Power distribution network equipment operation risk dynamic assessment method based on digital twinning and svdd

Abstract

This invention discloses a dynamic risk assessment method for distribution network equipment operation based on digital twin and SVDD, relating to the field of distribution network monitoring and maintenance technology. The method includes: processing multi-dimensional observation data of the equipment, completing missing data, and quantifying its reliability to construct a reliability matrix and equipment feature vectors; constructing a reliable noise covariance matrix based on the reliability matrix, and updating the calibrable parameters of the digital twin model using model residuals to obtain reliable model parameters; constructing an SVDD input vector based on the equipment feature vectors, the reliable noise covariance matrix, model residuals, and reliable model parameters, and calculating the SVDD anomaly distance and equipment health; calculating the predicted temperature rise based on the twin prediction state within a preset prediction window, and coupling the predicted electrical stress with the predicted temperature rise to obtain a predicted comprehensive stress index; calculating the cumulative risk probability of the equipment within the preset prediction window based on the equipment health and the predicted comprehensive stress index, and calculating the equipment risk value based on the cumulative risk probability.

Description

Technical Field

[0001] This invention relates to the field of distribution network monitoring and operation and maintenance, specifically to a method for dynamic assessment of the operational risks of distribution network equipment based on digital twins and SVDD. Background Technology

[0002] With the increasing integration of high-proportion renewable energy sources and enhanced load fluctuations, distribution networks are characterized by complex operation modes, power flow reversals, and increased difficulty in coordinating low voltage and protection. The operating conditions of distribution network equipment change frequently and drastically, making them prone to overload and overheating, which in turn accelerates insulation aging and significantly increases the risk of failure. This places higher demands on the real-time, accurate, and forward-looking nature of equipment operation risk assessment.

[0003] Existing technologies for equipment health diagnosis and risk warning typically include: alarm methods based on thresholds / rules and expert experience; data-driven supervised / semi-supervised fault identification methods; health assessment methods based on anomaly detection and single-classification; and state estimation methods based on digital twins and mechanistic models. Among these, alarm methods based on thresholds / rules and expert experience issue alarms by setting thresholds or rules for voltage, current, temperature, load rate, and the number of switching actions. These methods are simple to implement and highly interpretable. However, their fixed thresholds or experience-based rule settings are difficult to adapt to data noise or missing measurements, fluctuations in renewable energy sources, and changes in operating conditions across multiple scenarios in distribution networks, easily leading to false alarms and missed alarms. Data-driven supervised / semi-supervised fault identification methods train classifiers or sequence models using historical fault labels to identify and predict equipment faults. The effectiveness of these methods is highly dependent on the quality of the label data. However, in distribution network environments, fault samples are scarce and their distribution drift is significant, resulting in insufficient generalization ability. Furthermore, the model output is mostly a fault category judgment, making it difficult to provide future window risk probability quantification results that can be used for operation and maintenance. Health assessment methods based on single-class anomaly detection (SVDD) identify equipment anomalies and health decline trends by learning the distribution of normal samples, overcoming the problem of scarce fault samples. However, traditional SVDD often directly scores observed features, failing to explicitly characterize the impact of multi-source data quality differences (such as missing data, noise, and drift) on the scores. Therefore, it is difficult to explain whether the abnormal results come from actual equipment degradation or data errors, affecting the credibility of the assessment results. State estimation methods based on digital twins and mechanistic models achieve physical interpretability of the analysis results by establishing equipment, network mechanistic models, and / or equivalent models for state estimation and simulation analysis. However, they face problems such as parameter uncertainty, large individual differences in equipment, and difficulties in online calibration on the distribution network side. At the same time, twin simulations often only output state quantities and lack quantitative indicators of risk probability and consequences for operation and maintenance, making it difficult to directly guide operation and maintenance classification.

[0004] In summary, existing risk assessment methods for power distribution network equipment have shortcomings such as weak perception capabilities, limited adaptability to specific scenarios, and insufficient generalization and consequence quantification capabilities. These shortcomings often result in inadequate accuracy, reliability, and / or interpretability of the assessment results, making it difficult to meet the usability requirements of engineering projects. Summary of the Invention

[0005] The purpose of this invention is to provide a dynamic risk assessment method for the operation of distribution network equipment based on digital twins and SVDD, so as to solve the problem that the existing risk assessment methods for distribution network equipment mentioned in the background art have low accuracy and are difficult to guide the operation and maintenance classification.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A dynamic risk assessment method for distribution network equipment operation based on digital twins and SVDD includes the following steps:

[0008] S100: Obtain multi-dimensional observation data of key equipment in the power distribution network, perform data processing and missing data completion on the multi-dimensional observation data to obtain a multi-dimensional completion vector, calculate the confidence value of each dimension of the multi-dimensional completion vector, construct the confidence matrix of the multi-dimensional completion vector based on the confidence value of each dimension, and construct the equipment feature vector based on the multi-dimensional completion vector and the confidence matrix.

[0009] S200: Construct a digital twin model of key equipment in the distribution network; construct a reliable noise covariance matrix based on the reference noise covariance matrix of the key equipment in the distribution network and the reliability matrix; update the calibrable parameters of the digital twin model based on the model residuals of the digital twin model and the reliable noise covariance matrix to obtain reliable model parameters; correct the one-step prediction state of the digital twin model to obtain the current optimal twin state.

