Non-intrusive load monitoring method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to a non-intrusive load monitoring method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring various electric power characteristic data acquired by non-intrusive load monitoring equipment; performing feature importance analysis processing on the electric power feature data, dividing the electric power feature data into different importance categories, and determining the importance category corresponding to each type of electric power feature data; integrating the power feature data of the same importance category into channel input features; and correspondingly inputting each channel input feature into different channels of a preset multi-channel fusion model to perform feature fusion processing on the channel input features of the different channels to obtain a load monitoring result. Through an intelligent feature selection mode and a flexible structural design, the load identification performance is remarkably improved, and meanwhile, good interpretability and resource efficiency are kept, so that the non-intrusive load monitoring accuracy is effectively improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a non-intrusive load monitoring method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Art

[0002] Nowadays, energy demand is growing rapidly, and efficient energy saving has become an important issue. In order to achieve accurate monitoring of electricity consumption behavior, non-intrusive load monitoring (NILM) methods have attracted people's attention and achieved rapid development. Non-intrusive load monitoring technology only requires a single sensor to be installed at the power entrance of each household. It has low deployment cost and low installation complexity, and has broad application prospects. Non-intrusive load monitoring technology mainly focuses on low-frequency and high-frequency feature recognition. Low-frequency feature recognition focuses on load decomposition on a long time scale, while high-frequency feature recognition focuses on equipment start and stop events on a short time scale.

[0003] However, the neural network architecture used in current non-invasive load monitoring methods has poor flexibility and versatility, resulting in low accuracy of non-invasive load monitoring and poor load identification effect. Summary of the invention

[0004] Based on this, it is necessary to provide a non-invasive load monitoring method, device, computer equipment, computer-readable storage medium and computer program product that can improve the accuracy of non-invasive load monitoring in response to the above technical problems.

[0005] In a first aspect, the present application provides a non-intrusive load monitoring method, comprising:

[0006] Obtain various types of power characteristic data collected by non-intrusive load monitoring equipment;

[0007] Performing feature importance analysis on the power feature data, and classifying the power feature data into different importance categories, and determining the importance category corresponding to each type of power feature data;

[0008] Integrate the power feature data of the same importance category into channel input features;

[0009] The input features of each channel are input into different channels of a preset multi-channel fusion model. The channel input features of different channels are subjected to feature fusion processing through the preset multi-channel fusion model to obtain load monitoring results. The preset multi-channel fusion model is obtained by training the load monitoring data in the historical data as training samples.

[0010] In one embodiment, the power characteristic data is subjected to characteristic importance analysis and processing, and the power characteristic data is divided into different importance categories, and the importance categories corresponding to each type of power characteristic data are determined to include:

[0011] The power characteristic data is input into a preset feature importance measurement model, and importance measurement processing is performed on the power characteristic data through the preset feature importance measurement model to obtain importance parameters corresponding to different types of power characteristic data, wherein the preset feature importance measurement model is obtained by training the load monitoring data in the historical data as a training sample;

[0012] Sorting the different types of power characteristic data based on the importance parameter to obtain an importance sorting result;

[0013] Determine the position of the fault layer in the importance ranking result;

[0014] Different types of power characteristic data are divided into different importance categories according to the fault position, and the importance category corresponding to each type of power characteristic data is determined.

[0015] In one embodiment, the input features of each channel are input into different channels of a preset multi-channel fusion model, and the channel input features of different channels are subjected to feature fusion processing by the preset multi-channel fusion model to obtain the load monitoring result, which includes:

[0016] Inputting each channel input feature into different channels of a preset multi-channel fusion model, and performing feature integration processing on the channel input features of different channels through the preset multi-channel fusion model to obtain a feature fusion vector;

[0017] Performing feature extraction and nonlinear transformation processing on the feature fusion vector through the multi-layer perceptron of the preset multi-channel fusion model to obtain a weight optimization vector;

[0018] Performing feature adjustment processing on the weight optimization vector through the long short-term memory artificial neural network of the preset multi-channel fusion model to obtain a feature adjustment vector;

[0019] The feature adjustment vector is subjected to feature mapping processing through the linear layer of the preset multi-channel fusion model to obtain a load monitoring result.

[0020] In one embodiment, the method further comprises:

[0021] Obtain load monitoring data from historical data;

[0022] Based on the load category of the load monitoring data, the load monitoring data is labeled to obtain model training data;

[0023] The initial multi-channel fusion model is subjected to model training processing through the model training data to obtain a preset multi-channel fusion model.

[0024] In one embodiment, the method further comprises:

[0025] Determine the load type based on the load monitoring result, and determine the electrical equipment information corresponding to the non-intrusive load monitoring device;

[0026] The power consumption profile of the monitoring target of the non-intrusive load monitoring device is processed through the electrical equipment information to obtain the power consumption profile data of the monitoring target.

[0027] In one embodiment, the method further comprises:

[0028] Summarize the power consumption profile data of different monitoring targets in the region to obtain regional power consumption monitoring results;

[0029] Load scheduling is performed based on the regional electricity consumption monitoring results to generate a load scheduling plan for the region.

