Transformer fault diagnosis method, device, equipment and medium
Through a deep confidence network with dynamic adaptive knowledge transfer-interval two-type fuzzy set structure, a transformer fault diagnosis model was established, which solved the problem of insufficient transformer fault diagnosis efficiency and accuracy in the prior art, and achieved efficient and accurate fault diagnosis.
Patent Information
- Application Number
- CN202411180272.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-11-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing transformer fault diagnosis methods are difficult to meet high requirements in terms of efficiency and accuracy, especially with the increase in online monitoring system data, chromatographic analysis systems cannot effectively support the efficient diagnosis of transformer faults.
A deep confidence network with dynamic adaptive knowledge transfer-interval two-type fuzzy set structure is adopted to establish a sparse deep confidence network through the fusion of neural symbol models and knowledge, optimize network parameters, and form a fault diagnosis model, which is used to efficiently and accurately diagnose transformer failures.
It realizes efficient and accurate diagnosis of transformer faults, improves diagnostic efficiency and accuracy, can process large amounts of data, and improves the robustness and generalization capabilities of the model.
Smart Images

Figure CN119004111A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a transformer fault diagnosis method, device, equipment and medium. Background Art
[0002] The power transformer is one of the important equipment in the power system, and its stability is directly related to the safety of the power grid. Therefore, it is of great significance to monitor the normal / fault status of the transformer in a timely and effective manner.
[0003] In the related scheme, an online monitoring system is used to obtain the dissolved concentration in the transformer oil, and then a chromatography analysis system is used to perform a chromatographic analysis on the dissolved concentration in the oil, and finally the transformer fault diagnosis is performed based on the chromatographic analysis results.
[0004] However, with the widespread application of online monitoring systems, the chromatographic analysis system needs to analyze more and more data, which cannot meet the high requirements of transformer fault diagnosis in terms of efficiency and accuracy. Summary of the invention
[0005] The present invention provides a transformer fault diagnosis method, device, equipment and medium, establishes a deep belief network with a dynamic adaptive knowledge transfer-interval type-2 fuzzy set structure, and obtains a fault diagnosis model through model training based on this, which can use the fault diagnosis model to efficiently and accurately diagnose transformer faults.
[0006] According to one aspect of the present invention, a transformer fault diagnosis method is provided, the method comprising:
[0007] Acquire multiple operating data of the candidate transformers, and determine a training data set according to the operating data;
[0008] Establishing a first deep belief network based on a neural symbolic model and knowledge fusion; wherein the first deep belief network is a sparse deep belief network based on dynamic adaptive knowledge transfer;
[0009] Optimizing the network parameters of the first deep belief network based on an interval type-2 fuzzy set structure to obtain a second deep belief network; wherein the network parameters include weights and biases;
[0010] The second deep belief network is trained according to the training data set. When the second deep belief network converges, the second deep belief network is used as a fault diagnosis model to perform fault diagnosis on the target transformer based on the fault diagnosis model.
[0011] According to another aspect of the present invention, there is provided a transformer fault diagnosis device, comprising:
[0012] A training data set determination module, used to obtain a plurality of operating data of a candidate transformer and determine a training data set according to the operating data;
[0013] A first network model determination module, used to establish a first deep belief network based on a neural symbolic model and knowledge fusion; wherein the first deep belief network is a sparse deep belief network based on dynamic adaptive knowledge transfer;
[0014] A second network model determination module is used to optimize the network parameters of the first deep belief network based on an interval type-2 fuzzy set structure to obtain a second deep belief network; wherein the network parameters include weights and biases;
[0015] A fault diagnosis model determination module is used to perform model training on the second deep belief network according to the training data set. When the second deep belief network converges, the second deep belief network is used as a fault diagnosis model to perform fault diagnosis on the target transformer based on the fault diagnosis model.
[0016] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:
[0017] at least one processor; and
[0018] a memory communicatively connected to the at least one processor; wherein,
[0019] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the transformer fault diagnosis method described in any embodiment of the present invention.
[0020] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the transformer fault diagnosis method described in any embodiment of the present invention when executed.
[0021] The technical solution of the embodiment of the present invention obtains multiple operating data of candidate transformers, determines a training data set based on the operating data; establishes a first deep belief network based on a neural symbolic model and knowledge fusion; wherein the first deep belief network is a sparse deep belief network based on dynamic adaptive knowledge transfer; optimizes the network parameters of the first deep belief network based on an interval type-2 fuzzy set structure to obtain a second deep belief network; wherein the network parameters include weights and biases; performs model training on the second deep belief network based on the training data set, and when the second deep belief network converges, uses the second deep belief network as a fault diagnosis model to perform fault diagnosis on the target transformer based on the fault diagnosis model. This technical solution establishes a deep belief network with a dynamic adaptive knowledge transfer-interval type-2 fuzzy set structure, and obtains a fault diagnosis model based on this structure through model training, which can efficiently and accurately diagnose transformer faults using the fault diagnosis model.
