Network message analysis and diagnosis method and system for smart substation
Through multi-scale Fourier-wavelet transform and space-time convolution recursive neural network, combined with the improved hybrid Gaussian-Markov random field model, the problem of insufficient data feature capture in intelligent substations is solved, and high-accurate fault diagnosis is achieved and alarms improve operational reliability and maintenance efficiency.
Patent Information
- Application Number
- CN202410903007.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-08
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-07-08
AI Technical Summary
The network environment of smart substations is complex, and the existing technology cannot fully capture the global and local characteristics of data and the time-frequency characteristics, resulting in insufficient diagnostic accuracy and robustness, and cannot provide timely and accurate fault diagnosis and alarm information, affecting operational reliability. and maintenance efficiency.
The multi-scale Fourier-wavelet transform is used to extract time-frequency features, combined with space-time convolution recursive neural network and improved hybrid Gaussian-Markov random field model, capture the spatial and temporal features of time-frequency features, and optimize the fault diagnosis probability distribution. Generate a troubleshooting report.
It improves the accuracy and robustness of fault diagnosis, can fully understand the space-time sequence, provide timely and accurate fault diagnosis and alarm information, and improves the operational reliability and maintenance efficiency of smart substations.
Smart Images

Figure CN118445554B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power systems, and in particular to a network message analysis and diagnosis method and system for an intelligent substation. Background Art
[0002] With the development of smart grid, smart substation has gradually become an important part of the power system. Smart substation realizes efficient communication and collaborative operation between various devices through digital and networked means, improving the automation level and operation efficiency of the power grid. In smart substation, network messages are the key carriers for data exchange, status monitoring and control command transmission between devices. Through real-time collection and analysis of network messages, the operating status of the equipment can be understood in time, potential fault hazards can be discovered, and preventive measures can be taken to ensure the safe and stable operation of the power system.
[0003] However, the network environment of smart substations is complex, involving multiple communication protocols and data types. Traditional network message analysis methods are difficult to fully cover various data sources. At the same time, the amount of substation operation data is huge, and how to efficiently store and process this data is also an important challenge. In addition, the operating status and fault types of equipment in the substation are diverse, requiring network message analysis methods to have high accuracy and robustness, and be able to detect and diagnose faults in a timely and accurate manner.
[0004] There are at least the following technical problems in the existing technology: the existing technology is unable to fully capture the global and local characteristics of the data, resulting in the loss of feature information and reduced analysis effect; it is unable to simultaneously capture the spatial and temporal characteristics of the time-frequency characteristics, resulting in incomplete understanding of the space-time series; it is unable to effectively handle complex fault types and correlations, and the diagnostic accuracy and robustness are insufficient, and it is unable to provide timely and accurate fault diagnosis and alarm information, affecting the operational reliability and maintenance efficiency of the smart substation. Summary of the invention
[0005] The present invention provides a network message analysis and diagnosis method and system for a smart substation to solve the technical problems that the prior art cannot fully capture the global and local characteristics of data, resulting in the loss of characteristic information and reduced analysis effect; cannot simultaneously capture the spatial and temporal characteristics of time-frequency characteristics, resulting in incomplete understanding of space-time series; cannot effectively handle complex fault types and correlations, lacks diagnostic accuracy and robustness, cannot provide timely and accurate fault diagnosis and alarm information, and affects the operational reliability and maintenance efficiency of the smart substation.
[0006] The network message analysis and diagnosis method and system of the smart substation of the present invention specifically include the following technical solutions:
[0007] The network message analysis and diagnosis method of the smart substation includes the following steps:
[0008] S1. Collect message data in the substation network, denoise and normalize the message data; extract the time-frequency characteristics of the normalized data through multi-scale Fourier-wavelet transform, and use spatiotemporal convolutional recursive neural network to model the time-frequency characteristics to capture the spatial and temporal characteristics of the time-frequency characteristics;
[0009] S2. Input the spatial and temporal features into the improved mixed Gaussian-Markov random field model, optimize the fault diagnosis probability distribution, obtain the diagnosis results, and generate a fault diagnosis report.
[0010] Preferably, the S1 specifically includes:
[0011] An improved bilateral filter is used to denoise the message data based on spatial neighborhood and pixel value similarity.