[0010] S300: Construct an SVDD input vector based on the device feature vector, the reliable noise covariance matrix, the model residual, and the reliable model parameters; calculate the anomaly distance of the SVDD input vector; and calculate the device health based on the anomaly distance and a preset health threshold.

[0011] S400: Calculate the predicted electrical stress of key equipment in the distribution network within a preset future time period; calculate the twin prediction state of key equipment in the distribution network within a preset future time period based on the reliable model parameters and the current optimal twin state; calculate the predicted temperature rise of key equipment in the distribution network within a preset future time period based on the twin prediction state; calculate the predicted comprehensive stress index based on the predicted electrical stress and the predicted temperature rise.

[0012] S500: Based on the equipment health status and all the predicted comprehensive stress indicators within the preset future time period, calculate the cumulative risk probability of key equipment in the distribution network within the preset prediction window, and calculate the equipment risk value of key equipment in the distribution network within the preset prediction window based on the cumulative risk probability.

[0013] The principle of this invention, a dynamic risk assessment method for distribution network equipment operation based on digital twins and SVDD, is as follows: In the data acquisition and modeling stage, data observation dimensions are improved through data processing and missing data completion to reduce issues such as missing data and noise. A confidence matrix is ​​constructed by calculating the confidence values ​​of data in each dimension to quantify the data quality of the multi-dimensional completion vectors. Equipment feature vectors are constructed by fusing the confidence matrix and the multi-dimensional completion vectors, overcoming the shortcomings of traditional methods that assume all observed data are equally reliable. When constructing the digital twin model, a reliable noise covariance matrix is ​​constructed by fusing the equipment's baseline noise covariance matrix and the confidence matrix. The calibrable parameters are iteratively updated using the model residuals to obtain reliable model parameters adapted to real-time operating conditions. Simultaneously, the model's predicted state is corrected, outputting the optimal twin state that closely matches the actual operating conditions of the equipment, achieving dynamic adaptive optimization of the digital twin model based on operating conditions and data quality. The device feature vectors, reliable noise covariance matrix, model residuals, and reliable model are integrated into the model. The parameters are used together as the input vector of the SVDD model, enabling the SVDD-based equipment health score to simultaneously reflect equipment anomalies and model consistency deviations, thereby improving the interpretability and engineering usability of the health score. Based on the adaptively optimized optimal twin state, the twin prediction state is projected forward, and the predicted temperature rise is calculated. Combined with the predicted electrical stress, the future predicted comprehensive stress index of the equipment is calculated, which improves the accuracy of depicting the future stress trajectory. By combining the equipment health and comprehensive stress index over multiple future time periods, the cumulative risk probability of the equipment within a preset prediction window is coupled and further converted into the equipment risk value, taking into account both risk probability and consequence quantification, providing reliable technical support for differentiated operation and maintenance decisions of the power grid. This invention achieves dynamic and forward-looking assessment of the operational risks of distribution network equipment through data-trusted modeling, digital twin adaptive calibration, SVDD anomaly identification, and stress risk coupled deduction, improving the accuracy and sensitivity of risk assessment and providing quantitative indicators for operation and maintenance classification.

[0014] Preferably, to reduce the impact of missing data and noise fluctuations on the observation data and improve the data reliability in the observation data modeling stage, in step S100, the data processing and missing data completion of the multidimensional observation data to obtain a multidimensional completion vector specifically includes:

[0015] S101: Based on the multidimensional observation data, construct a multidimensional observation vector for each device at each observation time. The multidimensional observation vector is represented in the following form:

[0016] ;

[0017] In the formula, For equipment At the observation time The multidimensional observation vector, The total number of observed dimensions. For transpose operation, , ,……and respectively equipment At the observation time The first dimension, the second dimension, ... and the third dimension 3D observation data;

[0018] S102: Based on Kalman filtering, filter trend estimation is performed on each dimension of the observation data to obtain the estimated value of the trend component of each dimension of the observation data. The calculation formula for the filter trend estimation is as follows:

[0019] ;

[0020] ;

[0021] In the formula, For equipment At the observation time The 3D observation data, are positive integers and , and respectively equipment At the observation time The Trend and disturbance components of the observed data For equipment At the observation time The Estimated trend components of the dimensional observation data. This represents the Kalman filter operator. For equipment From the first observation time to the first All the observation times Historical observation sequences of dimensional observation data;

[0022] S103: Based on the amplitude limiting function, the disturbance component obtained from the filtered trend estimation, and the estimated value of the trend component, amplitude limiting and padding operations are performed on each dimension of the observation data to obtain the padded observation values. The expression for the amplitude limiting and padding operation is as follows:

[0023] ;

[0024] In the formula, For equipment At the observation time The After dimensional completion, the observed values ​​are... For the amplitude limiting function, For the first The preset amplitude limit threshold corresponding to the dimensional observation data;

[0025] S104: Combine the observed values ​​after all dimensions are filled in according to the dimensional order to obtain the multidimensional filling vector.

[0026] Preferably, to achieve differentiated weighting of observations with different data qualities, highlighting high-quality data while reducing the weight of low-quality data, in step S100, the formula for calculating the confidence value of each dimension of the multidimensional completion vector is as follows:

[0027] ;

[0028] In the formula, For equipment At the observation time The Credibility of dimensions For the first Credible weights of dimensions.

