Low-voltage intelligent switch cabinet fault diagnosis method and system based on deep learning

Abstract

The present invention discloses a method and system for fault diagnosis of a low-voltage intelligent switch cabinet based on deep learning, the method comprising: obtaining first monitoring data and second monitoring data of each monitoring point of the low-voltage intelligent switch cabinet; extracting global features from the first monitoring data, fusing the first monitoring data and the correlation relationship data between the monitoring points, and obtaining local features; inputting the global features and the local features into a first fault diagnosis model to obtain a first fault diagnosis result; when the first fault diagnosis result is an unrecognizable fault, extracting auxiliary local features from the second monitoring data, inputting the global features, the auxiliary local features and the local features into a second fault diagnosis model to obtain a second fault diagnosis result. The present invention can effectively improve the accuracy and efficiency of fault diagnosis of low-voltage intelligent switch cabinets, and enhance the stability and safety of power systems.

Description

Technical Field

[0001] The present invention relates to the technical field of switch cabinets, and in particular to a low-voltage intelligent switch cabinet fault diagnosis method and system based on deep learning. Background Art

[0002] As an important part of the power system, the stable operation of low-voltage intelligent switchgear is crucial to the reliability of the entire power system. However, due to the complex and changeable operating environment, the switchgear is prone to various faults. If they cannot be discovered and handled in time, they will have a serious impact on the stable operation of the power system.

[0003] Traditional fault diagnosis methods are usually based on single sensor data or simple threshold judgments, which often do not work well when dealing with complex faults. In recent years, the development of deep learning technology has provided new ideas for fault diagnosis. Through deep learning models, features in data can be automatically extracted, and faults can be classified and identified.

[0004] However, existing fault diagnosis methods based on deep learning often only focus on global features or local features, ignoring the correlation between the two, resulting in low accuracy in complex fault diagnosis. In addition, all multimodal monitoring data are fused, and the amount of fused data is too large, resulting in the inability to quickly judge some basic faults. Summary of the invention

[0005] In response to the defects in the prior art, the present invention provides a low-voltage intelligent switch cabinet fault diagnosis method and system based on deep learning. The method and system achieve rapid identification of simple faults and accurate identification of complex faults by integrating local features of different monitoring points and global features of the entire switch cabinet.

[0006] On the one hand, the present invention provides a low-voltage intelligent switch cabinet fault diagnosis method based on deep learning, the method comprising:

[0007] Acquire first monitoring data and second monitoring data of each monitoring point of the low-voltage intelligent switch cabinet;

[0008] Extracting global features from the first monitoring data, fusing the first monitoring data and the association relationship data between monitoring points, and obtaining local features;

[0009] Inputting the global feature and the local feature into a first fault diagnosis model to obtain a first fault diagnosis result;

[0010] When the first fault diagnosis result is an unrecognizable fault, auxiliary local features are extracted from the second monitoring data, and the global features, auxiliary local features and local features are input into a second fault diagnosis model to obtain a second fault diagnosis result.

[0011] Preferably, the first monitoring data includes time-stamped temperature and humidity data, current data, voltage data, sound data and ambient temperature and humidity data of key components at the monitoring point; the second monitoring data includes time-stamped temperature and humidity data, current data, voltage data and sound data of other components at the monitoring point.

[0012] Preferably, the association relationship data includes position data and weight data under different loads.

[0013] Preferably, extracting global features from the first monitoring data includes: normalizing and denoising the first monitoring data.

[0014] Preferably, extracting global features from the first monitoring data includes: extracting first time series features from the first monitoring data through a first attention module; extracting first multi-parameter features from the first monitoring data through a second attention module; and fusing the first time series features and the first multi-parameter features to obtain global features.

[0015] Preferably, the first monitoring data and the association relationship data between the monitoring points are fused to obtain local features, including:

[0016] The fourth attention module is used to extract time series correlation features from the association relationship data; the fifth attention module is used to extract single-point multi-parameter features from the first monitoring data; the time series correlation features and the single-point multi-parameter features are fused to obtain local features.

