Transformer risk assessment method and device, storage medium and electronic equipment

Abstract

The present disclosure relates to a transformer risk assessment method and device, a storage medium and an electronic device, and relates to the technical field of transformer risk assessment. The method comprises the following steps: first, obtaining different monitoring data of the transformer; then, respectively performing risk assessment on the different monitoring data, obtaining risk assessment probabilities corresponding to the different monitoring data, and confidence degrees corresponding to the different risk assessment probabilities; then, determining risk distribution information corresponding to the different monitoring data according to the different risk assessment probabilities and the confidence degrees corresponding to the different risk assessment probabilities, wherein the risk distribution information comprises monitoring probabilities of different risk levels of the transformer; finally, determining a risk assessment result of the transformer based on the risk distribution information. In this way, the present disclosure can improve the reliability of the risk distribution information, effectively overcome the one-sidedness and uncertainty caused by single sensor or a small number of fixed parameter analysis, and thus comprehensively reflect the real health condition of the transformer.

Description

Technical Field

[0001] This disclosure relates to the field of transformer risk assessment technology, specifically to a transformer risk assessment method, apparatus, storage medium, and electronic equipment. Background Technology

[0002] During their service life, power transformers inevitably experience various faults or damages. These faults not only affect the performance of the transformers but may also lead to power system interruptions or failures, thereby threatening the stability of the power grid and the reliability of power supply. Conducting risk assessments on power transformers helps ensure the safety of power grid transportation.

[0003] Related technologies are usually based on the analysis results of a single sensor or a fixed number of parameters to conduct risk assessments of transformers, which has a certain degree of bias and is difficult to fully reflect the true health status of transformers. Summary of the Invention

[0004] In view of this, this disclosure provides a method, apparatus, storage medium and electronic device for transformer risk assessment. The main purpose is to solve the technical problem that related technologies usually rely on the analysis results of a single sensor or a fixed number of parameters to conduct transformer risk assessment. However, this approach is difficult to fully reflect the true health status of the transformer.

[0005] According to a first aspect of this disclosure, a transformer risk assessment method is provided, the method comprising:

[0006] Acquire different monitoring data of the transformer, including dissolved gas data in oil, vibration data, oil temperature data, and magnetic field data;

[0007] Risk assessments were conducted on different monitoring data to obtain the risk assessment probabilities corresponding to different monitoring data, as well as the confidence levels corresponding to different risk assessment probabilities.

[0008] Based on different risk assessment probabilities and the corresponding confidence levels, the risk distribution information corresponding to different monitoring data is determined. The risk distribution information includes the monitoring probabilities of different risk levels of transformers.

[0009] Based on risk distribution information, the risk assessment results for the transformer are determined;

[0010] Before conducting risk assessments on different monitoring data and obtaining the risk assessment probabilities and confidence levels corresponding to different monitoring data, the method further includes: obtaining risk assessment models for different monitoring data; the risk assessment models include a first assessment model, a second assessment model, a third assessment model, and a fourth assessment model; the risk assessment probabilities include the first assessment probability, the second assessment probability, the third assessment probability, and the fourth assessment probability; the first assessment model is used to determine a sub-feature set using the minimum redundancy maximum correlation method, and inputs the sub-feature set into a Gaussian Naive Bayes model, outputting the first assessment probability corresponding to the dissolved gas data in oil from different monitoring data. The first evaluation model extracts time-frequency and time-domain features from vibration signals through wavelet packet decomposition and short-time Fourier transform. It then fuses the original, time-frequency, and time-domain sequence features using a multi-scale one-dimensional convolutional neural network to output the second evaluation probability corresponding to vibration data in different monitoring data. The second evaluation model obtains historical temperature curves distributed at multiple points and outputs the third evaluation probability corresponding to oil temperature data in different monitoring data based on the predicted temperature increase and its confidence interval. The third evaluation model utilizes internal magnetic field data from multiple locations for modeling and outputs the fourth evaluation probability corresponding to magnetic field data in different monitoring data based on the spatiotemporal distribution characteristics of magnetic flux density.

[0011] According to a second aspect of this disclosure, a transformer risk assessment apparatus is provided, the apparatus comprising:

[0012] The acquisition module is used to acquire different monitoring data of the transformer, including dissolved gas data in oil, vibration data, oil temperature data, and magnetic field data. Risk assessments are performed on each of the different monitoring data to obtain the risk assessment probabilities and corresponding confidence levels. Before performing risk assessments on the different monitoring data and obtaining the corresponding risk assessment probabilities and confidence levels, the method further includes: acquiring risk assessment models corresponding to the different monitoring data; the risk assessment models include a first assessment model, a second assessment model, a third assessment model, and a fourth assessment model; the risk assessment probabilities include the first assessment probability, the second assessment probability, the third assessment probability, and the fourth assessment probability; the first assessment model is used to apply the minimum redundancy maximum correlation method. A sub-feature set is determined and input into a Gaussian Naive Bayes model, outputting the first evaluation probability corresponding to the dissolved gas data in oil from different monitoring data. The second evaluation model is used to extract time-frequency and time-domain features from the vibration signal by wavelet packet decomposition and short-time Fourier transform, and then fuses the original, time-frequency and time-domain sequence features through a multi-scale one-dimensional convolutional neural network to output the second evaluation probability corresponding to the vibration data from different monitoring data. The third evaluation model is used to obtain the historical temperature curves distributed at multiple points, and outputs the third evaluation probability corresponding to the oil temperature data from different monitoring data based on the predicted temperature increase and its confidence interval. The fourth evaluation model is used to model using internal magnetic field data from multiple locations, and outputs the fourth evaluation probability corresponding to the magnetic field data from different monitoring data based on the spatiotemporal distribution characteristics of magnetic flux density.

[0013] The determination module is used to determine the risk distribution information corresponding to different monitoring data based on different risk assessment probabilities and the corresponding confidence levels. The risk distribution information includes the monitoring probabilities of different risk levels of the transformer. Based on the risk distribution information, the risk assessment result of the transformer is determined.

[0014] According to a third aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of the first aspect described above.

[0015] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause a computer to perform the method of the first aspect described above.

