Non-intrusive load identification method and system based on stable transient feature fusion

Abstract

The invention discloses a non-intrusive load identification method and system based on stable transient feature fusion, and belongs to the technical field of non-intrusive load identification. The method comprises the following steps: fusing a non-intrusive load identification result based on steady state characteristic data and a non-intrusive identification result based on transient state characteristic data into composite characteristic data; training a convolutional neural network CNN and a long short-term memory recurrent neural network LSTM to obtain a non-intrusive load identification model; and obtaining target steady-state feature data and target transient feature data of a target professional transformer, and based on the non-intrusive load identification model, identifying the target steady-state feature data and the target transient feature data to obtain a non-intrusive load identification result fused with the target professional transformer. According to the non-intrusive load identification method, the accuracy of non-intrusive load identification is effectively improved by identifying the transient and steady state characteristic data.

Description

Technical Field

[0001] The present invention relates to the technical field of non-intrusive load identification, and more specifically, to a non-intrusive load identification method and system based on steady-state and transient feature fusion. Background Art

[0002] As an important part of power users, dedicated transformer users (power users who use professional transformers) have a high proportion of diversified power load characteristics, which have an important impact on the upgrade, operation and maintenance of power grid sources, networks, loads and storage. Real-time perception of user power loads and operating conditions is of great significance to users in ensuring power safety, improving power efficiency, and responding to the dual-carbon strategy. Traditional intrusive load identification is not suitable for large-scale promotion due to factors such as high cost and difficulty in transformation. On this basis, non-intrusive load identification technology has developed rapidly.

[0003] However, the existing non-intrusive load identification technology often has slow calculation speed and low identification accuracy, and cannot be put into practical application in many scenarios. It is urgent to propose a new non-intrusive load identification technology to achieve fast and accurate identification of various types and characteristics of non-intrusive loads. Summary of the invention

[0004] In view of the above problems, the present invention proposes a non-intrusive load identification method based on steady-state and transient feature fusion, comprising:

[0005] Obtain the steady-state characteristic data and transient characteristic data of the professional transformers used by the professional transformer users, and fuse the non-intrusive load identification results based on the steady-state characteristic data and the non-intrusive identification results based on the transient characteristic data into composite characteristic data;

[0006] In a parallel pool, a convolutional neural network (CNN) and a long short-term memory (LSTM) recursive neural network (LSTM) are trained according to the steady-state characteristic data, the transient characteristic data and the composite characteristic data by parallel processing to obtain a non-intrusive load identification model;

[0007] The target steady-state characteristic data and the target transient characteristic data of the target professional transformer are obtained, and based on the non-intrusive load identification model, the target steady-state characteristic data and the target transient characteristic data are identified to obtain the non-intrusive load identification result of the target professional transformer fusion.

[0008] Optionally, a non-intrusive load identification result based on steady-state characteristic data is obtained, including:

[0009] The VI trajectory image and the power characteristics in the steady-state feature data are fused to generate three-dimensional image steady-state feature data, and the three-dimensional image steady-state feature data is identified based on a convolutional neural network (CNN) to obtain a non-intrusive load identification result based on the steady-state feature data.

[0010] Optionally, non-intrusive identification results based on transient characteristic data are obtained, including:

[0011] The transient feature matrix of the transient feature data is constructed by Fisher principal component analysis, and the transient feature matrix is ​​identified based on a long short-term memory recursive neural network (LSTM) to obtain a non-intrusive identification result based on the transient feature data.

[0012] Optionally, in the parallel pool, a convolutional neural network CNN and a long short-term memory recursive neural network LSTM are trained according to the steady-state characteristic data, transient characteristic data and load characteristic data by parallel processing to obtain a non-intrusive load identification model, including:

[0013] In different parallel pools at the same time, feature identification is performed on steady-state feature data through a convolutional neural network CNN to train a steady-state feature identification model, and feature identification is performed on the transient feature data through a long short-term memory recursive neural network LSTM to train a transient feature identification model. In the different parallel pools, the identification results obtained by the convolutional neural network CNN and the long short-term memory recursive neural network LSTM are fused, and the fused data of the identification results are identified using the long short-term memory recursive neural network LSTM to train a composite feature identification model. The steady-state feature identification model, the transient feature identification model and the composite feature identification model are fused to obtain a non-intrusive load identification model.