[0030] In a second aspect, the present application also provides a non-intrusive load monitoring device, comprising:

[0031] A data acquisition module is used to obtain various types of power characteristic data collected by non-intrusive load monitoring equipment;

[0032] An importance classification module is used to perform feature importance analysis on the power feature data, classify the power feature data into different importance categories, and determine the importance category corresponding to each type of power feature data;

[0033] A feature integration module, used for integrating power feature data of the same importance category into channel input features;

[0034] The feature fusion processing module is used to input the input features of each channel into different channels of a preset multi-channel fusion model, and perform feature fusion processing on the channel input features of different channels through the preset multi-channel fusion model to obtain load monitoring results. The preset multi-channel fusion model is obtained by training the load monitoring data in the historical data as training samples.

[0035] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0036] Obtain various types of power characteristic data collected by non-intrusive load monitoring equipment;

[0037] Performing feature importance analysis on the power feature data, and classifying the power feature data into different importance categories, and determining the importance category corresponding to each type of power feature data;

[0038] Integrate the power feature data of the same importance category into channel input features;

[0039] The input features of each channel are input into different channels of a preset multi-channel fusion model. The channel input features of different channels are subjected to feature fusion processing through the preset multi-channel fusion model to obtain load monitoring results. The preset multi-channel fusion model is obtained by training the load monitoring data in the historical data as training samples.

[0040] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0041] Obtain various types of power characteristic data collected by non-intrusive load monitoring equipment;

[0042] Performing feature importance analysis on the power feature data, and classifying the power feature data into different importance categories, and determining the importance category corresponding to each type of power feature data;

[0043] Integrate the power feature data of the same importance category into channel input features;

[0044] The input features of each channel are input into different channels of a preset multi-channel fusion model. The channel input features of different channels are subjected to feature fusion processing through the preset multi-channel fusion model to obtain load monitoring results. The preset multi-channel fusion model is obtained by training the load monitoring data in the historical data as training samples.

[0045] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0046] Obtain various types of power characteristic data collected by non-intrusive load monitoring equipment;

[0047] Performing feature importance analysis on the power feature data, and classifying the power feature data into different importance categories, and determining the importance category corresponding to each type of power feature data;

[0048] Integrate the power feature data of the same importance category into channel input features;

[0049] The input features of each channel are input into different channels of a preset multi-channel fusion model. The channel input features of different channels are subjected to feature fusion processing through the preset multi-channel fusion model to obtain load monitoring results. The preset multi-channel fusion model is obtained by training the load monitoring data in the historical data as training samples.

[0050] The above-mentioned non-invasive load monitoring method, device, computer equipment, computer-readable storage medium and computer program product obtain various types of power characteristic data collected by non-invasive load monitoring equipment; perform feature importance analysis on the power characteristic data, and divide the power characteristic data into different importance categories, determine the importance categories corresponding to each type of power characteristic data; integrate the power characteristic data of the same importance category into channel input features; input each channel input feature into different channels of a preset multi-channel fusion model, and perform feature fusion processing on the channel input features of different channels through the preset multi-channel fusion model to obtain load monitoring results. The present application classifies and integrates various types of power characteristic data through feature importance measurement, and then inputs the obtained channel input features into different channels of a preset multi-channel fusion model, thereby processing features of different importance through the preset multi-channel fusion model, making full use of the key information of the load, and processing secondary features at the same time, avoiding the under-fitting problem of traditional neural networks due to too many features, and increasing the adaptability and robustness of the model. This application achieves a significant improvement in load identification performance through intelligent feature selection methods and flexible structural design, while maintaining good interpretability and resource efficiency, thereby effectively improving the accuracy of non-invasive load monitoring and ensuring the effectiveness of load identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0052] FIG1 is a diagram of an application environment of a non-intrusive load monitoring method in one embodiment;

[0053] FIG2 is a schematic diagram of a flow chart of a non-intrusive load monitoring method in one embodiment;

[0054] FIG3 is a schematic diagram of the structure of a neural network model of multi-channel fusion in one embodiment;

[0055] FIG4 is a schematic diagram of a data stream processing method of a multi-channel fusion model in one embodiment;

[0056] FIG5 is a wiring diagram of a data acquisition system in one embodiment;

[0057] FIG6 is a schematic diagram of a current waveform under a load stable working state in one embodiment;

[0058] FIG7 is a block diagram of a non-intrusive load monitoring device according to an embodiment;

[0059] FIG. 8 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0061] The non-intrusive load monitoring method provided in the embodiment of the present application can be applied to the application environment shown in Figure 1. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. When the staff responsible for load monitoring wants to analyze the collected load monitoring data, they can submit a corresponding request to the server 104 through the terminal 102, and the server 104 performs corresponding monitoring data analysis and processing according to the request. The server 104 first obtains various types of power feature data collected by the non-intrusive load monitoring device; performs feature importance analysis and processing on the power feature data, and divides the power feature data into different importance categories, and determines the importance categories corresponding to each type of power feature data; integrates the power feature data of the same importance category into channel input features; inputs each channel input feature into different channels of the preset multi-channel fusion model, and performs feature fusion processing on the channel input features of different channels through the preset multi-channel fusion model to obtain the load monitoring result, and the preset multi-channel fusion model is obtained by training the load monitoring data in the historical data as a training sample. The terminal 102 may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart car devices, projection devices, etc. Portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. Head-mounted devices may be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. The server 104 may be an independent physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides cloud computing services.

