A power distribution equipment digital twin driven intelligent fault early warning method and system

Abstract

This application discloses a fault intelligent early warning method and system driven by digital twins for power distribution equipment, belonging to the field of power system digital twin automation. The method includes: acquiring measured data from physical equipment and simulation data from the twin model, calculating and standardizing the residual sequence; extracting statistical features of the residuals using a sliding window; triggering adaptive model correction when the residual evolution conforms to preset aging characteristics, and using a recursive least squares algorithm with a forgetting factor to identify and correct key physical characteristic parameters in the twin model online, this correction being based on the aging characteristic curve constructed from the Arrhenius equation; based on the corrected model, calculating the dynamic confidence interval through Monte Carlo simulation, and combining the mutation criterion and cumulative sum control chart algorithm for fault early warning. This application solves the problems of virtual-physical mismatch and false alarms caused by equipment aging in traditional methods, achieving accurate identification and early warning of early, minor faults.

Description

Technical Field

[0001] This application belongs to the field of digital twin automation of power systems, and specifically relates to a fault intelligent early warning method and system driven by digital twin of power distribution equipment. Background Technology

[0002] With the deepening of smart grid construction, digital twin technology has been widely applied in the operation, maintenance, and fault early warning of power distribution equipment. Digital twins achieve deep perception and accurate simulation of the operating status of power distribution equipment by constructing a real-time mapping of physical entities in digital space. This technology can integrate multi-source sensor data and combine it with physical mechanism models, providing crucial digital support for the safe and stable operation of power systems and has become a core means to improve the intelligence level of the power Internet of Things.

[0003] Among them, the fault early warning method based on digital twins aims to identify potential risks in power distribution equipment in advance through a two-way mapping between physical entities and virtual models. This technology typically combines the electromagnetic and thermal conductivity characteristics of the power distribution equipment to construct a high-fidelity simulation model, and captures abnormal fluctuations during equipment operation by comparing and analyzing real-time monitoring data with the model output. Ideally, this method can effectively identify equipment operational deviations, providing real-time and scientific reference for operation and maintenance decisions.

[0004] However, traditional power distribution equipment early warning systems often rely on fixed threshold settings, making it difficult to address the drift in physical characteristics caused by environmental stress and material aging during long-term operation. Existing digital twin early warning methods typically use initial design parameters to construct static models. As equipment service time increases, the residual between the measured physical values ​​and the twin simulation values ​​gradually increases due to performance degradation, leading to a significant mismatch between the virtual and real spaces. In practical applications, this deviation is easily misjudged as equipment failure, resulting in high-frequency false alarms, or the system may be forced to raise the warning threshold, masking early, weak faults such as partial discharge or poor contact. Furthermore, existing methods lack in-depth analysis of the residual evolution and online correction mechanisms for model parameters, making it difficult to achieve synchronous evolution between the twin model and the equipment state. Consequently, the accuracy of the early warning results cannot meet the precise operation and maintenance requirements under complex operating conditions. Therefore, a fault intelligent early warning solution based on digital twin-driven power distribution equipment is desired. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for intelligent fault early warning driven by digital twin of power distribution equipment, so as to solve the problems in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A fault intelligent early warning method driven by digital twin of power distribution equipment includes the following specific steps: Acquire real-time operating status data of physical power distribution equipment, and based on a pre-built high-fidelity digital twin model, acquire simulation status data at the same time scale as the real-time operating status data; Based on real-time operating status data and simulation status data, a multidimensional residual sequence is calculated, and the multidimensional residual sequence is standardized to obtain a standardized residual sequence. The sliding window technique is used to extract statistical features of the standardized residual series, including mean shift. When the mean offset exceeds the preset aging judgment threshold in multiple consecutive sliding windows, and the residual change rate matches the aging rate described by the aging characteristic curve constructed based on the Arrhenius equation, the model adaptive correction is triggered. The adaptive model correction includes: using a recursive least squares algorithm with a forgetting factor to identify and correct key physical characteristic parameters in the high-fidelity digital twin model online. Key physical characteristic parameters include the contact resistance of the conductor and the dielectric constant of the insulating medium. Based on the modified high-fidelity digital twin model, dynamic confidence intervals are calculated through Monte Carlo simulation, and combined with mutation criteria and cumulative sum control chart algorithms, fault warnings are provided for physical power distribution equipment.

[0007] Furthermore, constructing a high-fidelity digital twin model includes: Establish a refined three-dimensional geometric model with dimensional errors of the physical power distribution equipment within a predetermined range; Configure temperature-dependent electrical conductivity parameters, nonlinear magnetization permeability parameters, and thermal conductivity parameters for the three-dimensional geometric model; Based on the finite element analysis method, a two-way coupled calculation framework for electromagnetic field and temperature field is established. In the two-way coupled calculation framework, the loss distribution obtained from electromagnetic field calculation is used as a heat source and applied to temperature field calculation, and the convergence criterion for coupled iterative calculation is set. Refined modeling of key physical properties is carried out, including: establishing a contact resistance model using Hertzian contact theory and Holm conductivity theory, as well as establishing a convective heat transfer model and a thermal radiation model.

