CNN-DTW-based two-stage non-intrusive electric appliance category identification method, apparatus and device, medium and product

Abstract

The invention discloses a CNN-DTW-based two-stage non-intrusive electric appliance category identification method, device and equipment, a medium and a product, and relates to the field of electric appliance category identification, and the method comprises the steps: obtaining a power signal of a to-be-identified electric appliance; extracting a steady-state feature and a transient-state feature of the to-be-identified electric appliance from the power signal, and obtaining a transient-state active power waveform of the to-be-identified electric appliance based on the transient-state feature; inputting the steady-state characteristics of a to-be-identified electric appliance into the trained electric appliance category identification model to obtain probability distribution of various electric appliances and determine candidate electric appliances; extracting a standard transient active power waveform of the candidate electric appliance from a transient feature library according to the category of the candidate electric appliance; respectively calculating the DTW distance between the transient active power waveform of the electric appliance to be identified and each standard transient active power waveform; and taking the category of the candidate electric appliance with the minimum DTW distance as the category of the electric appliance to be identified. The precision of electric appliance identification is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of appliance category recognition, and in particular to a two-stage non-intrusive appliance category recognition method, device, equipment, medium and product based on CNN-DTW. BACKGROUND

[0002] Non-intrusive load monitoring (NILM) technology realizes the recognition of the running state and energy consumption of a single electrical device by analyzing the aggregate signals such as total current and voltage at the power inlet, and further identifies the category of the electrical device without the need to install a separate sensor on each electrical device, which has the advantages of low cost and convenient deployment. The existing non-intrusive load monitoring method has the following disadvantages: When relying solely on steady-state characteristics (such as power, harmonics, etc.), the recognition accuracy of electrical devices with similar structures and close steady-state characteristics (such as air conditioners and washing machines) is low; when only using transient characteristics (such as power fluctuations during the startup process), the recognition stability is insufficient due to the influence of power grid noise and measurement errors; Traditional machine learning methods (such as support vector machines and decision trees) have limited processing capacity for high-dimensional features, and deep learning methods such as single CNN (Convolutional Neural Network) models may lead to an increase in model complexity due to differences in feature dimensions if steady-state and transient characteristics are directly fused, and it is difficult to capture the temporal correlation of transient characteristics. SUMMARY

[0003] The purpose of the present application is to provide a two-stage non-intrusive appliance category recognition method, device, equipment, medium and product based on CNN-DTW, which can improve the accuracy of appliance recognition.

[0004] To achieve the above purpose, the present application provides the following solutions: In a first aspect, the present application provides a two-stage non-intrusive appliance category recognition method based on CNN-DTW, comprising: obtaining a power signal of an electrical device to be recognized; extracting steady-state characteristics and transient characteristics of the electrical device to be recognized from the power signal, and obtaining a transient active power waveform of the electrical device to be recognized based on the transient characteristics; inputting the steady-state characteristics of the electrical device to be recognized into a trained appliance category recognition model to obtain the probability distribution of each type of electrical device; wherein the appliance category recognition model comprises an attention mechanism network, a CNN network and a classification network; determining at least one candidate electrical device based on the probability distribution; extracting a standard transient active power waveform of the candidate electrical device from a transient characteristic library according to the category of the candidate electrical device; calculate DTW distances between the transient active power waveform of the to-be-identified electrical appliance and each standard transient active power waveform respectively; take the category of the candidate electrical appliance with the minimum DTW distance as the category of the to-be-identified electrical appliance.

[0005] In a second aspect, the present application provides a two-stage non-intrusive electrical appliance category identification device based on CNN-DTW, comprising: an acquisition module configured to acquire a power signal of a to-be-identified electrical appliance; an extraction module configured to extract steady-state features and transient features of the to-be-identified electrical appliance from the power signal, and obtain a transient active power waveform of the to-be-identified electrical appliance based on the transient features; an identification module configured to input the steady-state features of the to-be-identified electrical appliance into a trained electrical appliance category identification model to obtain probability distributions of electrical appliances of each category; wherein the electrical appliance category identification model comprises an attention mechanism network, a CNN network and a classification network; a candidate electrical appliance confirmation module configured to determine at least one candidate electrical appliance based on the probability distributions; a standard waveform extraction module configured to extract standard transient active power waveforms of the candidate electrical appliance from a transient feature library according to the category of the candidate electrical appliance; a distance calculation module configured to calculate DTW distances between the transient active power waveform of the to-be-identified electrical appliance and each standard transient active power waveform respectively; an identification result output module configured to take the category of the candidate electrical appliance with the minimum DTW distance as the category of the to-be-identified electrical appliance.