[0022] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0024] Figure 1 is a flow chart of a transformer fault diagnosis method provided according to Embodiment 1 of the present invention;
[0025] Figure 2 is a schematic diagram of a process of establishing a first deep belief network provided according to Embodiment 1 of the present invention;
[0026] Figure 3 is a flow chart of a transformer fault diagnosis method provided according to Embodiment 2 of the present invention;
[0027] Figure 4 is a curve diagram of the accuracy of a diagnosis result provided according to the second embodiment of the present invention;
[0028] Figure 5 is a schematic diagram of the structure of a transformer fault diagnosis device provided according to Embodiment 3 of the present invention;
[0029] Figure 6 It is a structural schematic diagram of an electronic device for implementing a transformer fault diagnosis method according to an embodiment of the present invention. DETAILED DESCRIPTION
[0030] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.
[0031] It should be noted that the terms "first", "second", "target", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0032] Embodiment 1
[0033] Figure 1 This is a flowchart of a transformer fault diagnosis method provided in the first embodiment of the present invention. This embodiment is applicable to the case of efficient and accurate diagnosis of transformer faults. The method can be executed by a transformer fault diagnosis device. The transformer fault diagnosis device can be implemented in the form of hardware and / or software. The transformer fault diagnosis device can be configured in an electronic device with data processing capabilities. Figure 1 As shown, the method includes:
[0034] S110, obtaining multiple operating data of candidate transformers, and determining a training data set according to the operating data.
[0035] Among them, the candidate transformer can be used as the collection object of the operation data in the training data set, usually including multiple transformers. Among them, the operation data can be collected using a multi-parameter coupled oil-immersed transformer intelligent sensor device based on a micro-electromechanical system. Specifically, the multi-parameter coupled oil-immersed transformer intelligent sensor device includes a sensor module, a data processing and storage module, and a communication module, wherein the sensor module can be used to monitor the concentration of specific gases (such as hydrogen, acetylene, methane, ethylene, ethane, etc.) in the transformer oil, and measure the pressure changes inside the transformer; the data processing and storage module can be used to pre-process, store and preliminarily analyze the sensor data; the communication module can be used to transmit the collected data to the central control unit or the remote monitoring center. Exemplarily, the operation data may include winding temperature, H2 dissolved gas in oil, C2H2 dissolved gas in oil, CO dissolved gas in oil, CO2 dissolved gas in oil, C2H4 dissolved gas in oil, CH4 dissolved gas in oil, oil temperature, oil pressure, ultrasonic partial discharge, high-frequency partial discharge, core grounding current, clamp grounding current and transformer operation soundprint, etc.
[0036] In this embodiment, after obtaining multiple operating data of the candidate transformer, in order to ensure the validity of the data, the operating data can be cleaned and filtered to remove invalid data (such as noise or abnormal data) therein, and then a training data set is constructed based on the retained valid operating data. Optionally, determining the training data set based on the operating data includes: determining the Lyapunov index of each operating data; filtering the operating data based on the Lyapunov index to obtain valid data; and constructing the training data set based on the valid data.
[0037] In this embodiment, the operating data can be screened according to the contraction characteristics of the chaotic phase space and the Lyapunov index in the chaos theory, so as to provide a method for selecting the optimal analysis data. Specifically, after n iterations, the distance between the initial value t and its adjacent point t+Δt can be expressed as: Among them, f (n) represents the function value after n iterations, where n is a positive integer representing the number of iterations; Δt represents the time step or can be understood as a small interval in the phase space; λ represents the separation rate between adjacent orbits in the chaotic system. Specifically, λ measures the growth or contraction of the distance between the initial value t and its adjacent point t+Δt per unit time. According to the differential law of composite functions The Lyapunov exponent is as follows: Where L represents the Lyapunov exponent.
[0038] After the Lyapunov index of each operating data is calculated based on the above method, the operating data can be screened according to the Lyapunov index to obtain valid data. Optionally, the operating data is screened according to the Lyapunov index to obtain valid data, including: taking the operating data with a Lyapunov index less than zero as valid data.