[0014] On the other hand, the present invention also proposes a non-intrusive load identification system based on steady-state and transient feature fusion, comprising:

[0015] A fusion unit is used to obtain steady-state characteristic data and transient characteristic data of a professional transformer used by a professional transformer user, and fuse the non-intrusive load identification result based on the steady-state characteristic data and the non-intrusive identification result based on the transient characteristic data into composite characteristic data;

[0016] A modeling unit is used to train a convolutional neural network (CNN) and a long short-term memory (LSTM) recursive neural network (LSTM) in a parallel pool according to the steady-state characteristic data, transient characteristic data and composite characteristic data through parallel processing to obtain a non-intrusive load identification model;

[0017] The identification unit is used to obtain the target steady-state characteristic data and the target transient characteristic data of the target professional transformer, and identify the target steady-state characteristic data and the target transient characteristic data based on the non-intrusive load identification model to obtain the non-intrusive load identification result of the target professional transformer fusion.

[0018] Optionally, the fusion unit obtains a non-intrusive load identification result based on the steady-state characteristic data, including:

[0019] The VI trajectory image and the power characteristics in the steady-state feature data are fused to generate three-dimensional image steady-state feature data, and the three-dimensional image steady-state feature data is identified based on a convolutional neural network (CNN) to obtain a non-intrusive load identification result based on the steady-state feature data.

[0020] Optionally, the fusion unit obtains a non-intrusive identification result based on the transient characteristic data, including:

[0021] The transient feature matrix of the transient feature data is constructed by Fisher principal component analysis, and the transient feature matrix is ​​identified based on a long short-term memory recursive neural network (LSTM) to obtain a non-intrusive identification result based on the transient feature data.

[0022] Optionally, the modeling unit trains a convolutional neural network (CNN) and a long short-term memory (LSTM) recursive neural network (LSTM) in a parallel pool according to the steady-state characteristic data, transient characteristic data and load characteristic data by parallel processing to obtain a non-intrusive load identification model, including:

[0023] In different parallel pools at the same time, feature identification is performed on steady-state feature data through a convolutional neural network CNN to train a steady-state feature identification model, and feature identification is performed on the transient feature data through a long short-term memory recursive neural network LSTM to train a transient feature identification model. In the different parallel pools, the identification results obtained by the convolutional neural network CNN and the long short-term memory recursive neural network LSTM are fused, and the fused data of the identification results are identified using the long short-term memory recursive neural network LSTM to train a composite feature identification model. The steady-state feature identification model, the transient feature identification model and the composite feature identification model are fused to obtain a non-intrusive load identification model.

[0024] In yet another aspect, the present invention further provides a computing device, comprising: one or more processors;

[0025] a processor for executing one or more programs;

[0026] When the one or more programs are executed by the one or more processors, the above-described method is implemented.

[0027] In yet another aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed, the method described above is implemented.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] The present invention provides a non-intrusive load identification method based on the fusion of steady-state and transient characteristics, including: obtaining steady-state characteristic data and transient characteristic data of a professional transformer used by a professional transformer user, fusing the non-intrusive load identification result based on the steady-state characteristic data and the non-intrusive identification result based on the transient characteristic data into composite characteristic data; training a convolutional neural network CNN and a long short-term memory recursive neural network LSTM in a parallel pool in a parallel processing manner according to the steady-state characteristic data, transient characteristic data and composite characteristic data to obtain a non-intrusive load identification model; obtaining target steady-state characteristic data and target transient characteristic data of a target professional transformer, and identifying the target steady-state characteristic data and target transient characteristic data based on the non-intrusive load identification model to obtain the non-intrusive load identification result fused by the target professional transformer. The present invention effectively improves the accuracy of non-intrusive load identification by identifying the transient and steady-state characteristic data. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of the method of the present invention;

[0031] Figure 2 is a flow chart of an embodiment of the method of the present invention;

[0032] Figure 3 This is a load steady-state feature identification model diagram based on CNN in an embodiment of the method of the present invention;

[0033] Figure 4 This is a load transient feature identification model diagram based on LSTM in an embodiment of the method of the present invention;

[0034] Figure 5 This is a diagram of a steady-transient load identification model based on LSTM in an embodiment of the method of the present invention;

[0035] Figure 6 It is a structural diagram of the system of the present invention. DETAILED DESCRIPTION

[0036] Now, exemplary embodiments of the present invention are described with reference to the accompanying drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided to disclose the present invention in detail and completely and to fully convey the scope of the present invention to those skilled in the art. The terms used in the exemplary embodiments shown in the accompanying drawings are not intended to limit the present invention. In the accompanying drawings, the same units / elements are marked with the same reference numerals.

