Comprehensive energy metering working condition extraction method and system
By combining one-dimensional and two-dimensional convolutional network models of working conditions with classification networks, the low rate and low accuracy problems of traditional methods in complex scenarios are solved, efficient working condition feature extraction and classification are achieved, and the accuracy of comprehensive energy metering is improved.
Patent Information
- Application Number
- CN202111571652.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-21
- Publication Date
- 2025-09-16
- Estimated Expiration
- 2041-12-21
Smart Images
Figure CN114398924B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of energy metering, and in particular relates to a comprehensive energy metering working condition extraction method and system. Background Art
[0002] With the integration of renewable energy power generation systems and the expansion of power grid services, nonlinear and other loads in energy metering have led to increasingly complex and variable operating conditions. Traditional operating condition feature extraction methods are no longer able to meet the needs of these complex scenarios. Currently, traditional energy metering operating condition feature extraction in my country primarily relies on wavelet transforms, Fourier transforms, S transforms, Hilbert-Huang transforms, singular value decomposition, and other variants. These traditional methods expand the methods and approaches for operating condition feature extraction through fitting or algorithms, but suffer from issues such as slow extraction speed and low accuracy. Machine learning algorithms have significantly advanced the development of operating condition feature extraction. Artificial intelligence algorithms such as PCA-SVM, ELM, PNN, and H-ELM have all been applied to feature extraction, improving accuracy and efficiency. However, due to the inevitable duality of algorithm research, research in these related fields generally focuses on one aspect of signal processing, feature selection, or classification methods, and lacks comprehensive consideration of feature engineering. For example, efficient classification methods generally employ systematic models rather than explicitly extracting features. Traditional feature extraction methods fail to link the feature extraction process with the classification results, resulting in poor classification performance. Summary of the Invention
[0003] In order to solve the problems existing in the prior art, the present invention provides a comprehensive energy metering working condition extraction method, which can improve the working condition classification accuracy.
[0004] The technical problem to be solved by the present invention is achieved through the following technical solutions:
[0005] In a first aspect, a method for extracting a comprehensive energy metering working condition is provided, comprising:
[0006] Inputting the integrated energy metering working condition signal into the working condition one-dimensional convolutional network model to obtain the first eigenvector of the integrated energy metering;
[0007] The imaged integrated energy metering working condition signal is input into the working condition one-dimensional convolutional network model to obtain the second eigenvector of the integrated energy metering;
[0008] A dimensionality reduction feature vector of the operating condition for comprehensive energy metering is obtained based on the first eigenvector and the second eigenvector.
[0009] In combination with the first aspect, further, obtaining the operating condition dimensionality reduction feature vector for comprehensive energy metering based on the first eigenvector and the second eigenvector includes:
[0010] Combining the first eigenvector and the second eigenvector to generate a working condition dimensionality reduction feature vector of comprehensive energy metering; establishing a working condition classification network model, using the working condition dimensionality reduction feature vector of comprehensive energy metering as input of the working condition classification network model, and outputting the working condition classification feature vector;
[0011] A working condition feature extraction model is constructed by combining the working condition signal input one-dimensional convolutional network model, the working condition signal input one-dimensional convolutional network model and the working condition classification network model. The model is trained through the back propagation algorithm, and the comprehensive energy metering working condition signal is input into the trained working condition feature extraction model to obtain the final working condition dimensionality reduction feature vector, which is the comprehensive energy metering working condition feature to be extracted.
[0012] In combination with the first aspect, further, the working condition dimension reduction feature vector is obtained by formula (1):
[0013]
[0014] Among them, F1 represents the first eigenvector, F2 represents the second eigenvector, and F represents the working condition dimensionality reduction eigenvector.
[0015] In combination with the first aspect, further, the working condition one-dimensional convolutional network model includes a pooling layer, three convolutional layers and corresponding activation functions; the working condition one-dimensional convolutional network model structure is input layer-convolutional layer-pooling layer-convolutional layer-convolutional layer-output layer; each convolutional layer has 130 filters, the first convolutional layer contains 96 cores, the second convolutional layer contains 152 cores, the third convolutional layer contains 196 cores, and the pooling layer adopts the mean pooling method.