[0029] Preferably, in step S200, the reliable noise covariance matrix is ​​constructed using the following formula:

[0030] ;

[0031] In the formula, For equipment At the observation time The reliable noise covariance matrix, For equipment The baseline noise covariance matrix, This indicates element-wise multiplication. For equipment At the observation time The credibility matrix;

[0032] The fixed noise covariance matrix of traditional digital twin models cannot adapt to the differences in data quality at different observation times, resulting in a large deviation between the filtering and residual calculation results. By inversely weighting the reliability matrix with the benchmark noise covariance matrix, an adaptive data quality reliable noise covariance matrix can be obtained, which can improve the noise matching accuracy of digital twin models.

[0033] Preferably, in step S200, the method for obtaining the reliable model parameters includes: constructing a parameter optimization objective function using a weighted least squares method with L2 regularization constraints, solving for the minimum value of the parameter optimization objective function, and obtaining the reliable model parameters; the expression of the parameter optimization objective function is:

[0034] ;

[0035] In the formula, For equipment At the observation time The vector representation of the parameters of the reliable model. For indexing historical observation times, For equipment In the The residual vector between the digital twin state at time step and the actual observed value. Indicates device In the Credible noise covariance matrix at time step The inverse matrix, The L2 regularization coefficient is... For the vector representation of the calibrable parameter variable to be updated, For equipment The vector representation of the initial calibrable parameters;

[0036] Traditional digital twin models have difficulty optimizing calibrable parameters based on real-time residuals and noise weighting, leading to a decrease in model accuracy over time. Using weighted least squares with L2 regularization constraints to solve for reliable model parameters can improve the accuracy of online updates of calibrable parameters in digital twin models and suppress overfitting.

[0037] Preferably, in step S300, the construction formula for the SVDD input vector is:

[0038] ;

[0039] In the formula, For equipment At the observation time SVDD input vector, For equipment At the observation time The device feature vector, The expression is:

[0040] ;

[0041] In the formula, For feature mapping function, For equipment At the observation time The multidimensional complement vector, and These are multidimensional complement vectors. The th wavelet decomposition Layer approximate components and the first Layer detail components, For the decomposition level index, This represents the total number of layers in the multi-scale wavelet decomposition.

[0042] Traditional SVDD-based equipment health assessment methods do not consider the quality differences of multi-source data, making it difficult to distinguish between actual equipment anomalies and measurement errors in the assessment results. By constructing a multi-scale SVDD input vector that integrates observations, wavelet multi-scale features, weighted residuals, and reliable model parameters, we can highlight the actual abnormal state characteristics of the equipment and reduce the impact of measurement errors and noise on subsequent health scores.

[0043] Preferably, in step S300, the formula for calculating the device health is:

[0044] ;

[0045] In the formula, For equipment At the observation time The health of the equipment. For the sigmoid function, This is a scaling factor for health. For equipment At the observation time abnormal distance, Set a preset health threshold.

[0046] Preferably, in step S400, the calculation formula for the predicted comprehensive stress index is:

[0047] ;

[0048] In the formula, For equipment In the Predicted comprehensive stress index at any time. Indexing future moments and These are the weighting coefficients for the electrical stress term and the thermal stress term, respectively. For equipment In the Predicted electrical stress at time, For equipment The maximum allowable current, For equipment In the Predicted temperature rise at any time For equipment The maximum allowable temperature rise.

[0049] Preferably, in step S500, the formula for calculating the cumulative risk probability is:

[0050] ;

[0051] In the formula, Indicates device From the observation time Starting from the preset prediction window, the cumulative risk probability. To preset the number of time steps included in the prediction window, For time step, For equipment In the Risk rate at any time The calculation formula is:

[0052] ;

[0053] In the formula, For equipment The benchmark risk rate, For health sensitivity coefficient, The comprehensive stress sensitivity coefficient, The coupling sensitivity coefficient;

[0054] Traditional risk assessment methods for distribution network equipment only consider the linear effect of equipment health and / or future operating stress, which makes it difficult to characterize the rapid deterioration of equipment caused by high stress and poor health in the distribution network. By introducing a coupling term of equipment health and comprehensive stress into the risk rate, the risk rate of equipment with poor health rises faster under the same stress, thereby enhancing the model's discriminative power and early warning sensitivity.

[0055] Preferably, to quantify the consequences of equipment failure and provide direct numerical guidance for operation and maintenance decisions, in step S500, the formula for calculating the equipment risk value is as follows:

[0056] ;

[0057] In the formula, For equipment From the observation time Starting with the risk value within the preset prediction window, For equipment The consequence coefficient.

[0058] One or more technical solutions provided by this invention have at least the following technical effects or advantages:

[0059] 1. High-quality modeling of field observation data is achieved through trend estimation, missing data completion, amplitude limiting filtering, and credibility quantification, which improves the resilience of subsequent risk assessment results to data fluctuations; the credibility matrix is ​​used as one of the core variables throughout subsequent twin modeling, parameter calibration, and health scoring, which enables the model to adaptively adjust according to data quality, thereby improving the ability to extrapolate and assess dynamic risks.

