CN-118656722-B - Low-frequency non-invasive load monitoring method and system based on self-adaptive event detection

CN118656722BCN 118656722 BCN118656722 BCN 118656722BCN-118656722-B

Abstract

The invention discloses a low-frequency non-invasive load monitoring method and system based on self-adaptive event detection, wherein the method comprises event detection, feature extraction and load identification; in event detection, a Bayesian information criterion is adopted as a detection window, a window threshold value is adaptively optimized through a power variable point weight model, power time sequences of different electrical equipment are decomposed and grouped by a variation modal decomposition method, and load identification and classification of power curve oscillograms of different electrical equipment are realized according to a load identification model. The method provided by the invention ensures that the event detection accuracy is ensured and higher load identification accuracy is also shown under the condition of low-frequency sampling.

Inventors

LUO ZHAO
LI ZHAO
WANG GANG
ZHANG TAO
LIN MINGLIANG
Deng Weiji
LI JIAHAO
SHEN XIN
ZHAO YITAO

Assignees

昆明理工大学

Dates

Publication Date: 20260508
Application Date: 20240515

Claims (6)

1. A low-frequency non-invasive load monitoring method based on self-adaptive event detection is characterized by comprising event detection, feature extraction and load identification, wherein a Bayesian information criterion is adopted as a detection window in the event detection, and a window threshold value is self-adaptively optimized through a power variable point weight model; During event detection, assuming that a power time sequence to be detected of the electrical equipment obeys poisson distribution, constructing the following two models, and constructing a maximum likelihood function in a Bayesian information criterion according to the following two models: no event occurs in the power time sequence, and a static model M s is constructed: ; Load events occur in the power time sequence, and a jump model M J is constructed: ; in the formula, Power time series indicating no event occurs Obeying the expectations and variances under poisson distribution; 、 Representing a power time series before a load event occurs in the power time series The expectations and variances of the poisson distribution are obeyed; representing a power time series after occurrence of a load event in the power time series The expectations and variances of the poisson distribution are obeyed; a marked point representing when a load event occurs, i.e., the nth sample of the power time series; The self-adaptive window threshold optimizing through the power variable point weight model comprises the following steps: In the power variable point weight model, the sequence of the hopping model M J is converted into a matrix form And : Static sequence Expressed as: ; Hopping sequence Expressed as: ; no event occurs in the static sequence, so the power increment corresponding to the sequence Zero, the jump sequence is accompanied by the state transition of the electrical equipment, so the power increment corresponding to the sequence Is not zero, is introduced into The minimum power increment representing the state switching of the electrical equipment is described as: ; the power delta set is: ; ; ; in the formula, An increment representing the nth sample of the power time series; And Respectively a static power increment set and a jump power increment set, wherein the power increment sets of the sequences to be detected The method comprises the following steps: ; Jump weight in power variable point weight model Assignment formula and jump information entropy The calculation formula of (2) is as follows: ; ; wherein, N 2 is the number of the jump sequence samples; and optimizing the detection window threshold h by combining with a Bayesian information criterion.
2. The adaptive event detection-based low frequency non-invasive load monitoring method according to claim 1, wherein the load recognition model is a CNN-SENet load recognition model that introduces a squeeze-stimulus attention mechanism into a convolutional neural network.
3. The low-frequency non-invasive load monitoring method based on adaptive event detection according to claim 2, wherein the CNN-SENet load identification model is characterized in that the model is divided into 1 input layer, 5 convolution layers, 1 pooling layer, 1 full connection layer and 1 output layer in the whole structure, wherein the first 4 convolution layers are conventional convolutions with convolution kernel size of 5x5, the 5 th convolution layer is 6x6 convolution kernel introducing SENet spatial attention module, the pooling layer performs maximum pooling and normalization processing based on a 5x5 digital matrix, the operation mode is "SAME", the loss function of the model is built based on cross entropy in the output layer, and Adam is selected as gradient optimizer to dynamically update CNN network parameters.
4. A low-frequency non-invasive load monitoring system based on self-adaptive event detection for implementing the method of claim 1 is characterized by comprising an event detection module, a feature extraction module and a load identification module, wherein a Bayesian information criterion is adopted in the event detection module as a detection window, the window threshold value is self-adaptively optimized through a power variable point weight model, the feature extraction module is used for decomposing and grouping load features of power time sequences of different electrical equipment by utilizing a variation modal decomposition method, and the load identification module is used for realizing load identification and classification of power curve oscillograms of the different electrical equipment according to the load identification model.
5. A processor, wherein the processor is operative to perform the low frequency non-intrusive load monitoring method of any of claims 1 to 3, based on adaptive event detection.
6. A computer readable storage medium, characterized in that the computer readable storage medium comprises a stored program, wherein the program when run controls a device in which the computer readable storage medium is located to perform the low frequency non-invasive load monitoring method based on adaptive event detection according to any of claims 1-3.