[0039] Specifically, by comparing the size of the Lyapunov exponent, the chaotic characteristics of the time series of the collected operation data can be judged. When L>0, it has local instability. If L=0, it corresponds to the bifurcation point, indicating a critical state, that is, the boundary between stability and instability. If L<0, it indicates that the orbit is locally stable. Therefore, L<0 can be used as a screening condition to filter out valid data from the original operation data, thereby reducing the amount of data to be processed and improving computational efficiency. After obtaining the valid data, each valid data can be combined into a set as a training data set.
[0040] S120, establishing a first deep belief network based on the neural symbolic model and knowledge fusion.
[0041] Among them, the first deep belief network is a sparse deep belief network based on dynamic adaptive knowledge transfer. Deep Belief Networks (DBN) is a neural network model in deep learning methods. It is built on restricted Boltzmann Machines (RBM) and realizes probability generation by stacking multiple layers of RBM. The neural symbolic model is an artificial intelligence technology that integrates neural networks and symbolic reasoning, combining the advantages of neural networks and symbolic reasoning. Knowledge fusion is a technical method that integrates knowledge and data from different sources and in different forms to obtain more comprehensive and accurate information.
[0042] In this embodiment, when constructing the network model, the dynamic operation mechanism of the DBN deep network is first extracted based on the neural symbolic model, and a new sparse deep belief network (i.e., the first deep belief network) based on adaptive knowledge transfer is established based on the knowledge fusion method of rule merging and deletion. Among them, the neural symbolic model integrates the generalization ability of the deep network and the logical reasoning ability of the symbolic rules, and can be used to describe the internal operation of the DBN while improving the feature learning performance. Optionally, the first deep belief network is established based on the neural symbolic model and knowledge fusion, including: pruning the network structure of the initial deep belief network to obtain a reference deep belief network; extracting the candidate symbol rules corresponding to each neuron from the restricted Boltzmann machine of the reference deep belief network; wherein the symbol rules include confidence rules and classification rules; merging the candidate symbol rules corresponding to each neuron according to the output and structure of the candidate symbol rules to obtain the target symbol rules; embedding the target symbol rules into the reference deep belief network to establish the first deep belief network.
[0043] Among them, the initial deep belief network and the reference deep belief network may refer to the deep belief network before and after the network structure pruning, respectively. The candidate symbol rule may refer to the symbol rule extracted directly from the restricted Boltzmann machine of the reference deep belief network (the symbol rule can be understood as a network layer), which may specifically include confidence rules and classification rules. The target symbol rule may refer to the symbol rule obtained after merging the candidate symbol rules.
[0044] When establishing the first deep belief network, firstly, based on the correlation analysis between neurons, a pruning algorithm is used to prune the network structure of the initial deep belief network to obtain a reference deep belief network. Specifically, given a neuron h j and its K connected neurons x in the previous layer jk (k=1,2,...,K), h j With each x jk The correlation coefficient between them can be expressed as Among them, μ j and μ jk , σ j and σ jk Respectively represent h j and x jk The mean and standard deviation of a variable, E represents the mathematical expectation of a variable.
[0045] After obtaining the reference deep belief network, the candidate symbol rules corresponding to each neuron can be extracted from the restricted Boltzmann machine of the reference deep belief network, including confidence rules and classification rules. Among them, the confidence rules represent the combination of numbers and symbols, which describe the internal operation of stacked RBMs in DBN. Hierarchical reasoning enables them to explain deep networks well. For the output of neurons in each hidden layer, the confidence rules can be extracted in the following form: in, Defines whether the jth neuron in the nth layer is activated, represents the confidence rule of the jth neuron in the nth layer, ∧ represents the logical "and", T and K define the sets of neurons that show positive and negative influence on the rule respectively, and l is the index of the hidden layer. are activated and all If both are not activated, then activate The probability of For each tag, the classification rules can be extracted in the following form: Among them, NumTrue() is the number of activated neurons in set A, which contains all possible category labels. 1 and A n are the specific category labels in set A, s is a threshold used to compare the similarity between the classification result and the actual label, is the connection weight after the nth clustering, bias is the bias of the output neuron, y is the final classification result (for example, it can be 0 or 1, representing negative and positive classes respectively), and C is a category label of the output.
