Method and system for detecting partial discharge of main transformer of booster station of new energy station

Abstract

The invention discloses a method and a system for detecting partial discharge of a main transformer of a booster station of a new energy field station, and relates to the technical field of power equipment detection, and the method comprises the steps: building a main transformer working condition electromagnetic field coupling model; acquiring main transformer operation condition data, a characteristic gas concentration signal and a multi-mode detection signal of the booster station main transformer in real time; constructing a multi-modal signal feature vector; determining a main transformer partial discharge first probability; determining a second probability of main transformer partial discharge; and performing deep partial discharge diagnosis under working condition fluctuation sensitive fusion to obtain a main transformer partial discharge diagnosis result. The technical problems that in the prior art, new energy station booster station main transformer partial discharge detection is prone to being interfered by working condition fluctuation, the detection dimension is single, and the accuracy and stability of the diagnosis result are insufficient are solved, accurate detection and multi-dimensional diagnosis of new energy station booster station main transformer partial discharge are achieved, and the accuracy and stability of the diagnosis result are improved. And the accuracy and stability of partial discharge detection are improved.

Description

Method and system for partial discharge detection of main transformer in booster substation of new energy power plants Technical Field

[0001] This invention relates to the field of power equipment testing technology, specifically to a method and system for detecting partial discharge in the main transformer of a new energy power station booster station. Background Technology

[0002] As the core equipment for power transmission, the main transformer of the substation in a new energy power plant directly affects the stability of the power supply and the safety of the power grid. Partial discharge is a typical precursor to insulation degradation in the main transformer. If not detected in time, it can easily lead to equipment failure or even power outages. Existing partial discharge detection technologies for main transformers mostly use single electrical signals or gas detection methods, which have limited detection dimensions and are easily affected by fluctuations in the operating conditions of the main transformer, such as load, voltage, and oil temperature. They also lack targeted operating condition coupling compensation mechanisms and lack reasonable fusion processing of multi-source detection signals, resulting in large deviations in discharge probability prediction and insufficient accuracy and stability of diagnostic results. These technologies are difficult to adapt to the complex and variable operating conditions of main transformers in new energy power plants and cannot meet the needs of refined and highly reliable on-site detection and maintenance.

[0003] Existing technologies for detecting partial discharge in the main transformer of the booster station of new energy power plants are susceptible to interference from operating condition fluctuations and have a single detection dimension, resulting in insufficient accuracy and stability of diagnostic results. Summary of the Invention

[0004] This application provides a method and system for detecting partial discharge in the main transformer of the booster station of new energy power plants, which addresses the technical problems in the prior art where the detection of partial discharge in the main transformer of the booster station of new energy power plants is easily affected by the fluctuation of operating conditions and the detection dimension is single, resulting in insufficient accuracy and stability of the diagnostic results.

[0005] In view of the above problems, this application provides a method and system for detecting partial discharge of the main transformer in the booster station of new energy power plants.

[0006] The first aspect of this application provides a method for detecting partial discharge in the main transformer of a booster station in a new energy power plant. The method includes: performing a certainty-based fusion modeling of the electromagnetic field distribution based on the main transformer's operating condition log set at the booster station, establishing an electromagnetic field coupling model for the main transformer's operating condition; acquiring, in real-time, the main transformer's operating condition data, characteristic gas concentration signals, and multimodal detection signals using a sensor network; performing dynamic compensation for the multimodal detection signals based on the main transformer's operating condition data and the electromagnetic field coupling model, constructing a multimodal signal feature vector; predicting the probability of partial discharge in the main transformer based on the multimodal signal feature vector, determining a first probability of partial discharge; predicting the probability of partial discharge in the main transformer based on the characteristic gas concentration signals, determining a second probability of partial discharge; and performing deep partial discharge diagnosis under condition fluctuation-sensitive fusion based on the first and second probabilities of partial discharge, obtaining a partial discharge diagnosis result for the main transformer.

[0007] A second aspect of this application provides a partial discharge detection system for the main transformer of a new energy power plant's booster station. The system includes: a coupling model establishment module, used to perform electromagnetic field distribution certainty fusion modeling based on the main transformer's operating condition log set from the booster station's main transformer, establishing a main transformer operating condition electromagnetic field coupling model; a main transformer operating condition acquisition module, used to acquire in real-time main transformer operating condition data, characteristic gas concentration signals, and multi-modal detection signals from the booster station's main transformer using a sensor network; and a multi-modal signal feature vector construction module, used to construct feature vectors based on the main transformer operating condition data and the main transformer operating condition electromagnetic field coupling model. The system performs dynamic compensation for operating condition coupling on the state detection signal to construct a multi-mode signal feature vector; a main transformer partial discharge first probability determination module is used to predict the partial discharge probability of the main transformer of the step-up substation based on the multi-mode signal feature vector to determine the first probability of main transformer partial discharge; a main transformer partial discharge second probability determination module is used to predict the partial discharge probability of the main transformer of the step-up substation based on the characteristic gas concentration signal to determine the second probability of main transformer partial discharge; a main transformer partial discharge diagnosis result acquisition module is used to perform deep partial discharge diagnosis under operating condition fluctuation sensitive fusion based on the first probability and the second probability of main transformer partial discharge to acquire the main transformer partial discharge diagnosis result.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: Electromagnetic field distribution is confidently fused and modeled based on the main transformer operating condition log set of the main transformer in the booster station of a new energy power plant, establishing a coupling model of the main transformer's operating condition electromagnetic field; the main transformer's operating condition data, characteristic gas concentration signals, and multi-modal detection signals are acquired in real time using a sensor network; the multi-modal detection signals are dynamically compensated for based on the main transformer's operating condition electromagnetic field coupling model, constructing a multi-modal signal feature vector; partial discharge probability prediction is performed on the main transformer in the booster station to determine the first probability of partial discharge; partial discharge probability prediction is performed on the main transformer in the booster station to determine the second probability of partial discharge; deep partial discharge diagnosis is performed under operating condition fluctuation sensitive fusion to obtain the main transformer's partial discharge diagnosis results. This achieves the technical effect of accurate detection and multi-dimensional diagnosis of partial discharge in the main transformer of the booster station in new energy power plants, improving the accuracy and stability of partial discharge detection. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 is a schematic flowchart of the partial discharge detection method for the main transformer of the new energy power station step-up substation provided in the embodiment of this application; Figure 2 is a schematic structural diagram of the partial discharge detection system for the main transformer of the new energy power station step-up substation provided in the embodiment of this application.

[0011] Figure labeling: Coupled model establishment module 10, main transformer operating condition acquisition module 20, multi-mode signal feature vector construction module 30, main transformer partial discharge first probability determination module 40, main transformer partial discharge second probability determination module 50, main transformer partial discharge diagnosis result acquisition module 60. Detailed Implementation

[0012] This application provides a method and system for detecting partial discharge in the main transformer of a new energy power plant step-up substation, which addresses the technical problems in the prior art where the detection of partial discharge in the main transformer of a new energy power plant step-up substation is easily affected by fluctuations in operating conditions and has a single detection dimension, resulting in insufficient accuracy and stability of diagnostic results.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] Example 1, as shown in Figure 1, provides a method for detecting partial discharge of the main transformer in the step-up substation of a new energy power station. The method includes: Step S100: Performing electromagnetic field distribution certainty fusion modeling based on the main transformer operating condition log set of the main transformer in the step-up substation of the new energy power station, and establishing an electromagnetic field coupling model of the main transformer operating condition.

