Non-intrusive Load Association Decomposition Technology For Energy Router

Energy router is to manage the orderly flow of energy,and the management is based on the extensive information transmission between various devices,so the energy router must obtain the detailed power consumption information of each device in order to operate efficiently.However,the traditional acquisition method needs to assemble a data acquisition module for each load and power supply equipment,which has high hardware cost and low practicability.In order to reduce the hardware cost and improve the practicability of the energy router,this paper uses non-intrusive load monitoring(NILM)to propose a correlation decomposition technology,which can separate the power consumption signals of independent loads from the total load signals obtained from an acquisition module,so as to realize the effective monitoring of each load.Firstly,the basic principle and structure of energy router and NILM are studied,including the basic idea,implementation feasibility and basic operation process.Aiming at the problems of low accuracy and less research on continuous load in the current load monitoring field,this paper realizes the model construction and workflow design of NILM in the framework of energy router system for different application scenarios.Then,the NILM technology based on low frequency data set is studied.In order to solve the problem of too few low-frequency sampling features,a load decomposition algorithm based on the combination of long-term memory network(LSTM)and attention mechanism is proposed.The low-frequency power sequence is coded,and the simulated annealing algorithm is used to segment the data set to obtain the appropriate power input sequence.The segmented power sequence is input into the trained recurrent neural network(RNN),and the attention mechanism is used to concentrate the output corresponding input near the key data,so as to improve the training speed and accuracy.After training,the new data is input into the network for prediction.In the experimental verification stage,the performance of RNN network is evaluated by absolute mean error,standardized decomposition error and total normalized signal error to verify the feasibility and effectiveness of the proposed algorithm.Finally,non-intrusive load monitoring technology based on high frequency data set is studied.High frequency sampling can improve the real-time performance of load decomposition,but the load characteristics are complex and there are many interferences in the extraction.In order to solve the problem that it is difficult to extract and identify high-frequency samples,a non-invasive load decomposition algorithm based on convolution and deep residual network is proposed in this paper.Firstly,independent component analysis(ICA)is used to separate independent signals from complex superimposed signals,and the current signals of independent loads are converted into waveform pictures and then input to convolution network for feature extraction,At the same time,the residual block is added to increase the depth of the network and improve the accuracy of load identification.In the experimental verification stage,we use the accuracy,accuracy,recall and the harmonic average of accuracy and recall to evaluate the algorithm,and verify the feasibility and effectiveness of the proposed algorithm.