[0016] The transformer risk assessment method, apparatus, storage medium, and electronic equipment disclosed herein, compared with related technologies, first acquire different monitoring data of the transformer; then, risk assessment is performed on each of the different monitoring data to obtain the risk assessment probability corresponding to each monitoring data, as well as the confidence level corresponding to each risk assessment probability; next, based on the different risk assessment probabilities and the corresponding confidence levels, risk distribution information corresponding to different monitoring data is determined, including the monitoring probability of different risk levels of the transformer; finally, based on the risk distribution information, the risk assessment result of the transformer is determined. In this way, this disclosure can perform independent risk assessments on different monitoring data of the transformer, improving the comprehensiveness of transformer risk assessment. By determining the risk distribution information of the transformer based on multiple risk assessment probabilities and their confidence levels, the reliability of the risk distribution information is improved, effectively overcoming the bias and uncertainty caused by analysis of a single sensor or a few fixed parameters, and thus comprehensively reflecting the true health status of the transformer. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0018] To more clearly illustrate the technical solutions in the embodiments or related technologies of this disclosure, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a transformer risk assessment method provided in an embodiment of this disclosure;

[0020] Figure 2 A schematic flowchart illustrating another transformer risk assessment method provided in this disclosure embodiment;

[0021] Figure 3 This illustration shows a sensor distribution diagram provided in an embodiment of this application;

[0022] Figure 4 A flowchart illustrating an example provided in an embodiment of this application is shown;

[0023] Figure 5 This is a schematic diagram of the structure of a transformer risk assessment device provided in an embodiment of this disclosure. Detailed Implementation

[0024] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description. It should be noted that, unless otherwise specified, the embodiments of this disclosure and the features described therein can be combined with each other.

[0025] The following description, with reference to the accompanying drawings, outlines a transformer risk assessment method, apparatus, storage medium, and electronic device according to embodiments of the present disclosure.

[0026] The healthy operation of transformers is crucial for ensuring the safe transmission of electricity and the conversion of electrical energy. However, due to various potential factors such as overload operation, changes in ambient temperature, electromagnetic field fluctuations, oil temperature rise, aging effects, and external mechanical shocks, power transformers will inevitably experience various faults or damages during their service life. These faults not only affect the performance of the transformer but may also lead to power system interruptions or failures, thereby threatening the stability of the power grid and the reliability of power supply.

[0027] In some embodiments, common methods for monitoring the operating status of transformers include dissolved gas analysis (DGA), temperature measurement, vibration detection, and magnetic field monitoring. However, these methods typically focus on the analysis of a single sensor or a fixed number of parameters, making it difficult to comprehensively reflect the true health status of the transformer. For example, relying solely on oil temperature or DGA indicators may fail to accurately distinguish between temperature fluctuations caused by load changes and anomalies caused by potential insulation aging or electrical faults; similarly, relying solely on vibration monitoring may be susceptible to external mechanical interference, leading to misjudgments of the transformer's condition.

[0028] To overcome the limitations of single-signal monitoring, a multi-modal data collaborative monitoring strategy can be adopted, integrating multi-dimensional information such as oil temperature, gas content, vibration characteristics, and internal magnetic field distribution for more accurate characterization of equipment health. However, in practical applications, the following problems often exist between multi-sensor signals: different sensors have significantly different sensitivities to different fault modes, which may lead to inconsistent inferences; signals are affected by external environmental interference or measurement noise, which may result in conflicts or outliers; different features have different time response rates, which may lead to inconsistent short-term results. These factors make it difficult for simple weighted averaging or fixed-weight methods to fully utilize multi-source information and dynamically adapt to complex operating environments, thus resulting in insufficient accuracy and robustness of condition identification and risk assessment results.

[0029] This disclosure provides a method, apparatus, storage medium, and electronic device for transformer risk assessment. The main purpose is to address the technical problem that related technologies typically rely on the analysis results of a single sensor or a fixed number of parameters to conduct transformer risk assessments, which are difficult to fully reflect the true health status of the transformer.

[0030] like Figure 1 As shown, embodiments of this disclosure provide a transformer risk assessment method, including:

[0031] Step 101: Obtain different monitoring data of the transformer.

[0032] The transformer may include, but is not limited to, oil-immersed transformers, and the different monitoring data include dissolved gas data in the oil, vibration data, oil temperature data, and magnetic field data.

[0033] In some embodiments, online monitoring data of the transformer can be obtained through sensors and monitoring devices installed on the transformer body or related circuits. Monitoring data can also be collected offline, such as dissolved gas data in oil, vibration data, oil temperature data, and magnetic field data (magnetic field distribution data).

[0034] Step 102: Conduct risk assessments on different monitoring data, obtain the risk assessment probabilities corresponding to different monitoring data, and the confidence levels corresponding to different risk assessment probabilities.

[0035] In some embodiments, for each type of independent monitoring data, a targeted professional analysis method or model can be used to conduct risk assessment to determine the probability of the specific fault mode indicated by the data occurring, thereby assessing the fault occurrence probability corresponding to each type of detection data and determining the risk assessment probability for each type of data. Secondly, to address the uncertainty of the assessed risk assessment probability, multiple methods can be used to calculate the confidence level of different risk assessment probabilities corresponding to different monitoring data from different analytical perspectives. This quantifies the reliability (credibility) of the risk assessment probability, reflecting the uncertainty of the assessment result, thereby improving the accuracy and comprehensiveness of transformer risk assessment and reducing assessment bias and uncertainty.

[0036] Before step 102, the process may further include: acquiring risk assessment models corresponding to different monitoring data; the risk assessment models include a first assessment model, a second assessment model, a third assessment model, and a fourth assessment model; the risk assessment probabilities include a first assessment probability, a second assessment probability, a third assessment probability, and a fourth assessment probability; the first assessment model is used to determine a sub-feature set using the minimum redundancy maximum correlation method, and input the sub-feature set into a Gaussian Naive Bayes model, outputting the first assessment probability corresponding to the dissolved gas data in the oil in different monitoring data; the second assessment model is used to extract time-frequency and time-domain features from the vibration signal by performing wavelet packet decomposition and short-time Fourier transform, and fuse the original, time-frequency, and time-domain sequence features through a multi-scale one-dimensional convolutional neural network, outputting the second assessment probability corresponding to the vibration data in different monitoring data; the third assessment model is used to acquire historical temperature curves distributed at multiple points, and outputs the third assessment probability corresponding to the oil temperature data in different monitoring data based on the predicted temperature increase amplitude and its confidence interval range; the fourth assessment model is used to model using internal magnetic field data from multiple locations, and outputs the fourth assessment probability corresponding to the magnetic field data in different monitoring data based on the spatiotemporal distribution characteristics of magnetic flux density.

[0037] Step 103: Determine the risk distribution information corresponding to different monitoring data based on different risk assessment probabilities and the corresponding confidence levels.

[0038] In some embodiments, these independent assessment results, each with a credibility weight, are integrated to determine the overall risk distribution information of the transformer. The risk distribution information may include the probability of occurrence of different risk levels of the transformer, such as low risk, medium risk, and high risk, etc., and may include the probability of occurrence of each risk level. For example, the probability of the transformer being in a "normal" state is 70%, in a "note" state (requiring attention) is 20%, in an "abnormal" state (requiring immediate inspection) is 8%, and in a "serious" state (requiring immediate handling) is 2%. This overall probability distribution can more clearly demonstrate the overall picture and concentration trend of transformer risks, providing a more comprehensive and accurate depiction of the various risk states faced by the transformer and their probabilities. This improves the reliability of risk distribution information, effectively overcomes the bias and uncertainty brought about by analysis of a single sensor or a few fixed parameters, and thus comprehensively reflects the true health status of the transformer.

