A cloud platform for full lifecycle management of power grid equipment

Abstract

This invention relates to the field of digital twin and intelligent operation and maintenance technology for power systems, specifically a cloud platform for the full lifecycle management of power grid equipment. The platform includes: a data acquisition module for acquiring operational monitoring data streams of the target power grid equipment; a state perception entropy calculation module for calculating the state perception entropy, which characterizes the degree of deviation between the digital twin model and the physical entity's state; a confidence assessment module for determining the state confidence of the current full lifecycle management system regarding the actual physical state of the target power grid equipment; an active verification decision module for acquiring high-fidelity calibration feedback data; and a model calibration module for restoring the system's model fit to the target power grid equipment. This invention solves the cognitive problem of the seemingly fitting but actually deviating cloud-based model from the physical entity, achieving globally optimal decision-making for the full lifecycle management of the power grid under the dual constraints of security and economy.

Description

Technical Field

[0001] This invention relates to the field of digital twin and intelligent operation and maintenance technology for power systems, specifically a cloud platform for the full lifecycle management of power grid equipment. Background Technology

[0002] In the current field of full life cycle management of power grid equipment, cloud-based digital twin models are widely used for physical entity status monitoring and operation and maintenance; however, due to the high-frequency impact of distributed energy in new power systems and the aging and drift of sensors during long-term operation, the power grid operating environment exhibits extremely high and complex random interference characteristics.

[0003] Existing solutions generally rely on traditional health scoring or data fitting methods, which involve acquiring conventional sampling data such as voltage and current through IoT gateways and performing model predictions in the cloud. Although such solutions have a certain fitting ability under normal stable operating scenarios, they lack quantitative analysis of the deep causal relationship between data features and physical states, resulting in the system's inability to effectively identify data smog caused by external environmental noise. This makes the cloud-based digital twin model and the physical entity state appear to fit each other at the data level, but in reality, there is a serious cognitive deviation. In addition, when faced with cognitive ambiguity, existing technologies are often in a passive state of waiting for data, lacking the ability to actively intervene in physical entities to obtain high-fidelity feedback, and making it difficult to close the loop and correct model parameters.

[0004] Therefore, how to quantify the ambiguity of system cognition under the dual impact of distributed energy shocks and sensor performance degradation, and improve the accuracy and self-healing ability of digital twin models to fit the real physical state of target devices, has become an urgent technical problem to be solved. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a cloud platform for the full lifecycle management of power grid equipment. Specifically, the technical solution of this invention includes:

[0006] The data acquisition module is used to acquire the operation monitoring data stream of the target power grid equipment. The operation monitoring data stream includes: sensor time series data, distributed energy impact characteristics, historical operation and maintenance logs, and environmental interference parameters.

[0007] The state-aware entropy calculation module is used to identify the nonlinear correlation between data features and physical state based on the operation monitoring data stream using a causal reasoning model, and to calculate the state-aware entropy that characterizes the degree of deviation between the digital twin model and the physical entity state.

[0008] The confidence assessment module is used to compare the state perception entropy with a preset deviation threshold to determine the state confidence of the current full life cycle management system in the true physical state of the target power grid equipment.

[0009] An active verification decision module is used to generate an active calibration verification command in response to the state perception entropy exceeding the deviation critical threshold. The active calibration verification command is used to forcibly trigger physical intervention operations on the target power grid equipment in order to obtain high-fidelity calibration feedback data.

[0010] The model calibration module is used to correct the internal parameters of the digital twin model based on the calibration feedback data, and reset the state-aware entropy to a preset safe range, so as to restore the system's model fitting accuracy to the target power grid equipment.

[0011] Preferably, the state-aware entropy calculation module is used to identify the nonlinear correlation between data features and physical states based on the operation monitoring data stream using a causal inference model, and to calculate the state-aware entropy characterizing the degree of deviation between the digital twin model and the physical entity state, including:

[0012] The sensor timing data and the environmental interference parameters are retrieved.