[0062] In an exemplary embodiment, as shown in FIG2 , a non-intrusive load monitoring method is provided, which is described by taking the method applied to the server 104 in FIG1 as an example, and includes the following steps 201 to 207. Among them:

[0063] Step 201, obtaining various types of power characteristic data collected by non-intrusive load monitoring equipment.

[0064] Among them, non-intrusive load monitoring equipment refers to equipment that collects load monitoring related data in a non-intrusive manner. Non-intrusive load monitoring equipment monitors and analyzes the voltage, current and other data at the entrance of the power user's load to realize functions such as user power equipment category identification and itemized power energy calculation. By analyzing and processing the total load meter data, it is possible to identify each power device and its working status. Various types of power characteristic data, including fundamental wave, first, second, ..., seventh harmonic amplitude, current effective value, active power and reactive power, can be collected by non-intrusive load monitoring equipment as basic data for load monitoring.

[0065] Exemplarily, for the load monitoring process, when the staff responsible for load monitoring wants to analyze the collected load monitoring data, they can submit a corresponding load monitoring processing request to the server 104 through the terminal 102, and specify the data they want to analyze in the request. The server 104 then performs data search processing according to the request to obtain various types of power characteristic data collected by the non-invasive load monitoring equipment as the basis for data analysis. In one embodiment, the processing process for non-invasive load monitoring can be automatically triggered on a periodic basis, such as collecting and analyzing data on a daily or weekly basis. When a new cycle is reached, the server 104 can obtain various types of power characteristic data collected by the non-invasive load monitoring equipment in the previous cycle, thereby performing load monitoring analysis and processing on the previous collection cycle.

[0066] Step 203 , performing feature importance analysis on the power feature data, and classifying the power feature data into different importance categories, and determining the importance category corresponding to each type of power feature data.

[0067] Step 205 , integrating the power feature data of the same importance category into channel input features.

[0068] Among them, feature importance analysis refers to analyzing and ranking the importance of different types of power feature data to load monitoring. By evaluating the importance of features, the model can understand which features have a significant impact on load state prediction and improve the transparency of the model. Through feature importance analysis, different types of power feature data can be divided into different importance categories.

[0069] Exemplarily, since different types of power characteristic data have different effects on the prediction of load state during load monitoring, in order to improve the accuracy of load prediction, different types of power characteristic data can be pre-processed for feature importance analysis. Feature importance analysis can be implemented through a machine learning model, by inputting various types of power characteristic data into the model, thereby predicting the importance categories corresponding to each type of power characteristic data. In the subsequent prediction process, features of different importance can be processed to make full use of key information with high importance. In order to analyze feature data of different importance categories in the future, load monitoring analysis can be implemented through multi-channel processing, and on the basis of emphasizing important features, the interaction between features is captured, thereby enhancing the accuracy of load identification by the model. Therefore, power characteristic data of the same importance category can be integrated into channel input features. In one specific embodiment, various types of power characteristic data collected by the non-invasive load monitoring device include fundamental wave, first, second, ..., seventh harmonic amplitude, current effective value, active power and reactive power. The results of ranking the importance of these features are as follows: the third and fifth harmonics, current effective value and active power (named as the first type of features) are almost the most critical features, while the first, fourth, seventh harmonics and reactive power (named as the second type of features) are relatively less important. The fundamental amplitude, second and sixth harmonics (named as the third type of features) are the least important, and by integrating the three types of features, the three channel input features corresponding to the three-channel fusion model can be obtained.

[0070] Step 207, input the input features of each channel into different channels of the preset multi-channel fusion model, perform feature fusion processing on the channel input features of different channels through the preset multi-channel fusion model to obtain the load monitoring result. The preset multi-channel fusion model is obtained by training the load monitoring data in the historical data as training samples.

[0071] Among them, the multi-channel fusion model refers to a model that contains multiple input channels and can perform fusion analysis on the feature vectors of different channel inputs. The channel input features integrated according to the importance category can be input into the corresponding channel for feature fusion analysis. As for the training process of the preset multi-channel fusion model, the load monitoring data in the historical data can be collected and then labeled to obtain training samples, and then the training of the preset multi-channel fusion model can be completed, so that the preset multi-channel fusion model can capture the characteristics of different types of power feature data and ensure the accuracy of load monitoring.

[0072] Exemplarily, the applicant found that the current load monitoring model mainly adopts a single-channel model, that is, the power characteristics (such as voltage P, current Q, current I and harmonics of each order) are integrated into a one-dimensional vector as input. Then, the load state is identified by using neural network models such as Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM). However, these models cannot enhance the feature representation capability, and their flexibility and generalization capabilities are poor. To this end, the applicant upgraded the traditional single-channel architecture to a multi-branch structure, that is, by presetting a multi-channel fusion model to analyze power characteristics of different importance, and by inputting the characteristics of different importance into the model for fusion analysis, it is possible to capture the interaction between features on the basis of emphasizing important features, and ensure the effect of load identification. For the process of feature fusion analysis, the load monitoring results can be obtained through feature weight adjustment, feature optimization, feature fusion and feature conversion, thereby completing the processing flow of load monitoring and load identification, and determining the power equipment corresponding to the detected power feature data.

