A distribution network automation equipment abnormal state detection method and device, electronic equipment and storage medium

Abstract

The application discloses a kind of distribution network automation equipment abnormal state detection method, device, electronic equipment and storage medium, the method comprises: obtaining the first time sequence abnormal signal of distribution network automation equipment;The first time sequence abnormal signal is decomposed, and the first feature map matrix for representing equipment working state is generated;The first time sequence abnormal signal and the first feature map matrix are input into multi-tensor convolution network model, so that multi-tensor convolution network model generates the abnormal state type of first time sequence abnormal signal. By implementing the application, the abnormal detection capability of distribution network automation equipment can be significantly improved, and the safe and efficient operation of distribution network is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of power distribution network security analysis technology, specifically to a method, device, electronic equipment, and storage medium for detecting abnormal states of power distribution network automation equipment. Background Technology

[0002] Distribution network automation equipment is a crucial component of modern power distribution networks. Through real-time monitoring and intelligent control of the power grid's operational status, it significantly improves the reliability and operational efficiency of the distribution system. However, due to complex operating environments, frequent load fluctuations, and long-term operation, distribution network automation equipment is prone to various abnormalities, such as equipment aging, excessive load, and communication failures. These abnormalities not only affect the stable operation of the distribution network but may also lead to partial power outages or equipment damage. Therefore, timely detection and identification of abnormal states in distribution network automation equipment is of paramount importance for ensuring the safe operation of the power system.

[0003] Existing anomaly detection technologies for distribution network automation equipment are typically based on preset rules or simple anomaly signal identification methods. These methods mainly rely on empirically derived thresholds to determine whether equipment malfunctions. For example, when current, voltage, or communication signal strength exceeds a set range, the system marks it as an abnormal state. However, these methods can only preliminarily identify the existence of anomalies and cannot further distinguish the specific type of anomaly. Due to the complex operating environment of distribution network automation equipment, different anomaly states, such as switch jamming, excessive equipment vibration, decreased communication signal strength, and aging electrical components, may produce similar anomaly signal characteristics under specific conditions. This makes it difficult to clearly diagnose the specific fault category of the equipment using only simple rules or signal thresholds. This not only limits the accuracy and efficiency of anomaly detection in distribution network automation equipment but may also lead to reduced grid reliability, increased maintenance costs, and increased safety hazards. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, and storage medium for detecting abnormal states of distribution network automation equipment. Implementing this invention can improve the accuracy and efficiency of abnormal state detection in distribution network automation equipment.

[0005] An embodiment of the present invention provides a method for detecting abnormal states of distribution network automation equipment, comprising:

[0006] Acquire the first timing anomaly signal from the distribution network automation equipment;

[0007] The first time-series anomaly signal is decomposed to generate a first feature map matrix used to characterize the working state of the equipment;

[0008] The first time-series anomalous signal and the first feature map matrix are input into a multi-tensor convolutional network model so that the multi-tensor convolutional network model generates the anomalous state type of the first time-series anomalous signal.

[0009] The training of the multi-tensor convolutional network model includes:

[0010] Acquire several second time-series anomalous signals and anomalous state labels corresponding to the second time-series anomalous signals for training; wherein, the anomalous state label is the anomalous state type corresponding to the second time-series anomalous signal;

[0011] The second time-series anomaly signal is decomposed to generate a second feature map matrix used to characterize the working state of the device;

[0012] A training set is constructed based on the second time-series anomalous signal, the second feature map matrix, and the anomalous state labels corresponding to the second time-series anomalous signal;

[0013] The training set is randomly divided into several batches of training samples according to a preset batch size;

[0014] Each batch of training samples is sequentially input into a multi-tensor convolutional network (MTL) model for training until a preset number of training iterations is reached. Each time the MTL model receives a batch of training samples, it outputs the corresponding abnormal state type. Based on the abnormal state type and its corresponding label, a loss function value is calculated. The MTL model is then updated based on the loss function value.