[0017] Preferably, the first fault diagnosis model and the second fault diagnosis model both use a BP neural network model, the BP neural network model includes an input layer, a hidden layer and an output layer, wherein the calculation formula for the number of neurons in the hidden layer is:

[0018]

[0019] Among them, u represents the number of neurons in the hidden layer, i represents the number of neurons in the input layer, o represents the number of neurons in the output layer, and γ is a constant.

[0020] Preferably, during training, the weights and thresholds of the model are initialized using a global search algorithm, and the initialized weights and thresholds are optimized using a BP algorithm.

[0021] On the other hand, an embodiment of the present invention provides a low-voltage intelligent switch cabinet fault diagnosis system based on deep learning, the system comprising:

[0022] A data acquisition module, used to obtain first monitoring data and second monitoring data of each monitoring point of the low-voltage intelligent switch cabinet;

[0023] A feature extraction module, used to extract global features from the first monitoring data, fuse the first monitoring data and the association relationship data between the monitoring points, and obtain local features;

[0024] A first diagnosis module, used for inputting the global feature and the local feature into a first fault diagnosis model to obtain a first fault diagnosis result;

[0025] The second diagnostic module is used to extract auxiliary local features from the second monitoring data when the first fault diagnosis result is an unrecognizable fault, input the global features, auxiliary local features and local features into the second fault diagnosis model, and obtain a second fault diagnosis result.

[0026] The beneficial effects of the present invention are embodied in that: the present invention extracts global features and local features from the first monitoring data and the association relationship data, integrates the local features of different monitoring points and the global features of the entire switch cabinet, realizes the rapid identification of simple faults, and introduces the second monitoring data when it cannot be identified, extracts auxiliary local features from the second monitoring data, and realizes the accurate identification of complex faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the description of the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0028] Figure 1 A flowchart of a low-voltage intelligent switch cabinet fault diagnosis method based on deep learning provided by the present invention;

[0029] Figure 2 A system block diagram of a low-voltage intelligent switch cabinet fault diagnosis system based on deep learning provided by the present invention. DETAILED DESCRIPTION

[0030] The following embodiments of the technical solution of the present invention are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.

[0031] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by those skilled in the art to which the invention belongs.

[0032] Example 1

[0033] like Figure 1As shown, the present invention provides a low-voltage intelligent switch cabinet fault diagnosis method based on deep learning, the method comprising:

[0034] Step 1: Obtain the first monitoring data and the second monitoring data of each monitoring point of the low-voltage intelligent switch cabinet.

[0035] The first monitoring data includes the time-stamped temperature and humidity data, current data, voltage data, sound data and ambient temperature and humidity data of the key components of the monitoring point; the second monitoring data includes the time-stamped temperature and humidity data, current data, voltage data and sound data of other components of the monitoring point. The key components include circuit breakers, contactors, transformers, etc., and the auxiliary components include wiring terminals, auxiliary contacts, etc. The key components and other components are divided according to the actual situation.

[0036] Step 2: extract global features from the first monitoring data, fuse the first monitoring data and the correlation relationship data between the monitoring points, and obtain local features.

[0037] The association relationship data includes position data and weight data under different loads.

[0038] In an embodiment of the present invention, extracting global features from the first monitoring data includes: normalizing and denoising the first monitoring data; extracting first time series features from the first monitoring data through a first attention module; extracting first multi-parameter features from the first monitoring data through a second attention module; and fusing the first time series features and the first multi-parameter features to obtain global features.

[0039] Specifically, to ensure the availability of the data, the first monitoring data is normalized and denoised before feature extraction to eliminate the influence of dimensions and outliers and make the data more regular. Among them, denoising includes median filtering, mean filtering, etc. After preprocessing, the first attention module uses the attention mechanism to automatically learn the time series dependencies in the data, and extracts the first time series features from the first monitoring data. Through the second attention module, focus on the relationship between multiple parameters, and extract the first multi-parameter features from the first monitoring data. Finally, the first time series features and the first multi-parameter features are fused, for example, by weighted averaging, splicing, and then passing through a fully connected layer to obtain global features.