[0060] 2. By continuously correcting the twin parameters and state estimates using the consistency deviation between real-time observation data and twin predicted states, and taking equipment observation characteristics, twin consistency deviation, and twin calibration parameters as inputs to the SVDD health score, the online calibration of the equipment digital twin model and the SVDD health score are coupled, so that the health score can not only reflect the surface observation anomalies, but also reflect the real degradation trend corresponding to the decline in model consistency, thereby improving the interpretability and robustness of the risk assessment results;

[0061] 3. Focusing on future time windows, the system utilizes the predicted future electrical stress of the equipment and combines it with the future thermal stress obtained by extrapolating from the calibrated twin model to form a unified comprehensive stress characterization. By coupling the current health status of the equipment with the future comprehensive stress, the system can extrapolate the probability of risks within the future window, obtaining equipment risk indicators that take into account both the probability of anomalies and their consequences. These indicators can be used to guide equipment risk ranking, graded early warning, and priority decision-making for operation and maintenance resources, thus upgrading the existing method of detecting anomalies and issuing alarms to a dynamic risk assessment method that proactively quantifies risks and provides early warnings. Attached Figure Description

[0062] The accompanying drawings, which are provided to further illustrate embodiments of the invention and constitute a part of this invention, are not intended to limit the scope of the invention.

[0063] Figure 1 This is a schematic diagram of the dynamic risk assessment method for power distribution network equipment operation based on digital twin and SVDD in this invention;

[0064] Figure 2 This refers to the completed observations obtained after trend estimation and missing data completion of multidimensional observation data in Embodiment 2 of the present invention, as well as the missing data distribution of observation data in each dimension.

[0065] Figure 3 This is a distribution diagram of the confidence value of each dimension of the observed data after multidimensional completion at each time step in Embodiment 2 of the present invention.

[0066] Figure 4 This is the multi-scale wavelet decomposition result in Embodiment 2 of the present invention, taking the temperature rise dimension as an example to complete the data;

[0067] Figure 5These are the current optimal twin state and reliable model parameters of the digital twin model in the historical time domain in Embodiment 2 of the present invention;

[0068] Figure 6 These are the abnormal distance distribution curves and equipment health distribution curves calculated by SVDD in Embodiment 2 of the present invention;

[0069] Figure 7 These are the historical distributions and predicted values ​​for future periods of equipment load and distributed energy output in Embodiment 2 of the present invention;

[0070] Figure 8 These are the predicted electrical stress distribution curve, predicted temperature rise distribution curve, and predicted comprehensive stress index distribution curve for the future time period in Embodiment 2 of the present invention.

[0071] Figure 9 This is the risk rate curve of the device in the future time period in Embodiment 2 of the present invention;

[0072] Figure 10 This is the cumulative risk probability of the device at future times in Embodiment 2 of the present invention;

[0073] Figure 11 It is the final risk value of the device at a future time in Embodiment 2 of the present invention. Detailed Implementation

[0074] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, where there is no conflict, the embodiments of the present invention and the features thereof can be combined with each other.

[0075] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0076] Example 1

[0077] Please refer to Figure 1 The present invention provides a method for dynamic assessment of the operational risk of distribution network equipment based on digital twins and SVDD, comprising the following steps:

[0078] S100: Obtain multi-dimensional observation data of key equipment in the power distribution network, perform data processing and missing data completion on the multi-dimensional observation data to obtain a multi-dimensional completion vector, calculate the confidence value of each dimension of the multi-dimensional completion vector, construct the confidence matrix of the multi-dimensional completion vector based on the confidence value of each dimension, and construct the equipment feature vector based on the multi-dimensional completion vector and the confidence matrix.

[0079] S200: Construct a digital twin model of key equipment in the distribution network; construct a reliable noise covariance matrix based on the reference noise covariance matrix of the key equipment in the distribution network and the reliability matrix; update the calibrable parameters of the digital twin model based on the model residuals of the digital twin model and the reliable noise covariance matrix to obtain reliable model parameters; correct the one-step prediction state of the digital twin model to obtain the current optimal twin state.

[0080] S300: Construct an SVDD input vector based on the device feature vector, the reliable noise covariance matrix, the model residual, and the reliable model parameters; calculate the anomaly distance of the SVDD input vector; and calculate the device health based on the anomaly distance and a preset health threshold.

[0081] S400: Calculate the predicted electrical stress of key equipment in the distribution network within a preset future time period; calculate the twin prediction state of key equipment in the distribution network within a preset future time period based on the reliable model parameters and the current optimal twin state; calculate the predicted temperature rise of key equipment in the distribution network within a preset future time period based on the twin prediction state; calculate the predicted comprehensive stress index based on the predicted electrical stress and the predicted temperature rise.

[0082] S500: Based on the equipment health status and all the predicted comprehensive stress indicators within the preset future time period, calculate the cumulative risk probability of key equipment in the distribution network within the preset prediction window, and calculate the equipment risk value of key equipment in the distribution network within the preset prediction window based on the cumulative risk probability.