[0073] The above-mentioned non-intrusive load monitoring method obtains various types of power feature data collected by non-intrusive load monitoring equipment; performs feature importance analysis on the power feature data, and divides the power feature data into different importance categories, determines the importance categories corresponding to each type of power feature data; integrates the power feature data of the same importance category into channel input features; inputs each channel input feature into different channels of a preset multi-channel fusion model, and performs feature fusion processing on the channel input features of different channels through a preset multi-channel fusion model to obtain load monitoring results. The present application classifies and integrates various types of power feature data through feature importance measurement, and then inputs the obtained channel input features into different channels of a preset multi-channel fusion model, thereby processing features of different importance through a preset multi-channel fusion model, making full use of the key information of the load, and processing secondary features at the same time, avoiding the underfitting problem that may be caused by too many features in traditional neural networks, and increasing the adaptability and robustness of the model. The present application achieves a significant improvement in load identification performance through intelligent feature selection methods and flexible structural design, while maintaining good interpretability and resource efficiency, thereby effectively improving the accuracy of non-intrusive load monitoring and ensuring the effect of load identification.

[0074] In an exemplary embodiment, step 203 includes: inputting the power characteristic data into a preset feature importance measurement model, performing importance measurement processing on the power characteristic data through the preset feature importance measurement model, and obtaining importance parameters corresponding to different types of power characteristic data, wherein the preset feature importance measurement model is obtained by training the load monitoring data in the historical data as training samples; sorting different types of power characteristic data based on the importance parameters to obtain importance sorting results; determining the fault position in the importance sorting results; dividing different types of power characteristic data into different importance categories according to the fault position, and determining the importance categories corresponding to each type of power characteristic data.

[0075] Among them, the preset feature importance measurement model refers to a pre-trained model for feature importance evaluation. Specifically, a variety of machine learning algorithms with feature importance measurement functions such as XGBoost, LightGBM, random forest and AdaBoost can be selected to construct a preset feature importance measurement model, so as to calculate the feature importance of each feature involved in the non-intrusive load detection process.

[0076] Exemplarily, for the process of feature importance analysis, the contribution of different power feature data to the load monitoring process can be measured by a pre-constructed feature importance measurement model, and the preset feature importance measurement model is trained by using the load monitoring data in the historical data as a training sample. The various power feature data collected in the actual scene can be input into the preset feature importance measurement model, and the model predicts the importance parameters corresponding to each power feature data, and then the different types of power feature data are sorted based on the importance parameters to obtain the importance sorting results. Since there will be faults in the importance sorting results, different types of power feature data can be divided into different importance categories through the faults, so as to determine the importance categories corresponding to each type of power feature data. For example, in one embodiment, the various types of power feature data collected by the non-invasive load monitoring equipment include fundamental wave, first, second, ..., seventh harmonic amplitude, current effective value, active power and reactive power. The results of ranking these features by feature importance parameters are as follows: the third and fifth harmonics, current effective value and active power (named as the first type of features) are almost the most critical features, while the first, fourth, seventh harmonics and reactive power (named as the second type of features) are relatively less important, and there is a clear gap between the importance of the second type of features and the first type of features. For example, the fundamental amplitude, the second and sixth harmonics (named as the third type of features) are the least important, and there is also a clear gap between the first and second types of features. By integrating the three types of features, three channel input features corresponding to the three-channel fusion model can be obtained. In a specific embodiment, for a fixed application scenario, the various types of power feature data collected can be determined in advance, and the importance classification of these power feature data can be completed. Then, in the subsequent non-invasive load monitoring process, the various types of power feature data collected can be directly integrated into channel input features according to the previous feature importance classification results. In this embodiment, the importance of various types of power feature data is evaluated and classified by machine learning algorithms, which can effectively ensure the efficiency and accuracy of feature importance classification, thereby improving the effect of load monitoring.

[0077] In an exemplary embodiment, step 207 includes: inputting the input features of each channel into different channels of a preset multi-channel fusion model, performing feature integration processing on the channel input features of different channels through the preset multi-channel fusion model to obtain a feature fusion vector; performing feature extraction and nonlinear transformation processing on the feature fusion vector through the multi-layer perceptron of the preset multi-channel fusion model to obtain a weight optimization vector; performing feature adjustment processing on the weight optimization vector through the long short-term memory artificial neural network of the preset multi-channel fusion model to obtain a feature adjustment vector; performing feature mapping processing on the feature adjustment vector through the linear layer of the preset multi-channel fusion model to obtain a load monitoring result.