[0040] In some embodiments, the first attention module includes: Input layer: receiving normalized and denoised first monitoring data, which are arranged in chronological order. Embedding layer: embedding the data into a higher dimensional space so that more complex patterns can be captured. This can be achieved by using techniques such as word embedding (such as Word2Vec or GloVe) or position embedding, especially for time series data, position information is very important. Attention mechanism layer: This is the core part of the module, which uses the attention mechanism to automatically learn the temporal dependencies in the data. Specifically, a self-attention mechanism (Self-Attention) can be adopted, such as the Multi-HeadAttention structure in the Transformer model. This mechanism can calculate the influence weight of each position in the input sequence on other positions, thereby capturing the long-term dependencies in the time series data. Output layer: output the first time series feature, which is a weighted feature vector that reflects the importance of each time point in the input sequence.

[0041] In some other embodiments, the second attention module includes an input layer: it also receives the first monitoring data that has been normalized and denoised, and these data contain information about multiple parameters. Feature extraction layer: for each parameter, use an appropriate feature extraction method (such as convolutional neural network CNN, recurrent neural network RNN ​​or its variants LSTM / GRU, etc.) to extract the internal features of the parameter. Attention mechanism layer: similar to the first attention module, but the attention mechanism here is used to capture the relationship between different parameters. A multi-head self-attention mechanism can be used to pay attention to multiple parameters at the same time and learn the correlation between them. Output layer: outputs the first set of parameter features, which is a feature vector that integrates information from multiple parameters.

[0042] In an embodiment of the present invention, single-point multi-parameter features are extracted from the first monitoring data through the fourth attention module, and association relationship time series features are extracted from a pre-constructed monitoring point association relationship knowledge graph according to different load states, and the single-point multi-parameter features are fused with the association relationship time series features to obtain local features.

[0043] Specifically, the fourth attention module includes: input data layer: the input of the fourth attention module is the multi-parameter data of a single monitoring point in the first monitoring data. Attention mechanism layer: similar to the previous attention module, the fourth attention module also uses the attention mechanism to focus on the parameters that are most important to the current task. By calculating the attention weights of different parameters, the module can emphasize the parameters that have the greatest impact on the current state or prediction. Feature extraction layer: under the guidance of the attention weights, key information is extracted from multiple parameters to form multi-parameter features of a single point. These features reflect the comprehensive state of the monitoring point at a specific time point. According to the association relationship defined in the knowledge graph, the time series data of other monitoring points related to the current monitoring point can be extracted. By analyzing these time series data, the characteristics of the associated monitoring points that change over time, namely, the association relationship time series features, can be captured. Before feature fusion, it is necessary to ensure that the single-point multi-parameter features extracted from the fourth attention module and the association relationship time series features extracted from the knowledge graph are aligned in the time and space dimensions, and then a variety of methods are used to fuse the two features, such as simple splicing, weighted averaging, or using more complex fusion networks (such as neural networks) to learn how to optimally combine these features.

[0044] The global feature is extracted from the entire first monitoring data, so it can capture the overall trend and pattern of the data. This helps to fully understand the overall status of the low-voltage intelligent switchgear. Since the global feature is extracted from the entire data set, it is relatively insensitive to the anomalies or noise of individual data points and has strong robustness. The local feature focuses on a specific monitoring point and its association, so it can more accurately reflect the state of the point and its relationship with the surrounding points. Compared with global features, local features pay more attention to details. The single-point multi-parameter features extracted by the fourth attention module can capture subtle changes in specific monitoring points, and combined with the corresponding global association relationships for optimization, more detailed local features can be obtained.

[0045] Step three, input the global features and the local features into a first fault diagnosis model to obtain a first fault diagnosis result; when the first fault diagnosis result is an unrecognizable fault, extract auxiliary local features from the second monitoring data, input the global features, auxiliary local features and local features into a second fault diagnosis model to obtain a second fault diagnosis result.