[0039] Step 104: Based on the risk distribution information, determine the risk assessment results for the transformer.

[0040] In some embodiments, risk distribution information can be further analyzed, and corresponding risk assessment results can be given based on the risk distribution. The risk assessment results may include, but are not limited to, target assessment level, risk change trend, risk assessment conclusion, risk assessment report, risk distribution details, risk source, confidence level, and operation and maintenance decision-making suggestions.

[0041] For example, the target assessment level can be the overall risk assessment level finally determined based on multiple risk levels, the risk source can be the main driving factors corresponding to different levels in the risk assessment level or risk distribution information, the confidence level can be the credibility of the risk assessment level, and the operation and maintenance decision suggestions can be the operation and maintenance suggestions given for the risk assessment level. The final risk level can be determined based on the principle of maximum probability (such as the risk level with the highest probability of occurrence as the final judgment result) or cumulative probability (such as if the cumulative probability of occurrence of different risk levels exceeds a certain preset threshold (such as 5% or 10%), then it can be judged as a higher level of risk).

[0042] In this way, the overall condition of the transformer can be assessed in a timely manner, so that corresponding preventive measures can be taken in advance for different risk levels, reducing the possibility of sudden equipment failure and ensuring the safe and stable operation of the transformer.

[0043] Compared with related technologies, this embodiment first acquires different monitoring data of the transformer; then, it performs risk assessments on each of the different monitoring data, obtaining the risk assessment probability corresponding to each monitoring data, and the confidence level corresponding to each risk assessment probability; next, based on the different risk assessment probabilities and the corresponding confidence levels, it determines the risk distribution information corresponding to the different monitoring data, including the monitoring probability of different risk levels of the transformer; finally, based on the risk distribution information, it determines the risk assessment result of the transformer. In this way, this disclosure can perform independent risk assessments on different monitoring data of the transformer, improving the comprehensiveness of the transformer risk assessment. By determining the risk distribution information of the transformer based on multiple risk assessment probabilities and their confidence levels, it improves the reliability of the risk distribution information, effectively overcoming the bias and uncertainty caused by analysis of a single sensor or a few fixed parameters, and thus comprehensively reflecting the true health status of the transformer.

[0044] Based on the technical implementation described in the above embodiments, in order to further illustrate the specific implementation process of the method in this embodiment, this embodiment provides the following: Figure 2 The specific method shown includes:

[0045] Step 201: Obtain different monitoring data of the transformer.

[0046] In some embodiments, the different monitoring data include dissolved gas data in the oil, vibration data, oil temperature data, and magnetic field data. Various types of sensor devices can be used, installed at key structural parts and monitoring channels of the oil-immersed transformer, to acquire different monitoring data during transformer operation, such as multi-source historical state data. For example, multi-source historical state data may include dissolved gas data in the oil, vibration data (mechanical vibration), oil temperature data, and magnetic field data. The specific data acquisition process S1 may include the following steps:

[0047] S11: Gas detection in the oil employs an online monitoring device installed in the bypass pipeline of the transformer oil circulation system. The device includes an automatic oil sample acquisition module, a gas-liquid separation membrane assembly, and a micro-chromatographic / infrared spectroscopy analysis unit. After being drawn by a diaphragm pump, the oil sample passes through the gas-liquid separation membrane. Dissolved gases in the oil are released under temperature control and negative pressure and enter the chromatographic column or infrared detection chamber. Based on the absorption spectral characteristics or residence time of gas molecules, the components of key gases such as hydrogen, methane, acetylene, ethylene, and carbon monoxide are separated and their concentrations quantified. The obtained gas concentration signals are then converted from analog to digital and uploaded to the data acquisition system. The sampling period can be set from 10 min to 30 min, thus forming continuous time-series gas concentration data.

[0048] Subsequently, based on the characteristics of key gas component extraction closely related to health status, they were used in subsequent risk assessment models to identify potential risks of electrical discharge, thermal failure, and insulation aging.

[0049] Examples are shown in Table 1.

[0050] Table 1

[0051]

[0052] S12: In this embodiment, mechanical vibration is monitored using a triaxial piezoelectric accelerometer. The sensor is bolted to the surface of the transformer core tie plate and winding pressure plate to ensure good rigid coupling. The charge signal output by the sensor is processed by a charge amplifier and an anti-aliasing filter before being input to the data acquisition system. The sampling frequency is set to 25.6 kHz to ensure the ability to capture high-order harmonics and local impact signals. The vibration signal is subjected to Fast Fourier Transform (FFT) and wavelet packet decomposition to obtain the spectral energy distribution, resonance peak position, and transient impact characteristics. In addition, the time-domain characteristics of the vibration signal are calculated, including eight features: root mean square value, peak value, peak-to-peak value, peak factor, skewness, kurtosis, waveform factor, and margin factor, which are used to identify structural anomalies such as winding loosening and core vibration.

[0053] S13: During transformer temperature monitoring, platinum resistance temperature sensors (Pt100 / Pt1000) or fiber Bragg grating temperature sensors are deployed at multiple key locations in the transformer tank, including the upper and lower layers of the oil, the cooling oil flow channels, and typical hot spots in the core and windings. The temperature measurement points for the high-voltage and low-voltage windings are densely distributed in the upper half and sparsely distributed in the lower half, such as... Figure 3 As shown in the figure, each temperature sensor periodically sends the collected temperature values ​​to the data acquisition system, forming a multi-point distributed historical temperature curve, which is used to analyze changes in heat load, cooling effect, and overheating risk.

[0054] S14: A Hall sensor array is used as a magnetic field sensor, installed in the end area of ​​the transformer windings, specifically in the oil gap near the leads of the high-voltage and low-voltage windings. The Hall sensor arrays are arranged symmetrically on both sides of the winding ends. The number of individual sensor elements in each sensor array corresponds to the number of turns in the winding, thus achieving a one-to-one correspondence between the sensor measurement points and the inter-turn positions. This arrangement can completely cover the magnetic field distribution of the winding cross-section in space, and the symmetrical arrangement enhances the sensitivity and anti-interference capability of anomaly detection.

[0055] Under normal operating conditions, the magnetic field distribution output by the sensors on both sides should remain approximately symmetrical. If an inter-turn short circuit or local arc discharge occurs at a certain location in the winding, it will cause a significant difference in the output of the sensors on both sides at the corresponding location, which can be used as a basis for identifying electrical anomalies. The sensor acquires the first sample at a fixed sampling period. magnetic flux density of individual sensors And calculate the first magnetic flux of each sensor ,in, For the first Each sensor effectively measures the area.

[0056] For a given phase of a winding, the distribution of its leakage magnetic field can be obtained by superimposing the leakage magnetic fields generated by the currents in each winding. This process allows us to establish a superposition model: where the current in each winding collectively generates the leakage magnetic field, and this relationship can be represented by a linear leakage magnetic field output matrix. Based on the principle of linear superposition, the magnetic flux density vector measured by all sensors and the winding current vector have the following transformation relationship:

[0057] .

