low frequency,the measured data is first processed by median filtering to reduce the impact of data redundancy and noise,and then the change of the operating status of electrical equipment is judged by the event detection method based on the bilateral cumulative sum of sliding windows.On the basis of event detection,active power,reactive power,fundamental power factor,electrical switching time,electrical operation duration,electrical operation periodicity Temperature constitutes the load characteristic library,which can characterize the electrical behavior of electrical equipment.(2)On the premise of feature extraction,this paper uses the Extreme Learning Machine(ELM)model for load identification.For the extracted load characteristics,the principal component analysis(PCA)is used to reduce the data dimension of the load characteristics to obtain the comprehensive variables.Because the weight and bias of the ELM model are random,Therefore,the sparrow search algorithm(SSA)is used to optimize the weight and bias of the limit learning machine network model to obtain the optimal ELM model.Finally,the PCA-SSA-ELM model is used to identify the load of 8 household appliances in the AMPds data set.The comparative analysis shows that the proposed method has high accuracy and stability.(3)To solve the problem of load decomposition,this paper adopts a non-intrusive load decomposition method based on full convolutional self-coder,which is improved on the traditional Denoising Auto Encoder(DAE)network.The full connection layer in the DAE network is replaced by the convolution layer,and the FCN-DAE network composed of full convolutional neural network is obtained,which reduces the computational load,The load power other than the target electrical appliance is regarded as noise,and the load decomposition model is established.Finally,the UKDALE data set is used for experiments.The experimental results show that the FCNDAE load decomposition model has better decomposition effect.