[0046] The confidence rules extracted from RBM have the following characteristics: the superposition features of the confidence rules correspond to the stacked RBMs, and the rule extraction process can be synchronized and alternating with the training of the RBM. Among them, the superposition features can be understood as updating the network parameters through the confidence rules during the RBM training process, so that the network can better capture the inherent laws of the data. The output of the hidden layer can be expressed as follows: Where σ() is an activation function (e.g., the sigmoid function, which can be expressed as σ(x) = 1 / [1+exp(-x)]), is the connection weight, and b j is the bias. Further, it can be updated as follows: Among them, c j is the confidence value. In order to make h′ j can represent the output of the network, which attempts to search for these positive and negative propositions (i.e. neurons) and the corresponding confidence values c by minimizing the following information loss j , c is calculated by minimizing the following information lossj value: Among them, s ij =sign(w ij ),s ij ∈{1,0,-1},w ij is the connection weight between the visible layer and the hidden layer, and x i and represent the activation value and the opposite activation value of the i-th neuron in the visible layer, respectively.
[0047] The classification rule extraction process can be regarded as the search condition for the activation of neurons in the classification layer of the DBN model. The logistic regression function is used as the activation function and the deep network is fine-tuned to obtain Among them, O j is the jth neuron of the classification layer h, and the logistic regression function σ can be expressed as Simplifying the output values of some neurons in the classifier to 0 or 1, we can get This simplification allows rule extraction to focus only on the weights and biases of these neurons in the classification layer, significantly reducing the complexity of rule extraction.
[0048] Due to the large scale of deep networks, the rules extracted from the DBN model contain a lot of redundancy. Rule merging is an effective method to filter noise and remove redundant rules. This is conducive to the subsequent insertion of knowledge into the deep network, and the sparse structure of the DBN can be adaptively determined. Therefore, after extracting the candidate symbol rules corresponding to each neuron, the candidate symbol rules corresponding to each neuron are merged according to the output and structure of the candidate symbol rules to obtain the target symbol rules. Optionally, the candidate symbol rules corresponding to each neuron are merged according to the output and structure of the candidate symbol rules to obtain the target symbol rules, including: for every two neurons, determining a first similarity according to the output of the candidate symbol rules; for every two neurons, determining a second similarity according to the structure of the candidate symbol rules; based on the first similarity and the second similarity, the candidate symbol rules corresponding to each neuron are merged to obtain the target symbol rules.
[0049] Specifically, first, the correlation between the rules is considered according to the output of the candidate symbol rule, and the correlation is used as the first similarity. For example, suppose two rules from the same layer and The outputs on P samples are and And their means are and but and The correlations between are as follows: Where σ represents the standard deviation. Then, the correlation between the rules is considered according to the structure of the candidate symbol rules, and the correlation is used as the second similarity. For example, assuming the same prior quantity rules and The number of elements between is e, and represents the total number of previous elements, then and The structural similarity between can be expressed as follows: when hour, and show high similarity and can be merged into one rule, where θ 1 and θ 2 is the threshold, and Can be judged Calculate it later to reduce the computational complexity. and The purpose is to delete the one with smaller structure and modify the next level rules related to them. The merged rules (i.e. the target symbol rule) is as follows:
[0050]
[0051] Knowledge transfer embeds these confidence rules and classification rules into the reference deep belief network and assigns weight parameters to DBN to adaptively establish a sparse network structure and improve the effectiveness of feature learning, so that the first deep belief network has a sparser network structure and better learning performance.
[0052] Figure 2 A schematic diagram of the process of establishing a first deep belief network provided in the first embodiment of the present invention. Figure 2 As shown in the figure, there are 5 solid boxes in total, and the arrows between the solid boxes indicate the order of the network establishment process. Specifically, the solid box in the upper left corner represents the initial deep belief network; the solid box in the lower left corner represents the reference deep belief network after network structure pruning; the solid box in the middle represents the knowledge extraction (i.e. candidate symbol rule extraction) process, and the red dotted box represents the confidence rule extraction process, and the blue dotted box represents the classification rule extraction process; the solid box in the upper right corner represents the target symbol rule after knowledge fusion (i.e. candidate symbol rule merging); the solid box in the lower right corner represents the first deep belief network (KT-SDBN) obtained after knowledge insertion (i.e. target symbol rule embedding).
[0053] S130, optimizing the network parameters of the first deep belief network based on the interval type-2 fuzzy set structure to obtain a second deep belief network.
[0054] In this embodiment, after the first deep belief network is established, the uncertain network parameters in the first deep belief network can be modeled by an interval type-2 fuzzy set (IT2F) structure, thereby achieving network parameter optimization. Wherein, the network parameters include weights and biases. Optionally, the network parameters of the first deep belief network are optimized based on the interval type-2 fuzzy set structure to obtain the second deep belief network, including: taking the network parameters as interval type-2 fuzzy sets; based on the interval type-2 fuzzy perceptual contrastive divergence learning framework, the network parameters of the first deep belief network are optimized to obtain the second deep belief network.