[0078] Among them, the multilayer perceptron (MLP) refers to stacking many fully connected layers together. Each layer outputs to the layer above until the final output is generated. The Long Short-Term Memory (LSTM) artificial neural network is a time recurrent neural network that is specially designed to solve the long-term dependency problem of general RNNs (recurrent neural networks). All RNNs have a chain form of repeated neural network modules.

[0079] Exemplarily, for the process of feature fusion processing, each channel input feature is input into different channels of a preset multi-channel fusion model, and each channel entrance corresponds to a feature encoder, and the input channel input features can be encoded. After the encoding is completed, the channel input features from different channels are first integrated and merged into a vector, and then the integrated vector passes through a multi-layer perceptron (MLP) module for further feature extraction and nonlinear transformation. By training the MLP neural layer parameters, the features from different channels are weighted and optimized, so as to highlight the important features and obtain the weight optimization vector. The weight optimization vector is then preset in the long short-term memory artificial neural network of the multi-channel fusion model to capture sequence information and model time dependency. Finally, the load monitoring result is obtained through the feature mapping processing of the linear layer. By integrating and transforming the information from different channels. It connects the channel features together to promote the interaction between the features. At the same time, the weight of the features from the first channel can be increased by the multi-layer perceptron, so as to highlight the influence of the important features on the load identification. In a specific embodiment, the structure of the neural network model based on multi-channel fusion can be referred to Figure 3, and its data flow processing method can be shown in Figure 4, where the calculation method of the long short-term memory artificial neural network structure is as follows.

[0080]

[0081]

[0082]

[0083]

[0084] in: is the input at the current moment; and are the outputs of the update gate and the reset gate respectively; tanh is the hyperbolic tangent function. In one embodiment, the third and fifth harmonics, current effective value and active power are the first-class features with the highest importance, the first, fourth, seventh harmonics and reactive power are the second-class features with the second highest importance, and the fundamental amplitude, second and sixth harmonics are the third-class features with the lowest importance. The three channel input features are input into the fusion model through three channels respectively. At this time, the input size of the first channel is 96×4, that is, the first-class feature data of 96 data points (third and fifth harmonics, current effective value and active power features); the input size of the second channel is 96×4, that is, the second-class feature data of 96 data points (first, fourth, seventh harmonics and reactive power features); the input size of the third channel is 96×3, and the input is the third-class feature data (fundamental amplitude, second and sixth harmonic features). In this embodiment, a multi-layer perceptron, a long short-term memory artificial neural network and a linear layer are combined to complete the fusion and prediction processing of multiple channel input features, so as to obtain accurate load monitoring results and determine the type of electrical appliance corresponding to the power characteristic data.

[0085] In an exemplary embodiment, the method further includes: obtaining load monitoring data from historical data; labeling the load monitoring data based on the load category of the load monitoring data to obtain model training data; and training the initial multi-channel fusion model using the model training data to obtain a preset multi-channel fusion model. The results show that the average load recognition accuracy of the multi-channel model reaches 85%, which is much higher than the existing commonly used methods. The recognition accuracy is particularly improved for electrical appliances such as air conditioners and rice cookers that are difficult to identify using conventional methods.

[0086] Exemplarily, before performing load monitoring-related processing through the preset multi-channel fusion model, the training process of the model needs to be completed. During training, it is necessary to collect load monitoring data from historical data. At the same time, for different load monitoring data, it is also necessary to mark the load category corresponding to each group of load monitoring data, such as which electrical appliance the load belongs to, such as the load monitoring data corresponding to the air conditioner, the load monitoring data corresponding to the refrigerator, etc. The model training data is obtained, and then the model training data is divided into a training group and a test group. The model training process of the initial multi-channel fusion model is completed through the training group, and the test process of the trained model is completed through the test group, until a usable preset multi-channel fusion model is obtained. Among them, the initial multi-channel fusion model uses the cross entropy function as the loss function during the training process, and the calculation formula is as follows:

[0087]

[0088] in: Represents the output of the model; represents the true label; is the number of categories; In this embodiment, the preset multi-channel fusion model is completed by collecting the load monitoring data in the historical data, thereby effectively ensuring the recognition effect of the preset multi-channel fusion model on different types of power feature data and ensuring the accuracy of load monitoring.

[0089] In an exemplary embodiment, the method also includes: determining the load type based on the load monitoring results, and determining the electrical equipment information corresponding to the non-invasive load monitoring device; performing power consumption profiling on the monitoring target of the non-invasive load monitoring device through the electrical equipment information to obtain the power consumption profile data of the monitoring target.

[0090] Exemplarily, after obtaining the load monitoring results for power characteristic data, it is possible to determine which electrical equipment is used by the monitoring target corresponding to the current non-invasive load monitoring device based on the power characteristic data of different electrical equipment collected by the non-invasive load monitoring device, and how these electrical equipment are used, including when they are used, how often they are used, etc., so as to complete the power consumption image processing of the monitoring target, obtain the power consumption image data of the monitoring target, so as to more conveniently serve the monitoring target and realize intelligent power dispatching and distribution. At the same time, it is also possible to realize regional load dispatching through macro-level non-invasive load monitoring, that is, to summarize the power consumption image data of different monitoring targets in the region and obtain regional power consumption monitoring results. In this way, the power consumption, peak power consumption period, peak power consumption and other data of the monitoring area are determined, and then the load dispatching processing is performed based on the regional power consumption monitoring results, and the load dispatching plan of the region is generated to obtain the power supply strategy that best suits the current monitoring area. In this embodiment, by dispatching the power consumption image of the monitoring target and the load of the region, it is possible to effectively ensure the accuracy of regional power supply and the utilization rate of electric energy.

