A transformer fault detection method, system, medium, and processor based on an improved chimpanzee optimization algorithm.

Abstract

This invention discloses a transformer fault detection method, system, processor, and medium based on an improved chimpanzee optimization algorithm. The method includes acquiring multi-source heterogeneous monitoring data of the transformer; performing source-specific preprocessing and feature extraction on the multi-source heterogeneous monitoring data to form a standardized feature set; performing cross-modal feature extraction and fusion on the standardized feature set based on a pre-constructed multi-branch deep neural network fusion model; employing a chimpanzee optimization algorithm based on chaotic mapping, with the diagnostic performance of the model as the optimization objective, to globally adaptively optimize the parameters of the model to obtain an optimized fault diagnosis model; inputting real-time monitoring data of the transformer to be diagnosed into the optimized fault diagnosis model, and outputting the fault type diagnosis result. This invention can fully identify the complementarity and deep correlation between multi-physics field signals, improving diagnostic accuracy and early fault identification capabilities.

Description

Technical Field

[0001] This invention relates to the field of transformer testing technology, and in particular to a transformer fault detection method, system, medium, and processor based on an improved chimpanzee optimization algorithm. Background Technology

[0002] As a core piece of equipment in the power system, the reliable operation of transformers is crucial to the safety of the power grid.

[0003] Currently, most mainstream fault detection methods rely on single or a few data sources, such as oil chromatography (DGA), partial discharge monitoring, or electrical protection signals. However, due to the complexity of transformer fault mechanisms, the state information reflected by a single data source is limited, easily leading to misjudgment or missed diagnosis. At the same time, existing methods that attempt to integrate multi-source data often use simple feature splicing combined with shallow classifiers or fixed-architecture deep learning models, failing to design differentiated deep feature extraction paths based on the physical characteristics of different data sources. This makes it difficult to achieve deep nonlinear correlation mining between multi-source information and optimization of model performance, limiting the accuracy of identifying complex faults and early weak faults. Summary of the Invention

[0004] To address the problem of insufficient diagnostic accuracy caused by the difficulty in mining deep nonlinear correlations between multi-source information and optimizing model performance in existing technologies, this invention provides a transformer fault detection method, system, medium, and processor based on an improved chimpanzee optimization algorithm. This method can fully identify the complementarity and deep correlations between multi-physics field signals, thereby improving diagnostic accuracy and early fault identification capabilities. The specific technical solution is as follows: This invention provides a transformer fault detection method based on an improved chimpanzee optimization algorithm, comprising the following steps: Acquire multi-source heterogeneous monitoring data of the transformer, including transformer power loss, differential current, dissolved gas chromatography data in oil, partial discharge ultrasonic signals, transient ground voltage or current signals, partial discharge ultra-high frequency signals, and load status data; perform source-specific preprocessing and feature extraction on the multi-source heterogeneous monitoring data to form a standardized feature set; Based on the constructed multi-branch deep neural network fusion model, cross-modal feature extraction and fusion are performed on the standardized feature set; The chimpanzee optimization algorithm based on chaotic mapping is adopted, and the diagnostic performance of the multi-branch convolutional neural network model is used as the optimization target. The parameters of the multi-branch convolutional neural network model are globally adaptively optimized to obtain the optimized fault diagnosis model. Input the real-time monitoring data of the transformer to be diagnosed into the optimized fault diagnosis model, and output the fault type diagnosis result; The constructed multi-branch deep neural network fusion model has multiple independent input branches, each used to process different heterogeneous data of the feature set. Within each branch, at least one convolutional layer and pooling layer are used for local feature abstraction. At at least two different depth levels of the network, the feature maps output by each branch are concatenated in the channel dimension to achieve multi-level feature fusion. The final fused features are mapped to the fault category space through a fully connected layer, and the probability distribution of fault types is output through a softmax function.

[0005] Preferably, the step of performing source-specific preprocessing and feature extraction on the multi-source heterogeneous monitoring data includes: The partial discharge UHF signal and the ultrasonic signal are subjected to wavelet denoising and converted into a phase-amplitude-frequency three-dimensional feature spectrum. Based on the dissolved gas chromatographic data in the oil, at least three sets of characteristic gas ratios are calculated to form a fault feature vector. The differential current, transient ground voltage / current, load status, and transformer power loss data are normalized after outlier removal and timing alignment.