[0083] In step S100, the multidimensional observation data is processed and missing data is imputed to obtain a multidimensional imputed vector, specifically including:

[0084] S101: Based on the multidimensional observation data, construct a multidimensional observation vector for each device at each observation time. The multidimensional observation vector is represented in the following form:

[0085] ;

[0086] In the formula, For equipment At the observation time The multidimensional observation vector, The total number of observed dimensions. For transpose operation, , ,……and respectively equipment At the observation time The first dimension, the second dimension, ... and the third dimension 3D observation data;

[0087] S102: Based on Kalman filtering, filter trend estimation is performed on each dimension of the observation data to obtain the estimated value of the trend component of each dimension of the observation data. The calculation formula for the filter trend estimation is as follows:

[0088] ;

[0089] ;

[0090] In the formula, For equipment At the observation time The 3D observation data, are positive integers and , and respectively equipment At the observation time The Trend and disturbance components of the observed data For equipment At the observation time The Estimated trend components of the dimensional observation data. This represents the Kalman filter operator. For equipment From the first observation time to the first All the observation times Historical observation sequences of dimensional observation data;

[0091] S103: Based on the amplitude limiting function, the disturbance component obtained from the filtered trend estimation, and the estimated value of the trend component, amplitude limiting and padding operations are performed on each dimension of the observation data to obtain the padded observation values. The expression for the amplitude limiting and padding operation is as follows:

[0092] ;

[0093] In the formula, For equipment At the observation time The After dimensional completion, the observed values ​​are... For the amplitude limiting function, For the first The preset amplitude limit threshold corresponding to the dimensional observation data. According to Criteria, Calculate the first Under normal operating conditions, the disturbance component of the historical observation data Standard deviation and take as ,in It can be either 2 or 3;

[0094] S104: Combine the observed values ​​after all dimensions are filled in according to the dimensional order to obtain a multidimensional filled vector.

[0095] In step S100, the formula for calculating the confidence value of each dimension of the multidimensional completion vector is as follows:

[0096] ;

[0097] In the formula, For equipment At the observation time The Credibility of dimensions For the first Credible weights of dimensions;

[0098] The expression for the constructed credibility matrix is:

[0099] ;

[0100] In the formula, For equipment At the observation time The credibility matrix This represents a diagonal matrix.

[0101] In step S100, the expression for the device feature vector is:

[0102] ;

[0103] ;

[0104] In the formula, For equipment At the observation time The device feature vector, For feature mapping function, For equipment At the observation time The multidimensional complement vector, and These are multidimensional complement vectors. The th wavelet decomposition Layer approximate components and the first Layer detail components, For the decomposition level index, This represents the total number of layers in the multi-scale wavelet decomposition.

[0105] In step S200, the expression for the constructed digital twin model is:

[0106] ;

[0107] In the formula, For equipment At the observation time The twin hidden state For equipment In the next moment The recursive twin hidden states obtained from the prediction deduction For equipment The calibrable parameters of the digital twin model, For equipment The parameterized state transition matrix, For equipment The input mapping matrix, For equipment At the observation time External control input, For equipment At the observation time Process noise or random disturbance terms;

[0108] in, By solving the observation equation, the expression of the observation equation is obtained as follows:

[0109] ;

[0110] ;

[0111] ;

[0112] In the formula, For equipment At the observation time System observations, For the observation mapping function, For equipment The observation matrix For equipment At the observation time Observation noise, For equipment At the observation time The reliable noise covariance matrix, This indicates that the mean is 0 and the covariance is... The multidimensional Gaussian normal distribution.

[0113] In step S200, the reliable noise covariance matrix is ​​constructed using the following formula:

[0114] ;

[0115] In the formula, For equipment The baseline noise covariance matrix, This indicates element-wise multiplication.

[0116] Wherein, the model residual of the digital twin model is the difference between the system observations of the device and the twin state variables, and the formula for calculating the model residual is:

[0117] ;

[0118] In the formula, For equipment At the observation time The model residuals of the digital twin model, For equipment At the observation time The one-step prediction state of the twin hidden state.

[0119] In step S200, the reliable model parameters are obtained by: constructing a parameter optimization objective function using a weighted least squares method with L2 regularization constraints, solving for the minimum value of the parameter optimization objective function, and obtaining the reliable model parameters; the expression of the parameter optimization objective function is:

[0120] ;

[0121] In the formula, For equipment At the observation time The vector representation of the parameters of the reliable model. For indexing historical observation times, Indicates device In the Credible noise covariance matrix at time step The inverse matrix, The L2 regularization coefficient is... Offline calibration can be performed based on the inherent physical characteristics of each device. For the vector representation of the calibrable parameter variable to be updated, For equipment The vector representation of the initial calibrable parameters.

[0122] In step S200, the expression for the current optimal twin state is:

[0123] ;

[0124] In the formula, For equipment At the observation time The current optimal twin state, For equipment At the observation time The Kalman gain matrix.

[0125] In step S300, the formula for constructing the SVDD input vector is:

[0126] ;

[0127] In the formula, For equipment At the observation time SVDD input vector.

[0128] In step S300, the formula for calculating the health status of the device is as follows:

[0129] ;

[0130] In the formula, For equipment At the observation time The health of the equipment. For the sigmoid function, This is a scaling factor for health. According to the equipment The alarm sensitivity needs to be tuned offline to control the steepness of the drop in health as the degree of degradation deviates. For equipment At the observation time abnormal distance, , To describe the distance function for support vector data, To preset the health threshold, Through the device The statistical distribution quantiles of historical deviations under normal operating conditions are used to classify the normal healthy range and the abnormal deterioration range.

[0131] In step S400, the calculation method for the predicted electrical stress includes: calculating the predicted load values ​​of key equipment in the distribution network at future times, and the predicted output values ​​of distributed energy resources in the distribution network at future times; and calculating the predicted electrical stress of the equipment based on the predicted load values ​​and the predicted output values ​​of distributed energy resources, combined with the distribution network topology information and mapping function. The formula for calculating the predicted electrical stress is as follows:

[0132] ;

[0133] In the formula, For equipment In the Predicted electrical stress at time, Indexing future moments For equipment The mapping function between the driving quantity and the current is used in this embodiment, which is the power flow mapping. For equipment In the The predicted load at any given time. For the distribution network in the first Predicted output of distributed energy resources at any given time. This indicates the topology information of the power distribution network.