[0008] Furthermore, the use of the sliding window technique to extract statistical features from the standardized residual series also includes: Within each sliding window, calculate the variance and rate of change of the standardized residual series; Data quality is assessed by analyzing changes in variance, and when variance increases sharply, the corresponding data segments are marked or removed to control data quality. By comparing the mean offset between adjacent windows, trend drift is identified when the mean offset of multiple consecutive windows shows a monotonically increasing or monotonically decreasing trend and the magnitude of the change exceeds the preset drift judgment threshold.

[0009] Furthermore, the aging characteristic curves constructed based on the Arrhenius equation are used to describe the evolution of key physical property parameters with operating time and operating temperature. The pre-exponential factor and activation energy of the Arrhenius equation are obtained by fitting accelerated aging experimental data. The residual change rate matches the aging rate described by the aging characteristic curve constructed based on the Arrhenius equation. Specifically, the ratio of the residual change rate to the theoretical aging rate calculated by the Arrhenius equation is within a preset range.

[0010] Furthermore, the objective function of the recursive least squares algorithm with a forgetting factor is to minimize the weighted sum of squares of the measured residuals. The forgetting factor takes a value in the range of 0.95 to 0.99 to balance the algorithm's ability to remember historical data with its tracking speed of parameter drift. The corrected contact resistance of the conductor is updated to the impedance boundary conditions of the contact interface in the high-fidelity digital twin model, and the corrected dielectric constant of the insulating medium is updated to the material property matrix of the insulating medium in the high-fidelity digital twin model.

[0011] Furthermore, the method for determining the preset aging judgment threshold is as follows: at the initial stage of equipment operation, a predetermined number of sampling points are selected to calculate the standard deviation of the mean offset, and the aging judgment threshold is set to a predetermined multiple of the standard deviation. The triggering conditions for adaptive model correction also include: the residual change rate is in the same direction as the aging rate described by the aging characteristic curve constructed based on the Arrhenius equation.

[0012] Furthermore, the dynamic confidence intervals calculated through Monte Carlo simulation include: A probability distribution model for Monte Carlo simulation input is constructed, which considers the probability distribution of current measurement error, voltage measurement error, ambient temperature measurement error, and load current fluctuation. Based on the modified high-fidelity digital twin model, a predetermined number of Monte Carlo simulations were performed to obtain the statistical distribution characteristics of the simulation output values. Based on the preset significance level, and combined with the mean and standard deviation of the simulation output values, the dynamic confidence interval is calculated. The dynamic confidence interval is adjusted in real time according to the changes in load current and ambient temperature.

[0013] Furthermore, the mutation criteria include: the number of consecutive sampling points of the measured residual monitored in real time exceeds the boundary of the dynamic confidence interval, and the instantaneous rate of change of the measured residual exceeds a predetermined multiple threshold of the normal aging rate of change predicted based on the Arrhenius equation.

[0014] Furthermore, the cumulative sum control chart algorithm identifies persistent weak anomalous offsets by calculating positive and negative cumulative sums. When the positive or negative cumulative sum exceeds a preset decision limit, it is determined that an early weak fault exists. The calculation of the positive cumulative sum and the negative cumulative sum is based on the measured residual value, the mean of the residual under normal conditions, and the detectable offset.

[0015] A fault intelligent early warning system driven by digital twin of power distribution equipment includes: The data acquisition module is configured to collect real-time operating status data of physical power distribution equipment through a sensor array and supports multi-channel synchronous sampling; The edge computing module communicates with the data acquisition module and is configured to receive operational status data and perform real-time residual monitoring and residual evolution analysis. The edge computing module adopts a multi-core parallel processing architecture. The digital twin simulation server communicates with the edge computing module and is configured to run a high-fidelity digital twin model and perform adaptive model correction. The digital twin simulation server is equipped with a graphics processor to accelerate matrix operations in finite element analysis. The early warning display terminal communicates with the digital twin simulation server and is configured to present fault early warning information based on three-dimensional visualization technology, mark the physical parts where abnormalities occur in real time, and provide graded early warnings according to the severity of the fault.

[0016] In summary, this application includes at least one of the following beneficial technical effects: 1. By introducing a model adaptive correction mechanism based on aging characteristic curves, this invention can identify and correct online drift in physical property parameters caused by environmental stress and material aging. This solves the problem of virtual-real mismatch caused by fixed model parameters in traditional technologies, enabling the twin model to maintain high-precision mapping capabilities throughout the entire lifecycle of the equipment and effectively eliminating residual accumulation caused by normal aging.