[0058] Wherein, coupling matrix M The matrix elements are determined by the geometry and spatial position of the windings. Quantitatively characterized the first The current of the first winding is related to the first winding current. The model determines the contribution of the magnetic field at each sensor location. It establishes a deterministic mapping from current to magnetic field. When the winding is operating normally, the coupling matrix exhibits a specific symmetry pattern due to the structural symmetry, ensuring symmetry in the sensor outputs on both sides. However, an inter-turn short circuit not only alters the resistance and inductance parameters of the faulty winding but also disrupts the spatial magnetic field symmetry by changing the current distribution. This manifests as changes in the corresponding elements of the coupling matrix, leading to significant differences in the sensor outputs. These differences can be accurately detected and identified by real-time monitoring of the magnetic field symmetry.

[0059] Step 202: Obtain the risk assessment model corresponding to different monitoring data.

[0060] For example, risk assessment models can be constructed for the collected historical dissolved gas data, historical vibration data, historical oil temperature data, and historical internal magnetic field distribution data. The risk assessment models may include, but are not limited to, the first assessment model (first risk assessment model), the second assessment model (second risk assessment model), the third assessment model (third risk assessment model), and the fourth assessment model (fourth risk assessment model), etc., to achieve independent assessment of different monitoring data and improve the reliability of the assessment of each data.

[0061] Optionally, the first evaluation model can be used to determine the sub-feature set through the minimum redundancy maximum correlation method, and input the sub-feature set into the Gaussian Naive Bayes model to output the first evaluation probability corresponding to the dissolved gas data in oil in different monitoring data.

[0062] The first risk assessment model first selects features using the minimum redundancy maximum correlation (mRMR) method to obtain the most discriminative sub-feature set. Using this sub-feature set as input, a Gaussian Naive Bayes (GNB) model based on the Gaussian distribution assumption is employed. This model assumes conditional independence between features and calculates the probability of the fault category corresponding to the current input based on Bayes' theorem. After training, it can output the first assessment probability of the current sample being in a "normal," "warning," or "fault" state. This method is lightweight, easily interpretable, highly stable with small sample data, and can quickly update the model distribution using prior probabilities.

[0063] Optionally, the second risk assessment model can be used to extract time-frequency and time-domain features (including eight time-domain features) from vibration signals by wavelet packet decomposition and short-time Fourier transform, and then fuse the original, time-frequency and time-domain sequence features through a multi-scale one-dimensional convolutional neural network to output the second assessment probability corresponding to the vibration data in different monitoring data.

[0064] Accordingly, a multi-scale convolutional neural network is employed for joint analysis of vibration signals in the time and time-frequency domains to construct a vibration risk assessment model, serving as the second risk assessment model. To enhance the model's ability to perceive different frequency disturbance modes, a 1D multi-scale convolutional structure is introduced, with convolutional kernels of different receptive field sizes used to extract local and global change information. The model structure includes three convolutional channels, processing the original sequence, time-domain feature sequence, and time-frequency domain feature sequence respectively, and feature concatenation is performed in the fusion layer. Subsequently, the fault state probability, i.e., the second assessment probability, is output through a fully connected layer. This method exhibits good vibration abnormal mode discrimination capability and can effectively identify vibration characteristics of different fault types such as mechanical loosening and core failure.

[0065] Optionally, the third risk assessment model can be used to obtain historical temperature curves distributed across multiple points, and output the third assessment probability corresponding to oil temperature data in different monitoring data based on the predicted temperature increase and its confidence interval range.

[0066] Accordingly, a Variational Bayesian Regression (VBR) model based on variational Bayesian inference is used to model the multi-point oil temperature sequence and establish a third risk assessment model. The original multi-point oil temperature sequence is detrended and normalized, and a time-lag feature vector is constructed. External variables such as ambient temperature and load current are introduced as fusion inputs, with the temperature increase within a future time window as the prediction target. A regression prediction model is constructed using VBR, and its inference process approximates the posterior parameter through a variational distribution, providing an uncertainty distribution while maintaining predictive performance. When the predicted temperature increase exceeds a preset threshold, or its lower confidence interval exceeds a safety limit, the current temperature operating state is determined to be abnormal, and the third assessment probability is output. This method exhibits good robustness when handling data with uncertainty and can reflect the risk evolution process behind the temperature rise trend.

[0067] Optionally, the fourth evaluation model is used to model using internal magnetic field data from multiple locations, and outputs the fourth evaluation probability corresponding to the magnetic field data in different monitoring data based on the spatiotemporal distribution characteristics of magnetic flux density.

[0068] Accordingly, a Graph Attention Network (GAT) mechanism is employed to model the spatial distribution characteristics of the magnetic field, enabling regional risk inference of magnetic field anomalies. First, a spatial sensing graph G=(V, E) is constructed using Hall sensor arrays deployed on both sides of the stator bar ends as nodes, where point V represents a sensor measurement point and edge E represents the spatial adjacency relationship between sensors. For the time-series magnetic flux signal input to each node, a weighted aggregation feature is calculated using the attention mechanism. The model outputs the probability value of overall structural anomalies, i.e., the fourth evaluation probability, through a global pooling layer. Simultaneously, by utilizing the magnetic field data at each point and its spatial correlation characteristics, local anomalies such as inter-turn short circuits are identified, making it suitable for spatially distributed sensing tasks.

[0069] Step 203: Using different risk assessment models, conduct risk assessments on different monitoring data to generate risk assessment probabilities corresponding to different monitoring data.

[0070] In some embodiments, different risk assessment models can be used to conduct risk assessments on different monitoring data collected in real time, and obtain the risk assessment probabilities corresponding to different monitoring data. The risk assessment probabilities may include, but are not limited to, the first assessment probability, the second assessment probability, the third assessment probability, the fourth assessment probability, etc., so as to achieve targeted risk assessment for different types of monitoring data.

[0071] For example, the first assessment probability can be the risk assessment probability obtained by the first risk assessment model based on dissolved gas data in oil for transformer risk assessment; the second assessment probability can be the risk assessment probability obtained by the second risk assessment model based on vibration data for transformer risk assessment; the third assessment probability can be the risk assessment probability obtained by the third risk assessment model based on oil temperature data for transformer risk assessment; and the fourth assessment probability can be the risk assessment probability obtained by the fourth risk assessment model based on internal magnetic field distribution data for transformer risk assessment.

[0072] Step 204: Using different confidence assessment strategies, assess the confidence level of the risk assessment probability of different monitoring data and obtain the different confidence levels corresponding to the risk assessment probability.

[0073] In some embodiments, to improve the accuracy and robustness of risk probability result fusion, a confidence assessment mechanism is introduced, which can assess the confidence of different assessment probabilities output by different assessment models and calculate their corresponding confidence levels.