[0012] Preferably, the S1 specifically includes:
[0013] Multiscale Fourier-wavelet transform captures global and local features by performing Fourier transform and wavelet transform at different scales simultaneously.
[0014] Preferably, the S1 specifically includes:
[0015] The spatiotemporal convolutional recurrent neural network consists of a convolutional layer, a pooling layer, a recurrent layer, and an output layer.
[0016] Preferably, the S1 specifically includes:
[0017] The spatiotemporal convolutional recurrent neural network uses convolutional layers to extract spatial features and recurrent layers to model temporal features.
[0018] Preferably, the S2 specifically includes:
[0019] The improved mixed Gaussian-Markov random field model first uses a Gaussian mixture model to preliminarily model the spatial and temporal characteristics, and then introduces spatial correlation through a Markov random field.
[0020] Preferably, the S2 specifically includes:
[0021] Markov random fields describe the dependencies between variables by defining potential functions on nodes and edges, and obtain the probability distribution of spatial and temporal characteristics.
[0022] Preferably, the S2 specifically includes:
[0023] Based on the probability distribution of spatial and temporal characteristics, an optimization objective function is constructed, and a regularization term including spatial smoothness is introduced. The optimal fault diagnosis probability distribution is obtained through iterative optimization.
[0024] The network message analysis and diagnosis system of the smart substation includes the following parts:
[0025] Data acquisition module, data storage module, data preprocessing module, feature extraction module, spatiotemporal sequence modeling module, hybrid fault diagnosis module, and report generation module;
[0026] The data acquisition module collects the message data in the substation network in real time and transmits the message data to the data storage module for storage; the data acquisition module is connected to the data storage module;
[0027] The data storage module is responsible for transmitting the stored message data to the data preprocessing module;
[0028] The data preprocessing module performs denoising and normalization on the stored message data. The denoising process uses an improved bilateral filter, and the normalization process normalizes the denoised data to The normalized data is transmitted to the feature extraction module;
[0029] The feature extraction module uses multi-scale Fourier-wavelet transform to extract features from the normalized data and transmits the extracted time-frequency features to the spatiotemporal sequence modeling module;
[0030] The spatiotemporal sequence modeling module uses a spatiotemporal convolutional recursive neural network to model the time-frequency features, captures the spatial and temporal features of the time-frequency features, outputs the spatiotemporal features, and transmits the spatiotemporal features to the hybrid fault diagnosis module;
[0031] The hybrid fault diagnosis module uses the Gaussian mixture model to preliminarily model the spatiotemporal features, introduces spatial correlation through the Markov random field, outputs the probability distribution of fault diagnosis, and transmits it to the report generation module;
[0032] The report generation module determines the fault type according to the probability distribution of the fault diagnosis, and selects the fault type with the highest probability as the diagnosis result; generates a fault diagnosis report according to the judgment result of the fault type, and transmits the fault diagnosis report to the data storage module for storage.
[0033] The beneficial effects of the technical solution of the present invention are:
[0034] 1. The improved bilateral filter is used for denoising, which can simultaneously consider the spatial neighborhood and pixel value similarity, smooth the noise and retain important edge information; combining the advantages of Fourier transform and wavelet transform, a multi-scale Fourier-wavelet transform is designed to extract features from the normalized data, which can simultaneously perform Fourier transform and wavelet transform at different scales to extract time-frequency features;
[0035] 2. Use spatiotemporal convolutional recursive neural network to model time-frequency features. It combines the advantages of convolutional neural network and recursive neural network, and can capture the spatial and temporal features of time-frequency features at the same time. It uses convolutional layer to extract spatial features and recursive layer to model temporal features, thus achieving a comprehensive understanding of spatiotemporal sequences.
[0036] 3. Through the improved mixed Gaussian-Markov random field model, the Gaussian mixture model is used to preliminarily model the output of the spatiotemporal convolutional recurrent neural network, and then the spatial correlation is introduced through the Markov random field to enhance the diagnostic effect. The Gaussian mixture model can effectively describe the probability distribution of the data, and the Markov random field can consider the spatial correlation, which comprehensively improves the accuracy and robustness of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 This is a structural diagram of the network message analysis and diagnosis system of the smart substation described in the present invention;
[0038] Figure 2 This is a flow chart of the network message analysis and diagnosis method for the smart substation described in the present invention. DETAILED DESCRIPTION
[0039] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0040] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0041] The specific scheme of the network message analysis and diagnosis method and system for the smart substation provided by the present invention is described in detail below with reference to the accompanying drawings.