[0015] Specifically, considering the operating characteristics of the main transformer in the booster station of a new energy power plant, a certainty fusion model of electromagnetic field distribution is carried out based on the main transformer operating condition log set. First, the log set is cleaned and the parameters are classified to split it into a main transformer operating condition sample area and an electromagnetic field sample area. Then, pairwise similarity evaluation is performed on the main transformer operating condition sample area to obtain the operating condition similarity distribution. Based on this distribution, the main transformer operating condition sample area is clustered and fused to obtain the main transformer operating condition distribution space. Based on a predetermined number of fusion rounds, pairwise similarity evaluation and clustering fusion are repeatedly performed on the main transformer operating condition distribution space to construct a typical main transformer operating condition space. Subsequently, the f-th typical main transformer operating condition vector is extracted from the typical main transformer operating condition space and matched with the electromagnetic field sample area to obtain the f-th matched electromagnetic field sample group. The frequency of occurrence of each sample in the group is counted to obtain the certainty of each matched sample. Based on this, data fusion is performed on the sample group to obtain the f-th typical electromagnetic field vector and construct the typical electromagnetic field space. Finally, through the mapping modeling of the typical main transformer operating condition space and the typical electromagnetic field space, the electromagnetic field coupling model of the main transformer operating condition is established.

[0016] Step S200: Acquire the main transformer operating condition data, characteristic gas concentration signal, and multi-modal detection signal of the main transformer of the booster station in real time according to the sensor network.

[0017] Specifically, relying on the sensor network deployed in the main transformer of the new energy power station step-up substation, multi-dimensional and real-time data acquisition of the main transformer's operating status is carried out. Simultaneously, the operating condition data of the main transformer that can reflect the actual operating status of the main transformer, characteristic gas concentration signals that can characterize the insulation status of the main transformer, and multi-mode detection signals including high-frequency pulse current signals, ultra-high frequency electromagnetic wave signals and ultrasonic signals are acquired, providing comprehensive and accurate raw data support for the subsequent detection and diagnosis of partial discharge of the main transformer.

[0018] Step S300: Based on the main transformer operating condition data, perform dynamic compensation for the multi-mode detection signal according to the main transformer operating condition electromagnetic field coupling model, and construct a multi-mode signal feature vector.

[0019] Specifically, the real-time collected operating condition data of the main transformer is first input into the established electromagnetic field coupling model of the main transformer operating condition, and the corresponding operating condition matching electromagnetic field characteristic parameters are obtained. Then, dynamic modeling is performed based on these parameters to obtain the operating condition matching electromagnetic field model. Subsequently, based on this model, the amplitude and phase attenuation offset analysis of the multi-mode detection signal is carried out, the attenuation offset features corresponding to each signal are extracted, and the multi-mode detection signal is subjected to reverse compensation processing based on these features to obtain the corrected multi-mode correction signal. Finally, feature extraction is performed on the multi-mode correction signal to complete the construction of the multi-mode signal feature vector.

[0020] Step S400: Based on the multi-mode signal feature vector, perform partial discharge probability prediction on the main transformer of the booster station to determine the first probability of partial discharge of the main transformer.

[0021] Specifically, the process first loads the multimodal signal feature sample set and the signal mapping partial discharge probability sample set corresponding to the main transformer of the substation. Based on this, the Bayesian network is trained through structure learning, parameter estimation, and probabilistic graph inference to establish a multimodal signal partial discharge probability graph mapping model. Then, the partial discharge state evolution model and state transition probability reconstruction are performed on the mapping model through a hidden Markov model to construct a multimodal signal-driven partial discharge prediction model. Finally, the constructed multimodal signal feature vector is input into the prediction model, and after variational inference and probability fusion calculation, the first probability of partial discharge of the main transformer of the substation is output.

[0022] Step S500: Based on the characteristic gas concentration signal, perform partial discharge probability prediction on the main transformer of the booster station to determine the second probability of partial discharge of the main transformer.

[0023] Specifically, the process begins by loading a characteristic gas concentration feature set and a gas concentration-mapped partial discharge probability sample set corresponding to the main transformer of the new energy power station's step-up substation. Using these two sample sets as the training foundation, the Bayesian network undergoes a full-process training process, including structure learning, parameter estimation, and probabilistic graphical inference. Structure learning determines the probabilistic correlation structure between gas concentration features and partial discharge probabilities. Parameter estimation optimizes the probability distribution parameters within the network. Combined with probabilistic graphical inference, the model's inference logic is trained, ultimately establishing a characteristic gas concentration partial discharge probability graphical mapping model. Then, a Hidden Markov Model is introduced to model the partial discharge state evolution of this probabilistic graphical mapping model, constructing a model of the main transformer's partial discharge from zero to present, and from mild to severe. The state evolution sequence simultaneously reconstructs the state transition probabilities between different partial discharge states, compensating for the shortcomings of static probability mapping, and constructs a characteristic gas concentration-driven partial discharge prediction model that combines concentration probability correlation and dynamic evolution of partial discharge states. Subsequently, feature extraction is performed on the characteristic gas concentration signals collected in real time by the sensor network to obtain standardized characteristic gas concentration feature vectors, which are input into the characteristic gas concentration-driven partial discharge prediction model. The potential correlation between gas concentration features and partial discharge probability is accurately derived through the variational inference algorithm inside the model. Then, the partial discharge probability results corresponding to multi-dimensional gas concentration features are integrated through probability fusion calculation, and finally the second probability of partial discharge of the main transformer of the step-up substation is output.

[0024] Step S600: Perform deep partial discharge diagnosis under the condition fluctuation sensitive fusion based on the first probability of partial discharge of the main transformer and the second probability of partial discharge of the main transformer, and obtain the diagnosis result of partial discharge of the main transformer.

[0025] Specifically, the initial conditions for partial discharge probability fusion, including initial weights for the first and second probabilities of partial discharge, are first obtained. Then, based on the main transformer operating condition data, the sensitivity to operating condition fluctuations is evaluated for the multi-modal signal feature vector and the characteristic gas concentration signal, respectively, to obtain the first and second sensitivities to operating condition fluctuations. The initial fusion conditions are then adaptively adjusted accordingly to obtain the current fusion weight conditions and complete the fusion calculation of the first and second probabilities of partial discharge in the main transformer, resulting in the main transformer partial discharge diagnosis probability. Next, it is determined whether this diagnosis probability is greater than or equal to the predetermined partial discharge diagnosis probability. If so, the partial discharge depth diagnosis model is activated. Subsequently, the characteristic gas concentration signal and the multi-modal signal feature vector are fused to construct the main transformer monitoring feature vector. Finally, the main transformer monitoring feature vector is input into the partial discharge depth diagnosis model to complete the deep partial discharge diagnosis under operating condition fluctuation-sensitive fusion, ultimately obtaining the main transformer partial discharge diagnosis result.

[0026] In one possible implementation, step S100 further includes: step S110: performing data cleaning and parameter classification on the main transformer operating condition log set to obtain the main transformer operating condition sample area and electromagnetic field sample area.

[0027] Step S120: Perform pairwise similarity evaluation based on the main transformer operating condition sample area to obtain the operating condition similarity distribution.

[0028] Step S130: Cluster and fuse the main transformer operating condition sample area according to the similarity distribution of the operating conditions to obtain the main transformer operating condition distribution space.

[0029] Step S140: Based on the predetermined number of fusion rounds, continue to perform pairwise similarity evaluation and cluster fusion on the main transformer operating condition distribution space to establish a typical main transformer operating condition space.

[0030] Step S150: Perform electromagnetic field distribution confirmation fusion on the electromagnetic field sample area based on the typical main transformer operating condition space to obtain the typical electromagnetic field space.

[0031] Step S160: Based on the typical main transformer operating condition space and the typical electromagnetic field space, perform mapping modeling to generate the electromagnetic field coupling model of the main transformer operating condition.

[0032] Specifically, the main transformer operating condition log set of the main transformer in the booster station of the new energy power plant is first cleaned and processed in all dimensions. By removing outliers, filling in missing values, and deduplicating duplicate values, effective, complete and standardized log data are selected. Then, the parameters are accurately classified according to their characterization attributes and correlation dimensions. The parameter data reflecting the actual operating status of the main transformer, such as load, voltage, current and oil temperature, are grouped into one category to construct the main transformer operating condition sample area. The parameter data related to the electromagnetic field strength, distribution pattern and phase characteristics during the operation of the main transformer are grouped into another category to construct the electromagnetic field sample area.