[0017] 2. This invention, through residual evolution analysis, can accurately distinguish between normal aging drift and sudden early failures in equipment. Based on the dynamic confidence interval calculated by the corrected model, it can automatically adapt to the current state of the equipment, avoiding the frequent false alarms caused by the natural increase of residuals in the later stages of equipment aging in traditional fixed threshold methods. This invention significantly reduces the false alarm rate compared to traditional methods while maintaining high sensitivity.

[0018] 3. Because the system can compensate for deviations caused by aging in real time, more stringent dynamic early warning boundaries can be set, thereby capturing weak abnormal signals such as partial discharge and slight increases in contact resistance that are easily masked in traditional systems. Combining the mutation criterion and the cumulative sum control chart algorithm, this invention can advance the fault detection time by a predetermined amount of time, providing sufficient decision-making time for preventive maintenance of the power system.

[0019] 4. The system architecture constructed in this invention supports efficient fusion and real-time simulation of multi-source data, realizing closed-loop management of the entire process from data acquisition, residual analysis, model correction to fault early warning. The three-dimensional visualization early warning display method enables operation and maintenance personnel to intuitively locate potential faults, reducing reliance on manual experience and significantly improving the safe operation and protection capabilities of the power distribution network. Attached Figure Description

[0020] Figure 1 This is an overall schematic diagram of a fault intelligent early warning method driven by digital twins for power distribution equipment; Figure 2 This is a schematic diagram of the core principle of adaptive correction of the model based on the aging characteristic curve; Figure 3 This is a flowchart of the main stages in building a high-fidelity digital twin model of power distribution equipment; Figure 4 It is a schematic diagram of the multi-level interaction relationship and data flow between physical entities and digital twin systems; Figure 5 This is a flowchart of the intelligent fault early warning logic based on dynamic confidence intervals and mutation criteria. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1-5 The present invention will be further described in detail with reference to specific embodiments.

[0022] The fault intelligent early warning method for power distribution equipment driven by digital twins in this application is implemented according to the following steps: First, for step S1, a high-fidelity digital twin model is constructed. Step S1 aims to establish a high-fidelity digital twin model that can accurately map the characteristics of physical entities, laying the foundation for subsequent residual analysis and fault early warning.

[0023] Step S101: Establish a refined three-dimensional geometric model; obtain the three-dimensional geometric structural parameters of the power distribution equipment, specifically including the equipment casing dimensions, the cross-sectional geometric features of the internal busbars, the contact surface geometric parameters of the contacts, and the layout parameters of the insulation support components. Based on the above parameters, a refined three-dimensional solid model is constructed using high-precision three-dimensional modeling technology, ensuring that the dimensional error between the geometric model and the physical entity is controlled within 1 mm.

[0024] Step S102: Configure material property parameters; based on the actual material composition of the physical entity, configure electrical conductivity, magnetic permeability, thermal conductivity, and specific heat capacity parameters for the three-dimensional solid model. Among them, the electrical conductivity parameter is configured as a function of temperature to reflect the characteristic of the resistivity increase of the metallic conductor due to the increase of temperature; the magnetic permeability parameter is used to describe the nonlinear magnetization characteristics of the ferromagnetic component under working conditions.

[0025] Step S103: Construct a multiphysics coupled calculation framework. Based on the finite element analysis method, establish a two-way coupled calculation architecture for electromagnetic and temperature fields in a three-dimensional solid model. Electromagnetic field calculations are used to solve for the loss distribution of the power distribution equipment under operating current, specifically including Joule losses of the busbars, eddy current losses of the casing, and localized concentrated losses due to contact resistance. The calculated loss distribution is used as a heat source in the temperature field calculation to simulate the real-time temperature distribution of the equipment under different load conditions.

[0026] Step S104: Set the convergence criterion for the coupled calculation. During the electromagnetic-thermal coupled iterative calculation, a clear convergence criterion is set: when the temperature field difference between two adjacent iterations is less than a preset temperature convergence threshold, the iteration terminates, and the steady-state or transient temperature field distribution data under the current operating condition is output. This step ensures the efficiency of the simulation calculation and the reliability of the results.

[0027] Step S105: Refine the key physical property model; to improve the fidelity of the model, the key physical properties need to be modeled in detail: For impedance characteristics, the focus is on establishing a contact resistance model. This model uses a combination of Hertzian contact theory and Holm conductivity theory to calculate the effective contact area of ​​the connection under a specific preload, thereby determining the initial value of the contact resistance. This simulates the nonlinear impedance characteristics of the connection point caused by factors such as pressure and surface condition.

[0028] For thermal conduction characteristics, convective heat transfer models and thermal radiation models are established. The convective heat transfer model comprehensively considers the effects of natural convection and forced convection, and calculates the convective heat transfer coefficient based on the Nusselt number of the equipment surface. The thermal radiation model is based on the Stefan-Boltzmann law and combines the emissivity parameters of the equipment surface material to accurately describe the heat exchange process between the equipment surface and the surrounding environment.