[0006] Preferably, the multi-branch deep neural network fusion model includes at least three independent input branches, each used to process different heterogeneous data of the feature set: A one-dimensional convolutional branch is used to process the combined electrical timing sequence consisting of transformer power loss, differential current, load state, and transient ground voltage / current data; A two-dimensional convolutional branch is used to process the three-dimensional feature spectrum generated from partial discharge ultrasonic signals and ultra-high frequency signals; A feature vector processing branch is used to process the fault feature vectors calculated from dissolved gas chromatography data in oil.

[0007] Preferably, the step of employing a chimpanzee optimization algorithm based on chaotic mapping, with the diagnostic performance of the multi-branch convolutional neural network model as the optimization objective, involves globally adaptively optimizing the parameters of the multi-branch convolutional neural network model to obtain an optimized fault diagnosis model, including: Initialize the population, with each individual representing a set of latent parameter vectors of the multi-branch convolutional neural network model; The fitness of each individual in the population is evaluated, and the fitness is determined based on the model's performance metrics on a fault diagnosis validation set; Based on fitness, individuals in the population were divided into attack, drive-away, interception, and pursuit groups, and a leader was selected for each group. Update the positions of other individuals in the population based on the information of each group leader and the introduced chaotic mapping perturbation; The fitness evaluation, grouping, and position update are iteratively performed until the termination condition is met. The model parameters corresponding to the historical best individual are output as the final optimized parameter configuration of the multi-branch convolutional neural network model, resulting in the optimized fault diagnosis model.

[0008] Preferably, the chaotic mapping is selected from one of the Logistic mapping, Tent mapping, or Chebyshev mapping.

[0009] Preferably, before the feature fusion layer of the multi-branch deep neural network fusion model, a cross-modal attention coordination module is set up. The execution steps of the cross-modal attention coordination module are as follows: Receive the intermediate layer feature maps extracted from each branch; Calculate the global average pooling vector of each branch feature map and concatenate them to form a joint representation vector; The joint representation vector is input into a lightweight fully connected network to generate a set of attention weights equal to the number of branches; The attention weights are used to channel-weight the feature maps of each branch.

[0010] Preferably, the improved chimpanzee optimization algorithm further includes: During the training phase, the number of samples for each fault category in the training dataset is counted, and the category weights are calculated. When calculating the fitness of an individual, the multi-branch deep neural network fusion model is initialized using the model parameters corresponding to that individual. Perform a forward propagation on the training set and compute the loss value of the current model using a weighted cross-entropy loss function that incorporates the class weights; The calculated loss value is converted and used as the fitness value for that individual.

[0011] This invention also provides a transformer fault detection system based on an improved chimpanzee optimization algorithm, which, using the aforementioned method, includes: The data acquisition unit is used to acquire multi-source heterogeneous monitoring data of the transformer, including transformer power loss, differential current, dissolved gas chromatography data in oil, partial discharge ultrasonic signal, transient ground voltage or current signal, partial discharge ultra-high frequency signal, and load status data; the data processing unit is used to perform source-specific preprocessing and feature extraction on the multi-source heterogeneous monitoring data to form a standardized feature set. The feature extraction unit is used to perform cross-modal feature extraction and fusion on the standardized feature set based on the constructed multi-branch deep neural network fusion model. The constructed multi-branch deep neural network fusion model has multiple independent input branches, each used to process different heterogeneous data from the feature set. Within each branch, at least one convolutional layer and pooling layer are used for local feature abstraction. At at least two different depth levels of the network, the feature maps output by each branch are concatenated along the channel dimension to achieve multi-level feature fusion. The final fused features are mapped to the fault category space through a fully connected layer, and the probability distribution of the fault type is output through a softmax function. The algorithm optimization unit is used to perform global adaptive optimization of the parameters of the multi-branch convolutional neural network model by adopting the chimpanzee optimization algorithm based on chaotic mapping, with the diagnostic performance of the multi-branch convolutional neural network model as the optimization target, to obtain the optimized fault diagnosis model. The fault output unit is used to input the real-time monitoring data of the transformer to be diagnosed into the optimized fault diagnosis model and output the fault type diagnosis result.