[0033] Using the core operating parameters such as load, voltage, current, and oil temperature of each operating condition sample within the main transformer operating condition sample area as feature dimensions, the cosine similarity algorithm is used as the specific implementation method to perform pairwise similarity calculations on all operating condition samples in the sample area. The cosine value between the feature vectors of two operating condition samples is used to quantify their similarity. The closer the cosine value is to 1, the higher the sample similarity. Then, the cosine similarity calculation results between all samples are systematically integrated and distributed statistically according to the sample correspondence relationship to form an operating condition similarity distribution that can accurately reflect the overall similarity characteristics and sample association rules of the main transformer operating condition sample area.

[0034] Using the obtained similarity distribution of operating conditions as the quantitative judgment basis, a similarity threshold is set as the clustering fusion standard. All operating condition samples in the main transformer operating condition sample area are clustered. Similar operating condition samples with cosine similarity calculation results higher than the threshold are grouped into the same cluster. Feature fusion and parameter normalization are performed on the operating condition samples in each cluster. Typical operating condition feature parameters of each cluster are extracted and corresponding operating condition feature vectors are constructed. Then, the typical operating condition feature vectors of all clusters are integrated to form a main transformer operating condition distribution space that can comprehensively cover various operating states of the main transformer and characterize the differences in different operating condition features.

[0035] First, considering the operating characteristics and modeling accuracy requirements of the main transformers in the booster stations of new energy power plants, a clustering and fusion iteration cycle of 3-5 rounds is preset. Then, using the already constructed main transformer operating condition distribution space as the initial data base, the cosine similarity algorithm and K-means clustering algorithm are used as the core implementation methods to perform iterative optimization in a cyclical manner according to the predetermined number of rounds: In each round, the cosine similarity algorithm is used to calculate the pairwise similarity of all operating condition feature vectors in the current operating condition distribution space, generating an operating condition similarity distribution matrix containing the degree of similarity between vectors; then, a dynamic similarity threshold is set based on this matrix, with the first round threshold being... The similarity is 0.8, and then reduced by 0.05 in each iteration. The K-means clustering algorithm is used to group the feature vectors of operating conditions with similarity higher than the threshold into the same cluster. The vectors in each cluster are fused by mean averaging, and the core operating condition parameters within the cluster are extracted and reconstructed into standardized typical main transformer operating condition vectors. After completing the predetermined number of iterations of fusion, redundant operating condition vectors with variance lower than 0.02 are removed. Finally, the typical main transformer operating condition space is obtained, which consists of multiple differentiated and highly representative typical main transformer operating condition vectors and can completely cover the core operating states of the main transformer such as no-load, light-load, full-load, and overload.

[0036] Using the established typical main transformer operating condition space as the core reference, the f-th typical main transformer operating condition vector (f being a positive integer) is extracted sequentially from this space. Each typical main transformer operating condition vector is precisely matched with the electromagnetic field sample area, and electromagnetic field sample data corresponding to each typical operating condition vector are selected to form the f-th matched electromagnetic field sample group. Subsequently, statistical analysis is performed on all samples within each matched electromagnetic field sample group to calculate the frequency of each matched electromagnetic field sample appearing within the group. This frequency is used as the confidence level of each matched sample to quantitatively characterize the representativeness of the sample. Then, based on the confidence level of each matched sample, data fusion processing is carried out on the corresponding matched electromagnetic field sample group to extract the core electromagnetic field characteristic parameters of each group and construct the f-th typical electromagnetic field vector. Finally, all typical electromagnetic field vectors are integrated and aggregated to construct a typical electromagnetic field space that corresponds one-to-one with the typical main transformer operating condition space.

[0037] First, data alignment processing is performed on the constructed typical main transformer operating condition space and typical electromagnetic field space to ensure that each typical main transformer operating condition vector in the typical main transformer operating condition space forms a one-to-one mapping relationship with the unique typical electromagnetic field vector in the typical electromagnetic field space, thus constructing a dataset corresponding to operating condition-electromagnetic field characteristic parameters. Then, using this dataset as the training basis, a multivariate nonlinear regression algorithm is used to perform mapping modeling. Nonlinear fitting is performed on the multidimensional operating characteristic parameters of the typical main transformer operating condition vector and the electromagnetic field distribution characteristic parameters of the corresponding typical electromagnetic field vector to explore the inherent coupling law between the two and establish a precise mapping relationship function from operating condition parameters to electromagnetic field parameters. Finally, this mapping relationship function is encapsulated into a model, incorporating core information such as the characteristic dimensions of operating conditions and electromagnetic fields, parameter thresholds, and correlation weights, to generate a main transformer operating condition electromagnetic field coupling model that can accurately match and output the corresponding electromagnetic field distribution characteristic parameters based on the real-time operating condition data of the main transformer.

[0038] In one possible implementation, step S150 further includes: step S151: extracting the f-th typical main transformer operating condition vector based on the typical main transformer operating condition space, where f is a positive integer.

[0039] Step S152: Match the electromagnetic field sample region according to the f-th typical main transformer operating condition vector to obtain the f-th matched electromagnetic field sample group.

[0040] Step S153: Calculate the frequency of each matching electromagnetic field sample in the f-th matching electromagnetic field sample group and obtain the confidence level of each matching sample.

[0041] Step S154: Perform data fusion on the f-th matched electromagnetic field sample group according to the confidence level of each matched sample to obtain the f-th typical electromagnetic field vector, and add the f-th typical electromagnetic field vector to the typical electromagnetic field space.

[0042] Specifically, for the typical main transformer operating condition space that has been iteratively optimized and includes various core typical operating states of the main transformer, the f-th typical main transformer operating condition vector is extracted from the space in a predetermined vector extraction order and rules. f is a positive integer, and the values ​​are taken in sequence as positive integers 1, 2, 3, ... until all vectors in the space are extracted. This vector integrates the core characteristic parameters such as load, voltage, current, and oil temperature under the corresponding typical operating conditions. It is a standardized feature vector after normalization and feature extraction, which can accurately represent the overall characteristics of a specific typical operating state of the main transformer.

[0043] Using the extracted f-th typical main transformer operating condition vector as the core matching basis, and employing precise timestamp association and Euclidean distance similarity matching as specific implementation methods, multiple matching electromagnetic field samples are screened from the electromagnetic field sample area constructed in the previous classification: First, the acquisition timestamps of all samples in the electromagnetic field sample area are precisely associated with the acquisition timestamps of the f-th typical main transformer operating condition vector to initially screen out the electromagnetic field sample set that matches in the time dimension; then, the operating condition parameters associated with each electromagnetic field sample in the initially screened sample set are converted into feature vectors, and Euclidean distance is calculated with the f-th typical main transformer operating condition vector. A second precise matching is completed by setting an Euclidean distance threshold to screen out all electromagnetic field samples that meet the similarity standard; finally, these multiple successfully matched electromagnetic field samples are uniformly collected and organized to form the f-th matching electromagnetic field sample group that uniquely corresponds to the f-th typical main transformer operating condition vector, providing a sample foundation with high matching degree and sufficient quantity for subsequent confidence statistics and data fusion.

[0044] For all electromagnetic field samples within the f-th matched electromagnetic field sample group, the confidence level is calculated using frequency statistics and normalization. First, all matched electromagnetic field samples within the sample group are traversed. The absolute number of occurrences of each unique electromagnetic field sample within the group is counted using frequency statistics. Then, this absolute number is divided by the total number of samples in the sample group to obtain the frequency of each sample. Subsequently, the frequency of all samples is normalized so that the confidence level of all samples falls within the range of 0 to 1. The normalized frequency value is the confidence level of each matched sample. The higher the confidence level, the stronger the representativeness and reliability of the electromagnetic field sample under the corresponding typical working condition. Finally, a dataset containing the identifier of each sample and its corresponding confidence level is output.

[0045] Using the confidence level of each matching sample within the f-th matching electromagnetic field sample group as the core weighting criterion, weighted data fusion processing is performed on all electromagnetic field samples within the group: First, the characteristic parameters of each electromagnetic field sample in each dimension, such as electromagnetic field strength, phase, and distribution coefficient, are multiplied by their own confidence level to obtain the weighted value of each sample's characteristic parameters. Then, the weighted characteristic parameters of the same dimension of all samples in the group are summed. Subsequently, the summation result is divided by the sum of the confidence levels of all samples in the group to obtain the fusion characteristic value of that dimension. After completing the weighted fusion calculation of all characteristic dimensions in sequence, the fusion characteristic values ​​of each dimension are integrated to construct a standardized f-th typical electromagnetic field vector. This vector accurately represents the core electromagnetic field distribution characteristics under the corresponding typical main transformer operating conditions. Finally, the constructed f-th typical electromagnetic field vector is directly added to the pre-created typical electromagnetic field space according to the vector correspondence relationship, completing the storage of the vector.