[0013] Construct a causal topology graph of device status based on a graph neural network, and extract high-frequency oscillation features and zero-point drift features from the operation monitoring data stream;

[0014] Quantify the interference weights of the high-frequency oscillation characteristics and the zero-point drift characteristics on the equipment insulation status assessment;

[0015] Based on the interference weights, the probability distribution of the evolution of the physical state of the device that cannot be explained by the current monitoring data is calculated, and the degree of dispersion of the probability distribution is defined as the state perception entropy.

[0016] Preferably, the active verification decision module is used to generate an active calibration verification instruction in response to the state perception entropy exceeding the deviation critical threshold, including:

[0017] Identify the current operating load rate and preset power supply continuity indicators of the target power grid equipment;

[0018] When the state perception entropy exceeds the deviation critical threshold, a preset intervention strategy library is retrieved;

[0019] Physical actions that can reduce the state perception entropy are selected from the intervention strategy library, including unplanned power outage maintenance or preventive component replacement.

[0020] The selected physical actions are encapsulated into the active calibration verification instruction, which is configured to sacrifice short-term operating efficiency in exchange for improved model cognitive accuracy.

[0021] Preferably, the model calibration module is used to correct the internal parameters of the digital twin model based on the calibration feedback data, including:

[0022] Receive physical disassembly data or offline test data generated after executing the active calibration verification command, and use it as the calibration feedback data;

[0023] Calculate the residual vector between the calibration feedback data and the predicted value of the digital twin model;

[0024] Using the backpropagation algorithm, the weight parameters of the digital twin model are adjusted according to the residual vector until the model output prediction state and the fitting degree of the calibration feedback data reach a preset standard.

[0025] Preferably, the distributed energy impact characteristics acquired by the data acquisition module include:

[0026] Voltage fluctuation frequency, reverse power flow amplitude, and harmonic distortion rate generated by photovoltaic or wind power access points;

[0027] The state-aware entropy calculation module is further used to establish a dynamic coupling model between the distributed energy impact characteristics and the equipment insulation aging rate, so as to remove data noise generated by external impacts.

[0028] Preferred options also include:

[0029] The cascaded fault risk prediction module is used to calculate the potential contribution rate of regional large-scale power outages caused by state deviations based on the state perception entropy and the node importance of the target power grid equipment in the power grid topology.

[0030] The active verification decision module is further configured to dynamically adjust the deviation critical threshold based on the potential contribution rate, wherein when the potential contribution rate increases, the deviation critical threshold is reduced to improve the system's sensitivity to data distortion.

[0031] Preferably, the process by which the active verification decision module generates the active calibration verification instruction follows the following optimization objective function:

[0032] Construct a joint optimization function that includes the first sub-objective and the second sub-objective;

[0033] The first sub-objective is to minimize the cumulative value of the state-aware entropy;

[0034] The second sub-objective is to minimize the physical resource consumption cost incurred in executing the active calibration verification command;

[0035] Provided that the state confidence level is not lower than a preset safety threshold, the joint optimization function is solved to determine the optimal active calibration verification command.

[0036] Preferably, the state-aware entropy calculation module is also used for:

[0037] Monitor the temporal correlation decay trend of the sensor's time-series data;

[0038] When the correlation coefficient between multi-source sensor data is detected to be lower than a preset consistency threshold, it is determined that a hidden data consistency failure has occurred, and a preset penalty factor is introduced to exponentially amplify the state perception entropy.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. This invention introduces a state-aware entropy calculation module and uses a causal reasoning model to deeply identify the nonlinear correlation between operating data and physical state. Unlike traditional health scoring, this technology can effectively quantify the degree to which cloud models cannot accurately perceive the true state of equipment, accurately identify false data anomalies caused by distributed energy shocks or sensor drift, and solve the cognitive problem that cloud models and physical entities seem to fit each other but actually deviate.