[0042] See attached Figure 1 , which shows a structural diagram of a network message analysis and diagnosis system for a smart substation provided by an embodiment of the present invention, the system includes the following parts:
[0043] Data acquisition module, data storage module, data preprocessing module, feature extraction module, spatiotemporal sequence modeling module, hybrid fault diagnosis module, and report generation module;
[0044] The data acquisition module includes 4 pairs of optical serial ports, 2 pairs of network ports, and 2 optical B-code interfaces. It collects message data in the substation network in real time and transmits the message data to the data storage module for storage, ensuring the comprehensiveness and diversity of data collection and being able to cope with different data types and communication protocols. The data acquisition module is connected to the data storage module;
[0045] The data storage module, which consists of SSD memory, DDR4 memory and FLASH memory, ensures the integrity and security of data and realizes high-speed data storage and access. The data storage module is responsible for transmitting the stored data to the data preprocessing module for further processing;
[0046] The data preprocessing module performs denoising and normalization on the collected message data. The denoising process uses an improved bilateral filter. The normalization process normalizes the denoised data to the range of [0, 1] and transmits the normalized data to the feature extraction module.
[0047] The feature extraction module uses multi-scale Fourier-wavelet transform to extract features from the normalized data. It captures global and local features by performing Fourier transform and wavelet transform at different scales simultaneously, and transmits the extracted time-frequency features to the spatiotemporal sequence modeling module.
[0048] The spatiotemporal sequence modeling module uses a spatiotemporal convolutional recursive neural network to model the time-frequency features, captures the spatial and temporal features of the time-frequency features, outputs the spatiotemporal features, and transmits the spatiotemporal features to the hybrid fault diagnosis module;
[0049] The hybrid fault diagnosis module uses a Gaussian mixture model to preliminarily model the output of the spatiotemporal convolutional recurrent neural network, introduces spatial correlation through the Markov random field to enhance the diagnosis effect, and finally outputs the probability distribution of fault diagnosis and transmits it to the report generation module;
[0050] The report generation module determines the fault type according to the probability distribution of fault diagnosis and selects the fault type with the highest probability as the diagnosis result; based on the fault type judgment result, a detailed fault diagnosis report is generated, including information such as fault type, occurrence time, probability value, etc., and the fault diagnosis report is transmitted to the data storage module for storage for subsequent analysis and trend prediction.
[0051] See attached Figure 2 , which shows a flow chart of a network message analysis and diagnosis method for a smart substation provided by an embodiment of the present invention, the method comprising the following steps:
[0052] S1. Collect message data in the substation network, denoise and normalize the message data; extract the time-frequency characteristics of the normalized data through multi-scale Fourier-wavelet transform, and use spatiotemporal convolutional recursive neural network to model the time-frequency characteristics to capture the spatial and temporal characteristics of the time-frequency characteristics;
[0053] The data acquisition module collects message data in the substation network in real time. The data acquisition module includes 4 pairs of optical serial ports, 2 pairs of network ports, and 2 optical B-code interfaces, supporting multiple data access methods. Each interface is designed to cope with different data types and communication protocols to ensure the comprehensiveness and diversity of data acquisition; the collected data is stored in the data storage module, which includes SSD memory, DDR4 memory and FLASH memory to ensure the integrity and security of the data and achieve high-speed data storage and access; SSD memory provides large-capacity and high-speed read and write storage space, DDR4 memory is used for temporary storage and fast access, and FLASH memory provides long-term data preservation function, ensuring flexible access and efficient storage of data, solving the problems of inflexible data access and low storage efficiency.
[0054] The collected message data is denoised and normalized. The denoising process uses an improved bilateral filter to denoise the message data to retain edge information. The core idea of the bilateral filter is to consider both the spatial neighborhood and the pixel value similarity, which can smooth the noise and retain important edge information. The denoising formula is as follows:
[0055] ,
[0056] in, is the denoised data, For location The message data at For location The message data at yes The neighboring pixels of and are row coordinates, and are the column coordinates, and are the spatial domain and intensity domain standard deviations of the bilateral filter, respectively. The bilateral filter removes noise by calculating the weighted average between pixels while maintaining the sharpness of the image edges.