[0046] In one possible implementation, step S300 further includes: step S310: inputting the main transformer operating condition data into the main transformer operating condition electromagnetic field coupling model to obtain the operating condition matching electromagnetic field characteristic parameters.

[0047] Step S320: Perform dynamic modeling based on the characteristic parameters of the matching electromagnetic field under the operating conditions to obtain the matching electromagnetic field model under the operating conditions.

[0048] Step S330: Analyze the amplitude and phase attenuation shift of the multimodal detection signal according to the operating condition matching electromagnetic field model to obtain the attenuation shift characteristics of each signal.

[0049] Step S340: Perform reverse compensation on the multimodal detection signal according to the attenuation offset characteristics of each signal to obtain the multimodal correction signal.

[0050] Step S350: Perform feature extraction on the multimodal correction signal to obtain the feature vector of the multimodal signal.

[0051] Specifically, the real-time collected and normalized main transformer operating condition data is first converted into a standard operating condition input vector. This vector is then input into a pre-trained main transformer operating condition electromagnetic field coupling model. Through the established nonlinear mapping relationship between operating condition and electromagnetic field within the model, the current main transformer operating condition is matched and inferred. The model directly outputs multi-dimensional parameters such as electromagnetic field intensity, spatial distribution, and phase characteristics that correspond one-to-one with the operating condition, and integrates them to form operating condition matching electromagnetic field characteristic parameters.

[0052] Using characteristic parameters of the electromagnetic field matching the operating conditions, such as electromagnetic field strength, phase, and distribution coefficient, as core inputs, dynamic modeling is carried out using the finite element method (FEM) combined with real-time parameter iterative updates. First, a geometric model for electromagnetic field simulation is constructed based on the three-dimensional structural drawings of the main transformer, defining the dielectric properties and boundary conditions of key areas such as the core, windings, and oil gap. Then, the characteristic parameters of the electromagnetic field matching the operating conditions are imported into the model as initial field source parameters. The electromagnetic field solution domain is discretized into multiple element meshes using the finite element method, and the Maxwell equations are solved to obtain the initial electromagnetic field distribution. Subsequently, the boundary conditions and field source parameters are dynamically iteratively updated based on the real-time operating conditions of the main transformer, such as load changes and oil temperature fluctuations, and the electromagnetic field distribution results are recalculated and corrected. Finally, an operating condition matching electromagnetic field model that can accurately reflect the amplitude, phase, and spatial distribution characteristics of the electromagnetic field under the current operating conditions of the main transformer is constructed.

[0053] Multimodal detection signals, such as UHF, ultrasound, and high-frequency pulse current, are input into the established working condition matching electromagnetic field model. Using the real-time electromagnetic field distribution obtained from the model simulation as the propagation environment, the amplitude attenuation and phase shift caused by electromagnetic interference during the propagation of each detection signal inside the transformer are quantitatively analyzed point by point. The amplitude attenuation coefficient, phase shift angle, and attenuation propagation law corresponding to each modal signal are calculated. Finally, the attenuation shift characteristics of each signal that can characterize the degree of signal distortion are extracted.

[0054] Based on the obtained amplitude attenuation coefficient, phase offset angle and other signal attenuation offset characteristics, the multi-mode detection signal is subjected to amplitude gain reverse compensation and phase reverse calibration: for amplitude attenuation, gain amplification compensation is performed according to the reciprocal of the attenuation coefficient, and for phase offset, reverse phase correction is performed according to the offset angle. The signal distortion caused by the electromagnetic field is eliminated point by point, the true amplitude and phase information of the signal is restored, and finally the multi-mode corrected signal after removing the electromagnetic field interference of the working condition is obtained, which provides a high-quality and high-fidelity signal foundation for subsequent feature extraction.

[0055] For multimodal correction signals that have undergone reverse compensation, refined feature extraction is carried out around four key indicators: amplitude, frequency, phase, and pulse characteristics. First, the peak value, effective value, and amplitude fluctuation range of the signal are extracted in the amplitude dimension. In the frequency dimension, the main frequency, frequency band energy distribution, and harmonic components are extracted through fast Fourier transform. In the phase dimension, the signal phase offset, phase consistency, and phase difference characteristics are statistically analyzed. In the pulse characteristic dimension, the pulse rise time, pulse width, pulse repetition rate, and pulse interval distribution are extracted. After normalizing all the above feature parameters in sequence, they are spliced ​​and fused in a fixed dimension order to finally form a standardized multimodal signal feature vector containing amplitude, frequency, phase, and pulse characteristics.

[0056] In one possible implementation, step S400 further includes: step S410: loading the multimode signal feature sample set and the signal mapping partial discharge probability sample set of the main transformer of the booster station.

[0057] Step S420: Based on the multimodal signal feature sample set and the signal mapping partial discharge probability sample set, perform structure learning, parameter estimation and probability graph inference training on the Bayesian network to establish a multimodal signal partial discharge probability graph mapping model.

[0058] Step S430: Based on the Hidden Markov Model, perform partial discharge state evolution modeling and state transition probability reconstruction on the multimodal signal partial discharge probability map mapping model to obtain the multimodal signal-driven partial discharge prediction model.

[0059] Step S440: Perform variational inference and probability fusion calculation on the feature vector of the multimodal signal based on the multimodal signal-driven partial discharge prediction model, and output the first probability of the main variable partial discharge.

[0060] Specifically, the model training database loads pre-collected, pre-processed, and labeled sample data of the main transformer of the booster station, and imports the multimodal signal feature sample set and the signal mapping partial discharge probability sample set respectively. The multimodal signal feature sample set contains a large number of standardized features such as amplitude, frequency, phase, and pulse characteristics of historical multimodal correction signals, while the signal mapping partial discharge probability sample set consists of partial discharge probability labels that correspond one-to-one with each feature sample and have been diagnosed by experts or calibrated experimentally. After the two sets of sample sets are aligned and loaded, they provide complete and reliable data support for the subsequent structure learning and parameter training of the Bayesian network.

[0061] Using a multimodal signal feature sample set and a signal mapping partial discharge probability sample set as core training data, a Bayesian network is trained in three orderly steps: structure learning, parameter estimation, and probabilistic graph inference training, ultimately establishing a multimodal signal partial discharge probability graph mapping model. In the structure learning stage, network node definitions are first clarified, with the amplitude, frequency, phase, and pulse characteristics of the multimodal signal set as observation nodes and the partial discharge probability set as target nodes. Then, by traversing the potential connections between nodes and calculating the node correlation degree in conjunction with sample data, the optimal directed acyclic topology with the best node dependencies is selected, clarifying the correlation logic between each observation node and the target node, and between each observation node, thus completing the network structure construction. In the parameter estimation stage, based on the determined network topology, the feature data in the sample set and the corresponding partial discharge probability graph are used to train the Bayesian network. Based on probability labels, the frequency of partial discharge probability under different feature combinations is statistically analyzed, the conditional probability table of each node is calculated, the dependence strength between nodes is quantified, and the statistical regularity of the sample data is transformed into parameters that the network can recognize, thus completing parameter fitting and calibration. In the probabilistic graph inference training stage, multimodal signal feature samples from the sample set are input into a Bayesian network with completed structure and parameter configuration. The inference algorithm calculates the predicted value of partial discharge probability corresponding to each feature combination, compares the predicted value with the actual partial discharge probability label in the sample set, calculates the error, and iteratively adjusts the network structure and conditional probability table parameters. The process is repeated until the prediction accuracy reaches the preset standard, and finally, the training is completed and a multimodal signal partial discharge probability graph mapping model that can accurately map multimodal signal features to partial discharge probability is established.