[0016] In combination with the first aspect, further, the working condition one-dimensional convolutional network model structure is expressed as follows by equations (2)-(5):
[0017]
[0018]
[0019]
[0020]
[0021] Among them, b1~b4 represent the bias of the 1st to 4th layer network, w1~w4 represent the convolution kernel of the 1st to 4th layer network, pool n×n Represents the maximum pooling layer function, n represents the pooling window size, and σ represents the ReLU function.
[0022] In combination with the first aspect, further, the visualization of the comprehensive energy metering working condition signal includes:
[0023] The comprehensive energy metering condition signal is expressed as an image matrix as shown in formula (6):
[0024]
[0025] Among them, x mn Represents the value of the signal image at the nth dimension of the mth dimension time axis of the time-varying parameter; the n-dimensional time axis of the signal image is compressed to m dimensions using numerical averaging to obtain the comprehensive energy metering operating condition matrix.
[0026] In combination with the first aspect, further, the process of obtaining the second eigenvector of comprehensive energy metering includes:
[0027] The comprehensive energy metering condition signal is convolved, and then the solution is input into the GoogleNet network to obtain the second eigenvector.
[0028] Combined with the first aspect, further, the working condition two-dimensional convolutional network model is expressed by formula (7):
[0029]
[0030] Among them, G represents the GoogleNet network solution.
[0031] In combination with the first aspect, further, the working condition classification network model includes a multi-layer perceptron and a softmax classifier, the multi-layer perceptron is a two-layer network, the softmax classifier is a one-layer network, and the output of the multi-layer perceptron serves as the input of the softmax classifier.
[0032] Combined with the first aspect, further, the network calculation of the multi-layer perceptron is shown as follows:
[0033] B 1 =σ(w′1·F+b′1) (8)
[0034] B 2 =σ(w′2·B 1 +b′2) (9)
[0035] Among them, B 1 、B 2 They represent the first and second hidden layers in the multilayer perceptron, w′1 and w′2 represent the weight coefficients in the first and second hidden layers, b′1 and b′2 represent the bias coefficients in the first and second hidden layers, F is the dimensionality reduction feature vector of the working condition of comprehensive energy metering, and σ represents the ReLU function.
[0036] In a second aspect, a comprehensive energy metering operating condition feature extraction system is provided, comprising:
[0037] A first eigenvector acquisition module inputs the integrated energy metering working condition signal into the working condition one-dimensional convolutional network model to obtain the first eigenvector of the integrated energy metering;
[0038] The second eigenvector acquisition module inputs the imaged integrated energy metering working condition signal into the working condition one-dimensional convolutional network model to obtain the second eigenvector of the integrated energy metering;
[0039] The operating condition dimensionality reduction feature vector acquisition module obtains the operating condition dimensionality reduction feature vector for comprehensive energy metering based on the first eigenvector and the second eigenvector.
[0040] Beneficial Effects of the Invention: This invention proposes a method for extracting operating condition features for comprehensive energy metering, achieving higher accuracy in operating condition classification. This method can specifically extract operating condition signal features and operating condition matrix features, and optimize feature data through a classification algorithm, thus possessing a stronger theoretical foundation. Furthermore, this method exhibits strong nonlinear capabilities, enabling effective feature extraction in a variety of operating scenarios. Its simple operation steps and fast computational speed suggest promising application prospects. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 is a flow chart of the present invention;
[0042] Figure 2 This is a flow chart of feature extraction of the one-dimensional convolutional network model for working conditions in the present invention;
[0043] Figure 3 This is a flow chart of feature extraction of the working condition two-dimensional convolutional network model in the present invention;
[0044] Figure 4 This is a training flow chart of the working condition feature extraction training model in the present invention;
[0045] Figure 5 The effect of a single working condition on the feature extraction method in the present invention;
[0046] Figure 6 Schematic diagram of the impact of complex abnormal working conditions on the feature extraction method in the present invention. DETAILED DESCRIPTION
[0047] In order to further illustrate the technical features and effects of the present invention, the present invention is further described below with reference to the accompanying drawings and specific implementation methods.