[0037] Unless otherwise specified, the terms (including technical terms) used herein have the commonly understood meanings to those skilled in the art. In addition, it is understood that the terms defined in commonly used dictionaries should be understood to have the same meanings as those in the context of the relevant fields, and should not be understood as idealized or overly formal meanings.

[0038] Embodiment 1:

[0039] The present invention proposes a non-intrusive load identification method based on steady-state and transient feature fusion. Figure 1 As shown, including:

[0040] Step 1: Obtain steady-state characteristic data and transient characteristic data of a professional transformer used by a professional transformer user, and fuse the non-intrusive load identification result based on the steady-state characteristic data and the non-intrusive identification result based on the transient characteristic data into composite characteristic data;

[0041] Step 2: In a parallel pool, a convolutional neural network (CNN) and a long short-term memory (LSTM) recursive neural network (LSTM) are trained according to the steady-state characteristic data, the transient characteristic data and the composite characteristic data by parallel processing to obtain a non-intrusive load identification model;

[0042] Step 3: Obtain target steady-state characteristic data and target transient characteristic data of the target professional transformer, and identify the target steady-state characteristic data and target transient characteristic data based on the non-intrusive load identification model to obtain a non-intrusive load identification result of the target professional transformer fusion.

[0043] Among them, obtaining non-intrusive load identification results based on steady-state characteristic data includes:

[0044] The VI trajectory image and the power characteristics in the steady-state feature data are fused to generate three-dimensional image steady-state feature data, and the three-dimensional image steady-state feature data is identified based on a convolutional neural network (CNN) to obtain a non-intrusive load identification result based on the steady-state feature data.

[0045] Among them, obtaining non-intrusive identification results based on transient characteristic data includes:

[0046] The transient feature matrix of the transient feature data is constructed by Fisher principal component analysis, and the transient feature matrix is ​​identified based on a long short-term memory recursive neural network (LSTM) to obtain a non-intrusive identification result based on the transient feature data.

[0047] Wherein, in the parallel pool, according to the steady-state characteristic data, transient characteristic data and load characteristic data, the convolutional neural network CNN and the long short-term memory recursive neural network LSTM are trained by parallel processing to obtain a non-intrusive load identification model, including:

[0048] In different parallel pools at the same time, feature identification is performed on steady-state feature data through a convolutional neural network CNN to train a steady-state feature identification model, and feature identification is performed on the transient feature data through a long short-term memory recursive neural network LSTM to train a transient feature identification model. In the different parallel pools, the identification results obtained by the convolutional neural network CNN and the long short-term memory recursive neural network LSTM are fused, and the fused data of the identification results are identified using the long short-term memory recursive neural network LSTM to train a composite feature identification model. The steady-state feature identification model, the transient feature identification model and the composite feature identification model are fused to obtain a non-intrusive load identification model.

[0049] The present invention will be further described below in conjunction with specific embodiments:

[0050] Implementation steps: Figure 2 As shown, including:

[0051] Step 1: Fuse the VI trajectory image and power features in the steady-state features to generate new three-dimensional image steady-state features, and use convolutional neural networks to perform non-intrusive load identification based on steady-state features.

[0052] Step 2: Construct the main transient characteristic matrix of the load based on Fisher principal component analysis method, and use the long short-term memory recursive neural network to perform non-intrusive load identification based on the main transient characteristics.

[0053] Step 3: The image recognition data of the steady-state feature and the numerical recognition data of the transient feature are fused into a composite feature, which is used as a new training feature to perform non-intrusive load identification using a long short-term memory recursive neural network.

[0054] Step 4: Improve the computational efficiency of the algorithm through parallel processing, thereby achieving rapid and accurate identification of multiple types and characteristics of non-intrusive loads.