Description

Low-frequency non-invasive load monitoring method and system based on self-adaptive event detection Technical Field The invention relates to a low-frequency non-invasive load monitoring method and system based on self-adaptive event detection, and belongs to the field of smart grids. Background With the continuous increase of energy demand and the increasing attention to energy efficiency in the modern society, the progress of technology and industry revolution towards informatization, automation and intellectualization is accelerated, and the development of the electric power internet of things puts higher demands on the perception technology of an electric power system. For intelligent management at the user demand side, load monitoring is a key link. Conventional load monitoring techniques have not been able to support the development of digital management on the consumer demand side, while NILM techniques can accurately detect and identify load anomalies without interrupting power operation and without violating resident consumer privacy. The method has the characteristics of high precision, low cost, high efficiency and the like, can effectively reduce the maintenance cost of the power system, realizes the on-line monitoring of resident user load data, further provides fine energy efficiency analysis for the interactive management of a power grid power supply enterprise and resident demand side, guides the power supply enterprise to optimize a power supply and distribution structure, plans a low-carbon energy-saving power consumption mode for users, and improves the energy efficiency of the power system. In recent years, the construction of home energy management systems and research on load monitoring have attracted extensive attention from related scholars, and various intelligent resident load monitoring methods using NILM technology as a core have gradually emerged. Currently, the event detection NILM-like algorithm has made a certain progress in the field of resident load monitoring. In the field of high-frequency NILM, documents Lin Shunfu, lin Yifeng, li Yi, and the like, a non-invasive load identification technology based on load high-frequency characteristic imaging is researched [ J/OL ]. A power grid technology is 1-11[2024-04-03] "on the basis of a gray level diagram of a U-I characteristic curve, a gram angle field algorithm and a Markov transfer field algorithm are utilized to add periodic steady-state current characteristics and periodic instantaneous active power characteristics into an image, so that the accuracy of load identification is improved, but the method is complex in calculation, and has high requirements on measurement dimension of resident load electrical quantity data. The patent documents Wang Ying and Yang Wei, shoprior brave, etc. A non-invasive resident load monitoring method based on the refined identification of U-I track curve [ J ]. Power grid technology, 2021,45 (10): 4104-4113. "utilize fitting goodness test to capture electrical appliance switching events, then respectively adopt k-means algorithm and convolutional neural network model considering initial optimization to conduct two-stage identification on load, and an event type NILM method based on the refined identification of U-I track curve is provided. The method improves the intellectualization and refinement level of the NILM-like method for detecting the event to a certain extent, however, the methods on one hand present remarkable computational complexity, on the other hand, provide very high sampling requirements (not only high sampling frequency is needed, but also high hardware performance requirements of the NILM acquisition device) so as to prevent the high-precision event detection method based on data driving from being deeply fused with the NILM load identification terminal with low cost and high efficiency. With the diversity of household electrical appliances and functions, the electricity utilization environment of residents is increasingly complex, and how to solve the problems of high equipment cost, complex calculation method and the like caused by high-frequency sampling while ensuring high-precision NILM load identification has become a challenging research task in the field of load monitoring. In view of this, the present invention has been made. Disclosure of Invention The invention provides a low-frequency non-invasive load monitoring method and system based on self-adaptive event detection, which can carry out event detection under the condition of low-frequency sampling so as to reduce the sampling cost and the complexity of a calculation method. The technical scheme of the invention is as follows: According to the first aspect of the invention, a low-frequency non-invasive load monitoring method based on self-adaptive event detection is provided, and the method comprises event detection, feature extraction and load identification, wherein Bayesian information criteria are adopted