[0015] Furthermore, the first timing anomaly signal of the distribution network automation equipment is acquired, including:

[0016] Acquire the original timing anomaly signals from the distribution network automation equipment;

[0017] Calculate the autocorrelation function of the original time-series anomalous signal;

[0018] Analyzing the autocorrelation function reveals random noise in the original time-series anomalous signal;

[0019] Random noise is removed from the original time-series abnormal signal based on a preset global threshold, generating the first time-series abnormal signal for the distribution network automation equipment.

[0020] Furthermore, the step of decomposing the first time-series anomalous signal to generate a first feature map matrix for characterizing the device's operating state includes:

[0021] The first time-series anomalous signal is decomposed into several first fixed mode components by using the ensemble empirical mode decomposition method.

[0022] A first feature map matrix is ​​generated based on several first fixed mode components to characterize the working state of the device; wherein each element in the first feature map matrix is ​​the value of the first fixed mode component at a certain moment.

[0023] Furthermore, the loss function value is calculated using the following formula:

[0024]

[0025] in, w These are the weights of a multi-tensor convolutional network model; b These are the bias coefficients of a multi-tensor convolutional network model; x This is the second feature map matrix; This refers to the actual type of abnormal status output. n The number of training samples; y This is an abnormal status label.

[0026] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0027] An embodiment of the present invention provides an abnormal state detection device for distribution network automation equipment, comprising: a time-series abnormal signal acquisition module, a time-series abnormal signal decomposition module, an abnormal state type detection module, and a model training module.

[0028] The timing anomaly signal acquisition module is used to acquire the first timing anomaly signal of the distribution network automation equipment;

[0029] The timing anomaly signal decomposition module is used to decompose the first timing anomaly signal and generate a first feature map matrix to characterize the working state of the device.

[0030] The abnormal state type detection module is used to input the first time-series abnormal signal and the first feature map matrix into a multi-tensor convolutional network model so that the multi-tensor convolutional network model generates the abnormal state type of the first time-series abnormal signal.

[0031] The model training module is used to acquire several second time-series anomalous signals and corresponding anomalous state labels for training; wherein the anomalous state label is the anomalous state type corresponding to the second time-series anomalous signal; decompose the second time-series anomalous signal to generate a second feature map matrix for characterizing the working state of the device; construct a training set based on the second time-series anomalous signal, the second feature map matrix, and the corresponding anomalous state label; randomly divide the training set into several batches of training samples according to a preset batch size; sequentially input each batch of training samples into a multi-tensor convolutional network model to train the multi-tensor convolutional network model until a preset number of training iterations is reached; wherein, when the multi-tensor convolutional network model receives a batch of training samples, it outputs the anomalous state type corresponding to the training sample; calculate the loss function value based on the anomalous state type and the corresponding anomalous state label; and update the multi-tensor convolutional network model based on the loss function value.

[0032] Furthermore, the timing anomaly signal acquisition module includes: an original timing anomaly signal acquisition unit, an autocorrelation function calculation unit, a noise identification unit, and a first timing anomaly signal generation unit;

[0033] The original timing anomaly signal acquisition unit is used to acquire the original timing anomaly signal of the distribution network automation equipment;

[0034] The autocorrelation function calculation unit is used to calculate the autocorrelation function of the original time-series anomaly signal;

[0035] The noise identification unit is used to analyze the autocorrelation function and identify random noise in the original time-series anomalous signal;

[0036] The first timing anomaly signal generation unit is used to remove random noise from the original timing anomaly signal according to a preset global threshold, and generate the first timing anomaly signal for the distribution network automation equipment.

[0037] Furthermore, the time-series anomaly signal decomposition module includes: a multi-frequency decomposition unit and a first feature map matrix generation unit;

[0038] The multi-frequency decomposition unit is used to perform multi-frequency decomposition on the first time-series anomalous signal using the ensemble empirical mode decomposition method to generate several first fixed mode components.

[0039] The first feature map matrix generation unit is used to generate a first feature map matrix for characterizing the working state of the device based on a plurality of first fixed mode components; wherein each element in the first feature map matrix is ​​the value of the first fixed mode component at a certain moment.

[0040] Furthermore, the loss function value is calculated using the following formula:

[0041]

[0042] in, w These are the weights of a multi-tensor convolutional network model; b These are the bias coefficients of a multi-tensor convolutional network model; x This is the second feature map matrix; This refers to the actual type of abnormal status output. n The number of training samples; y This is an abnormal status label.