[0012] The present invention also provides a computer-readable storage medium comprising a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to execute the aforementioned transformer fault detection method based on an improved chimpanzee optimization algorithm.

[0013] The present invention also provides a processor for running a program, wherein the program executes the steps of the aforementioned transformer fault detection method based on an improved chimpanzee optimization algorithm.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention discloses a transformer fault detection method based on an improved chimpanzee optimization algorithm. This method employs a multi-branch deep neural network architecture to perform parallel and multi-level fusion of multi-source heterogeneous data, and introduces a chimpanzee optimization algorithm with integrated chaotic mapping to jointly optimize network parameters and structure globally. It fully identifies the complementarity and deep correlation between multi-physics field signals, improving the feature representation of complex faults and early weak faults, thereby enhancing diagnostic accuracy and early fault identification capabilities. Simultaneously, the optimization algorithm automatically searches for the optimal model configuration adapted to different data characteristics and fault modes, reducing the model's dependence on specific datasets or manual parameter tuning, and improving its adaptability to different operating conditions and transformer models. Attached Figure Description

[0015] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. 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 to scale.

[0016] Figure 1 This is a flowchart of the transformer fault detection method based on the improved chimpanzee optimization algorithm of the present invention.

[0017] Figure 2 This is a flowchart illustrating the optimization process of the improved chimpanzee optimization algorithm of this invention.

[0018] Figure 3 This is a schematic diagram of the transformer fault detection system based on the improved chimpanzee optimization algorithm of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be understood that, when used in this specification, the terms “comprising” and “including” indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0021] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should also be further understood that the term "and / or" as used in this specification refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes such combinations.

[0023] Please refer to the following examples. Figures 1 to 3 .

[0024] Please see Figure 1 This invention provides a transformer fault detection method based on an improved chimpanzee optimization algorithm, comprising the following steps: Step S1: Acquire multi-source heterogeneous monitoring data of the transformer, including transformer power loss, differential current, dissolved gas chromatography data in oil, partial discharge ultrasonic signals, transient ground voltage or current signals, partial discharge ultra-high frequency signals, and load status data; the partial discharge ultra-high frequency signals and ultrasonic signals reflect the insulation status of the transformer; the differential current and transient ground voltage / current reflect the electrical connection status and fault characteristics of the transformer; the dissolved gas chromatography data in oil reflects the thermal faults and discharge faults of the transformer; the load status data reflects the operating conditions of the transformer; and the power loss data reflects the loss status of the transformer.

[0025] Step S2: Perform source-specific preprocessing and feature extraction on the multi-source heterogeneous monitoring data to form a standardized feature set; The source data are cleaned and aligned, and normalized using the min-max normalization method. Key features related to the fault are then extracted.

[0026] Specifically, the source-specific preprocessing and feature extraction of the multi-source heterogeneous monitoring data includes: The partial discharge UHF signal and the ultrasonic signal are subjected to wavelet denoising and converted into a phase-amplitude-frequency three-dimensional feature spectrum. Based on the dissolved gas chromatographic data in the oil, at least three sets of characteristic gas ratios are calculated to form a fault feature vector. The differential current, transient ground voltage / current, load status, and transformer power loss data are normalized after outlier removal and timing alignment.

[0027] Step S3: Based on the constructed multi-branch deep neural network fusion model, perform cross-modal feature extraction and fusion on the standardized feature set; The constructed multi-branch deep neural network fusion model has multiple independent input branches, each used to process different heterogeneous data of the feature set. Within each branch, at least one convolutional layer and pooling layer are used for local feature abstraction. At at least two different depth levels of the network, the feature maps output by each branch are concatenated in the channel dimension to achieve multi-level feature fusion. The final fused features are mapped to the fault category space through a fully connected layer, and the probability distribution of fault types is output through a softmax function.

[0028] Specifically, the multi-branch deep neural network fusion model includes at least three independent input branches, each used to process different heterogeneous data from the feature set. This is achieved by inputting three types of data from the standardized feature set into the three independent branches of the model in parallel. A one-dimensional convolutional branch is used to process the combined electrical timing sequence consisting of transformer power loss, differential current, load state, and transient ground voltage / current data; A two-dimensional convolutional branch is used to process the three-dimensional feature spectrum generated from partial discharge ultrasonic signals and ultra-high frequency signals; A feature vector processing branch is used to process the fault feature vectors calculated from dissolved gas chromatography data in oil.