[0029] In addition, a dielectric loss model of the insulating medium is established. By setting the loss tangent of the insulating material, the heating power per unit volume under the action of a strong electric field is calculated to simulate the heating characteristics inside the insulating medium.

[0030] In summary, step S1 constructed a high-fidelity digital twin model through geometric modeling, material property configuration, the establishment of a multiphysics coupled computational architecture, and independent modeling of key physical properties. This model is highly consistent with the physical entity in its static structure and accurately reflects the coupling effects of multiple physical quantities such as electromagnetics and heat in its dynamic response, providing a precise simulation benchmark for subsequent virtual-real comparison and fault diagnosis.

[0031] Step S2 aims to simultaneously collect the operating status data of the physical equipment and the simulation data of the digital twin model, and calculate the residual sequence of both at the same time scale, providing standardized input data for subsequent residual evolution analysis. It includes the following implementation steps: Step S201: Configure the sensor group and set the sampling parameters; install the sensor group at key nodes of the power distribution equipment to acquire real-time operating status data. Specifically, a high-precision current transformer with an accuracy class of 0.2 is used to collect three-phase current, a high-precision voltage transformer with an accuracy class of 0.2 is used to collect three-phase voltage, an infrared temperature sensor with a measurement accuracy of ±1 degree Celsius or better is used to collect the surface temperature of key connection parts, and an integrated temperature and humidity sensor with a measurement accuracy of ±2% relative humidity and ±0.5 degrees Celsius is used to collect ambient temperature and humidity.

[0032] The sampling frequency of the sensor group is set to a fixed value within the range of 1000 Hz to 5000 Hz. The specific value is determined according to the dynamic response characteristics of the monitored equipment in order to capture transient fluctuations during operation.

[0033] Step S202: Perform multi-sensor clock synchronization; the clocks of each sensor node are aligned using the IEEE 1588 precision time synchronization protocol to ensure that data acquisition at different locations has a unified time reference and the time synchronization error is controlled within 1 microsecond.

[0034] Step S203: Load data and run real-time simulation; use the real-time load current and environmental parameters collected by the physical equipment as inputs, and load them into the high-fidelity digital twin model for real-time simulation calculation. The simulation calculation engine adopts multi-threaded parallel acceleration technology, which decomposes the calculation task into multiple sub-tasks and allocates them to different threads for parallel execution, ensuring that the simulation calculation step size is synchronized with the actual data acquisition step size, that is, the simulation step size is consistent with the sampling period, thereby achieving real-time correspondence between simulation and actual measurement.

[0035] Step S204: Obtain the simulation state vector; extract the simulation output state vector from the digital twin model. This vector specifically includes the temperature value, electromagnetic field strength value, and power loss value predicted by the model for each monitoring point. The location of each monitoring point corresponds one-to-one with the key node locations of sensor installation in step S201.

[0036] Step S205: Calculate the multidimensional residual sequence; subtract the measured physical state vector from the simulated state vector at the same time scale to generate the multidimensional residual sequence. Specifically, for the... Each monitoring point, at any given time residual The calculation formula is ,in For a moment The physical measured value, For a moment The model simulation values. This residual sequence reflects the difference between the current operating state of the physical entity and the ideal model state.

[0037] Step S206, Standardize the residual sequence; Standardize the multidimensional residual sequence by subtracting the mean of the current residual sequence and dividing by the standard deviation of the sequence. The calculation formula is as follows: ,in The standardized residual value. The original residual value, The mean of the current residual sequence. This represents the standard deviation of the current residual sequence.

[0038] This process eliminates the influence of different physical dimensions on subsequent analysis. The standardized residual sequence has a mean of 0 and a standard deviation of 1, which can intuitively reflect the degree to which the physical entity deviates from the theoretical model. Moreover, the data in each dimension are on the same order of magnitude, which facilitates subsequent statistical analysis and feature extraction.

[0039] In summary, step S2 transforms the original residual data into a dimensionless standardized residual sequence, providing a unified scale for the extraction of statistical features in subsequent residual evolution analysis.

[0040] The next step is S3, which aims to extract the statistical characteristics of the standardized residual sequence using the sliding window technique. Through quantitative analysis of the residual evolution law, it provides a basis for distinguishing between normal aging drift and abnormal fluctuations, and lays the foundation for determining the triggering conditions for subsequent adaptive correction of the model.

[0041] Step S301: Set the sliding window parameters; set the sliding window length to 500 sampling points and the sliding step size to 50 sampling points. The window length determines the amount of historical data covered in each analysis, and the step size determines the analysis frequency. Together, they form the basis of the time window for continuous rolling analysis of the residual sequence. The specific values ​​of the window length and step size can be adaptively adjusted according to the dynamic response characteristics of the device and computing resources.

[0042] Step S302: Calculate the statistical characteristics within the window; within each sliding window, perform statistical analysis on the standardized residual series and calculate the following three statistical characteristics: Mean offset, or the first moment of the residual sequence within the window, is used to measure the degree of systematic bias in the residuals, reflecting whether the physical entity deviates continuously from the model in the current period.