[0062] A Hidden Markov Model (HMM) is introduced, and based on the partial discharge probability output by the multimodal signal partial discharge probability map mapping model, partial discharge state evolution modeling and state transition probability reconstruction are carried out. The specific process is as follows: First, partial discharge state evolution modeling is performed. Different development stages of partial discharge in the main transformer, such as no partial discharge, slight partial discharge, moderate partial discharge, and severe partial discharge, are abstracted into a sequence of hidden states in the HMM. The partial discharge probability output by the multimodal signal partial discharge probability map mapping model is used as the observation sequence. Combining the historical operating data of the main transformer and the development law of partial discharge, characteristic thresholds for each hidden state are defined, and the correspondence between the hidden states and the observation sequence is established to realize the time-series evolution modeling of partial discharge state and accurately characterize the partial discharge state. The model describes the dynamic development process of partial discharge (PD) from its initial state to its current state, from mild to severe. It then reconstructs the state transition probability based on historical PD state evolution data, statistically analyzing the transition frequencies between different hidden states and calculating the transition probabilities between each PD state, such as from mild to moderate, and from moderate to severe. Simultaneously, it combines the correlation between multimodal signal characteristics and PD states to correct the observation probability matrix, reconstructing a state transition probability matrix that reflects the dynamic transition patterns of PD states. Finally, it integrates the temporal evolution characteristics of the Hidden Markov Model with the static probability mapping capability of the multimodal signal PD probability graph mapping model, forming a multimodal signal-driven PD prediction model that combines static probability mapping and temporal state prediction functions.

[0063] The real-time multi-mode signal feature vectors are input into the constructed multi-mode signal-driven partial discharge prediction model. The partial discharge probability is calculated using variational inference and weighted probability fusion. First, based on the hidden states and observed variables in the model, an approximate variational distribution is constructed through variational inference. The KL divergence between the distribution and the posterior probability distribution of the partial discharge is minimized to achieve an efficient approximate solution for complex posterior probabilities and obtain the partial discharge probability distribution corresponding to each mode. Then, the probability results of different feature dimensions and detection modes are weighted and fused. Consistency correction is completed by combining the state transition probability. Finally, the first probability of partial discharge of the main transformer, which comprehensively reflects the possibility of partial discharge of the current main transformer, is output.

[0064] In one possible implementation, step S600 further includes: step S610: performing condition fluctuation sensitive fusion on the first probability of partial discharge of the main transformer and the second probability of partial discharge of the main transformer based on the main transformer operating condition data to obtain the diagnostic probability of partial discharge of the main transformer.

[0065] Step S620: Determine whether the main transformer partial discharge diagnosis probability is greater than or equal to the predetermined partial discharge diagnosis probability.

[0066] Step S630: If the partial discharge diagnosis probability of the main transformer is greater than or equal to the predetermined partial discharge diagnosis probability, activate the partial discharge depth diagnosis model.

[0067] Step S640: The characteristic gas concentration signal and the multimodal signal feature vector are fused to obtain the main transformer monitoring feature vector.

[0068] Step S650: Input the main transformer monitoring feature vector into the partial discharge depth diagnostic model to obtain the main transformer partial discharge diagnostic result.

[0069] Specifically, based on real-time collected operating condition data of the main transformer, including load rate, operating voltage, oil temperature, and cooling status, dynamic correction is performed on the first probability and second probability of partial discharge in the main transformer. First, an operating condition sensitivity evaluation mechanism is established. According to the degree of susceptibility of partial discharge probability to interference under different operating conditions, the operating condition adaptive weighting coefficients corresponding to the two probabilities are calculated respectively. Then, the first probability and second probability of partial discharge in the main transformer are weighted and summed. At the same time, the fusion result is dynamically corrected in combination with the current operating condition fluctuation amplitude to suppress the probability deviation caused by sudden changes in operating conditions, improve the stability and authenticity of the diagnostic results, and finally obtain the diagnostic probability of partial discharge in the main transformer after eliminating operating condition interference.

[0070] The obtained partial discharge diagnosis probability of the main transformer is compared with the pre-set partial discharge diagnosis probability threshold value by value. The two values ​​are directly compared according to the preset judgment rules to clearly determine whether the comprehensive probability of partial discharge of the current main transformer has reached or exceeded the abnormal warning threshold, providing a clear logical judgment basis for whether to start in-depth diagnosis.

[0071] If the probability of partial discharge diagnosis of the main transformer is determined to be greater than or equal to the predetermined probability of partial discharge diagnosis, indicating that the current main transformer has a risk of abnormal partial discharge, the pre-built and trained deep partial discharge diagnosis model is immediately activated. This model is built with a convolutional neural network (CNN) combined with a long short-term memory (LSTM) network as the core algorithm, and adopts a three-layer structure of feature extraction, temporal modeling, and classification diagnosis. Its construction and training process is as follows: In the model construction stage, the input layer receives high-dimensional data after fusing the feature gas concentration signal and the feature vector of the multimodal signal; the convolutional layer extracts the spatial features of the signal through the convolutional kernel; the pooling layer performs feature dimensionality reduction; and the LSTM layer captures the temporal evolution features of the partial discharge. The fully connected layer completes feature fusion and classification mapping, and the output layer corresponds to diagnostic labels for different partial discharge types and severity levels. During the model training phase, historical main transformer monitoring feature vectors, including samples of normal operating conditions and different types of partial discharge conditions, are used as the training set. The Adam optimization algorithm is used to minimize the cross-entropy loss function. Through multiple rounds of iteration, the convolution kernel parameters, the number of hidden units in the LSTM layer, and the weights of the fully connected layer are adjusted. Combined with an early stopping strategy to prevent overfitting, the model is ensured to have accurate partial discharge identification capabilities. After activation, the model can quickly receive the subsequently fused main transformer monitoring feature vectors and perform deep diagnostic operations, providing core support for the subsequent output of accurate partial discharge diagnostic results.

[0072] The characteristic gas concentration signal obtained from oil chromatography analysis is cascaded and standardized with the previously extracted multimodal signal feature vector. First, the two types of features are normalized to eliminate dimensional differences. Then, the characteristic gas concentration features and the amplitude, frequency, phase, and pulse characteristics of the multimodal signal are sequentially spliced ​​together according to a fixed dimensional order to form a high-dimensional joint feature containing electrical signals and chemical gas information. Finally, a main transformer monitoring feature vector with unified dimensions and complete information is obtained.

[0073] The integrated main transformer monitoring feature vector is input into the activated partial discharge deep diagnostic model. The model performs in-depth analysis and comprehensive reasoning on the information such as feature gas concentration, amplitude, frequency, phase, and pulse characteristics in the vector through internal convolution and temporal feature extraction structure. It classifies, identifies, and calculates the severity level, discharge type, and discharge location of the partial discharge, and finally outputs a complete main transformer partial discharge diagnostic result containing the discharge severity level, discharge type, and discharge location.

[0074] In one possible implementation, step S610 further includes: step S611: obtaining initial conditions for partial discharge probability fusion, wherein the initial conditions for partial discharge probability fusion include initial weights for the first probability of partial discharge and initial weights for the second probability of partial discharge.

[0075] Step S612: Evaluate the sensitivity of the multimodal signal feature vector to operating condition fluctuations based on the main transformer operating condition data, and obtain the first sensitivity to operating condition fluctuations.

[0076] Step S613: Evaluate the sensitivity of the characteristic gas concentration signal to operating condition fluctuations based on the main transformer operating condition data, and obtain the second sensitivity to operating condition fluctuations.

[0077] Step S614: Adaptively adjust the initial conditions for partial discharge probability fusion based on the first sensitivity to operating condition fluctuations and the second sensitivity to operating condition fluctuations to obtain the current fusion weight conditions.

[0078] Step S615: Perform fusion calculation on the first probability of the main variable partial discharge and the second probability of the main variable partial discharge according to the current fusion weight conditions to generate the diagnostic probability of the main variable partial discharge.