[0091] In one embodiment, the non-invasive load monitoring method of the present application includes: obtaining load monitoring data from historical data; based on the load category of the load monitoring data, labeling the load monitoring data to obtain model training data; performing model training on the initial multi-channel fusion model through the model training data to obtain a preset multi-channel fusion model. Obtain various types of power characteristic data collected by the non-invasive load monitoring device; input the power characteristic data into a preset feature importance measurement model, perform importance measurement on the power characteristic data through the preset feature importance measurement model, and obtain the importance parameters corresponding to different types of power characteristic data, the preset feature importance measurement model is obtained by training the load monitoring data in the historical data as training samples; sorting different types of power characteristic data based on the importance parameters to obtain importance sorting results; determining the fault position in the importance sorting results; dividing different types of power characteristic data into different importance categories according to the fault position, and determining the importance corresponding to each type of power characteristic data Category; integrate the power feature data of the same importance category into channel input features; input each channel input feature into different channels of the preset multi-channel fusion model, and integrate the channel input features of different channels through the preset multi-channel fusion model to obtain a feature fusion vector; extract and nonlinearly transform the feature fusion vector through the multi-layer perceptron of the preset multi-channel fusion model to obtain a weight optimization vector; adjust the weight optimization vector through the long short-term memory artificial neural network of the preset multi-channel fusion model to obtain a feature adjustment vector; perform feature mapping on the feature adjustment vector through the linear layer of the preset multi-channel fusion model to obtain the load monitoring result. Determine the load type based on the load monitoring result, and determine the electrical equipment information corresponding to the non-intrusive load monitoring device; perform power consumption portrait processing on the monitoring target of the non-intrusive load monitoring device through the electrical equipment information to obtain the power consumption portrait data of the monitoring target. Summarize the power consumption portrait data of different monitoring targets in the region to obtain the regional power consumption monitoring result; perform load scheduling based on the regional power consumption monitoring result to generate the regional load scheduling plan.

[0092] As for the experimental verification process of the non-intrusive load monitoring device of this application, the applicant built a small radial power system by using multiple electrical equipment, and the system was built in a residential area. The system can automatically collect the current and voltage data of the total equipment and realize data storage, and its wiring diagram is shown in Figure 5. By controlling the opening and closing of the switch, the current and voltage data of a single device running alone or multiple devices running in combination can be measured.

[0093] During the experimental verification process, nine types of electrical loads, including air conditioners, rice cookers, televisions, kettles, induction cookers, washing machines, projectors, refrigerators, and hair dryers, were specifically selected as the load identification objects of the method of this application. The current waveforms of various electrical loads in steady state are shown in Figure 6. In order to facilitate the description of the experimental results, these nine electrical appliances are numbered 1 to 9. When collecting data, the sampling frequency is 12.8kHz. The nine electrical appliances are randomly started and stopped, with a total of 250 starts and stops, and the collection time is 500 minutes. Based on the collected current and voltage data of the total load, by calculating its active power, the event detection method based on the sliding window is used to detect the switching event, and then the load characteristics are extracted, and the method described in this application is used to identify the load operation status, so as to realize non-invasive load monitoring.

[0094] When using the data acquisition tool for data acquisition, first change the operating state of a single power load by opening and closing a single switch, start data acquisition before the load state changes, and terminate data acquisition after the load operating state reaches stability. The system can collect voltage, current and other data including transient and steady-state processes when the load is running alone. Then, change the operating state of multiple power loads at the same time by opening and closing multiple switches, collect current, voltage and other data including transient and steady-state processes when multiple loads are running in combination, and collect enough test cases for subsequent algorithms.

[0095] For the feature importance identification process, the feature importance involved in non-intrusive load monitoring was analyzed and compared based on a variety of machine learning algorithms with feature importance measurement functions, such as XGBoost, LightGBM, random forest and AdaBoost. The results are shown in Table 1.

[0096] Table 1: Importance of features for non-intrusive load monitoring

[0097]

[0098] Among them, Harm0, Harm1,…,Harm7 represent the fundamental amplitude, the first, second,…, seventh harmonic amplitude. The results show that the third and fifth harmonics, current effective value and active power (named as the first type of features) are almost the most critical features, while the first, fourth, seventh harmonics and reactive power (named as the second type of features) are relatively less important, and the importance of the second type of features is significantly different from that of the first type of features. For example, the fundamental amplitude, the second and sixth harmonics (named as the third type of features) are the least important, and there is also an obvious gap between them and the first and second types of features. Therefore, in order to improve the model's ability to learn and represent important features, this application introduces a multi-channel fusion mechanism to upgrade the single-channel architecture of the traditional neural network to a three-branch structure.