[0029] In this step, the convolutional neural network model adopts a multi-branch input architecture. The input layer receives a standardized feature set. The data is grouped according to physical characteristics. Power loss, differential current, load state and transient ground current / voltage are organized into a one-dimensional time series. The phase-resolved spectrum generated after processing of ultrasonic signals and ultra-high frequency signals is used as a two-dimensional image input. Oil chromatography data is used as a feature vector. Each branch uses an independent sub-network for front-end feature extraction. Each branch extracts local features from the input data through a series of optimized convolutional layers. The kernel size is selected based on the characteristics of each data source. For the one-dimensional time-series branch, a one-dimensional convolutional layer is used to extract fault mode features in the time dimension. For the two-dimensional spectrogram branch, a two-dimensional convolutional layer is used to extract spatial features. This enables the network to extract fault mode features from time-series data and trend change features from spatial data. For the vector branch, features can be directly used as features in subsequent fusion. By setting three independent input branches, each branch is adapted to three completely different data structures: time-series sequences, image spectrograms, and feature vectors. This allows each branch to use the neural network layer most suitable for its data characteristics for feature extraction. This avoids information loss or structural distortion caused by forcibly converting heterogeneous data into a single format, ensuring that the most relevant fault features are extracted from each data source. Each branch acts as an independent feature learner, capable of deeply exploring the complex patterns and subtle anomalies within its corresponding data mode. For example, the one-dimensional convolutional branch can effectively capture transient or periodic changes in electrical quantities related to faults; the two-dimensional convolutional branch excels at identifying spatial distribution and texture patterns in partial discharge spectra.

[0030] A pooling layer is set after the convolutional layer to downsample the convolutional output, reduce the feature dimension, retain important local feature information, reduce computational complexity, and avoid overfitting; Feature fusion is performed in both the intermediate and decision layers of the network. A concatenation fusion method is adopted. In the intermediate layer, the abstract feature maps extracted from each branch to the same semantic level are concatenated in the channel dimension. In the decision layer, the high-level feature vectors finally output by each branch are concatenated and input into the subsequent fully connected network to integrate multiple information from various types of data, thereby enhancing feature representation capabilities and improving detection accuracy. After each layer, the ReLU nonlinear activation function is applied to activate the model, increasing its nonlinear expressive power and enabling the network to adapt to more complex fault detection tasks. The fused high-level features are mapped to the fault category space through a fully connected layer, and the final output layer generates a fault classification result, which represents the fault type of the transformer and its probability distribution. The result is output through the softmax activation function, and the presence and possible types of faults are determined based on the output.

[0031] Step S4: Using the Chimp Optimization Algorithm based on chaotic mapping, with the diagnostic performance of the multi-branch convolutional neural network model as the optimization objective, the parameters of the multi-branch convolutional neural network model are globally adaptively optimized to obtain the optimized fault diagnosis model. Specifically, the method employs a chimpanzee optimization algorithm based on chaotic mapping, using the diagnostic performance of the multi-branch convolutional neural network model as the optimization objective, to perform global adaptive optimization of the parameters of the multi-branch convolutional neural network model, thereby obtaining an optimized fault diagnosis model, including: Initialize the population, with each individual representing a set of latent parameter vectors of the multi-branch convolutional neural network model; The fitness of each individual in the population is evaluated, and the fitness is determined based on the model's performance metrics on a fault diagnosis validation set; Based on fitness, individuals in the population were divided into attack, drive-away, interception, and pursuit groups, and a leader was selected for each group. Update the positions of other individuals in the population based on the information of each group leader and the introduced chaotic mapping perturbation; The fitness evaluation, grouping, and position update are iteratively performed until the termination condition is met. The model parameters corresponding to the historical best individual are output as the final optimized parameter configuration of the multi-branch convolutional neural network model, resulting in the optimized fault diagnosis model.

[0032] The chaotic mapping is selected from one of the Logistic mapping, Tent mapping, or Chebyshev mapping.