[0048] Example 1
[0049] like Figure 1-6 As shown, the present invention proposes a comprehensive energy metering working condition extraction method, comprising the following steps:
[0050] Step 1: Find the first eigenvector
[0051] A one-dimensional convolutional network model of working condition is constructed based on the integrated energy metering working condition signal. The input layer of the network is the integrated energy metering working condition signal, and the output layer of the network is the first eigenvector of the integrated energy metering, such as Figure 2 shown.
[0052] The working condition one-dimensional convolutional network model includes a pooling layer, three convolutional layers and corresponding activation functions; the working condition one-dimensional convolutional network model structure is input layer-convolutional layer-pooling layer-convolutional layer-convolutional layer-output layer; each convolutional layer has 130 filters, the first convolutional layer contains 96 cores, the second convolutional layer contains 152 cores, and the third convolutional layer contains 196 cores. The pooling layer adopts the mean pooling method.
[0053] The structure of the one-dimensional convolutional network model for working conditions is expressed by equations (1)-(4) as follows:
[0054]
[0055]
[0056]
[0057]
[0058] Among them, b1~b4 represent the bias of the 1st to 4th layer network, w1~w4 represent the convolution kernel of the 1st to 4th layer network, pool n×n Represents the maximum pooling layer function, n represents the pooling window size, and σ represents the ReLU function.
[0059] Step 2: Obtain the second eigenvector
[0060] The imaged integrated energy metering working condition signal is convoluted to obtain the integrated energy metering two-dimensional convolution signal, and a working condition two-dimensional network convolution model is constructed. The input layer of the network is the integrated energy metering two-dimensional convolution signal, and the network output layer is the second eigenvector of the integrated energy metering. The working condition two-dimensional convolution network model is used for feature extraction of the imaged working condition signal, such as Figure 3 As shown;
[0061] The imaged integrated energy metering working condition signal is convolved to obtain the integrated energy metering two-dimensional convolution signal, and a working condition two-dimensional network convolution model is constructed. The input layer of the network is the integrated energy metering two-dimensional convolution signal, and the network output layer is the second eigenvector of the integrated energy metering.
[0062] The comprehensive energy metering working condition signal is visualized as an image matrix as shown in formula (1):
[0063]
[0064] Among them, x mn Represents the value of the signal image at the nth dimension of the mth dimension time axis of the time-varying parameter; the n-dimensional time axis of the signal image is compressed to m dimensions using numerical averaging to obtain the comprehensive energy metering operating condition matrix.
[0065] The matrix obtained by visualizing the integrated energy metering working condition signal is convolved, and then the solution is input into the GoogleNet network to obtain the second eigenvector.
[0066] The working condition two-dimensional convolutional network model is expressed by formula (7):
[0067]
[0068] Among them, G represents the GoogleNet network solution.
[0069] Step 3: Obtain the reduced dimension feature vector of the working condition for comprehensive energy metering based on the first and second feature vectors
[0070] A working condition classification network model is established based on the working condition dimensionality reduction feature vector of the comprehensive energy measurement. The input layer of the neural network is the working condition dimensionality reduction feature vector of the comprehensive energy measurement, and the output layer is the working condition classification feature vector. The working condition classification network model is used for classification training of the working condition dimensionality reduction feature vector;
[0071] Combine the working condition one-dimensional convolutional network model, the working condition two-dimensional convolutional network model and the working condition classification network model to form a working condition feature extraction training model. The network weights in the model are continuously updated through the back propagation algorithm to obtain a better trained network. The training process is as follows: Figure 4 As shown;
[0072] The integrated energy metering operating condition signal that needs to be extracted is input into the input layer of the operating condition feature extraction training model to obtain the reduced-dimensional feature vector of the integrated energy metering operating condition, which is the integrated energy metering operating condition feature that needs to be extracted.
[0073] The working condition classification network model includes a multi-layer perceptron and a softmax classifier. The multi-layer perceptron is a two-layer network, and the softmax classifier is a one-layer network. The output of the multi-layer perceptron serves as the input of the softmax classifier.
[0074] The network calculation of the multi-layer perceptron is shown as follows:
[0075] B 1 =σ(w′1·F+b′1) (8)
[0076] B 2 =σ(w′2·B 1 +b′2) (9)
[0077] Among them, B 1 、B 2 They represent the first and second hidden layers in the multilayer perceptron, w′1 and w′2 represent the weight coefficients in the first and second hidden layers, b′1 and b′2 represent the bias coefficients in the first and second hidden layers, F is the dimensionality reduction feature vector of the working condition of comprehensive energy metering, and σ represents the ReLU function.