[0099] For the final load identification effect, in the process of non-intrusive load monitoring, feature selection has a direct impact on the performance and economic cost of the system. If only focusing on basic power parameter features, such as current, power, reactive power and apparent power. Such features have low requirements for sampling frequency and low implementation costs, and these features are classified as "low-cost features". It should be noted that low-cost features are prone to low load identification accuracy due to insufficient details of load imprint features. In order to pursue higher identification accuracy, it is necessary to introduce advanced features such as harmonic features. Harmonic features can provide richer power quality information and help to more accurately parse and distinguish various load characteristics. However, the introduction of harmonic features requires an increase in sampling frequency, which leads to increased costs. These features are classified as "high-cost features". These two types of features are collectively referred to as "full features".

[0100] This application adopts a comparative analysis method to systematically evaluate and compare the load identification effects of the following three methods, aiming to highlight the superiority of the multi-channel fusion model proposed in this application. The three methods include: method (1) a single-channel model based on low-cost features; method (2) a single-channel model based on full features; method (3) a multi-channel fusion model based on full features. The comparison results can be seen in Table 2.

[0101] Table 2: Load identification accuracy (unit: %)

[0102]

[0103] Table 2 lists in detail their accuracy rates in various household appliance recognition tasks. Analysis of the table data shows that the three recognition algorithms show significant differences in recognition accuracy. The average accuracy of the single-channel model (low-cost features) of method (1) is 69%. Although the cost is low, the recognition effect is poor. Among them, the recognition accuracy of appliances such as rice cookers, kettles and washing machines is low. This is mainly because these appliances may produce similar current and voltage characteristics when working, or the current value is low, making the model difficult to distinguish. The average accuracy of the single-channel model (full features) of method (2) is improved to 79%, indicating that adding harmonic features can significantly improve the recognition accuracy. Compared with the previous two methods, the method (3) of the present application has the following advantages: First, higher average recognition accuracy: The average recognition accuracy of method (3) is 85%, which provides better appliance recognition effect overall. For most of the 9 types of appliances listed, the recognition accuracy of method (3) is higher than that of method (1) and method (2). Second, the recognition accuracy of some types of electrical appliances has been significantly improved: for example, for air conditioners, the recognition accuracy of method (3) is 20 percentage points higher than that of method (1) and 5 percentage points higher than that of method (2). For rice cookers, the recognition accuracy of method (3) is 22 percentage points higher than that of method (1) and 7 percentage points higher than that of method (2). This shows that method (3) has a more obvious advantage in some types of electrical appliances.

[0104] The above experimental results highlight the advantages of the multi-channel fusion model, which can make full use of multiple feature channels to more comprehensively capture the characteristics of electrical appliance signals, thereby improving recognition accuracy.

[0105] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0106] Based on the same inventive concept, the embodiment of the present application also provides a non-intrusive load monitoring device for implementing the non-intrusive load monitoring method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more non-intrusive load monitoring device embodiments provided below can refer to the limitations of the non-intrusive load monitoring method above, and will not be repeated here.

[0107] In an exemplary embodiment, as shown in FIG7 , a non-intrusive load monitoring device is provided, comprising:

[0108] The data acquisition module 702 is used to acquire various types of power characteristic data collected by non-intrusive load monitoring equipment.

[0109] The importance classification module 704 is used to perform feature importance analysis on the power feature data, classify the power feature data into different importance categories, and determine the importance category corresponding to each type of power feature data.

[0110] The feature integration module 706 is used to integrate the power feature data of the same importance category into channel input features.

[0111] The feature fusion processing module 708 is used to input the input features of each channel into different channels of the preset multi-channel fusion model, and perform feature fusion processing on the channel input features of different channels through the preset multi-channel fusion model to obtain the load monitoring results. The preset multi-channel fusion model is obtained by training the load monitoring data in the historical data as training samples.

[0112] In one embodiment, the importance classification module 704 is specifically used to: input the power characteristic data into a preset feature importance measurement model, perform importance measurement processing on the power characteristic data through the preset feature importance measurement model, and obtain the importance parameters corresponding to different types of power characteristic data, and the preset feature importance measurement model is obtained by training the load monitoring data in the historical data as a training sample; sort the different types of power characteristic data based on the importance parameters to obtain the importance sorting result; determine the fault position in the importance sorting result; divide the different types of power characteristic data into different importance categories according to the fault position, and determine the importance category corresponding to each type of power characteristic data.

[0113] In one embodiment, the feature fusion processing module 708 is specifically used to: input the input features of each channel into different channels of a preset multi-channel fusion model, perform feature integration processing on the channel input features of different channels through the preset multi-channel fusion model to obtain a feature fusion vector; perform feature extraction and nonlinear transformation processing on the feature fusion vector through the multi-layer perceptron of the preset multi-channel fusion model to obtain a weight optimization vector; perform feature adjustment processing on the weight optimization vector through the long short-term memory artificial neural network of the preset multi-channel fusion model to obtain a feature adjustment vector; perform feature mapping processing on the feature adjustment vector through the linear layer of the preset multi-channel fusion model to obtain a load monitoring result.