[0033] In this step, please refer to Figure 2 A certain number of chimpanzee individuals are randomly generated, and each individual is represented as a potential neural network weight vector. The population size and the initial range of individuals are set according to the complexity of the neural network and the scale of the training data. By inputting the weight vector of each individual into the neural network, forward propagation is performed on the fault detection dataset to calculate its corresponding fitness value. The fitness value represents the performance of the neural network in the fault detection task, and can be measured by error rate, accuracy, F1 score or other evaluation metrics. Based on their fitness values, the population is divided into four groups: attack, drive, intercept, and chase. The four individuals with the best fitness are selected and designated as the leaders of the attack, drive, intercept, and chase groups, respectively. These four leaders collectively guide the optimization process, controlling the search behavior of other individuals. An individual's fitness is typically represented by an error function, assuming the network output is... The actual value is The fitness function f can be calculated using the mean squared error (MSE): Where w is the network weight and N is the number of samples. It is the predicted value of the i-th sample; By simulating the group hunting behavior rules of chimpanzees, each individual continuously adjusts its weight vector based on the fitness information of its group leader, making the individual's search range more precise and moving closer to the optimal solution. The individual's position update is based on the information of the four group leaders, and the update rules are as follows: in, and For control parameters, It is a random perturbation used to introduce an exploratory component into the search space. k is the number of individuals participating in the update. It is the weight of the i-th individual; The four roles update their positions based on the fitness information of the individuals around them. Let i be the position of the i-th individual. The weights for the four group leaders are updated according to the following rules: By combining exploration and development strategies and introducing chaotic mapping, if the random number is ≥0.5, the chaotic value is used for updating, simulating motivational behavior, gradually improving the individual's weight vector, and performing global search and local optimization throughout the population to avoid getting trapped in local optima; Repeat the fitness evaluation, four-role update and position update until the fitness value reaches the preset threshold, thereby obtaining the optimized neural network weight parameters; The individual with the best historical fitness, usually the leader of the attack group, is selected as the optimal weight configuration for the neural network. This weight configuration, including but not limited to parameters such as kernel size, number of convolutional layers, and number of neurons per layer, is applied to the convolutional neural network model to complete the transformer fault detection task.

[0034] Due to the excessively large parameter space of multi-branch convolutional neural network models, traditional manual or grid search methods are extremely inefficient. The chimpanzee optimization algorithm, based on chaotic mapping, achieves global adaptation and automated optimization of model parameters, shortening the development cycle. The introduced chaotic perturbation effectively prevents the algorithm from getting trapped in local optima in the later stages of evolution, enhancing its global exploration capability in complex parameter spaces, thus making it more likely to find a better-performing model parameter configuration.

[0035] Step S5: Input the real-time monitoring data of the transformer to be diagnosed into the optimized fault diagnosis model and output the fault type diagnosis result; This invention discloses a transformer fault detection method based on an improved chimpanzee optimization algorithm. This method employs a multi-branch deep neural network architecture to perform parallel and multi-level fusion of multi-source heterogeneous data, and introduces a chimpanzee optimization algorithm with integrated chaotic mapping to jointly optimize network parameters and structure globally. It fully identifies the complementarity and deep correlation between multi-physics field signals, improving the feature representation of complex faults and early weak faults, thereby enhancing diagnostic accuracy and early fault identification capabilities. Simultaneously, the optimization algorithm automatically searches for the optimal model configuration adapted to different data characteristics and fault modes, reducing the model's dependence on specific datasets or manual parameter tuning, and improving its adaptability to different operating conditions and transformer models.

[0036] Specifically, in a preferred embodiment of this application, a cross-modal attention coordination module is set before the feature fusion layer of the multi-branch deep neural network fusion model. The execution steps of the cross-modal attention coordination module are as follows: Receive the intermediate layer feature maps extracted from each branch; Calculate the global average pooling vector of each branch feature map and concatenate them to form a joint representation vector; The joint representation vector is input into a lightweight fully connected network to generate a set of attention weights equal to the number of branches; The attention weights are used to channel-weight the feature maps of each branch.

[0037] In this preferred embodiment, to improve the diagnostic accuracy of multi-source information fusion, a cross-modal attention coordination module is added before the feature fusion layer of the multi-branch deep neural network fusion model. This module acts as a lightweight adaptive gateway, aiming to dynamically evaluate and adjust the contribution of each data source branch to the current diagnostic task. Its specific implementation process is as follows: The module receives intermediate layer feature maps extracted from various branches, such as the one-dimensional time series branch, the two-dimensional image branch, and the feature vector branch. The feature maps already contain primary fault characterization information abstracted from the original data, but they are not fused.