[0079] Specifically, the initial conditions for partial discharge probability fusion are obtained from the model parameter configuration library. These initial conditions are pre-set benchmark weighting parameters, which mainly include the initial weight of the first probability of partial discharge used to weight the first probability of partial discharge of the main transformer, and the initial weight of the second probability of partial discharge used to weight the second probability of partial discharge of the main transformer. These serve as the initial benchmark for adaptively adjusting the fusion weights according to the fluctuation of the operating conditions.

[0080] Using real-time transformer operating condition data acquired by a sensor network as the core evaluation criterion, this study evaluates the sensitivity to operating condition fluctuations of the multimodal signal feature vector constructed after dynamic compensation through operating condition coupling. First, key operating condition parameters are extracted from the transformer operating condition data, including load rate, operating voltage, top oil temperature, and cooling system operating status. Then, the multimodal signal feature vector is decomposed into characteristic parameters corresponding to high-frequency pulse current, ultra-high-frequency electromagnetic waves, and ultrasonic waves. By establishing the correlation mapping relationship between operating condition parameters and each modal characteristic parameter, the relative rate of change and fluctuation offset of each characteristic parameter with changes in operating condition parameters are calculated. Subsequently, the degree of operating condition disturbance of each dimension feature is normalized, and then... Based on the detection accuracy, response characteristics, and characterization effectiveness of three types of signals—high-frequency pulse current, ultra-high-frequency electromagnetic waves, and ultrasonic waves—in the partial discharge detection of main transformers, preset mode weighting coefficients are assigned to each of the three modes. At the same time, corresponding feature weighting coefficients are configured for each dimension of feature parameters under each mode. First, the normalized disturbance values ​​of each dimension under a single mode are multiplied by the corresponding feature weighting coefficients and then summed to obtain the comprehensive disturbance value of a single mode. Then, the comprehensive disturbance values ​​of each mode are multiplied by the corresponding mode weighting coefficients and accumulated to complete the multi-level weighted synthesis. Finally, the first sensitivity of the operating condition fluctuation, which can quantify the degree of interference of the overall feature vector of multi-mode signals by the operating condition fluctuation of the main transformer, is obtained.

[0081] Using real-time transformer operating condition data collected by a sensor network as the core evaluation basis, this study evaluates the sensitivity to operating condition fluctuations of characteristic gas concentration signals. First, key operating condition parameters such as load rate, operating voltage, top oil temperature, oil level, and cooling system operating status are extracted from the transformer operating condition data. Simultaneously, the concentration values ​​of core characteristic gases such as hydrogen, methane, ethane, ethylene, and acetylene, as well as the ratio characteristics between these gases, are extracted from the characteristic gas concentration signals to form characteristic gas evaluation dimensions. Then, a correlation mapping relationship is established between the key operating condition parameters and each characteristic gas evaluation dimension. This is achieved by calculating the relative change rate and fluctuation amplitude of each gas concentration value and gas ratio characteristic as the operating condition parameters dynamically change. The perturbation results of all characteristic gas evaluation dimensions are normalized, mapping the perturbation values ​​of each dimension to the same numerical range to eliminate calculation bias caused by differences in the concentration dimensions and numerical ranges of different gases. Finally, based on the characterization importance of each characteristic gas at different stages of partial discharge development in the main transformer, corresponding preset weight coefficients are assigned. The normalized perturbation values ​​of each dimension are multiplied by the corresponding weight coefficients and summed to complete the weighted synthesis. Finally, the second sensitivity of the operating condition fluctuation, which can quantify the degree of interference of the overall characteristic gas concentration signal with the main transformer operating condition fluctuation, is obtained.

[0082] Based on the acquired first and second sensitivity to operating condition fluctuations, dynamic adaptive adjustment is performed on the initial conditions for partial discharge probability fusion, which include the initial weights for the first and second partial discharge probabilities. This yields the current fusion weight conditions adapted to the actual operating conditions of the main transformer. First, the first and second sensitivity to operating condition fluctuations are normalized to obtain dimensionless relative sensitivity values. A negative correlation mapping rule between sensitivity and weight adjustment is established: the higher the fluctuation sensitivity of a feature, the lower its corresponding partial discharge probability weight should be appropriately reduced; conversely, the lower the sensitivity, the more stable or appropriately increased the corresponding partial discharge probability weight should be, thus avoiding excessive influence of the probability results corresponding to highly disturbed features on the fusion value. Subsequently, the initial conditions are determined using the first partial discharge probability... Using the initial weights of the partial discharge (PD) first probability and the second probability as adjustment benchmarks, the initial weights of the PD first probability and the relative sensitivity values ​​of the first sensitivity to operating condition fluctuations are calculated in conjunction with each other, resulting in preliminary adjustment weights for the two probabilities. These preliminary adjustment weights are then normalized to ensure that the sum of the current weights of the PD first and PD second probabilities is 1, satisfying the normalization constraint for weight allocation. Finally, the calibrated current weights of the PD first and PD second probabilities are integrated into the current fusion weight condition. This condition accurately adapts to the current operating conditions of the main transformer, providing a reasonable weight basis for the subsequent weighted fusion of PD probabilities.

[0083] Based on the current fusion weight conditions adapted to the current operating conditions of the main transformer, a weighted fusion calculation is performed on the first probability and the second probability of partial discharge (PD) of the main transformer. This generates a PD diagnostic probability that truly reflects the actual partial discharge situation of the main transformer. First, the current weight coefficients corresponding to the first probability and the second probability of PD of the main transformer are extracted from the current fusion weight conditions. The two weight coefficients are normalized and their weight sum is 1. Then, the first probability of PD of the main transformer is multiplied by the corresponding current weight coefficient to obtain the first probability weighted value. At the same time, the second probability of PD of the main transformer is multiplied by the corresponding current weight coefficient to obtain the second probability weighted value. Finally, the first probability weighted value and the second probability weighted value are summed. Through this weighted summation fusion method, the influence of the PD probability corresponding to high sensitivity to operating condition fluctuations on the result is weakened, and the characterization role of the low-sensitivity and more stable PD probability is strengthened. Ultimately, a PD diagnostic probability of the main transformer that eliminates the interference of operating condition fluctuations and closely matches the actual operating state of the main transformer is generated.

[0084] In one possible implementation, step S200 further includes: the multimodal detection signal includes a high-frequency pulse current signal, an ultra-high frequency electromagnetic wave signal, and an ultrasonic signal.

[0085] Specifically, the multimodal detection signals include three types: high-frequency pulse current signals, ultra-high-frequency electromagnetic wave signals, and ultrasonic signals. These three types of signals capture the physical characteristics of the partial discharge process of the main transformer in the substation from different detection dimensions, forming a multi-dimensional and complementary detection signal system. The high-frequency pulse current signal detects the high-frequency pulse current waveform generated by the partial discharge of the main transformer, capturing the electrical signal characteristics during the discharge process. The ultra-high-frequency electromagnetic wave signal collects the ultra-high-frequency electromagnetic waves generated during partial discharge, accurately capturing the electromagnetic radiation characteristics corresponding to the discharge and exhibiting strong anti-interference capabilities. The ultrasonic signal collects the ultrasonic vibration signals generated during the partial discharge of the main transformer, reflecting the physical process of the discharge through acoustic signal characteristics. The coordinated acquisition of these three types of signals enables multi-dimensional perception of the partial discharge state of the main transformer, providing comprehensive and accurate raw detection data for subsequent construction of multimodal signal feature vectors and partial discharge probability prediction.

[0086] In one possible implementation, step S600 further includes: generating a main transformer partial discharge early warning signal based on the main transformer partial discharge diagnosis result.

[0087] Specifically, based on the complete partial discharge diagnosis results of the main transformer obtained from the aforementioned process, including the severity level, discharge type, and discharge location, and in accordance with the preset early warning rules for the operation of the main transformer in the booster station of the new energy power station, the diagnosis results are graded and analyzed, and signals are generated. First, based on the discharge severity level in the diagnosis results, preset tiered early warning levels such as Level 1, Level 2, and Level 3 are matched. Then, the early warning level is calibrated again based on the degree of danger of the discharge type and the criticality of the discharge location to determine the final early warning level. Next, in accordance with the signal format requirements of the main transformer operation monitoring system, the core diagnostic information such as the early warning level, discharge type, discharge location, and severity is standardized and packaged, along with basic monitoring data such as the current operating condition data of the main transformer, characteristic gas concentration, and multimodal signal characteristics, to generate a structured main transformer partial discharge early warning signal. This early warning signal can be directly transmitted to the main transformer monitoring terminal, back-end management system, and related operation and maintenance alarm modules of the new energy power station, realizing real-time push of early warning information and providing accurate signal guidance and data support for operation and maintenance personnel to carry out targeted maintenance and handling work.