[0134] In step S400, the formula for calculating the predicted temperature rise is:

[0135] ;

[0136] ;

[0137] In the formula, For equipment In the Predicted temperature rise at any time For equipment The temperature rise readout vector, and respectively equipment In the Time and the Twin prediction state at any given moment For equipment In the Predicting external control inputs at specific times.

[0138] In step S400, the calculation formula for the predicted comprehensive stress index is as follows:

[0139] ;

[0140] In the formula, For equipment In the Predicted comprehensive stress index at any time. and These are the weighting coefficients for the electrical stress term and the thermal stress term, respectively. and According to the equipment The proportions of samples caused by current overload and overheating in the historical fault samples were determined to meet the following conditions. ,and , , For equipment The maximum allowable current, For equipment The maximum allowable temperature rise.

[0141] In step S500, the formula for calculating the cumulative risk probability is as follows:

[0142] ;

[0143] In the formula, Indicates device From the observation time Starting from the preset prediction window, the cumulative risk probability. To preset the number of time steps included in the prediction window, For time step, For equipment In the Risk rate at any time The calculation formula is:

[0144] ;

[0145] In the formula, For equipment The benchmark risk rate, For health sensitivity coefficient, The comprehensive stress sensitivity coefficient, The coefficient is the coupling sensitivity coefficient. , and Based on device The historical equipment health status, historical comprehensive stress index, and historical failure / outage / aging records were obtained by fitting using the least squares method.

[0146] In step S500, the formula for calculating the equipment risk value is as follows:

[0147] ;

[0148] In the formula, For equipment From the observation time Starting with the risk value within the preset prediction window, For equipment The consequence coefficient, According to the equipment The topology hierarchy, power supply load importance, and fault outage loss cost classification are calibrated offline.

[0149] Example 2

[0150] Based on Example 1, Example 2 will be described and illustrated with specific implementation cases.

[0151] Implementation Case 1: Verifying that the present invention can still stably assess equipment health and future risks under conditions of multi-source measurement deficiencies, noise disturbances, fluctuating operating conditions, and equipment degradation; please refer to... Figure 2-11 This embodiment constructs a set of repeatable synthetic data in MATLAB 2021b and performs a full-process simulation, with the following settings: historical time domain length. Discrete moments, time step Preset prediction window length That is, based on 600 steps of historical data, predict the risk rate for the next 48 steps, with a total number of observation dimensions. The simulation includes current, temperature rise, vibration / fluctuation characteristics, and environmental impact; missing and anomaly settings: different missing rates are set for each dimension, namely 10%, 8%, 15%, and 5%, with a small number of spikes and glitches superimposed to simulate acquisition anomalies; degradation range: a range is set from the historical sequence... arrive A gradual degradation window is defined, resulting in a larger current and easier temperature accumulation under the same drive conditions, to simulate degradation effects such as insulation aging, increased contact resistance, and poor heat dissipation; Risk extrapolation parameters: a baseline risk rate is set. Health sensitivity coefficient Comprehensive stress sensitivity coefficient Coupling sensitivity coefficient Consequence coefficient .

[0152] First, for each observation dimension Kalman filtering is used to filter and estimate the trend of historical observations, resulting in estimated values ​​of the trend components. Then, amplitude limiting and padding operations are performed on the observation data of each dimension to obtain the padded observation sequence. ,like Figure 2 As shown, Figure 2 (a) shows the Kalman filter trend estimation and the observations after missing measurements are filled, taking the current dimension as an example. Figure 2 (b) shows the missing data distribution for each dimension in the historical time domain; for each dimension's completed observations, the confidence value for each dimension is calculated based on the degree of deviation between the observed values ​​and the trend, forming a confidence matrix. ,like Figure 3 As shown, Figure 3 The distribution of confidence values ​​for each dimension at each time step is plotted; multi-scale wavelet decomposition is performed on the observations after dimension completion, such as... Figure 4 As shown, Figure 4 The results are shown in the multi-scale wavelet decomposition example using the temperature rise dimension; combined with multi-dimensional complement vectors. Multi-scale wavelet decomposition results and confidence matrix Constructing device feature vectors .

[0153] Then, a digital twin model of the device is constructed, and the credibility matrix is ​​added. Mapped to a reliable noise covariance matrix This method enables automatic weight reduction of low-confidence data and uses weighted least squares with L2 regularization constraints to update the twin parameters online, thereby obtaining the parameters of the reliable model. The current optimal twin state is obtained using Kalman filtering. ,like Figure 5 As shown, Figure 5 (a) represents the current optimal twin state within the historical time domain. Figure 5 (b) represents the reliable model parameters in the historical time domain.

[0154] Next, the device feature vector Confidential noise covariance matrix Model residuals and reliable model parameters Combined into SVDD input vectors Utilizing historical time domain from arrive and from arrive The data was used to train a compact descriptive boundary for normal samples, and the calculation was performed from... arrive Abnormal distance within the progressive degradation window ,Will Mapped to device health , to obtain Figure 6 The distribution curves of equipment anomaly distance and equipment health status are shown, where, Figure 6 (a) shows the abnormal distance distribution curve of the equipment. Figure 6 (b) is the equipment health distribution curve; from Figure 6 It can be seen that, in arrive During this period, the equipment health level was close to 1, indicating that the equipment had experienced abnormal degradation.