[0046] In the embodiment of the present invention, both the first fault diagnosis model and the second fault diagnosis model use a BP neural network model, the BP neural network model includes an input layer, a hidden layer and an output layer, wherein the calculation formula for the number of neurons in the hidden layer is:

[0047]

[0048] Among them, u represents the number of neurons in the hidden layer, i represents the number of neurons in the input layer, o represents the number of neurons in the output layer, and γ is a constant.

[0049] In the embodiment of the present invention, during training, the weights and thresholds of the model are initialized using a global search algorithm, and the initialized weights and thresholds are optimized using a BP algorithm.

[0050] The traditional BP neural network model uses the BP algorithm to modify the weights, which has problems such as slow convergence speed and easy to fall into local optimality. In order to solve this problem, the embodiment of the present invention provides an optimization strategy, which adds a global search algorithm before the traditional BP algorithm. In this embodiment, the global search algorithm is a weed algorithm, which simulates the population expansion process of weeds and performs a global search in a specific area. It allows seeds (i.e. potential solutions) to generate offspring seeds according to their own adaptability. This process involves the entire search space. The algorithm continuously explores and optimizes the search space through mechanisms such as seed growth and reproduction, spatial dispersion and competition to find the optimal solution to the problem.

[0051] Specifically, the weights from the input layer to the hidden layer, the weights from the hidden layer to the output layer, and the bias values ​​of the hidden layer are used as seeds. The dimensions of these seeds are i+o+u. The mean square error formula is used to calculate the fitness of each seed, and then a global search is performed based on the fitness. The optimal weights and thresholds are searched through the iterative evolution of the population, and the searched weights and thresholds are used as the initial weights and initial thresholds for further optimization.

[0052] After research, it was found that even if the above method is used to construct the first fault diagnosis model, it still has certain limitations in the diagnostic accuracy of some complex faults. This embodiment adds additional monitoring data to construct a second fault diagnosis model to make up for the shortcomings of the first fault diagnosis model. Although the second fault diagnosis model takes a certain amount of time to diagnose, since the diagnosis object is a low-voltage switchgear, its importance in the entire power grid system is relatively low compared to the high-voltage switchgear, and the reaction sensitivity requirements corresponding to its faults are also relatively low. Therefore, in view of the low importance of low-voltage switchgear in the power grid system and the relatively low requirements for fault reaction sensitivity, the present invention mainly optimizes the diagnostic accuracy under the premise of ensuring a certain diagnostic time, and better meets the actual needs of complex fault diagnosis of low-voltage switchgear.

[0053] It should also be noted that since the newly introduced monitoring data has a weaker impact on the fault diagnosis effect, this embodiment does not consider its global features, but only extracts its local features. The extraction method is the same as that in the above embodiment and will not be repeated here.

[0054] The present invention designs two levels of fault diagnosis models. The first fault diagnosis model is used to quickly identify simple faults, while the second fault diagnosis model performs more in-depth diagnosis for complex faults. This design can flexibly adjust the diagnosis strategy according to actual conditions, thereby improving the adaptability and efficiency of diagnosis.

[0055] Example 2

[0056] like Figure 2 As shown, an embodiment of the present invention provides a low-voltage intelligent switch cabinet fault diagnosis system based on deep learning, the system comprising:

[0057] A data acquisition module, used to obtain first monitoring data and second monitoring data of each monitoring point of the low-voltage intelligent switch cabinet;

[0058] A feature extraction module, used to extract global features from the first monitoring data, fuse the first monitoring data and the association relationship data between the monitoring points, and obtain local features;

[0059] A first diagnosis module, used for inputting the global feature and the local feature into a first fault diagnosis model to obtain a first fault diagnosis result;

[0060] The second diagnostic module is used to extract auxiliary local features from the second monitoring data when the first fault diagnosis result is an unrecognizable fault, input the global features, auxiliary local features and local features into the second fault diagnosis model, and obtain a second fault diagnosis result.