[0088] Example 2: Based on the same inventive concept as the method for partial discharge detection of main transformers in new energy power plant step-up substations described in the previous examples, as shown in Figure 2, this application provides a system for partial discharge detection of main transformers in new energy power plant step-up substations. The system and method embodiments in this application are based on the same inventive concept. The system includes a coupling model establishment module 10, used to perform electromagnetic field distribution certainty fusion modeling based on the main transformer operating condition log set of the main transformer in the new energy power plant step-up substation, and establish a coupling model of the main transformer operating condition electromagnetic field.

[0089] The main transformer operating condition acquisition module 20 is used to acquire, in real time, the main transformer operating condition data, characteristic gas concentration signals, and multi-modal detection signals of the main transformer of the booster station based on the sensor network.

[0090] The multimodal signal feature vector construction module 30 is used to construct multimodal signal feature vectors by performing dynamic compensation of the multimodal detection signal based on the main transformer operating condition data and the main transformer operating condition electromagnetic field coupling model.

[0091] The main transformer partial discharge first probability determination module 40 is used to predict the partial discharge probability of the main transformer of the step-up station based on the feature vector of the multi-mode signal, and determine the first probability of the main transformer partial discharge.

[0092] The main transformer partial discharge second probability determination module 50 is used to predict the partial discharge probability of the main transformer of the step-up station based on the characteristic gas concentration signal, and determine the main transformer partial discharge second probability.

[0093] The main transformer partial discharge diagnosis result acquisition module 60 is used to perform deep partial discharge diagnosis based on the first probability of the main transformer partial discharge and the second probability of the main transformer partial discharge under the condition fluctuation sensitive fusion, and to acquire the main transformer partial discharge diagnosis result.

[0094] Furthermore, the system is also used to perform the following functions: perform data cleaning and parameter classification on the main transformer operating condition log set to obtain the main transformer operating condition sample area and the electromagnetic field sample area; perform pairwise similarity evaluation on the main transformer operating condition sample area to obtain the operating condition similarity distribution; perform clustering and fusion on the main transformer operating condition sample area based on the operating condition similarity distribution to obtain the main transformer operating condition distribution space; continue to perform pairwise similarity evaluation and clustering and fusion on the main transformer operating condition distribution space based on a predetermined number of fusion rounds to establish a typical main transformer operating condition space; perform electromagnetic field distribution certainty fusion on the electromagnetic field sample area based on the typical main transformer operating condition space to obtain a typical electromagnetic field space; and perform mapping modeling on the typical main transformer operating condition space and the typical electromagnetic field space to generate the main transformer operating condition electromagnetic field coupling model.

[0095] Furthermore, the system is also used to perform the following functions: extracting the f-th typical main transformer operating condition vector based on the typical main transformer operating condition space, where f is a positive integer; matching the electromagnetic field sample region based on the f-th typical main transformer operating condition vector to obtain the f-th matched electromagnetic field sample group; statistically analyzing the frequency of each matched electromagnetic field sample in the f-th matched electromagnetic field sample group to obtain the confidence level of each matched sample; performing data fusion on the f-th matched electromagnetic field sample group based on the confidence level of each matched sample to obtain the f-th typical electromagnetic field vector, and adding the f-th typical electromagnetic field vector to the typical electromagnetic field space.

[0096] Furthermore, the system is also used to perform the following functions: inputting the main transformer operating condition data into the main transformer operating condition electromagnetic field coupling model to obtain the operating condition matching electromagnetic field characteristic parameters; performing dynamic modeling based on the operating condition matching electromagnetic field characteristic parameters to obtain the operating condition matching electromagnetic field model; performing amplitude and phase attenuation offset analysis on the multi-mode detection signal based on the operating condition matching electromagnetic field model to obtain the attenuation offset features of each signal; performing reverse compensation on the multi-mode detection signal based on the attenuation offset features of each signal to obtain the multi-mode correction signal; and performing feature extraction on the multi-mode correction signal to obtain the multi-mode signal feature vector.

[0097] Furthermore, the system is also used to perform the following functions: loading the multimodal signal feature sample set and the signal mapping partial discharge probability sample set of the main transformer of the substation; performing structure learning, parameter estimation and probabilistic graph inference training on the Bayesian network based on the multimodal signal feature sample set and the signal mapping partial discharge probability sample set to establish a multimodal signal partial discharge probability graph mapping model; performing partial discharge state evolution modeling and state transition probability reconstruction on the multimodal signal partial discharge probability graph mapping model based on the hidden Markov model to obtain a multimodal signal-driven partial discharge prediction model; performing variational inference and probability fusion calculation on the multimodal signal feature vector based on the multimodal signal-driven partial discharge prediction model to output the first probability of partial discharge of the main transformer.

[0098] Furthermore, the system is also used to perform the following functions: performing condition fluctuation-sensitive fusion of the first probability of partial discharge (PD) and the second probability of PD based on the main transformer operating condition data to obtain the PD diagnosis probability; determining whether the PD diagnosis probability is greater than or equal to a predetermined PD diagnosis probability; if the PD diagnosis probability is greater than or equal to the predetermined PD diagnosis probability, activating the partial discharge depth diagnosis model; fusing the characteristic gas concentration signal and the multi-modal signal feature vector to obtain the main transformer monitoring feature vector; and inputting the main transformer monitoring feature vector into the partial discharge depth diagnosis model to obtain the PD diagnosis result.

[0099] Furthermore, the system is also used to implement the following functions: obtaining initial conditions for partial discharge probability fusion, the initial conditions for partial discharge probability fusion including initial weights for a first probability of partial discharge and initial weights for a second probability of partial discharge; evaluating the sensitivity of the multi-mode signal feature vector to operating condition fluctuations based on the main transformer operating condition data, and obtaining a first sensitivity to operating condition fluctuations; evaluating the sensitivity of the characteristic gas concentration signal to operating condition fluctuations based on the main transformer operating condition data, and obtaining a second sensitivity to operating condition fluctuations; adaptively adjusting the initial conditions for partial discharge probability fusion based on the first sensitivity to operating condition fluctuations and the second sensitivity to operating condition fluctuations, and obtaining current fusion weight conditions; and performing fusion calculations on the first probability of partial discharge and the second probability of partial discharge in the main transformer based on the current fusion weight conditions to generate the diagnostic probability of partial discharge in the main transformer.

[0100] Furthermore, the system is also used to achieve the following functions: the multimodal detection signal includes high-frequency pulse current signal, ultra-high frequency electromagnetic wave signal and ultrasonic signal.

[0101] Furthermore, the system is also used to perform the following function: generate a main transformer partial discharge early warning signal based on the main transformer partial discharge diagnosis result.

[0102] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0103] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0104] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for detecting partial discharge in the main transformer of a new energy power station step-up substation, characterized in that, The method includes: performing electromagnetic field distribution certainty fusion modeling based on the main transformer operating condition log set of the main transformer in the booster station of the new energy power station, and establishing a main transformer operating condition electromagnetic field coupling model; acquiring the main transformer operating condition data, characteristic gas concentration signals, and multi-mode detection signals of the main transformer in real time using a sensor network; performing operating condition coupling dynamic compensation on the multi-mode detection signals based on the main transformer operating condition data and the main transformer operating condition electromagnetic field coupling model, and constructing a multi-mode signal feature vector; performing partial discharge probability prediction on the main transformer based on the multi-mode signal feature vector, and determining the first probability of main transformer partial discharge; performing partial discharge probability prediction on the main transformer based on the characteristic gas concentration signals, and determining the second probability of main transformer partial discharge; performing deep partial discharge diagnosis under operating condition fluctuation sensitive fusion based on the first probability and the second probability of main transformer partial discharge, and obtaining the main transformer partial discharge diagnosis result.