[0043] Based on the above method embodiments, the present invention provides corresponding electronic device embodiments.

[0044] An embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it can implement the abnormal state detection method for distribution network automation equipment described in any of the above-described method embodiments.

[0045] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments.

[0046] One embodiment of the present invention provides a storage medium storing a computer program thereon, which, when executed by a processor, can implement the abnormal state detection method for distribution network automation equipment as described in any of the above-described method embodiments.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] This invention provides a method, apparatus, electronic device, and storage medium for detecting abnormal states of distribution network automation equipment. The method acquires time-series abnormal signals from distribution network automation equipment and decomposes these signals to generate a series of feature map matrices to characterize the equipment's operating state, effectively extracting multi-dimensional information about the equipment's operating state. These time-series abnormal signals and feature map matrices are input into a multi-tensor convolutional network (MTL) model, enabling the MTL model to generate the abnormal state type. By jointly analyzing the equipment's time-series abnormal signals and their corresponding feature map matrices using the MTL model, the network can extract meaningful feature information from multiple dimensions and signal sources, effectively reducing information loss caused by single signals or feature dimensions, and accurately generating the abnormal state type based on the input time-series data and feature map matrices. During model training, training samples are constructed using historical time-series abnormal signals and their corresponding abnormal state labels. After multiple training iterations, the model can accurately identify the abnormal state type of the equipment. Implementing this invention significantly improves the abnormal detection capability of distribution network automation equipment, ensuring the safe and efficient operation of the distribution network. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating a method for detecting abnormal states of distribution network automation equipment according to an embodiment of the present invention.

[0050] Figure 2 This is a schematic diagram illustrating the training process of a multi-tensor convolutional network model according to an embodiment of the present invention.

[0051] Figure 3 This is a schematic diagram of the structure of an abnormal state detection device for distribution network automation equipment provided in an embodiment of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] like Figure 1 As shown, an embodiment of the present invention provides a method for detecting abnormal states of distribution network automation equipment, which includes at least the following steps:

[0054] Step S1: Obtain the first timing anomaly signal from the distribution network automation equipment;

[0055] Specifically, in a preferred embodiment, acquiring the first timing anomaly signal of the distribution network automation equipment includes:

[0056] Acquire the original timing anomaly signals from the distribution network automation equipment;

[0057] Calculate the autocorrelation function of the original time-series anomalous signal;

[0058] Analyzing the autocorrelation function reveals random noise in the original time-series anomalous signal;

[0059] Random noise is removed from the original time-series abnormal signal based on a preset global threshold, generating the first time-series abnormal signal for the distribution network automation equipment.

[0060] Understandably, this implementation step can effectively improve the quality and accuracy of abnormal signals from distribution network automation equipment. By acquiring the raw time-series abnormal signals from the distribution network automation equipment and calculating their autocorrelation function, the time dependence of the signals can be fully utilized to reveal their periodic characteristics and random noise. By analyzing the autocorrelation function, periodic signals and random noise can be accurately distinguished, thereby effectively removing noise from the signal. This denoising process, based on a preset global threshold, can flexibly handle different types of noise, thus retaining effective information related to the equipment status. Ultimately, the generated first time-series abnormal signal is cleaner, providing high-quality data input for subsequent feature extraction and anomaly detection.

[0061] Step S2: Decompose the first time-series abnormal signal to generate a first feature map matrix for characterizing the working state of the device;

[0062] Specifically, in a preferred embodiment, the decomposition of the first time-series anomalous signal to generate a first feature map matrix for characterizing the device's operating state includes:

[0063] The first time-series anomalous signal is decomposed into several first fixed mode components by using the ensemble empirical mode decomposition method.

[0064] A first feature map matrix is ​​generated based on several first fixed mode components to characterize the working state of the device; wherein each element in the first feature map matrix is ​​the value of the first fixed mode component at a certain moment.