[0055] The key idea is to perform alternating steps of "update" and "reconstruction" to train the stacked RBMs in a DBN. Specifically, for any RBM, the energy of the joint configuration of visible and hidden units with IT2-fuzzy weights and biases is given by: Among them, v represents the input vector of the visible layer, which contains the activation values of all visible neurons; h represents the output vector of the hidden layer, which contains the activation values of all hidden neurons; W represents the weight matrix, which represents the connection weights between the visible layer and the hidden layer; b represents the bias vector of the hidden layer; d represents the bias vector of the visible layer; is the IT2 fuzzy energy function. Each possible pair of visible and hidden states is assigned a probability. The joint probability with any vector argument o, denoted by p, can be defined as follows: Therefore, according to the definition of IT2 fuzzy energy function, the probability can be defined as: in, is the partition function, which sums over all possible hidden and visible vectors. Using the quantities explained above, we can derive the following equation:
[0056] The probability of the validity of the training data can be increased by fine-tuning the weights and biases to reduce the energy of the training data. In addition, the weights are adjusted to increase the energy of other data points with lower energy that have a significant impact on the partition function. To do this, the derivative of the log probability with respect to the weight is calculated as follows: The angle brackets represent the average (or expected value) of the training data identified by the subscript, and v and h represent the activation values of the visible layer and the hidden layer, respectively. Therefore, the above formula constitutes the learning rule of the training data.
[0057] Since RBM is a restricted version of BM, there is no connection between hidden units, so unbiased samples can be easily calculated. Given any randomly selected training case of visible units, the binary state of each hidden unit is set to 1 with probability given by: Furthermore, given the hidden vector, an unbiased sample of the states of the visible units can be easily computed using: Finally, the change in the weight of the interconnection between units i and j can be defined as follows:
[0058] In addition, interval type-2 fuzzy sets can be used in different ways. For example, the expected value of the probability is calculated before the activation function, and according to the expected value, the probability of setting the u+1th node to 1 is given by: However, it is better to apply the activation function before doing the expectation. Therefore, we can Infer the probability of the u+1th node. Optimizing the weights and biases of the first deep belief network based on the interval type-2 fuzzy set structure can better handle the uncertainty between input data and model parameters, thereby improving the robustness and generalization ability of the model.
[0059] S140, performing model training on the second deep belief network according to the training data set, and when the second deep belief network converges, using the second deep belief network as a fault diagnosis model to perform fault diagnosis on the target transformer based on the fault diagnosis model.
[0060] After obtaining the second deep belief network, the second deep belief network can be trained according to the training data set. The training process of DBN is completed by fine-tuning the network in a supervised manner using the classic back-propagation algorithm, in which the derivatives of each hidden unit are propagated and reach a local minimum. The parameters of the RBM are randomly initialized, and the entire DBN is finally initialized. Exemplarily, the RBM uses a Bernoulli distribution. In addition, the architecture of the DBN is also initialized. The DBN model has a standard visible layer and an output layer, with two hidden layers between them, and the number of nodes in each hidden layer is the same. The uncertain weights and biases are initialized using IT2 fuzzy weights and biases. When the second deep belief network converges, the second deep belief network can be used as a fault diagnosis model to perform fault diagnosis on the target transformer (transformer that needs to be fault diagnosed) based on the fault diagnosis model.
[0061] The technical solution of the embodiment of the present invention obtains multiple operating data of candidate transformers, determines a training data set based on the operating data; establishes a first deep belief network based on a neural symbolic model and knowledge fusion; wherein the first deep belief network is a sparse deep belief network based on dynamic adaptive knowledge transfer; optimizes the network parameters of the first deep belief network based on an interval type-2 fuzzy set structure to obtain a second deep belief network; wherein the network parameters include weights and biases; performs model training on the second deep belief network according to the training data set, and when the second deep belief network converges, uses the second deep belief network as a fault diagnosis model to perform fault diagnosis on the target transformer based on the fault diagnosis model. This technical solution establishes a deep belief network with a dynamic adaptive knowledge transfer-interval type-2 fuzzy set structure, and obtains a fault diagnosis model based on this structure through model training, and can use the fault diagnosis model to achieve efficient and accurate diagnosis of transformer faults.