[0114] In one embodiment, a model training module is also included, which is used to: obtain load monitoring data from historical data; label the load monitoring data based on the load category of the load monitoring data to obtain model training data; and perform model training on the initial multi-channel fusion model through the model training data to obtain a preset multi-channel fusion model.

[0115] In one embodiment, it also includes a power portrait module, which is used to: determine the load type based on the load monitoring results, and determine the electrical equipment information corresponding to the non-invasive load monitoring device; perform power consumption portrait processing on the monitoring target of the non-invasive load monitoring device through the electrical equipment information to obtain the power consumption portrait data of the monitoring target.

[0116] In one embodiment, it also includes a load scheduling module, which is used to: summarize the power consumption portrait data of different monitoring targets in the area to obtain regional power consumption monitoring results; perform load scheduling processing based on the regional power consumption monitoring results to generate a regional load scheduling plan.

[0117] Each module in the above non-intrusive load monitoring device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0118] In an exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG8. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data related to non-intrusive load monitoring. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a non-intrusive load monitoring method is implemented.

[0119] Those skilled in the art will understand that the structure shown in FIG. 8 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0120] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0121] In one embodiment, a computer-readable storage medium is provided, storing a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0122] In one embodiment, a computer program product or computer program is provided, the computer program product or computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the steps in the above-mentioned method embodiments.

[0123] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0124] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0125] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0126] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A non-intrusive load monitoring method, characterized in that: The method comprises: Obtain various types of power characteristic data collected by non-intrusive load monitoring equipment; Performing feature importance analysis on the power feature data, and classifying the power feature data into different importance categories, and determining the importance category corresponding to each type of power feature data; Integrate the power feature data of the same importance category into channel input features; The input features of each channel are input into different channels of a preset multi-channel fusion model. The channel input features of different channels are subjected to feature fusion processing through the preset multi-channel fusion model to obtain load monitoring results. The preset multi-channel fusion model is obtained by training the load monitoring data in the historical data as training samples.

2. The method according to claim 1, characterized in that The performing feature importance analysis on the power feature data, and classifying the power feature data into different importance categories, and determining the importance categories corresponding to each type of power feature data include: The power characteristic data is input into a preset feature importance measurement model, and importance measurement processing is performed on the power characteristic data through the preset feature importance measurement model to obtain importance parameters corresponding to different types of power characteristic data, wherein the preset feature importance measurement model is obtained by training the load monitoring data in the historical data as a training sample; Sorting the different types of power characteristic data based on the importance parameter to obtain an importance sorting result; Determine the position of the fault layer in the importance ranking result; Different types of power characteristic data are divided into different importance categories according to the fault position, and the importance category corresponding to each type of power characteristic data is determined.

3. The method according to claim 1, characterized in that The input features of each channel are input into different channels of a preset multi-channel fusion model, and the channel input features of different channels are subjected to feature fusion processing by the preset multi-channel fusion model to obtain the load monitoring result, which includes: Inputting each channel input feature into different channels of a preset multi-channel fusion model, and performing feature integration processing on the channel input features of different channels through the preset multi-channel fusion model to obtain a feature fusion vector; Performing feature extraction and nonlinear transformation processing on the feature fusion vector through the multi-layer perceptron of the preset multi-channel fusion model to obtain a weight optimization vector; Performing feature adjustment processing on the weight optimization vector through the long short-term memory artificial neural network of the preset multi-channel fusion model to obtain a feature adjustment vector; The feature adjustment vector is subjected to feature mapping processing through the linear layer of the preset multi-channel fusion model to obtain a load monitoring result.

4. The method according to claim 1, characterized in that: The method further comprises: Obtain load monitoring data from historical data; Based on the load category of the load monitoring data, the load monitoring data is labeled to obtain model training data; The initial multi-channel fusion model is subjected to model training processing through the model training data to obtain a preset multi-channel fusion model.

5. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: Determine the load type based on the load monitoring result, and determine the electrical equipment information corresponding to the non-intrusive load monitoring device; The power consumption profile of the monitoring target of the non-intrusive load monitoring device is processed through the electrical equipment information to obtain the power consumption profile data of the monitoring target.

6. The method according to claim 5, characterized in that The method further comprises: Summarize the power consumption profile data of different monitoring targets in the region to obtain regional power consumption monitoring results; Load scheduling is performed based on the regional electricity consumption monitoring results to generate a load scheduling plan for the region.

7. A non-intrusive load monitoring device, characterized in that: The device comprises: A data acquisition module is used to obtain various types of power characteristic data collected by non-intrusive load monitoring equipment; An importance classification module is used to perform feature importance analysis on the power feature data, classify the power feature data into different importance categories, and determine the importance category corresponding to each type of power feature data; A feature integration module, used for integrating power feature data of the same importance category into channel input features; The feature fusion processing module is used to input the input features of each channel into different channels of a preset multi-channel fusion model, and perform feature fusion processing on the channel input features of different channels through the preset multi-channel fusion model to obtain load monitoring results. The preset multi-channel fusion model is obtained by training the load monitoring data in the historical data as training samples.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images