[0065] In an optional embodiment, when decomposing the first time-series anomalous signal using the ensemble empirical mode decomposition (EEMD) method, multiple sets of different Gaussian white noise sequences are first superimposed on the first time-series anomalous signal to reduce mode aliasing and improve the stability of the decomposition. The amplitude of the Gaussian white noise is an important parameter for the decomposition and needs to be set reasonably to balance noise introduction and signal integrity. Next, empirical mode decomposition (EMD) is performed on each set of first time-series anomalous signals superimposed with noise. The specific process of EMD includes: determining the local extrema of the first time-series anomalous signal and generating upper and lower envelopes respectively by interpolation; then calculating the average of the two as a local trend term and subtracting it from the first time-series anomalous signal to obtain candidate first intrinsic mode components; the candidate first intrinsic mode components are screened multiple times until the definition conditions of the first intrinsic mode component are met, that is, the upper and lower envelopes are symmetrical about the zero point in the entire range of the first time-series anomalous signal, and the number of zero intersections is equal to or close to the number of extrema points, and finally a first intrinsic mode component is extracted. The above steps are repeated for the remaining first-series anomalous signals until the remaining terms become monotonic functions, completing the EMD decomposition of a set of first-series anomalous signals. EEMD eliminates the randomness introduced by Gaussian noise by averaging the EMD results of multiple sets of first-series anomalous signals, obtaining a set of first intrinsic mode components with stability and physical meaning. These first intrinsic mode components can comprehensively characterize the features of the first-series anomalous signals at different frequency components, providing crucial data support for subsequent feature extraction and equipment condition diagnosis.

[0066] Understandably, using the ensemble empirical mode decomposition (EEMD) method to decompose the first time-series anomalous signal can analyze the high and low frequency characteristics of the first time-series anomalous signal. For example, abnormal fluctuations in temperature and voltage are usually more obvious in low-frequency components, while high-frequency components may contain abnormal vibration information of the equipment. This method can more effectively distinguish the long-term trend, periodic changes and instantaneous anomalies of the equipment.

[0067] In a preferred embodiment, a first feature map matrix for characterizing the device's operating state is constructed based on these first fixed mode components. This first feature map matrix is ​​generated by concatenating all first intrinsic mode components in a time-series manner, where each column corresponds to one first intrinsic mode component, representing the evolution of the first time-series anomalous signal at a certain frequency component; each row represents the amplitude of all frequency components of the first time-series anomalous signal at a certain moment. Each element of the first feature map matrix represents the value of a certain first fixed mode component at a specific moment.

[0068] For example, taking the first time-series anomalous signal as an example, the N intrinsic mode components {IMF1(t), IMF1(t), ..., IMF1(t)} obtained by EEMD decomposition are...N (t)}, each IMF represents a part of the first time-series anomalous signal at different frequency components, which are concatenated into a two-dimensional matrix M, where each column represents an IMF component and each row represents the value of the signal at a certain time.

[0069]

[0070] Where each matrix element M i，j This indicates at time point t i At time t, the value of the first time-series anomalous signal on the j-th IMF component.

[0071] Understandably, this method of constructing feature map matrices can represent complex first-series anomaly signal features in a two-dimensional structured form, providing a clear feature description for subsequent analysis and processing. Furthermore, since each column in the matrix independently reflects changes in specific frequency components, it can better distinguish the operating status and potential anomalies of equipment across multiple frequency characteristics.

[0072] In a preferred embodiment, before performing multi-frequency decomposition on the first time-series anomalous signal using an ensemble empirical mode decomposition method to generate several first fixed mode components, the method further includes:

[0073] The hyperparameters such as the ensemble number, Gaussian white noise amplitude, and number of decomposed modes in the ensemble empirical mode decomposition (EEMD) method are treated as individuals in a population, and the cross-hatching algorithm is used to optimize the hyperparameters.