[0055] The following is a further explanation of steps 1-4:

[0056] Step 1: While considering the VI trajectory characteristics, combined with the power characteristics, an important parameter of load characteristics, linear transformation is used to extract VIP three-dimensional steady-state features, reducing the complexity of feature training while ensuring recognition accuracy.

[0057] Step 2: Extract the main transient features based on the Fisher principal component analysis method. The Fisher score can be used to score the separability of the transient features. The higher the score, the better the separability. Then, the features with poor separability are eliminated to obtain transient features with better classification effect. The scoring function of the Fisher score can be defined as:

[0058]

[0059] In the formula: c represents the number of categories; n i represents the number of samples in the i-th category; u i and σ i They represent the mean and standard deviation of feature f in the i-th sample respectively; u represents the mean of feature f in all samples.

[0060] Step 3: First, the generated VIP three-dimensional image steady-state features are used to identify the steady-state feature images of non-intrusive loads using CNN (Convolutional Neural Networks). Then, the main transient feature matrix of the load obtained based on FPCA (Fisher principal component analysis) is constructed, and the transient feature numerical value of the non-intrusive load is identified using the long short-term memory recursive neural network LSTM (Long Short Term Memory).

[0061] Then, the steady-state feature image recognition data obtained based on CNN and the transient feature numerical recognition data obtained based on LSTM are fused into composite features.

[0062] Finally, this is used as a new training feature for non-intrusive load identification using LSTM.

[0063] Step 4: In order to improve the efficiency of load identification, the training process of multiple loads is processed in data parallel: In the initial stage of data training, the steady-state data of various loads are first extracted by CNN model in different parallel pools according to the load category, and the transient data are extracted based on LSTM network in different parallel pools. Then, the data of the two models of different categories of loads are merged in different parallel pools, and the LSTM model is used to perform load identification based on steady-transient data fusion. The data parallel structure reduces the training time of multiple types of loads, thereby realizing the rapid identification of multi-source loads.

[0064] Further, the step 1 comprises the following steps:

[0065] Step 1.1: Extract power features and VI trajectory features as shown in equations (2)-(6). Assume that there are p sampling points in one cycle and the three-dimensional feature consists of a spatial shape of size q×q×q.

[0066] Power characteristics:

[0067] The power characteristics mainly include active power P and reactive power Q, as shown in equations (2)-(3):

[0068]

[0069] Where: T represents the number of sampling points in one cycle; V(t) represents the input voltage value of the load at time t; I(t) represents the current value at time t.

[0070] VI track features:

[0071] The VI trajectory feature is a high-frequency feature. It is a trajectory curve with voltage and current as coordinate axes, reflecting the coupling relationship between current and voltage. It is used to describe the cycle direction, area, slope, etc. of the trajectory shape feature. As shown in equations (4)-(6):

[0072]

[0073] I event (k)=|I(t occ +k·Δt)-I(t pre +k·Δt)| (5)

[0074]

[0075] Where: k = 1, 2, ..., n; V event (k) is the voltage average matrix before and after the event, with a length of n; I event (k) is the current difference matrix before and after the event, with a length of n; t pre is the voltage steady-state zero-crossing time before the event; t occ is the voltage steady-state zero-crossing moment after the event; V p.u. (k) and I p.u. (k) represents V event (k) and I event (k) Normalized results.

[0076] V p.u. is the horizontal axis, I p.u. As the vertical coordinate, you can draw the VI trajectory curve with unified image format.

[0077] Step 1.2: Linearly transform the voltage, current and power values ​​during the steady state into integers between 0 and q. The conversion formulas for the current, voltage and power at each sampling point are shown in equations (7) to (9):

[0078]

[0079] Where: m = 1, 2, ..., p is the one-dimensional size of the spatial shape; I m 、V m and P m are the current, voltage and power values ​​of the mth sampling point of the original data respectively; i m 、v m and p m are the conversion current, voltage and power values ​​of the mth sampling point respectively; minI, minV and minP are the minimum values ​​of current, voltage and power in the steady-state cycle respectively; maxI, maxV and maxP are the maximum values ​​of current, voltage and power in the steady-state cycle; Is the round-down sign.