[0041] 2. In view of the limitations of existing technologies that passively wait for data when cognition is ambiguous, this invention designs an active verification decision module. When the system's confidence in the device status is insufficient, it can break with convention and actively generate physical intervention commands. By sacrificing short-term operating efficiency in exchange for high-fidelity real feedback data, it provides a reliable basis for decision-making in extremely uncertain environments.

[0042] 3. Through the model calibration module, the system uses the physical disassembly or offline test data obtained through active intervention as the calibration benchmark, and uses the backpropagation algorithm to dynamically correct the internal parameters of the digital twin model. This supervised learning process based on real physical feedback effectively solves the parameter drift problem caused by the lack of negative samples in the long-term operation of the model, and enables the system to have the self-healing ability to automatically restore the model's fit to reality.

[0043] 4. This invention integrates cascaded fault risk prediction and joint optimization functions, which can dynamically adjust the early warning threshold according to the importance of the equipment in the power grid topology and balance the relationship between improving cognitive accuracy and consuming physical resources. This ensures zero-tolerance sensitive perception for critical hub equipment, while optimizing operation and maintenance costs for conventional scenarios, and realizes the global optimal decision-making for the whole life cycle management of the power grid under the dual constraints of safety and economy. Attached Figure Description

[0044] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0045] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0047] Example 1:

[0048] Please see Figure 1 A cloud platform for full lifecycle management of power grid equipment includes: a data acquisition module for acquiring operation monitoring data streams of target power grid equipment, the operation monitoring data streams including: sensor time-series data, distributed energy impact characteristics, historical operation and maintenance logs and environmental interference parameters;

[0049] The state-aware entropy calculation module is used to identify the nonlinear relationship between data features and physical state based on the operation monitoring data stream and the causal reasoning model, and to calculate the state-aware entropy that characterizes the degree of deviation between the digital twin model and the physical entity state.

[0050] The confidence assessment module is used to compare the state perception entropy with a preset deviation threshold to determine the state confidence of the current full life cycle management system in the true physical state of the target power grid equipment.

[0051] The active verification decision module is used to generate an active calibration verification command in response to the state perception entropy exceeding the deviation critical threshold. The active calibration verification command is used to forcibly trigger physical intervention operations on the target power grid equipment in order to obtain high-fidelity calibration feedback data.

[0052] The model calibration module is used to correct the internal parameters of the digital twin model based on calibration feedback data and reset the state-aware entropy to a preset safe range in order to restore the system's model fit to the target power grid equipment.

[0053] This embodiment provides a cloud platform for the full life cycle management of power grid equipment. The platform aims to solve the technical problem in the prior art where data haze caused by the superposition of sensor aging and drift and the high-frequency impact of distributed energy causes the cloud-based digital twin model and the physical entity state to appear to fit each other at the data level, but in fact, cognitive deviation has occurred.

[0054] The system utilizes a data acquisition module to construct a high-dimensional physical world mapping benchmark. This module not only acquires conventional sensor time-series data such as voltage, current, and oil temperature through IoT gateways, but also specifically targets the characteristics of new power systems, capturing the impact characteristics of distributed energy sources and environmental interference parameters in real time. Furthermore, it retrieves historical operation and maintenance logs throughout the entire lifecycle of the equipment via API interfaces, forming a multi-source heterogeneous operation monitoring data stream containing both structured and unstructured information. ;

[0055] Based on this, the state perception entropy calculation module quantifies the ambiguity of the system's perception of the current device state. Unlike traditional health scoring, this module calculates the state perception entropy. This entropy value does not indicate whether the device is damaged, but rather the degree to which the cloud platform is unsure of the device's true state; this module uses a causal reasoning model to analyze... Nonlinear relationships between variables;

[0056] To specifically implement causal reasoning, this embodiment uses a constraint-based PC algorithm to construct a causal graph structure between variables: constructing a fully connected undirected graph and utilizing... The transformation is used to perform a conditional independence test, with a significance level of [missing information]. If the variables are independent, the connections are deleted; the causal direction is determined by identifying the V-structure, thus constructing a directed acyclic graph; based on this directed acyclic graph structure, the nonlinear state-space equations are specifically constructed:

[0057]

[0058] in, For state vectors, For the input vector, It is a nonlinear mapping function. For model error; nonlinear mapping function A two-layer neural network structure is used for approximation, and its calculation formula is as follows:

[0059]

[0060] in, This is the weight matrix. For bias vectors, Activation function; model error The covariance matrix is ​​set as Zero-mean Gaussian white noise;

[0061] This equation is used to identify data anomalies that do not conform to the laws of physics; to calculate the entropy value, the residual between the monitored data and the model predictions must be obtained. and its probability density function;

[0062] This system discretizes and solves the above state-space equations to obtain... Predicted state vector at time step And through observation equations Mapped to the measurement space; here, the observation matrix Constructed as A sparse choice matrix of dimension, where, As an observation dimension, For the state dimension, its elements The determination rule is: if the first The sensor directly corresponds to the first one. If there are several state variables, then ,otherwise ; where superscript Represents the transpose of a vector or matrix; the dot product symbol appearing in formulas. Indicates matrix multiplication; all the above subscripts ,like All uniformly represent the first Each sampling moment; from the operational monitoring data stream Extract the corresponding measured vector The residual vector is calculated as follows:

[0063]

[0064] in, The model predicts the output; given that the residual vector is a multidimensional vector, a scalar residual is defined for scalar kernel density estimation. Let be the Euclidean norm of the residual vector, i.e. The probability density function is obtained from the scalar residual sequence using a non-parametric kernel density estimation method. The calculation formula is as follows:

[0065]

[0066] in, For the sample size, For bandwidth parameters, For Gaussian kernel function, For scalar residuals; calculate the state-aware entropy, which characterizes the degree of deviation between the digital twin model and the physical entity's state. Its theoretical definition is:

[0067]

[0068] in, scalar residual The probability density function is used to characterize cognitive bias by quantifying the disorder of the residual distribution; the confidence assessment module constructs a firewall for cognitive risk, transmitting the real-time calculated... Deviation threshold from preset Comparison; Deviation Critical Threshold The setting logic is based on statistical process control using historical data: collecting a set of historical entropy values ​​of the equipment during the health baseline period and calculating their mean. and standard deviation ,set up To cover 99.7% of the normal fluctuation range; in response to System state confidence Within the safe zone, continue to maintain the normal operation and maintenance mode; respond to The system determines that it has entered a cognitive blind spot, and at this point, any health diagnosis results output by the model are unreliable.

[0069] The proactive verification decision module, acting as the core decision-making hub, executes a counterintuitive strategy that sacrifices physical resources to obtain accurate information; in response to... This module no longer passively waits for data, but instead generates active calibration and verification commands. The instruction is configured to forcibly trigger physical intervention operations on the target power grid equipment; the model calibration module closes the loop to correct cognitive biases in the system, and based on the aforementioned high-fidelity data, corrects the internal parameters of the digital twin model, forcibly pulling the model back to a state synchronized with the physical entity, and... Reset to a lower safety range to restore the system's model fit to the target power grid equipment;

[0070] To verify the technical effectiveness of this system, a simulation experiment was conducted on the IEEE-39 node standard test system. The experiment involved introducing a gradual drift spoofing attack with an amplitude 1.2 times the normal value onto the voltage transformer at node 14. Experimental data showed that at the 50th time step after the attack, the residual change rate of the traditional state estimation method was only 2%, failing to trigger an alarm; while the state-aware entropy of this system... The value surged from 0.15 to 0.85, instantly exceeding the set threshold. The active verification command was successfully triggered; after physical intervention, the model calibration module reset the entropy value to 0.12, effectively avoiding catastrophic missed judgments caused by data deception.