[0006] In a third aspect, the present application provides a computer device, comprising a memory, a processor, a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the steps of the two-stage non-intrusive electrical appliance category identification method based on CNN-DTW according to any one of the above embodiments.

[0007] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the steps of the two-stage non-intrusive electrical appliance category identification method based on CNN-DTW according to any one of the above embodiments.

[0008] In a fifth aspect, the present application provides a computer program product comprising a computer program, and the computer program is executed by a processor to implement the steps of the two-stage non-intrusive electrical appliance category identification method based on CNN-DTW according to any one of the above embodiments.

[0009] According to the specific embodiments provided by the present application, the following technical effects are disclosed: The application provides a two-stage non-intrusive appliance category recognition method and device based on CNN-DTW, which recognizes the appliance category by using the steady-state characteristics and transient-state characteristics of the appliance in stages. In the first stage, the appliance category recognition model is used to preliminarily classify the high-dimensional steady-state characteristics, the feature extraction capability of CNN and the reinforcing effect of the attention mechanism on the key characteristics are used to quickly reduce the types of candidate appliances, determine the candidate appliances and reduce the subsequent calculation amount. In the second stage, the DTW algorithm is used to accurately match the time sequence correlation of the transient active power waveform for the candidate appliances with similar steady-state characteristics. DTW is good at processing time sequence data with variable length and time offset, and can effectively distinguish the appliances with similar steady-state characteristics but large differences in starting process waveform, thereby improving the accuracy of appliance category recognition. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0011] Figure 1 An application environment diagram of a two-stage non-intrusive appliance category recognition method based on CNN-DTW in an embodiment of the present application.

[0012] Figure 2 A flowchart of a two-stage non-intrusive appliance category recognition method based on CNN-DTW provided by an embodiment of the present application Figure 1 .

[0013] Figure 3 A flowchart of a two-stage non-intrusive appliance category recognition method based on CNN-DTW provided by an embodiment of the present application Figure 2 .

[0014] Figure 4 A structural diagram of an appliance category recognition model provided by an embodiment of the present application.

[0015] Figure 5 A flowchart of transient active power waveform matching provided by an embodiment of the present application.

[0016] Figure 6 A functional module diagram of a two-stage non-intrusive appliance category recognition device based on CNN-DTW provided by another embodiment of the present application.

[0017] Figure 7 A structural diagram of a computer device provided by an embodiment of the present application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] To make the objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] The two-stage non-invasive appliance category recognition method based on CNN-DTW provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 101 communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be set up independently, integrated into server 102, or placed in the cloud or on other servers. Terminal 101 can send the power signal of the appliance to be identified to server 102. Server 102 receives the power signal, extracts the steady-state and transient features of the appliance from the power signal, and obtains the transient active power waveform of the appliance based on the transient features. The steady-state features of the appliance are input into a trained appliance category recognition model to obtain the probability distribution of each type of appliance. The appliance category recognition model includes an attention mechanism network, a CNN network, and a classification network. At least one candidate appliance is determined based on the probability distribution. The standard transient active power waveform of the candidate appliance is extracted from the transient feature library according to its category. The DTW distance between the transient active power waveform of the appliance to be identified and each standard transient active power waveform is calculated. The category of the candidate appliance with the smallest DTW distance is taken as the category of the appliance to be identified. Server 102 can feed back the obtained category of the appliance to be identified to terminal 101. In addition, in some embodiments, the two-stage non-intrusive appliance category identification method based on CNN-DTW can also be implemented by server 102 or terminal 101 alone. For example, terminal 101 can directly process the power signal of the appliance to be identified, or server 102 can process the power signal of the appliance to be identified from the data storage system.

[0021] The terminal 101 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 102 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0022] In one exemplary embodiment, such as Figure 2 and Figure 3 As shown, a two-stage non-invasive appliance category recognition method based on CNN-DTW is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 102 as an example, the explanation includes the following steps 201 to 207. Wherein: Step 201: Obtain the power signal of the appliance to be identified.

[0023] Step 202: Extract the steady-state and transient features of the electrical appliance to be identified from the power signal, and obtain the transient active power waveform of the electrical appliance to be identified based on the transient features.

[0024] Step 203: Input the steady-state features of the electrical appliance to be identified into the trained electrical appliance category recognition model to obtain the probability distribution of each type of electrical appliance; wherein, the electrical appliance category recognition model includes an attention mechanism network, a CNN network and a classification network.

[0025] Step 204: Determine at least one candidate electrical appliance based on the probability distribution.