[0043] Variance, or the second moment of the residual sequence within a window, is used to measure the degree of random fluctuation in the residuals and reflects the intensity of random disturbances caused by factors such as measurement noise and load fluctuations during operation.

[0044] The rate of change, or the slope of the residual sequence within the window, is obtained by linearly fitting the residual sequence within the window. It is used to measure the trend of deviation over time and reflects whether the residual tends to stabilize, grows slowly, or changes rapidly.

[0045] Step S303: Identify trend drift; by comparing the mean offsets between adjacent windows, identify whether there is a trend drift in the residuals. Specifically, when the mean offsets of multiple consecutive windows show a monotonically increasing or monotonically decreasing pattern, and the magnitude of the change exceeds a preset drift judgment threshold, it is determined that there is a trend drift in the residuals.

[0046] Step S304: Perform data quality control; assess the severity of residual signal fluctuations by analyzing variance changes. When variance increases sharply in a short period of time, it indicates that the data for the current period is strongly interfered with by measurement noise or sudden load fluctuations. Such data segments are marked or removed in subsequent analyses to ensure that the data used for model correction and fault identification have sufficient signal-to-noise ratio and stability.

[0047] In summary, step S3 extracts the mean offset, variance, and rate of change of the residual sequence using the sliding window technique, thereby achieving a quantitative description of the residual evolution law and data quality control, and providing a reliable statistical basis for determining the triggering conditions for model adaptive correction in step S4.

[0048] For step S4, the model is adaptively corrected. Step S4 aims to correct the key physical characteristic parameters in the twin model online by using the parameter identification operator when the residual evolution law conforms to the preset aging characteristics, so that the model can follow the performance drift of the physical entity and thus eliminate the virtual-real mismatch problem caused by equipment aging.

[0049] Step S401: Establish an aging characteristic curve model; the preset aging characteristic curve is constructed based on the Arrhenius equation and is used to describe the evolution of the physical parameters of the power distribution equipment with operating time and operating temperature. The mathematical expression of the Arrhenius equation is as follows: in It represents the rate of change of a physical parameter, expressed in negative first power of seconds or in a unit that matches the rate of change of a specific physical parameter. The exponential factor is a constant related to material properties, expressed in seconds to the power of negative one. The activation energy, expressed in joules per mole, reflects the energy barrier during the aging process of a material. Its value is obtained by fitting accelerated aging test data. For copper contact materials used in typical power distribution equipment, the activation energy ranges from 0.5 to 1.2 electron volts. The molar gas constant is 8.314 joules per mole Kelvin. This is absolute temperature, measured in Kelvin.

[0050] This formula can be used to predict the growth trend of contact resistance over time or the degradation trend of the dielectric constant of the insulating medium under specific operating temperature conditions. Specifically, for contact resistance, its relationship with time is as follows: ,in The initial contact resistance. This is a proportionality coefficient related to the material. This refers to the runtime.

[0051] Step S402: Set the physical characteristic parameters to be corrected; these parameters include the contact resistance of the conductor and the dielectric constant of the insulating medium. These two types of parameters are the key parameters most significantly affected by aging during equipment operation and have a major impact on equipment condition assessment. By correcting them online, the twin model can effectively track the aging process of the equipment.

[0052] Step S403: Configure the parameter identification operator; the parameter identification operator uses a recursive least squares algorithm with a forgetting factor. The objective function of this algorithm is to minimize the weighted sum of squares of the measured residuals, and its recursive form is as follows: in For a moment The parameter estimation vector contains the contact resistance and dielectric constant to be identified; For a moment The measured residual value; The regression vector is composed of the model's sensitivity coefficients to the parameters; It is the gain vector; It is the covariance matrix; The forgetting factor, set between 0.95 and 0.99, is used to balance the algorithm's ability to remember historical data with its tracking speed in response to parameter drift. A forgetting factor closer to 0.95 results in a higher sensitivity to new data and faster tracking speed, but relatively weaker noise resistance; a value closer to 0.99 increases the algorithm's reliance on historical data and enhances noise resistance, but correspondingly reduces its response speed to parameter drift.

[0053] The specific value is adjusted according to the equipment aging rate and the on-site noise level. A smaller forgetting factor is used for equipment with a faster aging rate, and a larger forgetting factor is used for environments with high noise levels.

[0054] Step S404: Set the triggering conditions for model adaptive correction; when the mean offset of the residual sequence exceeds the preset aging judgment threshold within 10 consecutive sliding windows, and the residual change rate conforms to the slope characteristics described by the aging characteristic curve in step S401, start the parameter identification operator.

[0055] The method for determining the aging judgment threshold is as follows: in the initial stage of equipment operation, the standard deviation of the mean offset of the first 1000 sampling points is calculated, and the aging judgment threshold is set to 3 times the standard deviation.