[0074] Step S3: Input the first temporal anomaly signal and the first feature map matrix into the multi-tensor convolutional network model so that the multi-tensor convolutional network model generates the anomaly state type of the first temporal anomaly signal;

[0075] like Figure 2 As shown, in a preferred embodiment of the present invention, the training of the multi-tensor convolutional network model includes:

[0076] Step S3.1: Obtain several second time-series anomalous signals for training and anomalous state labels corresponding to the second time-series anomalous signals; wherein, the anomalous state label is the anomalous state type corresponding to the second time-series anomalous signal;

[0077] Step S3.2: Decompose the second time-series anomaly signal to generate a second feature map matrix used to characterize the working state of the equipment;

[0078] Step S3.3: Construct a training set based on the second time-series anomalous signal, the second feature map matrix, and the anomalous state label corresponding to the second time-series anomalous signal;

[0079] Step S3.4: Randomly divide the training set into several batches of training samples according to the preset batch size;

[0080] Step S3.5: Input the training samples of each batch into the multi-tensor convolutional network model in sequence, and train the multi-tensor convolutional network model until the preset number of training times is reached; wherein, when the multi-tensor convolutional network model receives a batch of training samples, it outputs the abnormal state type corresponding to the training samples; according to the abnormal state type and the corresponding abnormal state label, the loss function value is calculated through the loss function; and the multi-tensor convolutional network model is updated according to the loss function value.

[0081] Optionally, the multi-tensor convolutional network model consists of a one-dimensional convolutional neural network and a two-dimensional convolutional neural network. In this study, the convolutional network mainly includes a four-layer structure: convolutional layers, pooling layers, fusion layers, and fully connected layers.

[0082] The convolutional layer performs one-dimensional and two-dimensional convolution operations using two convolutional kernels with different weights. Each kernel extracts features from the input first feature map matrix by moving a window. The process of feature extraction by the convolutional kernel is described as follows:

[0083] y 2d =a 2d (W 2d *H+b 2d )

[0084] y 1d =a 1d (W 1d *V+b 1d )

[0085] Where * represents the convolution operator; y 2d With y 1d To output two-dimensional and one-dimensional features after the convolution operation; W 2d A two-dimensional convolution kernel; W 1d b is a one-dimensional convolution kernel; 2d b represents the bias coefficients for the two-dimensional convolution operation. 1d is the bias coefficient of the one-dimensional convolution operation; H is the first feature map matrix; V is the first temporal anomaly signal.

[0086] The pooling layer is used to further remove redundant information, improving the robustness and generalization ability of the overall network. The pooling layer in the convolutional neural network of this study employs max pooling, specifically applying a single max pooling operation. Its mathematical description is as follows:

[0087] z i =max{z i-1 [mL,(m+1)L]}

[0088]

[0089] z i The feature matrix after max pooling; z i-1 is the feature matrix output by the previous convolutional layer; L is the window size of the pooling operation; S is the stride of the pooling operation; m is the number or index of the pooling window, used to gradually slide the pooling window, select a local region and calculate the maximum value.

[0090] The fusion layer is used to fuse one-dimensional and two-dimensional features after convolutional kernel pooling operations. The operation of the fusion layer is as follows:

[0091] F fused =Concat(z 1d , z 2d )

[0092] Among them, z 1d With z 2d The output of the pooling layers for the main channel and auxiliary channel is shown. Concat is the concatenation operation along the feature dimension.

[0093] The fully connected layer, serving as the last hidden layer in the convolutional neural network, is flattened into a one-dimensional feature vector as its input. Neurons in this layer are fully connected to all neurons in the previous layer, and the fused features are used as input to the classifier Softmax, outputting the probability of the abnormal state in the form of probabilistic encoding. Based on the probability of the abnormal state, the type of abnormal state corresponding to the first time-series signal is determined.

[0094] Specifically, the loss function value is calculated using the following formula:

[0095]

[0096] Where w represents the weights of the multi-tensor convolutional network model; b represents the bias coefficients of the multi-tensor convolutional network model; and x represents the second feature map matrix. y represents the type of abnormal state in the actual output; n represents the number of training samples; and y represents the label of the abnormal state.

[0097] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0098] like Figure 3 As shown, an embodiment of the present invention provides an abnormal state detection device for distribution network automation equipment, including: a time-series abnormal signal acquisition module, a time-series abnormal signal decomposition module, an abnormal state type detection module, and a model training module;

[0099] The timing anomaly signal acquisition module is used to acquire the first timing anomaly signal of the distribution network automation equipment;

[0100] The timing anomaly signal decomposition module is used to decompose the first timing anomaly signal and generate a first feature map matrix to characterize the working state of the device.