[0026] Step 205: Extract the standard transient active power waveform of the candidate electrical appliance from the transient feature library according to the category of the candidate electrical appliance.

[0027] Step 206: Calculate the DTW distance between the transient active power waveform of the electrical appliance to be identified and each standard transient active power waveform.

[0028] Step 207: Select the category of the candidate appliance with the smallest DTW distance as the category of the appliance to be identified.

[0029] Implementing steps 201 to 207 above can improve the accuracy of electrical appliance category identification.

[0030] In one exemplary embodiment, step 201 specifically includes: The power signals of the electrical appliances to be identified in the target scene are collected. The power signals include voltage signals and current signals. They are converted into digital signals through signal conditioning circuits, and the sampling frequency is set to 5kHz~20kHz.

[0031] In this embodiment, common household appliances are selected as the targets to be identified, including but not limited to air conditioners, refrigerators, washing machines, televisions, microwave ovens, electric water heaters, table lamps, electric fans, etc.

[0032] In one exemplary embodiment, step 202 specifically includes: When the electrical appliance is operating stably, eight characteristic parameters are extracted as steady-state features. These steady-state features include: active power, RMS current, peak current factor, power factor, total current distortion rate, third harmonic component of current, fifth harmonic component of current, and ninth harmonic component of current. Specifically: 1. Active power (unit: W).

[0033] 2. Effective value of current (unit: A).

[0034] 3. Peak current factor (the ratio of peak current to effective current, without unit).

[0035] 4. Power factor (the ratio of active power to apparent power, without unit).

[0036] 5. Total current distortion (the ratio of total harmonic content to the effective value of the fundamental frequency, in %, calculated by FFT (Fast Fourier Transform)).

[0037] 6. Third harmonic component of current (unit: A, current amplitude corresponding to the third harmonic frequency after FFT transformation).

[0038] 7. 5th harmonic component of current (unit: A, current amplitude corresponding to the 5th harmonic frequency after FFT transformation).

[0039] 8. 9th harmonic component of current (unit: A, current amplitude corresponding to the 9th harmonic frequency after FFT transformation).

[0040] During the start-up process of the electrical appliance (starting from the moment of start-up), 30 active power values ​​are continuously collected as transient characteristics, and the collected transient characteristics are used to construct a transient active power waveform.

[0041] In an exemplary embodiment, step 203 specifically includes steps 2031-2033: Step 2031: Input the steady-state features of the electrical appliance to be identified into the attention mechanism network to enhance key features and obtain the enhanced key feature matrix.

[0042] Step 2032: Input the enhanced key feature matrix into the CNN network for feature extraction to obtain key features.

[0043] Step 2033: Input the key features into the classification network for category recognition and obtain the probability distribution of each appliance category.

[0044] like Figure 4 As shown, the appliance category recognition model includes an attention mechanism network, a CNN network, and a classification network.

[0045] 1) Attention mechanism network.

[0046] 1.1 Dimensional transformation: Convert the 5×8 two-dimensional feature matrix into an 8×5 channel-feature format (8 features as channel dimensions, 5 sets of data as feature dimensions).

[0047] 1.2 Global Average Pooling: Compresses 8×5 features into 8×1 channel descriptors (each channel takes the average of 5 sets of data).

[0048] 1.3 Fully connected layer 1:8→4 (reduction rate 0.5), ReLU activated.

[0049] 1.4 Fully connected layer 2: 4→8, Sigmoid activation, generating 8 channel weights (corresponding to 8 steady-state features).

[0050] 1.5 Weighted Calculation: Multiply the channel weights by the input 5×8 feature matrix according to the feature dimension (e.g., increase the feature weights of harmonic components), and output the weighted 5×8 feature matrix.

[0051] 2) A CNN network consists of an input layer, a convolutional layer, a batch normalization layer, and a ReLU activation function. The convolutional layer, batch normalization layer, and ReLU activation function are used as feature extraction modules to extract features.

[0052] 2.1. Convolutional layer 1: 3×3 convolution kernels are used, with a total of 16 kernels, a stride of 1, and padding of 1. The weighted 5×8 feature matrix is ​​convolved, and the output feature map size is 5×8×16.

[0053] 2.2. Batch Normalization Layer: The output of Convolutional Layer 1 is normalized (mean 0, variance 1), and nonlinearity is enhanced by the ReLU activation function.

[0054] 2.3. Convolutional layer 2: Uses 3×3 convolutional kernels, 32 kernels, stride 1, padding 1, output feature map size 5×8×32.