[0056] The slope characteristic refers to the residual change rate being of the same order of magnitude and in the same direction as the aging rate predicted by the Arrhenius equation. Specifically, when the ratio of the residual change rate to the theoretical aging rate calculated by the Arrhenius equation is within the range of 0.5 to 2.0, it is considered to meet the slope characteristic. This triggering condition can effectively distinguish between slow drift caused by normal equipment aging and transient fluctuations caused by occasional disturbances.

[0057] Step S405: Perform online model parameter updates; the corrected model parameters are fed back into the high-fidelity digital twin model, realizing online updates of model parameters. The update process includes modifying the material property matrix in the finite element calculation and the impedance boundary conditions of the contact interface.

[0058] Specifically, the identified contact resistance value is updated to the impedance boundary condition of the contact interface, which is represented as the equivalent resistance element of the contact surface in the finite element model; the identified dielectric constant is updated to the material property matrix of the insulating medium, which is used to calculate dielectric loss and electric field distribution in the finite element model.

[0059] This allows the output of the twin model to dynamically follow the performance drift caused by normal equipment aging, thereby eliminating the accumulation of residuals caused by virtual-real mismatch and ensuring the accuracy of the model throughout its entire life cycle.

[0060] In summary, step S4 achieves adaptive tracking of physical equipment aging drift by the twin model through constructing an aging characteristic curve based on the Arrhenius equation, employing a recursive least squares algorithm with a forgetting factor for online parameter identification, and setting explicit triggering conditions and update mechanisms. The corrected model provides a reliable model foundation for the accurate calculation of the dynamic confidence interval in step S5, thereby ensuring the accuracy of fault warning.

[0061] Finally, regarding step S5, intelligent fault early warning; step S5 aims to calculate the dynamic confidence interval based on the modified digital twin model, and combine it with the mutation criterion to achieve accurate identification and early warning of early minor faults, providing clear maintenance guidance for operation and maintenance personnel.

[0062] Step S501: Construct a probability distribution model for the Monte Carlo simulation input; based on the modified digital twin model, perform multiple Monte Carlo simulations under the current operating conditions. During the simulation, the probability distribution of environmental interference and measurement errors must be considered.

[0063] Specifically, the current measurement error is set to follow a normal distribution with a mean of 0 and a standard deviation corresponding to the error limit of the current transformer's accuracy class; the voltage measurement error follows a normal distribution with a mean of 0 and a standard deviation corresponding to the error limit of the voltage transformer's accuracy class; the ambient temperature measurement error follows a normal distribution with a mean of 0 and a standard deviation corresponding to the error limit of the temperature and humidity sensor's accuracy; and the load current fluctuation follows a probability distribution model fitted based on historical data. The distributions of the above error sources are independent of each other.

[0064] Step S502: Perform Monte Carlo simulation and extract statistical characteristics; set the number of Monte Carlo simulations to at least 1000. In each simulation, randomly sample each input quantity according to the probability distribution set in step S501, input it into the corrected digital twin model for simulation calculation, and obtain a set of simulation output values. After completing all simulations, obtain the statistical distribution characteristics of the simulation results, including the mean and standard deviation of the simulation output values ​​at each monitoring point.

[0065] Step S503: Calculate the dynamic confidence interval; based on the statistical characteristics obtained from the Monte Carlo simulation, calculate the dynamic confidence interval. Set a preset significance level. The value is typically taken as 0.05. The formula for calculating the dynamic confidence interval is as follows: in Indicates the dynamic confidence interval; This is the corrected simulation prediction value of the model under the current input, i.e., the mean of the Monte Carlo simulation results; The quantiles of the standard normal distribution, when the significance level is... hour, ; The overall standard deviation is obtained from the Monte Carlo simulation, reflecting the degree of uncertainty in the propagation of input error to output.

[0066] This confidence interval is adjusted in real time according to changes in load current and ambient temperature, reflecting the residual range allowed for normal operation fluctuations under the current aging state of the equipment.

[0067] Step S504: Set the triggering conditions for the mutation criterion; the mutation criterion for triggering the fault warning includes the following two conditions, which must be met simultaneously: First, monitor in real time whether the measured residual exceeds the boundary of the dynamic confidence interval calculated in step S503. Specifically, when the number of five consecutive sampling points of the measured residual exceeds the upper or lower limit of the dynamic confidence interval, the first condition is deemed met.

[0068] Second, the instantaneous rate of change of the residual sequence exceeds 10 times the threshold of the normal aging rate of change. The instantaneous rate of change is calculated by dividing the difference between the residual of the current sampling point and the previous sampling point by the sampling period; the normal aging rate of change is predicted by the Arrhenius equation in step S401 and reflects the normal aging rate of the equipment at the current temperature.