[0078] Take the output vector B of the multilayer perceptron 2 As the input of the softmax classifier, assuming that there are K types of working conditions to be classified, the normalized probability conversion method is shown in (10):
[0079]
[0080] Wherein, X represents the operating condition classification feature vector, and j=1…K represents the elements of the operating condition classification feature vector.
[0081] This implementation plan creates 13,000 pieces of comprehensive energy metering operating condition data in a simulation environment, including standard operating conditions, single abnormal operating conditions, and compound abnormal operating conditions, and adds random noise to simulate real scenarios.
[0082] Table 1 compares the effects of applying the comprehensive energy metering operating condition features extracted by different artificial intelligence algorithms in classification. It can be found that the method used in the present invention is superior to other algorithms in terms of sensitivity, accuracy and specificity.
[0083] Table 1
[0084]
[0085] Figure 5 This figure shows the classification quality of features extracted under a single operating condition for integrated energy metering. Overall, the quality of feature extraction under a single operating condition is relatively high, with features extracted from pure sine waves performing very well. Furthermore, it can be seen that the specificity performance of this framework for a single operating condition is quite high, but the accuracy value decreases due to the impact of sensitivity.
[0086] Figure 6 The classification quality of the extracted features under compound abnormal conditions is shown. It can be seen that the model has a high classification accuracy for compound abnormal conditions, and is also affected by the sensitivity performance, with the accuracy value decreasing.
[0087] Example 2
[0088] The present invention also provides a comprehensive energy metering working condition feature extraction system, comprising:
[0089] A first eigenvector acquisition module inputs the integrated energy metering working condition signal into the working condition one-dimensional convolutional network model to obtain the first eigenvector of the integrated energy metering;
[0090] The second eigenvector acquisition module inputs the imaged integrated energy metering working condition signal into the working condition one-dimensional convolutional network model to obtain the second eigenvector of the integrated energy metering;
[0091] The operating condition dimensionality reduction feature vector acquisition module obtains the operating condition dimensionality reduction feature vector for comprehensive energy metering based on the first eigenvector and the second eigenvector.
[0092] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.
[0093] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes 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 steps in the process. 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.
[0094] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work 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 The function specified in one or more boxes.
[0095] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.
Claims
1. A method for extracting characteristics of comprehensive energy metering working conditions, characterized in that: include: Inputting the integrated energy metering working condition signal into the working condition one-dimensional convolutional network model to obtain the first eigenvector of the integrated energy metering; The imaged integrated energy metering working condition signal is input into the working condition two-dimensional convolutional network model to obtain the second eigenvector of the integrated energy metering; generating a working condition classification feature vector for comprehensive energy metering according to the first feature vector and the second feature vector; Generating the working condition classification feature vector for comprehensive energy metering according to the first feature vector and the second feature vector includes: Combining the first eigenvector and the second eigenvector to generate a working condition dimensionality reduction feature vector of comprehensive energy metering; establishing a working condition classification network model, using the working condition dimensionality reduction feature vector of comprehensive energy metering as input of the working condition classification network model, and outputting the working condition classification feature vector; A working condition feature extraction model is constructed by combining a one-dimensional convolutional network model for working condition signal input, a two-dimensional convolutional network model for working condition signal input based on images, and a working condition classification network model. This model is trained using a backpropagation algorithm, and the integrated energy metering working condition signal is input into the trained working condition feature extraction model to obtain the final working condition classification feature vector, which is the integrated energy metering working condition feature to be extracted. The working condition one-dimensional convolutional network model includes a pooling layer, three convolutional layers and corresponding activation functions; The process of obtaining the second eigenvector of comprehensive energy measurement includes: The integrated energy metering condition signal is convolved, and the result is then input into the GoogleNet network to obtain the second eigenvector; The working condition classification network model includes a multi-layer perceptron and a softmax classifier, the multi-layer perceptron is a two-layer network, the softmax classifier is a one-layer network, and the output of the multi-layer perceptron serves as the input of the softmax classifier; The visualization of comprehensive energy metering working condition signals includes: The comprehensive energy metering condition signal is expressed as an image matrix as shown in formula (6): （6） in, Represents the time-varying parameter Dimensional Timeline The value of the signal image when the dimension is The dimensional time axis is compressed using numerical averaging to The comprehensive energy metering condition matrix is obtained.