[0055] 2.4. Batch normalization layer + ReLU activation.

[0056] 3) The classification network includes a global average pooling layer, a fully connected layer, and a softmax layer.

[0057] 3.1 Global Average Pooling: Converts the 5×8×32 feature map into a 3-dimensional feature vector.

[0058] 3.2 Fully connected layer: 32→64, ReLU activated.

[0059] 3.3 The softmax layer outputs the probability distribution of each type of appliance (the sum is 1).

[0060] The training process for the appliance category recognition model includes: Extract steady-state and transient features for each type of electrical appliance. Steady-state feature extraction: Collect 5 sets of steady-state features continuously under stable operating conditions (each set is spaced at a certain interval), and standardize the steady-state features (normalize to the [0,1] interval) to construct a dataset containing electrical appliance category labels.

[0061] All steady-state features are normalized using min-max normalization and mapped to the [0,1] interval, as shown in the formula: in, The steady-state characteristics after normalization are... This is a steady-state characteristic. and These are the minimum and maximum values ​​of the corresponding steady-state characteristics, respectively.

[0062] Transient feature extraction: Record 30 consecutive active power values ​​after the appliance starts.

[0063] A training dataset is constructed based on the extracted steady-state features and corresponding labels. Each category of electrical appliances contains 100 samples, with 70 used as the training set and 30 as the validation set. Each sample contains a 5×8 steady-state feature matrix. In addition, for each category of electrical appliances in the dataset, its standard transient active power waveform is stored to construct a transient feature library.

[0064] The appliance category recognition model was trained using a training set. Batch size and a learning rate decay strategy were set, and the Adam optimizer was used. PyTorch's DataLoader was used to load the dataset, supporting batch processing and multithreading. Cross-entropy loss was used as the loss function. The steady-state feature matrix and appliance labels were input into the appliance category recognition model to obtain the appliance categories and the loss. The weights were updated based on the loss, and training was stopped after a preset number of iterations to obtain the trained appliance category recognition model. The model performance was evaluated using a validation set, and the NME (Normalized Mean Error) was calculated. The NME formula is as follows. ; Where d is the normalization factor, x i It is the predicted coordinate, y i These are the actual coordinates.

[0065] In an exemplary embodiment, the trained appliance category recognition model is exported as a pth format, a real-time recognition program is written in conjunction with the DTW algorithm, and deployed to a cloud platform. Specifically, this includes: using Python 3.9, building the appliance category recognition model through the Python framework, and loading the trained pth model file; using the Flask framework to build an API interface to receive steady-state and transient feature data uploaded by edge devices.

[0066] In an exemplary embodiment, the number of candidate electrical appliances is determined to be 2 based on the probability distribution. The DTW distance between the transient active power waveforms of these two candidate electrical appliances and their standard transient active power waveforms is calculated. The category of the candidate electrical appliance with the smallest DTW distance is selected as the category of the electrical appliance to be identified.

[0067] DTW Algorithm Implementation: Using the fastdtw library, the DTW algorithm matches transient active power waveforms as follows: Figure 5 As shown.

[0068] The edge acquisition device is activated to check the power signal access of the electrical appliance to be identified in real time and extract steady-state and transient features. The steady-state and transient feature information is then uploaded to the cloud platform to achieve the identification of the appliance category.

[0069] Based on the same inventive concept, this application also provides an apparatus for implementing the above-described two-stage non-invasive appliance category recognition method based on CNN-DTW. The solution provided by this apparatus is similar to the implementation described in the above method. Therefore, the specific limitations of one or more embodiments of the two-stage non-invasive appliance category recognition apparatus based on CNN-DTW provided below can be found in the limitations of the two-stage non-invasive appliance category recognition method based on CNN-DTW described above, and will not be repeated here.

[0070] In one exemplary embodiment, such as Figure 6 As shown, a two-stage non-invasive appliance category recognition device based on CNN-DTW is provided, comprising: The signal acquisition module 61 is used to acquire the power signal of the electrical appliance to be identified.

[0071] The feature extraction module 62 is used to extract the steady-state features and transient features of the electrical appliance to be identified from the power signal, and to obtain the transient active power waveform of the electrical appliance to be identified based on the transient features.

[0072] The recognition module 63 inputs the steady-state features of the electrical appliances to be recognized into the trained electrical appliance category recognition model to obtain the probability distribution of each type of electrical appliance; wherein, the electrical appliance category recognition model includes an attention mechanism network, a CNN network and a classification network.

[0073] Candidate appliance confirmation module 64 is used to determine at least one candidate appliance based on a probability distribution.