[0155] Then, based on the load distribution and distributed energy output distribution over the historical time domain, the load forecast for the next 48 steps is predicted. and distributed energy output forecast ,like Figure 7 As shown, combining the topology / power flow mapping function Calculate the predicted electrical stress distribution over the next 48 steps using information on the distribution network topology. reuse Reliable model parameters at time step and the current optimal twin state By projecting the twin state 48 steps ahead and combining it with the temperature rise readout vector, the predicted temperature rise distribution within the next 48 steps can be calculated. This will predict the distribution of electrical stress. Compared with the predicted temperature rise distribution Unified into a sequence of predicted comprehensive stress indices for the next 48 steps. ,like Figure 8 As shown, Figure 8 (a) shows the predicted electrical stress distribution curve over the next 48 steps. Figure 8 (b) shows the predicted temperature rise distribution curve over the next 48 steps. Figure 8 (c) is the distribution curve of the predicted comprehensive stress index within the next 48 steps.

[0156] Finally, according to Device health at all times And the predicted comprehensive stress index sequence within the next 48 steps The risk rate sequence of the computing device over the next 48 steps , to obtain Figure 9 The risk rate curve of the device over the next 48 steps is shown. Then, the risk rates over the next 48 steps are discretely accumulated to obtain the cumulative risk probability of the device within the next 48-step window. Combined with the consequences coefficient of the equipment The final risk value of the device at a future time in step 648 is calculated. ,like Figure 10 and Figure 11 As shown, Figure 10 The cumulative risk probability of the device at a future time in step 648. Figure 11 This represents the final risk value of the device at a future moment in step 648.

[0157] according to Figure 6 (b) Figure 10 and Figure 11 Based on the calculation results, the device simulated in this embodiment is... The current device health is 0.6253, within the prediction window length. The cumulative risk probability at the future moment (i.e., moment 648) is 0.4729, and the corresponding equipment risk value is 3.7835.

[0158] Figure 2-4 The results show that the observation data of a certain device simulated in this embodiment at 600 historical observation times, after trend estimation, missing measurement completion, amplitude limiting filtering and credibility quantification, achieves high-quality modeling of the observation data; Figure 5The results show that by mapping the confidence matrix to a confidence noise covariance matrix and further applying it to the parameter update and twin calibration of the equipment digital twin model, consistent modeling and updating of the twin model and equipment observations are achieved, thus improving the accuracy of the equipment digital twin model. Figure 6 The results show that by using equipment observation characteristics, twin consistency bias, and twin calibration parameters as inputs to the SVDD health score, the health score can track and reflect the actual degradation trend of the equipment, thus improving the interpretability and engineering usability of the health score. Figure 7-11 The results show that by combining future load forecasting, output forecasting, power flow mapping and twin forward projection, a future comprehensive stress trajectory can be constructed. Combined with the current health score of the equipment, the risk rate and cumulative risk probability of the equipment in the future time window can be deduced. Finally, the risk value of the equipment is obtained by combining the consequence coefficient. This achieves a balance between risk probability and consequences, and provides a graded and forward-looking quantitative risk indicator for resource scheduling and operation and maintenance.

[0159] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0160] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A dynamic assessment method for the operational risk of distribution network equipment based on digital twins and SVDD, characterized in that, Includes the following steps: S100: Obtain multi-dimensional observation data of key equipment in the power distribution network, perform data processing and missing data completion on the multi-dimensional observation data to obtain a multi-dimensional completion vector, calculate the confidence value of each dimension of the multi-dimensional completion vector, construct the confidence matrix of the multi-dimensional completion vector based on the confidence value of each dimension, and construct the equipment feature vector based on the multi-dimensional completion vector and the confidence matrix. S200: Construct a digital twin model of key equipment in the distribution network; construct a reliable noise covariance matrix based on the reference noise covariance matrix of the key equipment in the distribution network and the reliability matrix; update the calibrable parameters of the digital twin model based on the model residuals of the digital twin model and the reliable noise covariance matrix to obtain reliable model parameters; correct the one-step prediction state of the digital twin model to obtain the current optimal twin state. S300: Construct an SVDD input vector based on the device feature vector, the reliable noise covariance matrix, the model residual, and the reliable model parameters, and calculate the anomaly distance of the SVDD input vector; The device health status is calculated based on the abnormal distance and the preset health threshold. S400: Calculates the predicted electrical stress of key equipment in the power distribution network within a preset future time period; Based on the reliable model parameters and the current optimal twin state, calculate the twin prediction state of key equipment in the distribution network within a preset future time period; based on the twin prediction state, calculate the predicted temperature rise of key equipment in the distribution network within a preset future time period; based on the predicted electrical stress and the predicted temperature rise, calculate the predicted comprehensive stress index. S500: Based on the equipment health status and all the predicted comprehensive stress indicators within the preset future time period, calculate the cumulative risk probability of key equipment in the distribution network within the preset prediction window, and calculate the equipment risk value of key equipment in the distribution network within the preset prediction window based on the cumulative risk probability.