[0080] Step 1.3: Generate a spatial shape with all elements in the interval 0 to q and a size of q×q×q, select each sampling point of the steady-state period from the first cycle to the last cycle, and set i m 、v m 、p m The block elements at the corresponding positions are assigned to 1 to obtain features of size q×q×q.

[0081] The step 2 comprises the following steps:

[0082] Step 2.1: Extract transient features. This paper considers transient power features and transient current features, where transient power features include transient active power and transient reactive power.

[0083] 1) Transient power characteristics:

[0084] In the transient process of the state change of the electrical equipment, the instantaneous power values ​​of different types of loads vary greatly. Therefore, the transient active power change △P, transient reactive power change △Q, and transient active power maximum value P are selected. max , the maximum transient reactive power Q max 4 as transient power characteristics.

[0085] 2) Transient current characteristics:

[0086] The current waveform of the load turning on transient process has rich characteristics, and there are obvious differences in the transient current waveforms of different loads. Therefore, the maximum value of the load transient current I max As a non-intrusive load identification transient characteristics.

[0087] Step 2.2: Obtain the initial data matrix of transient characteristics. Assume that the original data has n groups of data, each group of data contains k transient characteristic variables X1, X2, ..., X k , then the original data matrix can be obtained as:

[0088]

[0089] Where: x nk Represents the sample represented by the kth feature in the nth group of data.

[0090] Step 2.3: Obtain the Fisher data matrix. According to the Fisher score results, extract the top p features with the highest scores from all features. The Fisher data matrix at this time is as follows:

[0091]

[0092] Where: x np It represents the sample represented by the pth feature in the nth group of data. Where p is usually k / 2.

[0093] Step 2.4: Principal component analysis.

[0094] 1) The Fisher data matrix X p Standardization:

[0095]

[0096] Where: i = 1, 2, ..., n, j = 1, 2, ..., p; With δ j are the sample mean and standard deviation respectively. The standardized data matrix can be expressed as X * .

[0097] 2) Establish the standardized variable correlation coefficient matrix R.

[0098]

[0099] 3) Calculate the eigenvalues ​​and eigenvectors of R.

[0100] From |R-λE|=0, calculate the p eigenvalues ​​of R, and use λ j (j=1, 2, ..., p) represents, where λ1≥λ2...≥λ p ≥0.

[0101] The corresponding eigenvector W j =[W 1j ,W 2j ,…,W pj ](j=1,2,…,p).

[0102] The principal component can be expressed as: j =X j * W j .

[0103] 4) Calculate the variance contribution rate β of each principal component i And the cumulative variance contribution rate β(i) can be expressed as:

[0104]

[0105] The principal components with cumulative variance contribution rate β(i)≤85% are extracted. The transient characteristics at this time are the characteristics that have the greatest impact on the load.

[0106] The step 3 comprises the following steps:

[0107] Step 3.1: The VIP three-dimensional steady-state features are trained and extracted through the CNN model. The convolutional neural network model used is as follows: Figure 3 As shown, the convolution extraction formula is shown in formula (9).

[0108] U i =f(w*X i,i +b) (9)

[0109] The full connection and activation function expressions of the fully connected layer and the output layer are shown in Equation (10) and Equation (11), respectively.

[0110]

[0111] Step 3.2: Load transient features are trained and extracted through the LSTM model. The single-class load transient feature identification process based on LSTM is as follows: Figure 4 As shown. Through Fisher principal component analysis, transient characteristics that can fully express load information are obtained, and the load characteristic matrix is ​​constructed considering the time series characteristics of the load. The behavior of the matrix is ​​the transient characteristics of the load, and the columns of the matrix are the characteristics of the same type of load at different times. After data preprocessing of the sample data, a digital data set with a value between [0,1] is obtained.

[0112] The feature vector corresponding to the sampled data is input into the LSTM network to construct the feature matrix, as shown in formula (12):

[0113]

[0114] Step 3.3: Combine the steady-state feature image recognition data obtained by CNN and the transient feature numerical recognition data obtained by LSTM into composite features, construct a steady-transient feature vector, and use LSTM to train and make decisions. If the feature vector has past historical data, the load type will be directly identified based on the previous load characteristics. If the extracted feature vector does not belong to historical data, retraining is required. The steady-transient load identification model based on LSTM is as follows: Figure 5 shown.