[0062] Embodiment 2
[0063] Figure 3 A flowchart of a transformer fault diagnosis method provided in the second embodiment of the present invention, this embodiment is optimized based on the above embodiment. The specific optimization is: after using the second deep belief network as the fault diagnosis model, it also includes: determining the model evaluation parameters according to the diagnosis results of the fault diagnosis model; wherein the model evaluation parameters include accuracy, sensitivity and comprehensive reliability; performing performance verification on the fault diagnosis model according to the model evaluation parameters; if the fault diagnosis model fails the performance verification, the fault diagnosis model is adjusted and updated until the fault diagnosis model passes the performance verification.
[0064] like Figure 3 As shown, the method of this embodiment specifically includes the following steps:
[0065] S210, obtaining multiple operating data of the candidate transformers, and determining a training data set according to the operating data.
[0066] S220, establishing a first deep belief network based on the neural symbolic model and knowledge fusion.
[0067] Among them, the first deep belief network is a sparse deep belief network based on dynamic adaptive knowledge transfer.
[0068] S230, optimizing network parameters of the first deep belief network based on the interval type-2 fuzzy set structure to obtain a second deep belief network.
[0069] Among them, network parameters include weights and biases.
[0070] S240: Perform model training on the second deep belief network according to the training data set, and when the second deep belief network converges, use the second deep belief network as a fault diagnosis model.
[0071] The specific implementation of S210 - S240 may refer to the detailed description of S110 - S140 in the above embodiment, which will not be described in detail here.
[0072] S250: determining a model evaluation parameter according to a diagnosis result of the fault diagnosis model.
[0073] Among them, the model evaluation parameters include accuracy, sensitivity and comprehensive reliability. Specifically, according to the fault type, the diagnosis results can usually be divided into three categories, namely, confirmed diagnosis, missed diagnosis or misdiagnosis. For a pre-known fault type T1, if it is correctly diagnosed as T1, it is a confirmed diagnosis. If it is diagnosed as another fault type T2, it is a missed diagnosis for fault type T1 and a misdiagnosis for fault type T2. Based on the above diagnostic results, the accuracy rate can be defined as the ratio of the number of confirmed cases to the total number of cases diagnosed with a certain fault type, then the accuracy rate of fault type i can be expressed as: Among them, N(A i ) represents the number of confirmed cases of fault type i, N(B i ) represents the number of misdiagnosed cases. The accuracy rate reflects the accuracy of a specific fault type. The higher the accuracy rate, the higher the accuracy in diagnosing a specific fault type. The sensitivity rate can be defined as the ratio of confirmed diagnosis cases of a certain fault type to the total number of cases. The sensitivity rate of fault type i can be expressed as: Among them, N(C i ) is the number of missed cases of fault type i. The sensitivity rate measures the ability to correctly identify all cases that are actually of this fault type, and the higher the sensitivity rate, the more effective it is in identifying all actual faults and reducing missed cases. The comprehensive reliability can be defined as the sum of the number of confirmed cases and the total number of cases in the database, and the comprehensive reliability can be expressed as: It can be seen that the comprehensive reliability calculated by accuracy and sensitivity rate is consistent.
[0074] The model evaluation parameters proposed in this embodiment include both the accuracy and sensitivity rate for various fault types and the comprehensive reliability, which not only makes the model performance evaluation more accurate and comprehensive, but also lays the foundation for the combination of fault diagnosis methods.
[0075] S260: Verify the performance of the fault diagnosis model according to the model evaluation parameters.
[0076] Exemplarily, the accuracy rate threshold, sensitivity rate threshold and comprehensive reliability threshold can be preset according to actual application requirements. After the accuracy rate, sensitivity rate and comprehensive reliability are determined according to the diagnosis result of the fault diagnosis model, they are compared with the accuracy rate threshold, sensitivity rate threshold and comprehensive reliability threshold respectively. If the accuracy rate is greater than the accuracy rate threshold, the sensitivity rate is greater than the sensitivity rate threshold and the comprehensive reliability is greater than the comprehensive reliability threshold, then it can be determined that the fault diagnosis model has passed the performance verification, otherwise it is determined that the fault diagnosis model has not passed the performance verification.
[0077] S270: If the fault diagnosis model fails the performance verification, the fault diagnosis model is adjusted and updated until the fault diagnosis model passes the performance verification, so as to perform fault diagnosis on the target transformer based on the fault diagnosis model that passes the performance verification.
[0078] Figure 4 The following is a curve diagram of the accuracy of a diagnosis result provided by the second embodiment of the present invention, wherein the diagnosis result is obtained based on a fault diagnosis model that has been verified by performance. Figure 4 It can be seen that the average recognition rate of transformer fault diagnosis using the fault diagnosis model is 97.43%, indicating that the fault diagnosis model has a high accuracy rate for transformer fault diagnosis, which can prove the effectiveness of the fault diagnosis model.