2. The method for dynamic assessment of operational risks of distribution network equipment based on digital twin and SVDD as described in claim 1, characterized in that, In step S100, the data processing and missing data completion of the multidimensional observation data to obtain a multidimensional completion vector specifically includes: S101: Based on the multidimensional observation data, construct a multidimensional observation vector for each device at each observation time. The multidimensional observation vector is represented in the following form: ； In the formula, For equipment At the observation time The multidimensional observation vector, The total number of observed dimensions. For transpose operation, , ,……and respectively equipment At the observation time The first dimension, the second dimension, ... and the third dimension 3D observation data; S102: Based on Kalman filtering, filter trend estimation is performed on each dimension of the observation data to obtain the estimated value of the trend component of each dimension of the observation data. The calculation formula for the filter trend estimation is as follows: ； ； In the formula, For equipment At the observation time The 3D observation data, are positive integers and , and respectively equipment At the observation time The Trend and disturbance components of the observed data For equipment At the observation time The Estimated trend components of the dimensional observation data. This represents the Kalman filter operator. For equipment From the first observation time to the first All the observation times Historical observation sequences of dimensional observation data; S103: Based on the amplitude limiting function, the disturbance component obtained from the filtered trend estimation, and the estimated value of the trend component, amplitude limiting and padding operations are performed on each dimension of the observation data to obtain the padded observation values. The expression for the amplitude limiting and padding operation is as follows: ； In the formula, For equipment At the observation time The After dimensional completion, the observed values ​​are... For the amplitude limiting function, For the first The preset amplitude limit threshold corresponding to the dimensional observation data; S104: Combine the observed values ​​after all dimensions are filled in according to the dimensional order to obtain the multidimensional filling vector.

3. The method for dynamic assessment of distribution network equipment operation risk based on digital twin and SVDD as described in claim 2, characterized in that, In step S100, the formula for calculating the confidence value of each dimension of the multidimensional completion vector is as follows: ； In the formula, For equipment At the observation time The Credibility of dimensions For the first Credible weights of dimensions.

4. The method for dynamic assessment of distribution network equipment operation risk based on digital twin and SVDD as described in claim 3, characterized in that, In step S200, the reliable noise covariance matrix is ​​constructed using the following formula: ； In the formula, For equipment At the observation time The reliable noise covariance matrix, For equipment The baseline noise covariance matrix, This indicates element-wise multiplication. For equipment At the observation time The credibility matrix.

5. The method for dynamic assessment of distribution network equipment operation risk based on digital twin and SVDD as described in claim 4, characterized in that, In step S200, the reliable model parameters are obtained by: constructing a parameter optimization objective function using a weighted least squares method with L2 regularization constraints, solving for the minimum value of the parameter optimization objective function, and obtaining the reliable model parameters; the expression of the parameter optimization objective function is: ； In the formula, For equipment At the observation time The vector representation of the parameters of the reliable model. For indexing historical observation times, For equipment In the The residual vector between the digital twin state at time step and the actual observed value. Indicates device In the Credible noise covariance matrix at time step The inverse matrix, The L2 regularization coefficient is... For the vector representation of the calibrable parameter variable to be updated, For equipment The vector representation of the initial calibrable parameters.

6. The method for dynamic assessment of operational risks of distribution network equipment based on digital twin and SVDD as described in claim 5, characterized in that, In step S300, the formula for constructing the SVDD input vector is: ； In the formula, For equipment At the observation time SVDD input vector, For equipment At the observation time The device feature vector, The expression is: ； In the formula, For feature mapping function, For equipment At the observation time The multidimensional complement vector, and These are multidimensional complement vectors. The th wavelet decomposition Layer approximate components and the first Layer detail components, For the decomposition level index, This represents the total number of layers in the multi-scale wavelet decomposition.

7. The method for dynamic assessment of operational risks of distribution network equipment based on digital twin and SVDD as described in claim 6, characterized in that, In step S300, the formula for calculating the health status of the equipment is: ； In the formula, For equipment At the observation time The health of the equipment. For the sigmoid function, This is a scaling factor for health. For equipment At the observation time abnormal distance, Set a preset health threshold.

8. The method for dynamic assessment of distribution network equipment operation risk based on digital twin and SVDD as described in claim 7, characterized in that, In step S400, the formula for calculating the predicted comprehensive stress index is: ； In the formula, For equipment In the Predicted comprehensive stress index at any time. Indexing future moments and These are the weighting coefficients for the electrical stress term and the thermal stress term, respectively. For equipment In the Predicted electrical stress at time, For equipment The maximum allowable current, For equipment In the Predicted temperature rise at any time For equipment The maximum allowable temperature rise.

9. The method for dynamic assessment of operational risks of distribution network equipment based on digital twin and SVDD as described in claim 8, characterized in that, In step S500, the formula for calculating the cumulative risk probability is: ； In the formula, Indicates device From the observation time Starting from the preset prediction window, the cumulative risk probability. To preset the number of time steps included in the prediction window, For time step, For equipment In the Risk rate at any time The calculation formula is: ； In the formula, For equipment The benchmark risk rate, For health sensitivity coefficient, The comprehensive stress sensitivity coefficient, This is the coupling sensitivity coefficient.

10. The method for dynamic assessment of operational risks of distribution network equipment based on digital twin and SVDD according to claim 9, characterized in that, In step S500, the formula for calculating the equipment risk value is: ； In the formula, For equipment From the observation time Starting with the risk value within the preset prediction window, For equipment The consequence coefficient.

Images