[0069] Step S505: Enhance the ability to identify weak anomalies by combining the cumulative sum control chart algorithm; to further improve the sensitivity of identifying early weak faults, the mutation criterion is also combined with the cumulative sum control chart algorithm. The calculation method of the cumulative sum control chart algorithm is as follows: in and They are time points Positive cumulative sum and negative cumulative sum, initial value , ; For a moment The measured residual value; This represents the mean of the residuals under normal conditions, and its value is 0. The detectable offset is set according to the equipment type and fault sensitivity, and is usually 0.5 to 1.0 times the standard deviation of the residual.

[0070] when or Exceeding the preset decision limit At that time, a persistent, weak anomalous offset was determined. Decision Limit The desired false alarm rate is set, typically four to five times the standard deviation of the residuals. This method identifies subtle but persistent anomalies that are difficult to detect in a single sampling by accumulating the amount of residual deviation from the mean.

[0071] Step S506: Trigger a fault warning and output location information; when the mutation criterion meets the triggering condition, the system triggers a fault warning. The warning information is presented through the warning display terminal, which uses 3D visualization technology to construct a virtual scene of the power distribution room based on a graphics engine, and presents the virtual-real mapping status of the equipment in real time.

[0072] When a fault warning is triggered, the terminal automatically pops up a warning window, marking the physical location of the abnormality in real time, and classifying the probability according to the severity of the fault, specifically into Level 1, Level 2, and Level 3 warnings, providing clear maintenance personnel with clear repair suggestions.

[0073] In summary, step S5 constructs a dynamic confidence interval through Monte Carlo simulation, combines the mutation criterion with the cumulative sum control chart algorithm to achieve a dual anomaly identification mechanism, and finally outputs accurate fault warning information through a three-dimensional visualization terminal, providing reliable technical support for the preventive maintenance of the power distribution network.

[0074] This embodiment also includes a digital twin-driven intelligent fault early warning system for power distribution equipment, and an intelligent fault early warning method for execution. The system includes a data acquisition module, an edge computing module, a digital twin simulation server, and an early warning display terminal. The data acquisition module transmits the operating status data of the physical equipment to the edge computing module via an industrial Ethernet or 5G communication network. The data acquisition module has multi-channel synchronous sampling capability and supports multiple power communication protocols such as Modbus TCP and IEC 61850.

[0075] The edge computing module employs a multi-core parallel processing architecture, possessing powerful floating-point arithmetic capabilities. It is responsible for real-time residual monitoring and residual evolution analysis, with a computation latency of less than 10 milliseconds for a single residual analysis. The edge computing module integrates lightweight data processing algorithms capable of filtering, denoising, and feature extraction of raw sampled data.

[0076] The digital twin simulation server is responsible for running and adaptively correcting the high-fidelity digital twin model. The server is equipped with a high-performance graphics processing unit (GPU) to accelerate large-scale matrix operations in finite element analysis. Data exchange between the server and the edge computing module is achieved via a high-speed network at 10 gigabits per second, ensuring real-time alignment between simulation and measured data. The system supports access to more than 128 power distribution devices, meeting the predetermined scale requirement, with an overall early warning response time controlled within 500 milliseconds.

[0077] The early warning display terminal uses 3D visualization technology to construct a virtual scene of the power distribution room based on a graphics engine. The terminal presents the virtual-to-real mapping status of the equipment in real time, intuitively displaying the temperature distribution, current flow direction, and residual intensity of the equipment through color cloud maps. When a fault warning is triggered, the terminal automatically pops up an warning window, marking the physical location of the anomaly in real time, and classifying the probability according to the severity of the fault (such as Level 1, Level 2, and Level 3 warnings), providing clear maintenance personnel with clear repair suggestions.

[0078] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0079] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A fault intelligent early warning method for power distribution equipment driven by digital twin, characterized in that, Includes the following steps: Acquire real-time operating status data of physical power distribution equipment, and based on a pre-built high-fidelity digital twin model, acquire simulation status data at the same time scale as the real-time operating status data; Based on real-time operating status data and simulation status data, a multidimensional residual sequence is calculated, and the multidimensional residual sequence is standardized to obtain a standardized residual sequence. The sliding window technique is used to extract statistical features of the standardized residual series, including mean shift. When the mean offset exceeds the preset aging judgment threshold in multiple consecutive sliding windows, and the residual change rate matches the aging rate described by the aging characteristic curve constructed based on the Arrhenius equation, the model adaptive correction is triggered. The adaptive model correction includes: using a recursive least squares algorithm with a forgetting factor to identify and correct key physical property parameters in the high-fidelity digital twin model online. Key physical property parameters include the contact resistance of the conductor and the dielectric constant of the insulating medium. Based on the modified high-fidelity digital twin model, dynamic confidence intervals are calculated through Monte Carlo simulation, and combined with mutation criteria and cumulative sum control chart algorithms, fault warnings are provided for physical power distribution equipment.