[0074] The standard waveform extraction module 65 is used to extract the standard transient active power waveform of the candidate electrical appliances from the transient feature library according to the category of the candidate electrical appliances.

[0075] The distance calculation module 66 is used to calculate the DTW distance between the transient active power waveform of the electrical appliance to be identified and each standard transient active power waveform.

[0076] The recognition result output module 67 is used to select the category of the candidate appliance with the smallest DTW distance as the category of the appliance to be identified.

[0077] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database contains the power signals of the electrical appliances to be identified. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a two-stage non-intrusive appliance category identification method based on CNN-DTW.

[0078] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0079] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0080] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0081] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0082] 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, data stored, data displayed, 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 the relevant data must comply with relevant regulations.

[0083] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile 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. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0084] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0085] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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 specification.

[0086] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A two-stage non-intrusive appliance category recognition method based on CNN-DTW, characterized in that, The method comprises: acquiring a power signal of an electric appliance to be identified; extracting steady-state features and transient-state features of the electric appliance to be identified from the power signal, and obtaining a transient-state active power waveform of the electric appliance to be identified based on the transient-state features; inputting the steady-state features of the electric appliance to be identified into a trained electric appliance category identification model to obtain probability distributions of various electric appliance categories; wherein the electric appliance category identification model comprises an attention mechanism network, a CNN network and a classification network; determining at least one candidate electric appliance based on the probability distributions; extracting a standard transient-state active power waveform of the candidate electric appliance from a transient-state feature library according to the category of the candidate electric appliance; calculating DTW distances between the transient-state active power waveform of the electric appliance to be identified and each standard transient-state active power waveform; taking the category of the candidate electric appliance with the minimum DTW distance as the category of the electric appliance to be identified.

2. The CNN-DTW based two-stage non-intrusive appliance class recognition method according to claim 1, characterized in that, The steady-state features comprise active power, current effective value, current peak value factor, power factor, current total distortion rate, current 3rd harmonic component, current 5th harmonic component and current 9th harmonic component.

3. The CNN-DTW based two-stage non-intrusive appliance classification method of claim 1, wherein, The transient-state features comprise a preset number of active power values continuously after the electric appliance is started.

4. The CNN-DTW based two-stage non-intrusive appliance classification method of claim 1, wherein, Inputting the steady-state features of the electric appliance to be identified into the trained electric appliance category identification model to obtain the probability distributions of various electric appliance categories specifically comprises: inputting the steady-state features of the electric appliance to be identified into the attention mechanism network for key feature enhancement to obtain an enhanced key feature matrix; inputting the enhanced key feature matrix into the CNN network for feature extraction to obtain key features; inputting the key features into the classification network for category identification to obtain the probability distributions of various electric appliance categories.

5. The CNN-DTW based two-stage non-intrusive appliance classification method of claim 1, wherein, The CNN network comprises an input layer, a convolution layer, a batch normalization layer and a ReLU activation function.

6. The CNN-DTW based two-stage non-intrusive appliance classification method of claim 1, wherein, The classification network comprises a global average pooling layer, a full connection layer and a softmax layer.

7. A two-stage non-intrusive appliance category recognition device based on CNN-DTW, characterized in that, The method comprises: a signal acquisition module configured to acquire a power signal of an electric appliance to be identified; a feature extraction module configured to extract steady-state features and transient-state features of the electric appliance to be identified from the power signal, and obtain a transient-state active power waveform of the electric appliance to be identified based on the transient-state features; an identification module configured to input the steady-state features of the electric appliance to be identified into a trained electric appliance category identification model to obtain probability distributions of various electric appliance categories; wherein the electric appliance category identification model comprises an attention mechanism network, a CNN network and a classification network; a candidate electric appliance confirmation module configured to determine at least one candidate electric appliance based on the probability distributions; a standard waveform extraction module configured to extract a standard transient-state active power waveform of the candidate electric appliance from a transient-state feature library according to the category of the candidate electric appliance; a distance calculation module configured to calculate DTW distances between the transient-state active power waveform of the electric appliance to be identified and each standard transient-state active power waveform; an identification result output module configured to take the category of the candidate electric appliance with the minimum DTW distance as the category of the electric appliance to be identified.

8. A computer device comprising: A memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the CNN-DTW-based two-stage non-intrusive electric appliance category identification method according to any one of claims 1-6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the CNN-DTW based two-stage non-intrusive appliance category recognition method according to any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, The computer program, when executed by a processor, implements the CNN-DTW based two-stage non-intrusive appliance category recognition method according to any one of claims 1-6.

Images