[0074] Specifically, the four risk assessment model outputs obtained in stage S2, such as the first assessment probability, the second assessment probability, the third assessment probability, and the fourth assessment probability, are used to calculate the corresponding confidence values ​​for each risk assessment model. For each risk assessment model, four types of confidence assessment factors can be constructed to evaluate the relative reliability of each model's prediction in the current state, guiding the subsequent weighted fusion process to obtain multiple confidence levels. For example, for the first assessment model, multiple confidence levels of the first assessment probability output by the first assessment model can be calculated, such as first confidence level, second confidence level, third confidence level, and fourth confidence level.

[0075] Optionally, the first confidence level corresponding to the risk assessment probability can be determined based on the probability difference index and the distribution information entropy inverse index corresponding to the risk assessment probability. The first confidence level can be used to assess the discrimination clarity of the risk assessment probability.

[0076] For example, risk assessment probabilities may include the monitoring probabilities of each assessment model monitoring the three risk levels, and the monitoring probability distribution of the three risk levels output by the current model. It can be represented as ,in, These can represent the monitoring probability of low-risk level, medium-risk level, and high-risk level, respectively.

[0077] For example, the first confidence assessment may specifically include the following two sub-indicators: the maximum-minimum probability difference index. Inverse index of distributed information entropy The formulas for calculating the two indicators are as follows:

[0078] ;

[0079] .

[0080] Among these, a larger maximum-minimum probability difference index indicates a significant bias in the model's judgment, resulting in concentrated output and lower reliability. The distribution information entropy inverse index quantifies the dispersion of the current probability distribution, where the logarithm is the natural logarithm, and a smaller entropy value indicates greater concentration. Therefore, the overall value is incremented by 1 to obtain the first confidence level (positive confidence vector). It can be represented as:

[0081] .

[0082] In this way, we can measure whether the probability distribution of risk levels output by the current risk assessment model has a clear bias, thereby judging the clarity and credibility of the model's current judgment results.

[0083] Optionally, a second confidence level can be determined based on the change range of the risk assessment probability at different times. The second confidence level can be used to conduct a time-series consistency assessment of the risk assessment probability.

[0084] For example, the assessment probabilities output by each assessment model at multiple time points can be obtained. Based on the magnitude of change between the risk assessment probability output of each assessment model at the current time point and the risk assessment probability output at the previous time point, it can be determined whether the prediction results of each assessment model are continuous and smooth. If there are sharp jumps, it is judged as unstable, and the confidence level is reduced. For example, if the risk level probability assessed by the first assessment model at the current time point is... The probability of the risk level assessed in the previous time step at the current moment is: Then the change in the model evaluation result can be defined as follows: Further construct the second confidence level. The calculation formula is as follows:

[0085] .

[0086] in, To adjust the parameters, The larger this value, the stronger the transition penalty. A transition threshold can be set to limit... When the confidence level exceeds a certain range, the confidence level is directly lowered.

[0087] Optionally, the third confidence level of the risk assessment probability corresponding to the current assessment model can be determined based on the average probability of the risk assessment probability corresponding to different risk assessment models. The third confidence level is used to assess the model consistency of the risk assessment probability.

[0088] For example, the third confidence level can be used to measure the consistency between the current model output and the outputs of other models. If the current model output deviates significantly from the average judgment result of other models, its confidence level should be reduced. Specifically, the third confidence level can be... The model output probability distribution As discrete probability variables At the same time, except for the first Other models besides the first model (the first model) The average output probability distribution is formed by averaging the output probabilities of each model. :

[0089] .

[0090] Correspondingly, the third confidence level The calculation formula is as follows:

[0091] .

[0092] Optionally, based on the risk assessment probability corresponding to the preset time step, a fourth confidence level is determined, which is used to assess the historical stability of the risk assessment probability.

[0093] For example, the method of this embodiment can be used to evaluate the performance stability and judgment accuracy of various risk assessment models over a period of time, in order to measure their long-term reliability. Assuming historical... Within the first time step, the risk assessment model... The output probability result at time step is The final result after fusion is Fourth confidence level The calculation formula is as follows:

[0094] .

[0095] Step 205: Determine the risk distribution information corresponding to different monitoring data based on different risk assessment probabilities and the corresponding confidence levels.

[0096] Optionally, step 205 may specifically include: determining the comprehensive confidence score corresponding to different risk assessment models based on the different confidence levels corresponding to the risk assessment probabilities; determining the model weights of different risk assessment models based on the comprehensive confidence scores corresponding to different risk assessment models and the conflict detection results of different risk assessment models; and determining the risk distribution information corresponding to different monitoring data based on the model weights of different risk assessment models.

[0097] In some embodiments, risk distribution information may include a comprehensive risk probability distribution, which can be weighted and aggregated based on different confidence scores and their weights from different risk assessment models to synthesize a comprehensive confidence score at the current time. :

[0098] ;

[0099] .

[0100] In the formula, The weights can represent the weights of the first confidence level, the second confidence level, the third confidence level, and the fourth confidence level, respectively. Each weight can be set based on historical statistical performance or experience. Finally, the confidence scores of each risk assessment model are normalized and used as dynamic input variables in the game weighting strategy.

[0101] Optionally, a weighted fusion stage can be performed using the four model evaluation results and their confidence values ​​obtained in the confidence calculation stage. The fusion process may specifically include the following steps:

[0102] S41: Conflict Detection and Direct Weighting: Detect the maximum difference in probability distributions for each of the four three-class classifications. If the maximum difference is less than a preset threshold, the results are considered to have no significant conflict. In this case, the confidence scores of the four models are normalized to obtain the weight vector. ,in, These can be represented by the confidence levels of the first, second, third, and fourth assessment models, respectively. The overall risk probability distribution is then calculated based on the confidence levels of these multiple models. :

[0103] .

[0104] In the formula, It can represent the first A risk assessment model.

[0105] S42: Conflict and Game Trigger: If the maximum difference is greater than or equal to a preset threshold, the game weighting phase begins. At this point, a four-partner non-cooperative game model is established, where each participant corresponds to a risk assessment model and a payoff function. Determined by confidence level This return value reflects the overall credibility of the current model under multidimensional reliability evaluation.

[0106] S43: Game Iteration and Weight Update: In the game weighting phase, initial weights are first assigned to the four risk assessment models. The initial weights are calculated based on the confidence percentage of each model. Then, an iterative update process begins, with the weights being calculated at the [number]th [number]th [number]. In each iteration, the average return of all models is calculated. And based on the difference between the profit value of each model and the average profit value, the first Model weights Adjustments were made to increase or decrease the weights. It can be represented as follows:

[0107] .