[0057] The denoised data is normalized so that the data value range is between [0, 1]. The normalization process eliminates the dimensional differences between different data, making the subsequent processing steps more robust and consistent. The normalization process formula is as follows:
[0058] ,
[0059] in, is the normalized data, and are the minimum and maximum values of the normalized data, respectively.
[0060] Combining the advantages of Fourier transform and wavelet transform, a multi-scale Fourier-wavelet transform is designed to extract the features of the normalized data and extract the time-frequency features of the normalized data; the design idea of the multi-scale Fourier-wavelet transform is to perform Fourier transform and wavelet transform at different scales simultaneously to capture the global and local features of the signal. The formula is as follows:
[0061] ,
[0062] in, is the Fourier-wavelet transform result, i.e., the time-frequency feature; is the mother wavelet function; Represents the mother wavelet function The complex conjugate of is the scale factor; It’s time; is the frequency, is the time-integrated variable; By combining Fourier transform and wavelet transform, the signal is analyzed in two dimensions, time and frequency, to extract richer feature information.
[0063] The spatiotemporal convolutional recurrent neural network is used to model the time-frequency features. It combines the advantages of convolutional neural networks and recurrent neural networks, and can simultaneously capture the spatial and temporal features of the time-frequency features. The convolutional layer is used to extract spatial features, and the recurrent layer is used to model temporal features, thereby achieving a comprehensive understanding of the spatiotemporal sequence. The spatiotemporal convolutional recurrent neural network includes the following steps and formulas:
[0064] Convolution layer: Perform convolution operation on the input time-frequency features to extract spatial features. The formula is as follows:
[0065] ,
[0066] in, For the Feature map after layer convolution; For the convolution kernels; Represents the convolution operation; is the activation function; is the number of convolution kernels; For the The bias term of the convolution kernel.
[0067] Pooling layer: The feature map after layer convolution is pooled to reduce the size and computational complexity of the feature map. The formula is as follows:
[0068] ,
[0069] in, For the Feature map after layer pooling; Indicates The convolutional layer outputs the local window The value inside.
[0070] Recursive layer: The features after layer pooling are recursively operated to capture the time dependency of the input features. The hidden state at the current moment is calculated through the current input and the hidden state at the previous moment to capture the time features. The formula is as follows:
[0071] ,
[0072] in, For the Layer in time Hidden state after recursion; is the recursive layer weight matrix; is the weight matrix of the hidden state of the recurrent layer; is the bias term; For the Layer in time Hidden state after recursion.
[0073] Output layer: Perform a full connection operation on the hidden state of the last layer to obtain the final spatiotemporal feature output. The formula is as follows:
[0074] ,
[0075] in, is the final output of the spatiotemporal convolutional recurrent neural network, i.e., spatial and temporal features; is the weight matrix of the output layer; is the last layer, i.e. Layer, in time The hidden state of is the bias term.
[0076] S2. Input the spatial and temporal features into the improved mixed Gaussian-Markov random field model, optimize the fault diagnosis probability distribution, obtain the diagnosis results, and generate a fault diagnosis report.
[0077] Final output of the spatiotemporal convolutional recurrent neural network The input is introduced into the improved mixed Gaussian-Markov random field model. The improved mixed Gaussian-Markov random field model first uses the Gaussian mixture model to calculate the final output of the spatiotemporal convolutional recurrent neural network. Preliminary modeling was performed, and then spatial correlation was introduced through Markov random fields to enhance the diagnostic effect.
[0078] Final output using the spatiotemporal convolutional recurrent neural network As input, a Gaussian mixture model is modeled, which assumes that spatial and temporal features are mixed by multiple Gaussian distributions, each of which is parameterized by a mean and covariance matrix. The formula is as follows:
[0079] ,
[0080] in, It is The mixing coefficient of Gaussian distributions represents the weight of each Gaussian distribution in spatial and temporal features, satisfying ; Indicates Gaussian distribution with mean , represents the center position of each Gaussian distribution; the covariance matrix is , representing the spread and shape of each Gaussian distribution.