[0079] The technical solution of the embodiment of the present invention, after using the second deep belief network as the fault diagnosis model, determines the model evaluation parameters according to the diagnosis results of the fault diagnosis model; wherein the model evaluation parameters include accuracy, sensitivity and comprehensive reliability; the performance of the fault diagnosis model is verified according to the model evaluation parameters; if the fault diagnosis model fails the performance verification, the fault diagnosis model is adjusted and updated until the fault diagnosis model passes the performance verification. This technical solution can ensure the effectiveness of the fault diagnosis model, help improve the accuracy of transformer fault diagnosis, and enhance the versatility and flexibility of the fault diagnosis model.
[0080] Embodiment 3
[0081] Figure 5 This is a schematic diagram of the structure of a transformer fault diagnosis device provided in the third embodiment of the present invention. The device can execute the transformer fault diagnosis method provided in any embodiment of the present invention and has the corresponding functional modules and beneficial effects of the execution method. Figure 5 As shown, the device comprises:
[0082] A training data set determination module 310 is used to obtain a plurality of operating data of a candidate transformer and determine a training data set according to the operating data;
[0083] A first network model determination module 320, configured to establish a first deep belief network based on a neural symbolic model and knowledge fusion; wherein the first deep belief network is a sparse deep belief network based on dynamic adaptive knowledge transfer;
[0084] A second network model determination module 330 is used to optimize the network parameters of the first deep belief network based on an interval type-2 fuzzy set structure to obtain a second deep belief network; wherein the network parameters include weights and biases;
[0085] The fault diagnosis model determination module 340 is used to perform model training on the second deep belief network according to the training data set, and when the second deep belief network converges, use the second deep belief network as a fault diagnosis model to perform fault diagnosis on the target transformer based on the fault diagnosis model.
[0086] Optionally, the training data set determination module 310 includes:
[0087] a calculation unit, for determining a Lyapunov exponent of each of the operating data;
[0088] A data screening unit, used for screening the operation data according to the Lyapunov exponent to obtain valid data;
[0089] The training data set construction unit is used to construct a training data set according to the valid data.
[0090] Optionally, the data screening unit is specifically used to:
[0091] The operating data whose Lyapunov exponent is less than zero is regarded as valid data.
[0092] Optionally, the first network model determining module 320 includes:
[0093] A network structure pruning unit, used to prune the network structure of the initial deep belief network to obtain a reference deep belief network;
[0094] A symbol rule extraction unit, used to extract candidate symbol rules corresponding to each neuron from the restricted Boltzmann machine of the reference deep belief network; wherein the symbol rules include confidence rules and classification rules;
[0095] A symbol rule merging unit, configured to merge the candidate symbol rules corresponding to the neurons according to the output and structure of the candidate symbol rules to obtain a target symbol rule;
[0096] The first deep belief network establishing unit is used to embed the target symbol rule into the reference deep belief network to establish a first deep belief network.
[0097] Optionally, the symbol rule merging unit is specifically used to:
[0098] For every two neurons, determining a first similarity according to the output of the candidate symbol rule;
[0099] For every two neurons, determining a second similarity according to the structure of the candidate symbol rule;
[0100] The candidate symbol rules corresponding to the neurons are merged based on the first similarity and the second similarity to obtain a target symbol rule.
[0101] Optionally, the second network model determining module 330 is configured to:
[0102] Taking the network parameters as interval type-2 fuzzy sets;
[0103] Based on the interval type-2 fuzzy perceptual contrastive divergence learning framework, the network parameters of the first deep belief network are optimized to obtain the second deep belief network.
[0104] Optionally, the device further comprises:
[0105] A model evaluation parameter determination module, used to determine the model evaluation parameters according to the diagnosis results of the fault diagnosis model after the second deep belief network is used as the fault diagnosis model; wherein the model evaluation parameters include accuracy, sensitivity and comprehensive reliability;
[0106] A model performance evaluation module, used to perform performance verification on the fault diagnosis model according to the model evaluation parameters;
[0107] The model adjustment and updating module is used to adjust and update the fault diagnosis model if the fault diagnosis model fails the performance verification until the fault diagnosis model passes the performance verification.
[0108] A transformer fault diagnosis device provided in an embodiment of the present invention can execute a transformer fault diagnosis method provided in any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method.
[0109] Embodiment 4
[0110] Figure 6A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.
[0111] like Figure 6 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.
[0112] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.