[0061] It should be understood that the low-voltage intelligent switch cabinet fault diagnosis system based on deep learning provided in the embodiment of the present invention and the low-voltage intelligent switch cabinet fault diagnosis method based on deep learning provided in the above embodiment are based on the same inventive concept. For more specific working principles of each module in the embodiment of the present invention, please refer to the above embodiment and will not be repeated in the embodiment of the present invention.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A low-voltage intelligent switch cabinet fault diagnosis method based on deep learning, characterized in that: include: Acquire first monitoring data and second monitoring data of each monitoring point of the low-voltage intelligent switch cabinet, wherein the first monitoring data includes monitoring data of key components of the monitoring point with timestamps; The second monitoring data includes monitoring data of other devices at the monitoring point with time stamps; Extracting global features from the first monitoring data, fusing the first monitoring data and the association relationship data between monitoring points, and obtaining local features; Inputting the global feature and the local feature into a first fault diagnosis model to obtain a first fault diagnosis result; When the first fault diagnosis result is an unrecognizable fault, extracting auxiliary local features from the second monitoring data, inputting the global features, the auxiliary local features and the local features into a second fault diagnosis model, and obtaining a second fault diagnosis result; Among them, global features are extracted from the first monitoring data, and the first monitoring data and the association relationship data between the monitoring points are fused to obtain local features, including: extracting a first time series feature from the first monitoring data through a first attention module; extracting a first multi-parameter feature from the first monitoring data through a second attention module; fusing the first time series feature and the first multi-parameter feature to obtain a global feature; extracting a time series association feature from the association relationship data through a fourth attention module; extracting a single-point multi-parameter feature from the first monitoring data through a fifth attention module; and fusing the time series association feature with the single-point multi-parameter feature to obtain a local feature.

2. A low-voltage intelligent switch cabinet fault diagnosis method based on deep learning according to claim 1, characterized in that: The first monitoring data includes time-stamped temperature and humidity data, current data, voltage data, sound data and ambient temperature and humidity data of key components at the monitoring point; the second monitoring data includes time-stamped temperature and humidity data, current data, voltage data and sound data of other components at the monitoring point.

3. According to a method for fault diagnosis of low-voltage intelligent switch cabinet based on deep learning according to claim 1, it is characterized in that: The global features are extracted from the first monitoring data, which includes: normalizing and denoising the first monitoring data.

4. A low-voltage intelligent switch cabinet fault diagnosis method based on deep learning according to claim 1, characterized in that: The first fault diagnosis model and the second fault diagnosis model both use a BP neural network model, the BP neural network model includes an input layer, a hidden layer and an output layer, wherein the calculation formula for the number of neurons in the hidden layer is: Among them, u represents the number of neurons in the hidden layer, i represents the number of neurons in the input layer, o represents the number of neurons in the output layer, and γ is a constant.

5. A low-voltage intelligent switch cabinet fault diagnosis method based on deep learning according to claim 4, characterized in that: During training, the global search algorithm is used to initialize the weights and thresholds of the model, and the BP algorithm is used to optimize the initialized full values ​​and thresholds.

6. A low-voltage intelligent switch cabinet fault diagnosis system based on deep learning, applicable to the method described in any one of claims 1 to 5, characterized in that: include: A data acquisition module, used to obtain first monitoring data and second monitoring data of each monitoring point of the low-voltage intelligent switch cabinet; A feature extraction module, used to extract global features from the first monitoring data, fuse the first monitoring data and the association relationship data between the monitoring points, and obtain local features; A first diagnosis module, used for inputting the global feature and the local feature into a first fault diagnosis model to obtain a first fault diagnosis result; The second diagnostic module is used to extract auxiliary local features from the second monitoring data when the first fault diagnosis result is an unrecognizable fault, input the global features, auxiliary local features and local features into the second fault diagnosis model, and obtain a second fault diagnosis result.

Images