[0108] Accordingly, the weights are normalized after the adjustment. This update mechanism ensures that the weights of models with returns above the average are gradually increased, while the weights of models with returns below the average are gradually decreased, thereby dynamically optimizing the weight allocation during the iteration process. The iteration process stops when the change in the weights of each model is below a preset convergence threshold in multiple consecutive iterations, and the weight distribution at this point is taken as the final game result.

[0109] Optionally, after obtaining the convergent weight distribution of the game, the weights are weighted and calculated with the corresponding three-category risk probability distribution to obtain the comprehensive risk probability distribution. To prevent drastic fluctuations in the comprehensive result due to short-term conflicts, the comprehensive risk probability distribution is smoothed using an exponential weighting method, i.e., the weighted average of the current weighted result and the smoothed result at the previous moment is used as the final output probability distribution. The final smoothed comprehensive risk probability distribution is used for risk level classification and early warning, ensuring a balance between accuracy and stability in risk assessment results. After smoothing, the obtained comprehensive risk probability distribution can be used to reflect the probability values ​​of the current operating status of oil-immersed transformers at three levels: low risk, medium risk, and high risk.

[0110] By introducing dynamic confidence calculation and game-theoretic weighting mechanism, the weights can be adjusted in real time when there may be conflicts in the results of multiple sensors, which significantly improves the stability and accuracy of risk assessment results. At the same time, the output risk probability results have quantitative interpretability, which is more convenient for operation and maintenance decision-making and risk management in engineering practice.

[0111] Step 206: Based on the risk distribution information, determine the risk assessment results of the transformer.

[0112] Optionally, the risk assessment results may include, but are not limited to, the target risk level. Step 206 may specifically include: determining the target risk level of the transformer based on the risk distribution information and the level threshold corresponding to the risk level.

[0113] The target risk level can be the final determined risk level of the transformer at the current moment. Specifically, after obtaining the comprehensive risk probability distribution, the system can classify the operating status of the oil-immersed transformer into three levels—high risk, medium risk, and low risk—based on preset threshold values ​​corresponding to different risk levels, thus determining the target risk level at the current moment. The different threshold values ​​can be set based on statistical analysis of historical operating data and expert experience. For example, if the probability corresponding to a high risk level is greater than the first threshold, then the current moment can be classified as a high-risk state; if the high-risk probability is lower than the first threshold but higher than the second threshold, then it can be classified as a medium-risk state; otherwise, it can be classified as a low-risk state.

[0114] Optionally, the method in this embodiment may further include: generating risk warning information corresponding to the target risk level. If the target risk level is a high-risk level, then generating risk warning information corresponding to the high-risk level.

[0115] The risk warning information may include, but is not limited to, the confidence level of the target risk level, the risk source, and the risk change trend. The risk source may include key monitoring factors that affect the risk assessment result. Through this risk warning information, the interpretability of the judgment result can be effectively improved. While outputting the target risk level, the system also provides a confidence level explanation, indicating the credibility of the current level determination and the key monitoring factors that affect the result.

[0116] Optionally, the system can also compare and analyze the comprehensive risk probability distribution with historical monitoring data to generate a risk change trend chart. When the risk level increases, approaches the high-risk threshold, or the target risk level is high, the system can trigger an early warning function to generate corresponding risk warning information and send alarm information and suggested maintenance plans to the operation and maintenance personnel so that preventive measures can be taken in advance to reduce the possibility of sudden equipment failure.

[0117] For example, such as Figure 4 As shown, the method in this embodiment may specifically include the following:

[0118] S1: Collect historical data on dissolved gas in oil, vibration, oil temperature, and internal magnetic field distribution of oil-immersed transformers;

[0119] S2: Using historical dissolved gas data, historical vibration data, historical oil temperature data, and historical magnetic field distribution data, respectively, construct the corresponding first risk assessment model, second risk assessment model, third risk assessment model, and fourth risk assessment model;

[0120] S3: Input the real-time collected data on dissolved gas in oil, vibration data, oil temperature data, and internal magnetic field distribution into each risk assessment model, and output the first risk probability distribution (first assessment probability), the second risk probability distribution (second assessment probability), the third risk probability distribution (third assessment probability), and the fourth risk probability distribution (fourth assessment probability), respectively.

[0121] S4: Based on the model output results, perform confidence assessment, combining discrimination clarity assessment, temporal consistency assessment, cross-model consistency assessment, and historical stability assessment to calculate the first confidence level, second confidence level, third confidence level, and fourth confidence level; adopt an adaptive game-theoretic weighted fusion strategy to dynamically adjust the weights of each model according to the confidence level, and fuse them to obtain a comprehensive risk probability distribution; specifically, this may include the following steps:

[0122] S41: Calculate the confidence level for the first assessment probability, the second assessment probability, the third assessment probability, and the fourth assessment probability, respectively. The confidence level calculation includes:

[0123] Based on the discrimination clarity of the model output, assess the risk level tendency of the current risk probability distribution, and adjust the first confidence level according to the strength of the tendency;

[0124] Based on temporal consistency analysis, the change in risk probability output between the current time and the previous time is calculated, and the second confidence level is adjusted according to the change.

[0125] Based on cross-model consistency analysis, the similarity index between the output of this model and the output of other models is calculated, and the third confidence level is adjusted according to the similarity.

[0126] Based on historical performance statistics, the matching degree between the model's historical output and the fusion results or expert judgments is calculated, and the fourth confidence level is adjusted accordingly.

[0127] S42: Based on the first confidence level, second confidence level, third confidence level, and fourth confidence level, the first evaluation probability, second evaluation probability, third evaluation probability, and fourth evaluation probability are weighted using an adaptive game-theoretic weighted fusion strategy to obtain a comprehensive risk probability distribution. The adaptive game-theoretic weighted fusion strategy includes:

[0128] When the risk probability results are consistent or the deviation is within the set threshold, they are directly weighted and fused according to the confidence level ratio.

[0129] When the risk probability results conflict or the difference exceeds the set threshold, a multi-participant weighted non-cooperative game model is constructed, with the risk probability results as the strategy and the confidence level as the input of the payoff function. The weights of each model are updated iteratively until convergence.

[0130] The convergence weights and the outputs of each model are weighted and calculated to obtain the game fusion result;

[0131] The game fusion results are smoothed to avoid drastic changes in weights caused by short-term fluctuations;

[0132] S5: Based on the comprehensive risk probability distribution, risk levels are classified and early warnings are issued, and the current operating status of the oil-immersed transformer is analyzed.

[0133] In this way, the system can further classify risk levels based on the set risk probability thresholds and output the corresponding target risk level judgment results. Simultaneously, the system can also provide explanatory notes based on confidence analysis, historical comparison trends, and dynamic early warning suggestions, offering operations and maintenance personnel intuitive, reliable, and interpretable risk assessment auxiliary reports.