[0081] The specific steps are as follows:
[0082] Initialization: Using the output of the spatiotemporal convolutional recurrent neural network As input, initialize the Gaussian mixture model. Set the initial mean , covariance matrix and the mixing coefficient .
[0083] The expectation maximization algorithm (EM algorithm) is implemented as follows:
[0084] Steps: Calculate the posterior probability that each data point belongs to each Gaussian distribution;
[0085]
[0086] in, is the posterior probability, indicating Data points Belong to The probability of a Gaussian distribution.
[0087] Steps: Update the parameters of the Gaussian mixture model;
[0088] ,
[0089] ,
[0090] ,
[0091] in, is the total number of data points, Indicates transpose.
[0092] After completing the Gaussian mixture model, the Markov random field is introduced to consider the spatial correlation. The Markov random field describes the dependency between variables by defining potential functions on nodes and edges. The formula is as follows:
[0093] ,
[0094] in, Represents the final output of the convolutional recurrent neural network at a given time and space The probability distribution at the time of , reflects the final output of the spatiotemporal convolutional recurrent neural network after being processed by the Gaussian mixture model and the Markov random field in the improved mixed Gaussian-Markov random field model. The overall probability distribution of is used to determine and diagnose the fault type in the smart substation; is a normalization constant that ensures that the sum of the probability distribution is 1; is the group potential function, describing the group Internal variables the interactions between them; represents a clique, which is a subset of multiple interconnected nodes in a Markov random field; is the field potential function, describing the node Upper variable potential energy; Represents the field set, which is a single set of nodes in the Markov random field.
[0095] Construct the objective function. To optimize the objective function, introduce a regularization term containing spatial smoothness, and solve the following objective function through iterative optimization:
[0096] ,
[0097] in, Indicated in Data points The probability distribution of is the regularization parameter, which controls the weight of the spatial smoothness term; is a set of adjacent node pairs, describing the spatial correlation between variables; Indicates The method for finding the optimal solution of the objective function is relatively mature and can be applied using existing technologies.
[0098] Specific solution process:
[0099] 1. Initialization parameters: set the initial mean , covariance matrix and the mixing coefficient and the initial potential function parameters of the Markov random field;
[0100] 2. Expectation-Maximization Algorithm ( algorithm):
[0101] a. Steps: Calculate the posterior probability of each data point belonging to each Gaussian distribution based on the current Gaussian mixture model parameters ;
[0102] b. Step: Update Gaussian mixture model parameters , , ;
[0103] 3. Introducing Markov random field:
[0104] a. Initialization: Define the potential function of the Markov random field and ;
[0105] b. Optimization: Use iterative optimization methods to solve the objective function and update each data point The value of
[0106] 4. Iterative process: Repeated execution algorithm and Markov random field optimization steps until the Gaussian mixture model parameters converge;
[0107] 5. Final output: The optimal fault diagnosis probability distribution is obtained through the optimized Gaussian mixture model and Markov random field .
[0108] According to the optimal fault diagnosis probability distribution , determine each time point The fault type with the highest probability is selected as the diagnosis result; based on the fault type identification result, a detailed fault diagnosis report is generated, including fault type, occurrence time, probability value and other information. For high-probability faults, alarm information is generated and corresponding processing suggestions are provided to guide operation and maintenance personnel to take action. The diagnosis results and related data records are stored for subsequent analysis and trend prediction.
[0109] In summary, the network message analysis and diagnosis method and system for smart substations have been completed.
[0110] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0111] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.
[0112] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.