2. A comprehensive energy metering working condition feature extraction method according to claim 1, characterized in that: The working condition dimension reduction feature vector is obtained by formula (1): （1） in, represents the first eigenvector, represents the second eigenvector, Represents the reduced dimension feature vector of the working condition.
3. The method for extracting operating condition characteristics of comprehensive energy metering according to claim 1, characterized in that: The one-dimensional convolutional network model structure of the working condition is input layer-convolution layer-pooling layer-convolution layer-convolution layer-output layer; each convolution layer has 130 filters, the first convolution layer contains 96 cores, the second convolution layer contains 152 cores, and the third convolution layer contains 196 cores. The pooling layer adopts the mean pooling method.
4. The method for extracting operating condition characteristics of comprehensive energy metering according to claim 1, characterized in that: The one-dimensional convolutional network model structure of the working condition is expressed as follows by equations (2)-(5): （2） （3） （4） （5） in, Indicates the The bias of the layer network, Indicates the Convolution kernel of layer network, pool n×n represents the maximum pooling layer function, represents the pooling window size, and σ represents the ReLU function.
5. The method for extracting operating condition characteristics of comprehensive energy metering according to claim 4, characterized in that: The working condition two-dimensional convolutional network model is expressed by formula (7): （7） in, Represents GoogleNet network solution.
6. A method for extracting operating condition characteristics of comprehensive energy metering according to claim 1 or 3, characterized in that: The network calculation of the multi-layer perceptron is shown as follows: （8） （9） in, 、 Respectively represent the first and second hidden layers in the multi-layer perceptron, 、 Represent the weight coefficients in the first and second hidden layers respectively, 、 Represent the bias coefficients in the first and second hidden layers respectively, is the reduced dimension feature vector of the working condition of comprehensive energy metering, and σ represents the ReLU function.
7. A comprehensive energy metering working condition feature extraction system, characterized in that: include: A first eigenvector acquisition module inputs the integrated energy metering working condition signal into the working condition one-dimensional convolutional network model to obtain the first eigenvector of the integrated energy metering; The second eigenvector acquisition module inputs the imaged integrated energy metering working condition signal into the working condition two-dimensional convolutional network model to obtain the second eigenvector of the integrated energy metering; A working condition dimensionality reduction feature vector acquisition module generates a working condition classification feature vector for comprehensive energy metering based on the first feature vector and the second feature vector; The step of generating the operating condition classification feature vector for comprehensive energy metering based on the first feature vector and the second feature vector includes: Combining the first eigenvector and the second eigenvector to generate a working condition dimensionality reduction feature vector of comprehensive energy metering; establishing a working condition classification network model, using the working condition dimensionality reduction feature vector of comprehensive energy metering as input of the working condition classification network model, and outputting the working condition classification feature vector; A working condition feature extraction model is constructed by combining the working condition signal input one-dimensional convolutional network model, the working condition signal input two-dimensional convolutional network model, and the working condition classification network model. The model is trained using the back-propagation algorithm, and the integrated energy metering working condition signal is input into the trained working condition feature extraction model to obtain the final working condition classification feature vector, which is the integrated energy metering working condition feature to be extracted. The working condition one-dimensional convolutional network model includes a pooling layer, three convolutional layers and corresponding activation functions; The process of obtaining the second eigenvector of comprehensive energy measurement includes: The integrated energy metering condition signal is convolved, and the result is then input into the GoogleNet network to obtain the second eigenvector; The working condition classification network model includes a multi-layer perceptron and a softmax classifier, the multi-layer perceptron is a two-layer network, the softmax classifier is a one-layer network, and the output of the multi-layer perceptron serves as the input of the softmax classifier; The visualization of comprehensive energy metering working condition signals includes: The comprehensive energy metering condition signal is expressed as an image matrix as shown in formula (6): （6） in, Represents the time-varying parameter Dimensional Timeline The value of the signal image when the dimension is The dimensional time axis is compressed using numerical averaging to The comprehensive energy metering condition matrix is obtained.