[0134] Compared with related technologies, this embodiment can collect multimodal historical monitoring data of oil-immersed transformers, including dissolved gas content in the oil, vibration characteristics, oil temperature information, and internal magnetic field distribution. Independent risk assessment models are constructed based on various data sources, yielding multiple risk assessment results. The confidence level of each risk assessment result is calculated according to signal characteristics and model performance. An adaptive game-theoretic weighted fusion strategy is used to dynamically adjust conflicting or consistent results based on the confidence level, ultimately outputting a quantified risk probability result. Therefore, by utilizing multimodal monitoring data of oil-immersed transformers, intelligent risk assessment for transformer operation risk alarms is achieved, improving the accuracy of transformer risk assessment and the reliability of decision support.

[0135] Based on the above Figure 1 and Figure 2 The specific implementation of the method shown in this embodiment provides a transformer risk assessment device, such as... Figure 5 As shown, the device includes: an acquisition module 31 and a determination module 32;

[0136] The acquisition module 31 is used to acquire different monitoring data of the transformer, including dissolved gas data in oil, vibration data, oil temperature data, and magnetic field data; to perform risk assessment on different monitoring data, and to acquire the risk assessment probability corresponding to different monitoring data, as well as the confidence level corresponding to different risk assessment probabilities;

[0137] The determination module 32 is used to determine the risk distribution information corresponding to different monitoring data based on different risk assessment probabilities and the confidence levels corresponding to different risk assessment probabilities. The risk distribution information includes the monitoring probabilities of different risk levels of the transformer; and based on the risk distribution information, the risk assessment result of the transformer is determined.

[0138] In some examples of this embodiment, the acquisition module 31 is specifically configured to acquire risk assessment models corresponding to different monitoring data; use different risk assessment models to perform risk assessments on different monitoring data respectively, and generate risk assessment probabilities corresponding to different monitoring data; use different confidence assessment strategies to perform confidence assessments on the risk assessment probabilities of different monitoring data respectively, and acquire different confidence levels corresponding to the risk assessment probabilities.

[0139] In some examples of this embodiment, the determining module 32 is specifically configured to: determine a first confidence level corresponding to the risk assessment probability based on the probability difference index and the distribution information entropy inverse index corresponding to the risk assessment probability; the first confidence level is used to assess the discrimination clarity of the risk assessment probability; determine a second confidence level corresponding to the risk assessment probability at the current time based on the change range of the risk assessment probability at different times; the second confidence level is used to assess the temporal consistency of the risk assessment probability; determine a third confidence level corresponding to the risk assessment probability of the current assessment model based on the average probability of the risk assessment probability corresponding to different risk assessment models; the third confidence level is used to assess the model consistency of the risk assessment probability; and determine a fourth confidence level corresponding to the risk assessment probability based on the risk assessment probability corresponding to a preset time step; the fourth confidence level is used to assess the historical stability of the risk assessment probability.

[0140] In some examples of this embodiment, the determining module 32 is specifically configured to determine the comprehensive confidence score corresponding to different risk assessment models based on the different confidence levels corresponding to the risk assessment probabilities; determine the model weights of different risk assessment models based on the comprehensive confidence scores corresponding to different risk assessment models and the conflict detection results of different risk assessment models; and determine the risk distribution information corresponding to different monitoring data based on the model weights of different risk assessment models.

[0141] In some examples of this embodiment, the acquisition module 31 is specifically configured as follows: the risk assessment model includes a first assessment model, a second assessment model, a third assessment model, and a fourth assessment model; the risk assessment probability includes a first assessment probability, a second assessment probability, a third assessment probability, and a fourth assessment probability; the first assessment model is used to determine a sub-feature set using the minimum redundancy maximum correlation method, and input the sub-feature set into a Gaussian Naive Bayes model, outputting the first assessment probability corresponding to the dissolved gas data in the oil in different monitoring data; the second assessment model is used to extract time-frequency and time-domain features from the vibration signal by performing wavelet packet decomposition and short-time Fourier transform, and fuse the original, time-frequency, and time-domain sequence features through a multi-scale one-dimensional convolutional neural network, outputting the second assessment probability corresponding to the vibration data in different monitoring data; the third assessment model is used to acquire the historical temperature curves distributed at multiple points, and output the third assessment probability corresponding to the oil temperature data in different monitoring data based on the predicted temperature increase amplitude and its confidence interval range; the fourth assessment model is used to model using internal magnetic field data at multiple locations, and output the fourth assessment probability corresponding to the magnetic field data in different monitoring data based on the spatiotemporal distribution characteristics of magnetic flux density.

[0142] In some examples of this embodiment, the risk assessment result includes a target risk level; the determination module 32 is further configured to determine the target risk level of the transformer based on the risk distribution information and the level threshold corresponding to the risk level; and generate risk warning information corresponding to the target risk level, which includes the confidence level of the target risk level, the risk source, and the risk change trend.

[0143] In some examples of this embodiment, the determining module 32 is further configured to generate risk warning information corresponding to the high-risk level if the target risk level is high-risk level.

[0144] Based on the above, Figure 1 and Figure 2 Accordingly, this embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. Figure 1 and Figure 2 The method shown.

[0145] Based on this understanding, the technical solution disclosed herein can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, or portable hard drive) and includes several instructions to cause a computer device (such as a personal computer, server, or network device) to execute the methods of various implementation scenarios of this disclosure.

[0146] Based on the above, Figure 1 and Figure 2 The method shown, and Figure 5 To achieve the above objectives, this disclosure also provides an electronic device, comprising a storage medium and a processor; the storage medium for storing a computer program; and the processor for executing the computer program to implement the above-described virtual device embodiments. Figure 1 and Figure 2 The method shown.

[0147] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0148] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0149] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0150] Through the above description of the embodiments, those skilled in the art can clearly understand that this disclosure can be implemented using software plus necessary general-purpose hardware platforms, or it can be implemented through hardware. Through the solution of this disclosure, this embodiment can collect multimodal historical monitoring data of oil-immersed transformers, including dissolved gas content in the oil, vibration characteristics, oil temperature information, and internal magnetic field distribution; construct independent risk assessment models based on multiple data sources to obtain multiple risk assessment results; calculate the confidence level corresponding to each risk assessment result based on signal characteristics and model performance; and dynamically adjust conflicting or consistent results according to the confidence level using an adaptive game-theoretic weighted fusion strategy, ultimately outputting a quantified risk probability result. Therefore, by using multimodal monitoring data of oil-immersed transformers, intelligent risk assessment for transformer operation risk alarms can be achieved, improving the accuracy of transformer risk assessment and the reliability of decision support.