[0101] The abnormal state type detection module is used to input the first time-series abnormal signal and the first feature map matrix into a multi-tensor convolutional network model so that the multi-tensor convolutional network model generates the abnormal state type of the first time-series abnormal signal.

[0102] The model training module is used to acquire several second time-series anomalous signals and corresponding anomalous state labels for training; wherein the anomalous state label is the anomalous state type corresponding to the second time-series anomalous signal; decompose the second time-series anomalous signal to generate a second feature map matrix for characterizing the working state of the device; construct a training set based on the second time-series anomalous signal, the second feature map matrix, and the corresponding anomalous state label; randomly divide the training set into several batches of training samples according to a preset batch size; sequentially input each batch of training samples into a multi-tensor convolutional network model to train the multi-tensor convolutional network model until a preset number of training iterations is reached; wherein, when the multi-tensor convolutional network model receives a batch of training samples, it outputs the anomalous state type corresponding to the training sample; calculate the loss function value based on the anomalous state type and the corresponding anomalous state label; and update the multi-tensor convolutional network model based on the loss function value.

[0103] Specifically, the timing anomaly signal acquisition module includes: an original timing anomaly signal acquisition unit, an autocorrelation function calculation unit, a noise identification unit, and a first timing anomaly signal generation unit;

[0104] The original timing anomaly signal acquisition unit is used to acquire the original timing anomaly signal of the distribution network automation equipment;

[0105] The autocorrelation function calculation unit is used to calculate the autocorrelation function of the original time-series anomaly signal;

[0106] The noise identification unit is used to analyze the autocorrelation function and identify random noise in the original time-series anomalous signal;

[0107] The first timing anomaly signal generation unit is used to remove random noise from the original timing anomaly signal according to a preset global threshold, and generate the first timing anomaly signal for the distribution network automation equipment.

[0108] Specifically, the time-series anomaly signal decomposition module includes: a multi-frequency decomposition unit and a first feature map matrix generation unit;

[0109] The multi-frequency decomposition unit is used to perform multi-frequency decomposition on the first time-series anomalous signal using the ensemble empirical mode decomposition method to generate several first fixed mode components.

[0110] The first feature map matrix generation unit is used to generate a first feature map matrix for characterizing the working state of the device based on a plurality of first fixed mode components; wherein each element in the first feature map matrix is ​​the value of the first fixed mode component at a certain moment.

[0111] For example, the loss function value is calculated using the following formula:

[0112]

[0113] Where w represents the weights of the multi-tensor convolutional network model; b represents the bias coefficients of the multi-tensor convolutional network model; and x represents the second feature map matrix. y represents the type of abnormal state in the actual output; n represents the number of training samples; and y represents the label of the abnormal state.

[0114] It should be noted that the embodiments of the device described above correspond to the embodiments of the present invention described above, and can realize any of the abnormal state detection methods for distribution network automation equipment described above. Furthermore, the embodiments of the device described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Additionally, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without creative effort.

[0115] Based on the above-described method embodiments of the present invention, a corresponding embodiment of an electronic device is provided.

[0116] An embodiment of the present invention provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the abnormal state detection method for distribution network automation equipment according to any one of the present invention, or, when the processor executes the computer program, it implements the functions of each module in the above-described device embodiments.

[0117] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the terminal device.

[0118] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0119] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0120] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0121] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments;

[0122] Another embodiment of the present invention provides a storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the storage medium is located to execute any of the above-described distribution network automation equipment abnormality detection methods of the present invention.