[0113] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a transformer fault diagnosis method.
[0114] In some embodiments, the transformer fault diagnosis method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the transformer fault diagnosis method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the transformer fault diagnosis method in any other appropriate manner (e.g., by means of firmware).
[0115] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
[0116] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.
[0117] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
[0118] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).
[0119] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.
[0120] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.
[0121] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.
[0122] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A transformer fault diagnosis method, characterized in that: The method comprises: Acquire multiple operating data of the candidate transformers, and determine a training data set according to the operating data; Establishing a first deep belief network based on a neural symbolic model and knowledge fusion; wherein the first deep belief network is a sparse deep belief network based on dynamic adaptive knowledge transfer; Optimizing the network parameters of the first deep belief network based on an interval type-2 fuzzy set structure to obtain a second deep belief network; wherein the network parameters include weights and biases; The second deep belief network is trained according to the training data set. When the second deep belief network converges, the second deep belief network is used as a fault diagnosis model to perform fault diagnosis on the target transformer based on the fault diagnosis model.
2. The method according to claim 1, characterized in that: Determining a training data set according to the operating data includes: determining a Lyapunov exponent for each of the operating data; Screening the operating data according to the Lyapunov exponent to obtain valid data; A training data set is constructed according to the valid data.
3. The method according to claim 2, characterized in that The operation data is screened according to the Lyapunov exponent to obtain valid data, including: The operating data whose Lyapunov exponent is less than zero is regarded as valid data.
4. The method according to claim 1, characterized in that: The first deep belief network is established based on neural symbolic model and knowledge fusion, including: Prune the network structure of the initial deep belief network to obtain a reference deep belief network; Extracting candidate symbol rules corresponding to each neuron from the restricted Boltzmann machine of the reference deep belief network; wherein the symbol rules include confidence rules and classification rules; Merging the candidate symbol rules corresponding to the neurons according to the output and structure of the candidate symbol rules to obtain a target symbol rule; The target symbol rule is embedded into the reference deep belief network to establish a first deep belief network.
5. The method according to claim 4, characterized in that The candidate symbol rules corresponding to each neuron are merged according to the output and structure of the candidate symbol rules to obtain a target symbol rule, including: For every two neurons, determining a first similarity according to the output of the candidate symbol rule; For every two neurons, determining a second similarity according to the structure of the candidate symbol rule; The candidate symbol rules corresponding to the neurons are merged based on the first similarity and the second similarity to obtain a target symbol rule.
6. The method according to claim 1, characterized in that The network parameters of the first deep belief network are optimized based on the interval type-2 fuzzy set structure to obtain a second deep belief network, including: Taking the network parameters as interval type-2 fuzzy sets; Based on the interval type-2 fuzzy perceptual contrastive divergence learning framework, the network parameters of the first deep belief network are optimized to obtain the second deep belief network.
7. The method according to any one of claims 1 to 6, characterized in that After using the second deep belief network as a fault diagnosis model, the method further includes: Determining model evaluation parameters according to the diagnosis results of the fault diagnosis model; wherein the model evaluation parameters include accuracy, sensitivity and comprehensive reliability; Performing performance verification on the fault diagnosis model according to the model evaluation parameters; If the fault diagnosis model fails the performance verification, the fault diagnosis model is adjusted and updated until the fault diagnosis model passes the performance verification.
8. A transformer fault diagnosis device, characterized in that: The device comprises: A training data set determination module, used to obtain a plurality of operating data of a candidate transformer and determine a training data set according to the operating data; A first network model determination module, used to establish a first deep belief network based on a neural symbolic model and knowledge fusion; wherein the first deep belief network is a sparse deep belief network based on dynamic adaptive knowledge transfer; A second network model determination module is used to optimize the network parameters of the first deep belief network based on an interval type-2 fuzzy set structure to obtain a second deep belief network; wherein the network parameters include weights and biases; A fault diagnosis model determination module is used to perform model training on the second deep belief network according to the training data set. When the second deep belief network converges, the second deep belief network is used as a fault diagnosis model to perform fault diagnosis on the target transformer based on the fault diagnosis model.
9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the transformer fault diagnosis method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the transformer fault diagnosis method according to any one of claims 1 to 7 when executed.
Citation Information
Patent Citations
A multi-dimensional information fusion method for transformer fault diagnosis based on depth learning
CN109214416A
Power transformer fault diagnosis method
CN113326881A
Power transformer fault prediction method and system driven by state holographic sensing data
CN114397526A
Reactor protection system dynamic reliability analysis method, system, equipment and medium
CN116244965A