2. The method according to claim 1, characterized in that, Building a high-fidelity digital twin model includes: Establish a refined three-dimensional geometric model with dimensional errors of the physical power distribution equipment within a predetermined range; Configure temperature-dependent electrical conductivity parameters, nonlinear magnetization permeability parameters, and thermal conductivity parameters for the three-dimensional geometric model; Based on the finite element analysis method, a two-way coupled calculation framework for electromagnetic field and temperature field is established. In the two-way coupled calculation framework, the loss distribution obtained from electromagnetic field calculation is used as a heat source and applied to temperature field calculation, and the convergence criterion for coupled iterative calculation is set. Refined modeling of key physical properties is carried out, including: establishing a contact resistance model using Hertzian contact theory and Holm conductivity theory, as well as establishing a convective heat transfer model and a thermal radiation model.

3. The method according to claim 1, characterized in that, The sliding window technique is used to extract statistical features of standardized residual sequences, and also includes: Within each sliding window, calculate the variance and rate of change of the standardized residual series; Data quality is assessed by analyzing changes in variance, and when variance increases sharply, the corresponding data segments are marked or removed to control data quality. By comparing the mean offset between adjacent windows, trend drift is identified when the mean offset of multiple consecutive windows shows a monotonically increasing or monotonically decreasing trend and the magnitude of the change exceeds the preset drift judgment threshold.

4. The method according to claim 1, characterized in that, The aging characteristic curves constructed based on the Arrhenius equation are used to describe the evolution of key physical property parameters with operating time and operating temperature. The pre-exponential factor and activation energy of the Arrhenius equation are obtained by fitting accelerated aging experimental data. The residual change rate matches the aging rate described by the aging characteristic curve constructed based on the Arrhenius equation. Specifically, the ratio of the residual change rate to the theoretical aging rate calculated by the Arrhenius equation is within a preset range.

5. The method according to claim 1, characterized in that, The objective function of the recursive least squares algorithm with a forgetting factor is to minimize the weighted sum of squares of the measured residuals. The forgetting factor takes a value in the range of 0.95 to 0.99 to balance the algorithm's ability to remember historical data with its tracking speed of parameter drift. The corrected contact resistance of the conductor is updated to the impedance boundary conditions of the contact interface in the high-fidelity digital twin model, and the corrected dielectric constant of the insulating medium is updated to the material property matrix of the insulating medium in the high-fidelity digital twin model.

6. The method according to claim 1, characterized in that, The method for determining the preset aging judgment threshold is as follows: at the initial stage of equipment operation, a predetermined number of sampling points are selected to calculate the standard deviation of the mean offset, and the aging judgment threshold is set to a predetermined multiple of the standard deviation. The triggering conditions for adaptive model correction also include: the residual change rate is in the same direction as the aging rate described by the aging characteristic curve constructed based on the Arrhenius equation.

7. The method according to claim 1, characterized in that, The dynamic confidence intervals calculated using Monte Carlo simulations include: A probability distribution model for Monte Carlo simulation input is constructed, which considers the probability distribution of current measurement error, voltage measurement error, ambient temperature measurement error, and load current fluctuation. Based on the modified high-fidelity digital twin model, a predetermined number of Monte Carlo simulations were performed to obtain the statistical distribution characteristics of the simulation output values. Based on the preset significance level, and combined with the mean and standard deviation of the simulation output values, the dynamic confidence interval is calculated. The dynamic confidence interval is adjusted in real time according to the changes in load current and ambient temperature.

8. The method according to claim 1, characterized in that, The mutation criteria include: the number of consecutive sampling points of the measured residuals monitored in real time exceeds the boundary of the dynamic confidence interval, and the instantaneous rate of change of the measured residuals exceeds a predetermined multiple threshold of the normal aging rate of change predicted based on the Arrhenius equation.

9. The method according to claim 1, characterized in that, The cumulative sum control chart algorithm identifies persistent weak anomalous offsets by calculating positive and negative cumulative sums. When the positive or negative cumulative sum exceeds a preset decision limit, it is determined that there is an early weak fault. The calculation of the positive cumulative sum and the negative cumulative sum is based on the measured residual value, the mean of the residual under normal conditions, and the detectable offset.

10. A fault intelligent early warning system driven by a digital twin for power distribution equipment, used to execute the method as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is configured to collect real-time operating status data of physical power distribution equipment through a sensor array and supports multi-channel synchronous sampling; The edge computing module communicates with the data acquisition module and is configured to receive operational status data and perform real-time residual monitoring and residual evolution analysis. The edge computing module adopts a multi-core parallel processing architecture. The digital twin simulation server communicates with the edge computing module and is configured to run a high-fidelity digital twin model and perform adaptive model correction. The digital twin simulation server is equipped with a graphics processor to accelerate matrix operations in finite element analysis. The early warning display terminal communicates with the digital twin simulation server and is configured to present fault early warning information based on three-dimensional visualization technology, mark the physical parts where abnormalities occur in real time, and provide graded early warnings according to the severity of the fault.

Images