[0123] The aforementioned storage medium is a computer-readable storage medium. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0124] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0125] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for detecting abnormal states of distribution network automation equipment, characterized in that, include: Acquire the original timing anomaly signals from the distribution network automation equipment; Calculate the autocorrelation function of the original time-series anomalous signal; Analyzing the autocorrelation function reveals random noise in the original time-series anomalous signal; Random noise is removed from the original time-series abnormal signal based on a preset global threshold, generating the first time-series abnormal signal for the distribution network automation equipment. The first time-series anomalous signal is decomposed into several first fixed mode components by using the ensemble empirical mode decomposition method. Based on several first fixed mode components, a first feature map matrix is ​​generated to characterize the working state of the device; wherein each element in the first feature map matrix is ​​the value of the first fixed mode component at a certain moment; The first time-series anomalous signal and the first feature map matrix are input into a multi-tensor convolutional network model so that the multi-tensor convolutional network model generates the anomalous state type of the first time-series anomalous signal. The training of the multi-tensor convolutional network model includes: Acquire several second time-series anomalous signals and anomalous state labels corresponding to the second time-series anomalous signals for training; wherein, the anomalous state label is the anomalous state type corresponding to the second time-series anomalous signal; The second time-series anomaly signal is decomposed to generate a second feature map matrix used to characterize the working state of the device; A training set is constructed based on the second time-series anomalous signal, the second feature map matrix, and the anomalous state labels corresponding to the second time-series anomalous signal; The training set is randomly divided into several batches of training samples according to a preset batch size; Each batch of training samples is sequentially input into a multi-tensor convolutional network (MTL) model for training until a preset number of training iterations is reached. Each time the MTL model receives a batch of training samples, it outputs the corresponding abnormal state type. Based on the abnormal state type and its corresponding label, a loss function value is calculated. The MTL model is then updated based on the loss function value.

2. The method for detecting abnormal states of distribution network automation equipment as described in claim 1, characterized in that, The loss function value is calculated using the following formula: in, These are the weights of a multi-tensor convolutional network model; These are the bias coefficients of a multi-tensor convolutional network model; This is the second feature map matrix; This refers to the actual type of abnormal status output. The number of training samples; This is an abnormal status label.

3. A device for detecting abnormal states of distribution network automation equipment, characterized in that, include: The module includes a time-series abnormal signal acquisition module, a time-series abnormal signal decomposition module, an abnormal state type detection module, and a model training module. The timing anomaly signal acquisition module is used to acquire the original timing anomaly signals of the distribution network automation equipment; Calculate the autocorrelation function of the original time-series abnormal signal; analyze the autocorrelation function to identify the random noise of the original time-series abnormal signal; remove the random noise of the original time-series abnormal signal according to a preset global threshold to generate the first time-series abnormal signal of the distribution network automation equipment. The time-series abnormal signal decomposition module is used to perform multi-frequency decomposition on the first time-series abnormal signal using a set empirical mode decomposition method to generate several first fixed mode components; and to generate a first feature map matrix to characterize the working state of the device based on the several first fixed mode components; wherein, each element in the first feature map matrix is ​​the value of the first fixed mode component at a certain moment. The abnormal state type detection module is used to input the first time-series abnormal signal and the first feature map matrix into a multi-tensor convolutional network model so that the multi-tensor convolutional network model generates the abnormal state type of the first time-series abnormal signal. The model training module is used to acquire several second time-series anomalous signals and corresponding anomalous state labels for training; wherein the anomalous state label is the anomalous state type corresponding to the second time-series anomalous signal; decompose the second time-series anomalous signal to generate a second feature map matrix for characterizing the working state of the device; construct a training set based on the second time-series anomalous signal, the second feature map matrix, and the corresponding anomalous state label; randomly divide the training set into several batches of training samples according to a preset batch size; sequentially input each batch of training samples into a multi-tensor convolutional network model to train the multi-tensor convolutional network model until a preset number of training iterations is reached; wherein, when the multi-tensor convolutional network model receives a batch of training samples, it outputs the anomalous state type corresponding to the training sample; calculate the loss function value based on the anomalous state type and the corresponding anomalous state label; and update the multi-tensor convolutional network model based on the loss function value.

4. The abnormal status detection device for distribution network automation equipment as described in claim 3, characterized in that, The loss function value is calculated using the following formula: in, These are the weights of a multi-tensor convolutional network model; These are the bias coefficients of a multi-tensor convolutional network model; This is the second feature map matrix; This refers to the actual type of abnormal status output. The number of training samples; This is an abnormal status label.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it can implement the abnormal state detection method for distribution network automation equipment as described in any one of claims 1 to 2.

6. A storage medium having a computer program stored thereon, characterized in that: When executed by a processor, the computer program can implement the abnormal state detection method for distribution network automation equipment as described in any one of claims 1 to 2.

Images