[0115] Embodiment 2:

[0116] The present invention also proposes a non-intrusive load identification system 200 based on steady-state and transient feature fusion, such as Figure 6 As shown, including:

[0117] The fusion unit 201 is used to obtain the steady-state characteristic data and transient characteristic data of the dedicated transformer used by the dedicated transformer user, and fuse the non-intrusive load identification result based on the steady-state characteristic data and the non-intrusive identification result based on the transient characteristic data into composite characteristic data;

[0118] A modeling unit 202 is used to train a convolutional neural network (CNN) and a long short-term memory (LSTM) recursive neural network (LSTM) in a parallel pool according to the steady-state characteristic data, transient characteristic data and composite characteristic data through parallel processing to obtain a non-intrusive load identification model;

[0119] The identification unit 203 is used to obtain the target steady-state characteristic data and the target transient characteristic data of the target professional transformer, and identify the target steady-state characteristic data and the target transient characteristic data based on the non-intrusive load identification model to obtain the non-intrusive load identification result of the target professional transformer fusion.

[0120] The fusion unit 201 obtains the non-intrusive load identification result based on the steady-state characteristic data, including:

[0121] The VI trajectory image and the power characteristics in the steady-state feature data are fused to generate three-dimensional image steady-state feature data, and the three-dimensional image steady-state feature data is identified based on a convolutional neural network (CNN) to obtain a non-intrusive load identification result based on the steady-state feature data.

[0122] The fusion unit 201 obtains a non-intrusive identification result based on transient feature data, including:

[0123] The transient feature matrix of the transient feature data is constructed by Fisher principal component analysis, and the transient feature matrix is ​​identified based on a long short-term memory recursive neural network (LSTM) to obtain a non-intrusive identification result based on the transient feature data.

[0124] The modeling unit 203 trains the convolutional neural network CNN and the long short-term memory recursive neural network LSTM in a parallel pool according to the steady-state characteristic data, transient characteristic data and load characteristic data through parallel processing to obtain a non-intrusive load identification model, including:

[0125] In different parallel pools at the same time, feature identification is performed on steady-state feature data through a convolutional neural network CNN to train a steady-state feature identification model, and feature identification is performed on the transient feature data through a long short-term memory recursive neural network LSTM to train a transient feature identification model. In the different parallel pools, the identification results obtained by the convolutional neural network CNN and the long short-term memory recursive neural network LSTM are fused, and the fused data of the identification results are identified using the long short-term memory recursive neural network LSTM to train a composite feature identification model. The steady-state feature identification model, the transient feature identification model and the composite feature identification model are fused to obtain a non-intrusive load identification model.

[0126] The present invention effectively improves the accuracy of non-intrusive load identification by identifying transient steady-state characteristic data.

[0127] Embodiment 3:

[0128] Based on the same inventive concept, the present invention also provides a computer device, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding functions, so as to implement the steps of the method in the above embodiment.

[0129] Embodiment 4:

[0130] Based on the same inventive concept, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It is understandable that the computer-readable storage medium here can include both a built-in storage medium in a computer device and an extended storage medium supported by the computer device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by a processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the steps of the method in the above embodiment.

[0131] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0132] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0133] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0134] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0135] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0136] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A non-intrusive load identification method based on steady-state and transient feature fusion, characterized in that: include: Obtain the steady-state characteristic data and transient characteristic data of the professional transformers used by the professional transformer users, and fuse the non-intrusive load identification results based on the steady-state characteristic data and the non-intrusive identification results based on the transient characteristic data into composite characteristic data; In a parallel pool, a convolutional neural network (CNN) and a long short-term memory (LSTM) recursive neural network (LSTM) are trained according to the steady-state characteristic data, the transient characteristic data and the composite characteristic data by parallel processing to obtain a non-intrusive load identification model; The target steady-state characteristic data and the target transient characteristic data of the target professional transformer are obtained, and based on the non-intrusive load identification model, the target steady-state characteristic data and the target transient characteristic data are identified to obtain the non-intrusive load identification result of the target professional transformer fusion.

2. The non-intrusive load identification method according to claim 1, characterized in that: Obtain non-intrusive load identification results based on steady-state characteristic data, including: The VI trajectory image and the power characteristics in the steady-state feature data are fused to generate three-dimensional image steady-state feature data, and the three-dimensional image steady-state feature data is identified based on a convolutional neural network (CNN) to obtain a non-intrusive load identification result based on the steady-state feature data.

