Non-intrusive Load Disaggregation Based On Deep Temporal Convolutional Network

Smart grid and energy-saving emission reduction are important means and requirements for realizing the intelligence of China's power grid.In order to meet the requirements of smart measurement and demand-side management in smart grid,high-efficiency and low-cost load monitoring has become an urgent technical problem to be solved.Non-intrusive load monitoring(NILM)came into being.NILM refers to the installation of smart meters at the user's power supply entrance to monitor the total electricity consumption data,and the current load operation conditions in the system are obtained through the load disaggregation algorithm.This method can realize the monitoring of the user's power consumption information,which is conducive to the grid to formulate the operation plan;and for the user,the power consumption behavior can be adjusted according to the specific conditions of the electrical energy consumption of each electrical appliance.The current analysis method mainly relies on manually extracting the characteristics of the load,using traditional combination optimization(CO)and factorial hidden Markov model(FHMM)method to achieve disaggregation.With the development of artificial intelligence(AI)technology,technologies such as machine learning and deep learning have also been gradually applied to non-intrusive load decomposition problems.This kind of method does not require a high sampling rate,only uses low-end hardware to collect data,and uses deep neural networks to learn the load characteristics,so as to establish a corresponding load model to carry out load identification and decomposition.However,the lack of event detection in general deep learning methods leads to the repeated recognition of the extracted load features due to the sliding window,which increases the training difficulty and the amount of calculation.In view of the above problems,this paper proposes a NILM algorithm based on deep temporal convolutional network(TCN).First,an event detection method based on change point recognition is used to detect load events and extract load activation.Secondly,the load activation data is combined to generate sufficient training data.Next,a TCN is used to build a model of each load,and the corresponding relationship between the total power and the power consumed by a single load is constructed to achieve load disaggregation.This paper uses REDD data set to verify the algorithm,and uses evaluation indexes such as Precision,Recall,F-score and so on.Finally,a set of load measurement and decomposition device is designed,and the algorithm is applied to actual load decomposition to verify the effectiveness of the algorithm.