[0038] For each branch's input feature map, a global average pooling operation is performed. This operation compresses each 2D or 1D feature map into a scalar vector that represents the global statistics and overall activation level of that branch's feature map. Then, the pooling vectors from all branches are concatenated along the feature dimension to form a joint global representation vector.

[0039] The joint global representation vector obtained in the previous step is input into a lightweight fully connected neural network. Through learning, based on the current global context information, it outputs a numerical vector with a dimension equal to the number of branches. Each value in this vector is the attention weight of the corresponding branch. These weights are normalized by the Softmax function, and their sum is 1, intuitively reflecting the importance distribution of information in each branch in the current fault diagnosis context.

[0040] The calculated attention weights are applied to the original intermediate layer feature maps of the corresponding branches. Specifically, the weight value of each branch is multiplied by each channel of its feature map. This is equivalent to recalibrating the features of each branch at the channel level, increasing the amplitude of features from important branches and suppressing the amplitude of features from minor or irrelevant branches.

[0041] By introducing this cross-modal attention coordination module, the traditional fusion method's strategy of using fixed weights or simple concatenation for each data source is changed. The module can automatically learn and assign optimal fusion weights based on specific input samples. For example, when a sample suggests a possible partial discharge fault, the module automatically assigns higher weights to the branch reflecting partial discharge characteristics. Through a weighting mechanism, the strength of the signal most relevant to the current fault is enhanced, while signals that may contain noise or irrelevant information are weakened. This results in higher quality features received by the fusion layer with less noise interference, thus making it easier for the classifier to make accurate judgments, especially when distinguishing between complex faults with similar symptoms or weak early faults, where the accuracy improvement is more significant.

[0042] Specifically, in a preferred embodiment of this application, the improved chimpanzee optimization algorithm further includes: During the training phase, the number of samples for each fault category in the training dataset is counted, and the category weights are calculated. During the training phase, each fault category in the training dataset is statistically analyzed. Number of samples Based on the stated sample size, the weight of each category is calculated using the following formula. : in, The total number of samples in the training set. This represents the total number of fault categories. According to this formula, rarer fault categories with fewer samples receive a higher calculated weight. The larger the value.

[0043] When calculating the fitness of an individual, the multi-branch deep neural network fusion model is initialized using the model parameters corresponding to that individual. Perform a forward propagation on the training set and compute the loss value of the current model using a weighted cross-entropy loss function that incorporates the class weights; Perform one forward propagation on the training set, and calculate the loss value of the current model using a weighted cross-entropy loss function that incorporates the class weights. The formula for calculating the weighted cross-entropy loss function is as follows: in, For batch size, For the sample The actual fault category, The preset weight for this category, Predict samples for the model The probability of belonging to its true category.

[0044] The calculated loss value is converted and used as the fitness value for that individual.

[0045] In this preferred embodiment, during the evolution of the optimization algorithm, the model parameters will be automatically guided to evolve in a direction that can reduce the classification error of rare fault samples, thereby directly improving the overall robustness of the diagnostic model to imbalanced datasets.

[0046] Please see Figure 3 This invention also provides a transformer fault detection system based on an improved chimpanzee optimization algorithm, which applies the aforementioned method and includes: The data acquisition unit is used to acquire multi-source heterogeneous monitoring data of the transformer, including transformer power loss, differential current, dissolved gas chromatography data in oil, partial discharge ultrasonic signal, transient ground voltage or current signal, partial discharge ultra-high frequency signal, and load status data; the data processing unit is used to perform source-specific preprocessing and feature extraction on the multi-source heterogeneous monitoring data to form a standardized feature set. The feature extraction unit is used to perform cross-modal feature extraction and fusion on the standardized feature set based on the constructed multi-branch deep neural network fusion model. The constructed multi-branch deep neural network fusion model has multiple independent input branches, each used to process different heterogeneous data from the feature set. Within each branch, at least one convolutional layer and pooling layer are used for local feature abstraction. At at least two different depth levels of the network, the feature maps output by each branch are concatenated along the channel dimension to achieve multi-level feature fusion. The final fused features are mapped to the fault category space through a fully connected layer, and the probability distribution of the fault type is output through a softmax function. The algorithm optimization unit is used to perform global adaptive optimization of the parameters of the multi-branch convolutional neural network model by adopting the chimpanzee optimization algorithm based on chaotic mapping, with the diagnostic performance of the multi-branch convolutional neural network model as the optimization target, to obtain the optimized fault diagnosis model. The fault output unit is used to input the real-time monitoring data of the transformer to be diagnosed into the optimized fault diagnosis model and output the fault type diagnosis result.