[0151] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "comprising" or any other variations thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0152] The above are merely specific embodiments of this disclosure, enabling those skilled in the art to understand or implement this disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to these embodiments, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A transformer risk assessment method, characterized by, The method comprises the following steps: acquiring different monitoring data of a transformer, wherein the different monitoring data comprises dissolved gas in oil data, vibration data, oil temperature data, and magnetic field data; respectively performing risk assessment on the different monitoring data to obtain risk assessment probabilities corresponding to the different monitoring data, and confidence degrees corresponding to different risk assessment probabilities; determining risk distribution information corresponding to the different monitoring data according to the different risk assessment probabilities and the confidence degrees corresponding to the different risk assessment probabilities, wherein the risk distribution information comprises monitoring probabilities of different risk levels of the transformer; determining a risk assessment result of the transformer based on the risk distribution information; before the step of respectively performing risk assessment on the different monitoring data to obtain risk assessment probabilities corresponding to the different monitoring data, and confidence degrees corresponding to different risk assessment probabilities, the method further comprises the following steps: acquiring risk assessment models corresponding to the different monitoring data, wherein the risk assessment models comprise a first assessment model, a second assessment model, a third assessment model, and a fourth assessment model; the risk assessment probabilities comprise a first assessment probability, a second assessment probability, a third assessment probability, and a fourth assessment probability; the first assessment model is used to determine a sub-feature set by a minimum redundancy maximum relevance method, and input the sub-feature set into a Gaussian naive Bayes model to output the first assessment probability corresponding to the dissolved gas in oil data in the different monitoring data; the second assessment model is used to perform wavelet packet decomposition and short-time Fourier transform on a vibration signal to extract time-frequency features and time-domain features, and fuse original, time-frequency, and time-domain sequence features by a multi-scale one-dimensional convolutional neural network to output the second assessment probability corresponding to the vibration data in the different monitoring data; the third assessment model is used to acquire a historical temperature curve of a multi-point distribution, and output the third assessment probability corresponding to the oil temperature data in the different monitoring data according to a predicted temperature increase amplitude and a confidence interval range thereof; the fourth assessment model is used to model by using multi-position internal magnetic field data, and output the fourth assessment probability corresponding to the magnetic field data in the different monitoring data according to a time-space distribution characteristic of magnetic flux density.

2. The method of claim 1, wherein, the step of respectively performing risk assessment on the different monitoring data to obtain risk assessment probabilities corresponding to the different monitoring data, and confidence degrees corresponding to different risk assessment probabilities comprises the following steps: respectively performing risk assessment on the different monitoring data by using different risk assessment models to generate risk assessment probabilities corresponding to the different monitoring data; respectively performing confidence degree assessment on the risk assessment probabilities of the different monitoring data by using different confidence degree assessment strategies to obtain different confidence degrees corresponding to the risk assessment probabilities.

3. The method of claim 2, wherein, the step of respectively performing confidence degree assessment on the risk assessment probabilities of the different monitoring data by using different confidence degree assessment strategies to obtain different confidence degrees corresponding to the risk assessment probabilities comprises the following steps: determining a first confidence degree corresponding to a risk assessment probability according to a probability difference value index and a distribution information entropy reverse index corresponding to the risk assessment probability, wherein the first confidence degree is used to perform risk assessment probability discrimination clarity assessment; Determine a second confidence degree corresponding to the risk assessment probability of the current time according to a change range of the risk assessment probability at different times, and the second confidence degree is used for time consistency evaluation of the risk assessment probability. Determine a third confidence degree of the risk assessment probability corresponding to the current evaluation model based on an average probability of the risk assessment probability corresponding to different risk assessment models, and the third confidence degree is used for model consistency evaluation of the risk assessment probability. Determine a fourth confidence degree corresponding to the risk assessment probability according to the risk assessment probability corresponding to a preset time step, and the fourth confidence degree is used for historical stability evaluation of the risk assessment probability.

4. The method of claim 2, wherein, The determination of the risk distribution information corresponding to the different monitoring data according to the different risk assessment probabilities and the confidence degrees corresponding to the different risk assessment probabilities comprises: Determine a comprehensive confidence score corresponding to the different risk assessment models according to the different confidence degrees corresponding to the risk assessment probabilities; Determine a model weight of the different risk assessment models according to the comprehensive confidence scores corresponding to the different risk assessment models and the conflict detection results of the different risk assessment models; Determine the risk distribution information corresponding to the different monitoring data based on the model weights of the different risk assessment models.

5. The method as claimed in claim 1, wherein, The risk assessment result comprises a target risk level; The determination of the risk assessment result of the transformer based on the risk distribution information comprises: Determine a target risk level of the transformer according to a level threshold corresponding to the risk distribution information and a risk level; After the determination of the risk assessment result of the transformer based on the risk distribution information, the method further comprises: Generate risk prompt information corresponding to the target risk level, and the risk prompt information comprises a confidence degree of the target risk level, a risk source, and a risk change trend.

6. The method of claim 5, wherein, The method further comprises: If the target risk level is a high risk level, generate risk warning information corresponding to the high risk level.

7. A transformer risk assessment apparatus, characterized by, Comprise: An acquisition module is configured to acquire different monitoring data of a transformer, and the different monitoring data comprises dissolved gas in oil data, vibration data, oil temperature data, and magnetic field data. The different monitoring data are respectively subjected to risk assessment, and risk assessment probabilities corresponding to the different monitoring data and confidence degrees corresponding to different risk assessment probabilities are obtained. Before the different monitoring data are respectively subjected to risk assessment, and the risk assessment probabilities corresponding to the different monitoring data and the confidence degrees corresponding to different risk assessment probabilities are obtained, the method further comprises: obtaining a risk assessment model corresponding to the different monitoring data; the risk assessment model comprises a first assessment model, a second assessment model, a third assessment model and a fourth assessment model; the risk assessment probabilities comprise a first assessment probability, a second assessment probability, a third assessment probability and a fourth assessment probability; the first assessment model is used to determine a sub-feature set by a minimum redundancy maximum relevance method, and input the sub-feature set into a Gaussian naive Bayes model to output a first assessment probability corresponding to dissolved gas data in the different monitoring data; the second assessment model is used to perform wavelet packet decomposition and short-time Fourier transform to extract time-frequency features and time-domain features of a vibration signal, and fuse original, time-frequency and time-domain sequence features by a multi-scale one-dimensional convolutional neural network to output a second assessment probability corresponding to vibration data in the different monitoring data; the third assessment model is used to obtain a historical temperature curve of multi-point distribution, and output a third assessment probability corresponding to oil temperature data in the different monitoring data according to a predicted temperature increase amplitude and a confidence interval range thereof; and the fourth assessment model is used to model by using multi-position internal magnetic field data, and output a fourth assessment probability corresponding to magnetic field data in the different monitoring data according to a time-space distribution characteristic of magnetic flux density. A determination module is configured to determine risk distribution information corresponding to the different monitoring data according to the different risk assessment probabilities and the confidence degrees corresponding to the different risk assessment probabilities, wherein the risk distribution information comprises monitoring probabilities of different risk levels of the transformer; and determine a risk assessment result of the transformer based on the risk distribution information.

8. An electronic device, comprising: The method comprises: at least one processor; and a memory connected in communication with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.

9. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to enable the computer to perform the method of any one of claims 1-6.

Images