3. The non-intrusive load identification method according to claim 1, characterized in that: Obtain non-intrusive identification results based on transient characteristic data, including: The transient feature matrix of the transient feature data is constructed by Fisher principal component analysis, and the transient feature matrix is ​​identified based on a long short-term memory recursive neural network (LSTM) to obtain a non-intrusive identification result based on the transient feature data.

4. The non-intrusive load identification method according to claim 1, characterized in that: The parallel pool is used to train the convolutional neural network CNN and the long short-term memory recursive neural network LSTM according to the steady-state characteristic data, transient characteristic data and load characteristic data in a parallel processing manner to obtain a non-intrusive load identification model, including: In different parallel pools at the same time, feature identification is performed on steady-state feature data through a convolutional neural network CNN to train a steady-state feature identification model, and feature identification is performed on the transient feature data through a long short-term memory recursive neural network LSTM to train a transient feature identification model. In the different parallel pools, the identification results obtained by the convolutional neural network CNN and the long short-term memory recursive neural network LSTM are fused, and the fused data of the identification results are identified using the long short-term memory recursive neural network LSTM to train a composite feature identification model. The steady-state feature identification model, the transient feature identification model and the composite feature identification model are fused to obtain a non-intrusive load identification model.

5. A non-intrusive load identification system based on steady-state and transient feature fusion, characterized in that: include: A fusion unit is used to obtain steady-state characteristic data and transient characteristic data of a professional transformer used by a professional transformer user, and fuse the non-intrusive load identification result based on the steady-state characteristic data and the non-intrusive identification result based on the transient characteristic data into composite characteristic data; A modeling unit is used to train a convolutional neural network (CNN) and a long short-term memory (LSTM) recursive neural network (LSTM) in a parallel pool according to the steady-state characteristic data, transient characteristic data and composite characteristic data through parallel processing to obtain a non-intrusive load identification model; The identification unit is used to obtain the target steady-state characteristic data and the target transient characteristic data of the target professional transformer, and identify the target steady-state characteristic data and the target transient characteristic data based on the non-intrusive load identification model to obtain the non-intrusive load identification result of the target professional transformer fusion.

6. The non-intrusive load identification system according to claim 5, characterized in that: The fusion unit obtains a non-intrusive load identification result based on the steady-state characteristic data, including: The VI trajectory image and the power characteristics in the steady-state feature data are fused to generate three-dimensional image steady-state feature data, and the three-dimensional image steady-state feature data is identified based on a convolutional neural network (CNN) to obtain a non-intrusive load identification result based on the steady-state feature data.

7. The non-intrusive load identification system according to claim 5, characterized in that: The fusion unit obtains a non-intrusive identification result based on transient feature data, including: The transient feature matrix of the transient feature data is constructed by Fisher principal component analysis, and the transient feature matrix is ​​identified based on a long short-term memory recursive neural network (LSTM) to obtain a non-intrusive identification result based on the transient feature data.

8. The non-intrusive load identification system according to claim 5, characterized in that: The modeling unit trains the convolutional neural network (CNN) and the long short-term memory (LSTM) recursive neural network (LSTM) in a parallel pool according to the steady-state characteristic data, transient characteristic data and load characteristic data through parallel processing to obtain a non-intrusive load identification model, including: In different parallel pools at the same time, feature identification is performed on steady-state feature data through a convolutional neural network CNN to train a steady-state feature identification model, and feature identification is performed on the transient feature data through a long short-term memory recursive neural network LSTM to train a transient feature identification model. In the different parallel pools, the identification results obtained by the convolutional neural network CNN and the long short-term memory recursive neural network LSTM are fused, and the fused data of the identification results are identified using the long short-term memory recursive neural network LSTM to train a composite feature identification model. The steady-state feature identification model, the transient feature identification model and the composite feature identification model are fused to obtain a non-intrusive load identification model.

9. A computer device, characterized in that: include: one or more processors; a processor for executing one or more programs; When the one or more programs are executed by the one or more processors, the method according to any one of claims 1 to 4 is implemented.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed, the method according to any one of claims 1 to 4 is implemented.

Images