[0047] The functional explanation of each unit in this embodiment is the same as that of a transformer fault detection method based on an improved chimpanzee optimization algorithm, and the technical effects are the same, so it will not be repeated here.

[0048] This invention also provides a computer-readable storage medium, which includes a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to execute the aforementioned transformer fault detection method based on an improved chimpanzee optimization algorithm.

[0049] The technical effect of this embodiment is the same as that of a transformer fault detection method based on an improved chimpanzee optimization algorithm in the embodiment, and will not be repeated here.

[0050] This invention can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0051] This invention also provides a processor for running a program, wherein the program executes the steps of the aforementioned transformer fault detection method based on an improved chimpanzee optimization algorithm.

[0052] The technical effects of this embodiment are the same as those of the transformer fault detection method based on the improved chimpanzee optimization algorithm in Embodiment 1, and will not be repeated here.

[0053] In this embodiment, the processor may be a central processing unit (CPU), a controller, a microcontroller, or other data processing chip.

[0054] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0055] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0056] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0057] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the specification of the present invention.

Claims

1. A transformer fault detection method based on an improved chimpanzee optimization algorithm, characterized in that, Includes the following steps: Acquire multi-source heterogeneous monitoring data of the transformer, including transformer power loss, differential current, dissolved gas chromatography data in oil, partial discharge ultrasonic signals, transient ground voltage or current signals, partial discharge ultra-high frequency signals, and load status data; perform source-specific preprocessing and feature extraction on the multi-source heterogeneous monitoring data to form a standardized feature set; Based on the constructed multi-branch deep neural network fusion model, cross-modal feature extraction and fusion are performed on the standardized feature set; The chimpanzee optimization algorithm based on chaotic mapping is adopted, and the diagnostic performance of the multi-branch convolutional neural network model is used as the optimization target. The parameters of the multi-branch convolutional neural network model are globally adaptively optimized to obtain the optimized fault diagnosis model. Input the real-time monitoring data of the transformer to be diagnosed into the optimized fault diagnosis model, and output the fault type diagnosis result; The constructed multi-branch deep neural network fusion model has multiple independent input branches, each used to process different heterogeneous data of the feature set. Within each branch, at least one convolutional layer and pooling layer are used for local feature abstraction. At at least two different depth levels of the network, the feature maps output by each branch are concatenated in the channel dimension to achieve multi-level feature fusion. The final fused features are mapped to the fault category space through a fully connected layer, and the probability distribution of fault types is output through a softmax function.

2. The transformer fault detection method based on the improved chimpanzee optimization algorithm according to claim 1, characterized in that, The process of performing source-specific preprocessing and feature extraction on the multi-source heterogeneous monitoring data includes: The partial discharge UHF signal and the ultrasonic signal are subjected to wavelet denoising and converted into a phase-amplitude-frequency three-dimensional feature spectrum. Based on the dissolved gas chromatographic data in the oil, at least three sets of characteristic gas ratios are calculated to form a fault feature vector. The differential current, transient ground voltage / current, load status, and transformer power loss data are normalized after outlier removal and timing alignment.

3. The transformer fault detection method based on the improved chimpanzee optimization algorithm according to claim 2, characterized in that, The multi-branch deep neural network fusion model includes at least three independent input branches, each used to process different heterogeneous data in the feature set: A one-dimensional convolutional branch is used to process the combined electrical timing sequence consisting of transformer power loss, differential current, load state, and transient ground voltage / current data; A two-dimensional convolutional branch is used to process the three-dimensional feature spectrum generated from partial discharge ultrasonic signals and ultra-high frequency signals; A feature vector processing branch is used to process the fault feature vectors calculated from dissolved gas chromatography data in oil.