Claims
1. A network message analysis and diagnosis method for a smart substation, characterized in that: The following steps are involved: S1. Collect message data in the substation network, use an improved bilateral filter, and perform denoising and normalization on the message data based on spatial neighborhood and pixel value similarity; extract the time-frequency characteristics of the normalized data through multi-scale Fourier-wavelet transform. Multi-scale Fourier-wavelet transform captures global and local features by performing Fourier transform and wavelet transform at different scales simultaneously; the formula is as follows: , in, is the Fourier-wavelet transform result, i.e., the time-frequency feature; is the normalized data; is the mother wavelet function; represents the mother wavelet function The complex conjugate of is the scale factor; It’s time; is the frequency; is the time-integrated variable; represents the imaginary part; The time-frequency features are modeled using a spatiotemporal convolutional recursive neural network. The spatiotemporal convolutional recursive neural network uses a convolutional layer to extract spatial features. The formula is as follows: , in, For the Feature map after layer convolution; For the convolution kernels; Represents the convolution operation; is the activation function; is the number of convolution kernels; For the The bias term of the convolution kernel; The temporal features are modeled using recursive layers, and the formula is as follows: , in, For the Layer in time Hidden state after recursion; is the recursive layer weight matrix; For the Feature map after layer pooling; is the weight matrix of the hidden state of the recurrent layer; is the bias term; For the Layer in time Hidden state after recursion; S2. Input the spatial and temporal features into the improved mixed Gaussian-Markov random field model. The improved mixed Gaussian-Markov random field model first uses the Gaussian mixture model to preliminarily model the spatial and temporal features, and then introduces spatial correlation through the Markov random field; the Markov random field describes the dependency between variables by defining the potential function on the nodes and edges. The formula is as follows: , in, Represents the final output of the convolutional recurrent neural network at a given time and space The probability distribution when ; is the final output of the spatiotemporal convolutional recurrent neural network, i.e., spatial and temporal features; is the normalization constant; is the group potential function; Indicates agglomeration; is the field potential function; represents the field set, which is a single node set in the Markov random field; Based on the probability distribution of spatial and temporal characteristics, an optimization objective function is constructed, a regularization term including spatial smoothness is introduced, and the following objective function is solved through iterative optimization: , in, Indicated in Data points The probability distribution of is the total number of data points; is the regularization parameter; is the set of adjacent node pairs; Indicates data points; The optimal fault diagnosis probability distribution is obtained; thereby optimizing the fault diagnosis probability distribution, obtaining the diagnosis result, and generating a fault diagnosis report.
2. The network message analysis and diagnosis method of the smart substation according to claim 1 is characterized in that: The S1 specifically includes: The spatiotemporal convolutional recurrent neural network consists of a convolutional layer, a pooling layer, a recurrent layer, and an output layer.
3. A network message analysis and diagnosis system for a smart substation, applied to the network message analysis and diagnosis method for a smart substation according to claim 1, characterized in that: Includes the following sections: Data acquisition module, data storage module, data preprocessing module, feature extraction module, spatiotemporal sequence modeling module, hybrid fault diagnosis module, and report generation module; The data acquisition module collects the message data in the substation network in real time and transmits the message data to the data storage module for storage; The data acquisition module is connected to the data storage module; The data storage module is responsible for transmitting the stored message data to the data preprocessing module; The data preprocessing module performs denoising and normalization on the stored message data. The denoising process uses an improved bilateral filter, and the normalization process normalizes the denoised data to The normalized data is transmitted to the feature extraction module; The feature extraction module uses multi-scale Fourier-wavelet transform to extract features from the normalized data and transmits the extracted time-frequency features to the spatiotemporal sequence modeling module; The spatiotemporal sequence modeling module uses a spatiotemporal convolutional recursive neural network to model the time-frequency features, captures the spatial and temporal features of the time-frequency features, outputs the spatiotemporal features, and transmits the spatiotemporal features to the hybrid fault diagnosis module; The hybrid fault diagnosis module uses the Gaussian mixture model to preliminarily model the spatiotemporal features, introduces spatial correlation through the Markov random field, outputs the probability distribution of fault diagnosis, and transmits it to the report generation module; The report generation module determines the fault type according to the probability distribution of the fault diagnosis, and selects the fault type with the highest probability as the diagnosis result; generates a fault diagnosis report according to the judgment result of the fault type, and transmits the fault diagnosis report to the data storage module for storage.
Citation Information
Patent Citations
Fault diagnosis method for rolling bearing based on Bayesian residual transform, singular value decomposition and Gaussian mixture hidden Markov model framework
CN107817106A
Low-voltage distribution system network topology identification method based on Markov random field
CN111313403A
Motor health diagnosis system based on Markov random field algorithm
CN117169716A
Aviation obstruction beacon remote monitoring method and system
CN117972580A
Equipment vibration fault diagnosis method based on artificial intelligence
CN118114186A