2. The method for detecting partial discharge of the main transformer in a new energy power station step-up substation as described in claim 1, characterized in that, Based on the main transformer operating condition log set of the booster substation of the new energy power plant, a certainty fusion model of electromagnetic field distribution is performed to establish a coupling model of electromagnetic field for the main transformer operating condition. This includes: cleaning and classifying the main transformer operating condition log set to obtain a main transformer operating condition sample area and an electromagnetic field sample area; evaluating pairwise similarity based on the main transformer operating condition sample areas to obtain a similarity distribution of operating conditions; clustering and fusion of the main transformer operating condition sample areas based on the similarity distribution of operating conditions to obtain a distribution space of main transformer operating conditions; continuing pairwise similarity evaluation and clustering and fusion of the distribution space of main transformer operating conditions based on a predetermined number of fusion rounds to establish a typical main transformer operating condition space; performing certainty fusion of electromagnetic field distribution on the electromagnetic field sample areas based on the typical main transformer operating condition space to obtain a typical electromagnetic field space; and performing mapping modeling based on the typical main transformer operating condition space and the typical electromagnetic field space to generate the coupling model of electromagnetic field for the main transformer operating condition.

3. The method for detecting partial discharge of the main transformer in a new energy power station step-up substation as described in claim 2, characterized in that, Based on the typical main transformer operating condition space, electromagnetic field distribution confidence fusion is performed on the electromagnetic field sample area to obtain a typical electromagnetic field space, including: extracting the f-th typical main transformer operating condition vector from the typical main transformer operating condition space, where f is a positive integer; matching the electromagnetic field sample area with the f-th typical main transformer operating condition vector to obtain the f-th matched electromagnetic field sample group; counting the frequency of each matched electromagnetic field sample in the f-th matched electromagnetic field sample group to obtain the confidence level of each matched sample; performing data fusion on the f-th matched electromagnetic field sample group based on the confidence level of each matched sample to obtain the f-th typical electromagnetic field vector, and adding the f-th typical electromagnetic field vector to the typical electromagnetic field space.

4. The method for detecting partial discharge of the main transformer in a new energy power station step-up substation as described in claim 1, characterized in that, Based on the main transformer operating condition data, dynamic compensation for the multimodal detection signal is performed according to the main transformer operating condition electromagnetic field coupling model to construct a multimodal signal feature vector. This includes: inputting the main transformer operating condition data into the main transformer operating condition electromagnetic field coupling model to obtain operating condition matching electromagnetic field feature parameters; performing dynamic modeling based on the operating condition matching electromagnetic field feature parameters to obtain an operating condition matching electromagnetic field model; performing amplitude and phase attenuation offset analysis on the multimodal detection signal according to the operating condition matching electromagnetic field model to obtain attenuation offset features for each signal; performing reverse compensation on the multimodal detection signal according to the attenuation offset features for each signal to obtain a multimodal correction signal; and extracting features from the multimodal correction signal to obtain the multimodal signal feature vector.

5. The method for detecting partial discharge of the main transformer in a new energy power station step-up substation as described in claim 1, characterized in that, The partial discharge probability prediction of the main transformer of the substation is performed based on the multimodal signal feature vector to determine the first probability of partial discharge of the main transformer. This includes: loading the multimodal signal feature sample set and the signal-mapped partial discharge probability sample set of the main transformer of the substation; training a Bayesian network for structure learning, parameter estimation and probability graph inference based on the multimodal signal feature sample set and the signal-mapped partial discharge probability sample set to establish a multimodal signal partial discharge probability graph mapping model; performing partial discharge state evolution modeling and state transition probability reconstruction on the multimodal signal partial discharge probability graph mapping model based on a hidden Markov model to obtain a multimodal signal-driven partial discharge prediction model; and performing variational inference and probability fusion calculation on the multimodal signal feature vector based on the multimodal signal-driven partial discharge prediction model to output the first probability of partial discharge of the main transformer.

6. The method for detecting partial discharge of the main transformer in a new energy power station step-up substation as described in claim 1, characterized in that, Deep partial discharge diagnosis based on the first probability and the second probability of partial discharge in the main transformer, using a condition fluctuation-sensitive fusion method, is performed to obtain the partial discharge diagnosis result of the main transformer. This includes: performing condition fluctuation-sensitive fusion on the first probability and the second probability of partial discharge in the main transformer based on the main transformer operating condition data to obtain the main transformer partial discharge diagnosis probability; determining whether the main transformer partial discharge diagnosis probability is greater than or equal to a predetermined partial discharge diagnosis probability; if the main transformer partial discharge diagnosis probability is greater than or equal to the predetermined partial discharge diagnosis probability, activating the partial discharge deep diagnosis model; fusing the characteristic gas concentration signal and the multimodal signal feature vector to obtain the main transformer monitoring feature vector; and inputting the main transformer monitoring feature vector into the partial discharge deep diagnosis model to obtain the main transformer partial discharge diagnosis result.

7. The method for detecting partial discharge of the main transformer in a new energy power station step-up substation as described in claim 6, characterized in that, Based on the main transformer operating condition data, the first probability and the second probability of partial discharge (PD) of the main transformer are fused using operating condition fluctuation sensitivity to obtain the PD diagnosis probability of the main transformer. This includes: obtaining initial conditions for PD probability fusion, which include initial weights for the first and second PD probabilities; evaluating the operating condition fluctuation sensitivity of the multi-mode signal feature vector based on the main transformer operating condition data to obtain a first operating condition fluctuation sensitivity; evaluating the operating condition fluctuation sensitivity of the characteristic gas concentration signal based on the main transformer operating condition data to obtain a second operating condition fluctuation sensitivity; adaptively adjusting the initial conditions for PD probability fusion based on the first and second operating condition fluctuation sensitivities to obtain current fusion weight conditions; and performing fusion calculation on the first and second PD probabilities of the main transformer based on the current fusion weight conditions to generate the PD diagnosis probability of the main transformer.

8. The method for detecting partial discharge of the main transformer in a new energy power station step-up substation as described in claim 1, characterized in that, The multimodal detection signals include high-frequency pulse current signals, ultra-high frequency electromagnetic wave signals, and ultrasonic signals.

9. The method for detecting partial discharge of the main transformer in a new energy power station step-up substation as described in claim 1, characterized in that, Obtaining the diagnostic results of the main transformer partial discharge also includes: generating a warning signal for the main transformer partial discharge based on the diagnostic results of the main transformer partial discharge.

10. A partial discharge detection system for the main transformer of a new energy power station step-up substation, characterized in that, The system is used to implement the partial discharge detection method for the main transformer of a new energy power station step-up substation as described in any one of claims 1-9. The system includes: a coupling model establishment module, used to perform electromagnetic field distribution certainty fusion modeling based on the main transformer operating condition log set of the main transformer of the new energy power station step-up substation, and establish a main transformer operating condition electromagnetic field coupling model; a main transformer operating condition acquisition module, used to acquire the main transformer operating condition data, characteristic gas concentration signals, and multimodal detection signals of the main transformer of the step-up substation in real time based on a sensor network; and a multimodal signal feature vector construction module, used to construct a feature vector based on the main transformer operating condition data and the main transformer operating condition electromagnetic field coupling model. The multimodal detection signal undergoes dynamic compensation under operating conditions to construct a multimodal signal feature vector. A main transformer partial discharge first probability determination module is used to predict the partial discharge probability of the main transformer at the substation based on the multimodal signal feature vector, determining the first probability of main transformer partial discharge. A main transformer partial discharge second probability determination module is used to predict the partial discharge probability of the main transformer at the substation based on the characteristic gas concentration signal, determining the second probability of main transformer partial discharge. A main transformer partial discharge diagnosis result acquisition module is used to perform deep partial discharge diagnosis under operating condition fluctuation sensitive fusion based on the first and second probabilities of main transformer partial discharge, acquiring the main transformer partial discharge diagnosis result.

Images