4. The transformer fault detection method based on the improved chimpanzee optimization algorithm according to claim 1, characterized in that, The method employs a chimpanzee optimization algorithm based on chaotic mapping, with the diagnostic performance of the multi-branch convolutional neural network model as the optimization objective. It performs global adaptive optimization of the parameters of the multi-branch convolutional neural network model to obtain an optimized fault diagnosis model, including: Initialize the population, with each individual representing a set of latent parameter vectors of the multi-branch convolutional neural network model; The fitness of each individual in the population is evaluated, and the fitness is determined based on the model's performance metrics on a fault diagnosis validation set; Based on fitness, individuals in the population were divided into attack, drive-away, interception, and pursuit groups, and a leader was selected for each group. Update the positions of other individuals in the population based on the information of each group leader and the introduced chaotic mapping perturbation; The fitness evaluation, grouping, and position update are iteratively performed until the termination condition is met. The model parameters corresponding to the historical best individual are output as the final optimized parameter configuration of the multi-branch convolutional neural network model, resulting in the optimized fault diagnosis model.

5. A transformer fault detection method based on an improved chimpanzee optimization algorithm according to claim 4, characterized in that, The chaotic mapping is selected from one of the Logistic mapping, Tent mapping, or Chebyshev mapping.

6. A transformer fault detection method based on an improved chimpanzee optimization algorithm according to claim 3, characterized in that, Before the feature fusion layer of the multi-branch deep neural network fusion model, a cross-modal attention coordination module is set up. The execution steps of the cross-modal attention coordination module are as follows: Receive the intermediate layer feature maps extracted from each branch; Calculate the global average pooling vector of each branch feature map and concatenate them to form a joint representation vector; The joint representation vector is input into a lightweight fully connected network to generate a set of attention weights equal to the number of branches; The attention weights are used to channel-weight the feature maps of each branch.

7. A transformer fault detection method based on an improved chimpanzee optimization algorithm according to claim 4, characterized in that, The improved chimpanzee optimization algorithm also includes: During the training phase, the number of samples for each fault category in the training dataset is counted, and the category weights are calculated. When calculating the fitness of an individual, the multi-branch deep neural network fusion model is initialized using the model parameters corresponding to that individual. Perform a forward propagation on the training set and compute the loss value of the current model using a weighted cross-entropy loss function that incorporates the class weights; The calculated loss value is converted and used as the fitness value for that individual.

8. A transformer fault detection system based on an improved chimpanzee optimization algorithm, characterized in that, The method described by any one of claims 1 to 7 includes: The data acquisition unit is used to acquire multi-source heterogeneous monitoring data of the transformer, including transformer power loss, differential current, dissolved gas chromatography data in oil, partial discharge ultrasonic signal, transient ground voltage or current signal, partial discharge ultra-high frequency signal, and load status data; the data processing unit is used to perform source-specific preprocessing and feature extraction on the multi-source heterogeneous monitoring data to form a standardized feature set. The feature extraction unit is used to perform cross-modal feature extraction and fusion on the standardized feature set based on the constructed multi-branch deep neural network fusion model. The constructed multi-branch deep neural network fusion model has multiple independent input branches, each used to process different heterogeneous data from the feature set. Within each branch, at least one convolutional layer and pooling layer are used for local feature abstraction. At at least two different depth levels of the network, the feature maps output by each branch are concatenated along the channel dimension to achieve multi-level feature fusion. The final fused features are mapped to the fault category space through a fully connected layer, and the probability distribution of the fault type is output through a softmax function. The algorithm optimization unit is used to perform global adaptive optimization of the parameters of the multi-branch convolutional neural network model by adopting the chimpanzee optimization algorithm based on chaotic mapping, with the diagnostic performance of the multi-branch convolutional neural network model as the optimization target, to obtain the optimized fault diagnosis model. The fault output unit is used to input the real-time monitoring data of the transformer to be diagnosed into the optimized fault diagnosis model and output the fault type diagnosis result.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform a transformer fault detection method based on an improved chimpanzee optimization algorithm as described in any one of claims 1 to 7.

10. A processor, characterized in that, The processor is used to run a program, wherein the program executes the steps of a transformer fault detection method based on an improved chimpanzee optimization algorithm as described in any one of claims 1 to 7.

Images