[0071] Example 2:

[0072] The State-Aware Entropy Calculation Module is used to identify the nonlinear correlation between data features and physical state based on the operation monitoring data stream using a causal inference model, and to calculate the State-Aware Entropy, which characterizes the degree of deviation between the digital twin model and the physical entity state. This includes: calling sensor time-series data and environmental interference parameters; constructing a causal topology graph of the equipment state based on a graph neural network; extracting high-frequency oscillation features and zero-point drift features from the operation monitoring data stream; quantifying the interference weights of high-frequency oscillation features and zero-point drift features on the equipment insulation state assessment; and, based on the interference weights, calculating the probability distribution of the evolution of the equipment physical state that the current monitoring data cannot explain, and defining the degree of dispersion of the probability distribution as the State-Aware Entropy.

[0073] This embodiment refines the execution logic of the state-aware entropy calculation module, employing deep causal mining technology based on graph neural networks. This module uses sensor time-series data and environmental interference parameters as input tensors and constructs a device state causal topology graph based on graph neural networks. The specific construction rules are as follows: Define node feature vectors. compute nodes Inter-information The specific calculation uses the discretized probability and statistics method as follows:

[0074]

[0075] in, For joint probability distribution, For marginal probability distribution; when Establish connecting edges at time, where, Causal threshold; causal threshold The Otsu's method is adopted, and the optimal segmentation threshold is calculated by the mutual information distribution histogram to adaptively distinguish between strong and weak causal relationships; the edge weights are defined, and their calculation formula is as follows:

[0076]

[0077] in, For activation function, For learning parameters; learning parameters In the model initialization phase, He initialization is used, and during training, the backpropagation algorithm and Adam optimizer are employed to minimize the prediction error, specifically using the mean squared error loss function. Iterative updates are performed; in this topology graph, nodes represent sensor variables, and edges represent physical causal relationships between variables; the module utilizes a graph convolutional network to extract features from the operational monitoring data stream; here, the inter-layer propagation model of the graph convolutional network is disclosed to clarify the feature extraction process, and its calculation formula is as follows:

[0078]

[0079] in, The node feature matrix, To renormalize the adjacency matrix, for The degree matrix, whose diagonal elements , This is the inter-layer weight matrix. For layer index, The activation function is defined; to separate specific physical quantities from abstract graph features, a readout mapping function based on global pooling is defined to extract the high-frequency oscillation features caused by distributed energy access. and zero-point drift characteristics caused by sensor aging :

[0080]

[0081]

[0082] Among them, superscript Represents the transpose of a vector. The weights of the high-frequency oscillation feature regression layer, The weights of the zero-point drift feature regression layer, This is the bias of the high-frequency oscillation characteristic regression layer. The bias of the zero-point drift feature regression layer, For the number of nodes, This serves as a single-layer index; the system utilizes an attention mechanism to dynamically quantify the interference weights of the aforementioned features on the equipment insulation status assessment; key variables in the specific calculation formula... and This embodiment discloses the specific calculation process:

[0083] For variables It employs a multi-head self-attention mechanism for computation; and constructs a set of input sequences containing extracted features. ,make Traversing a collection Elements in; calculate the energy score between eigenvectors. The attention weight matrix is ​​obtained by normalization using the Softmax function. In the denominator of the formula Indicates a node Iterate through all neighboring nodes and sum the values, then take the maximum weight value for the noise feature as the sum. The calculation formula is as follows:

[0084]

[0085] in, To query the key weight matrix, For dimensional scaling factor, For from set Input features; express It belongs to the node The set of neighboring nodes One of the elements;

[0086] For variables A deep residual network is used for solution; the input feature tensor of the deep residual network is defined. The high-frequency oscillation features extracted in the aforementioned steps With zero-point drift characteristics To splice, that is The input feature tensor After processing the residual block containing skip connections, the output logits vector is then mapped to a probability distribution via a normalized exponential function. The calculation formula is as follows:

[0087]

[0088] in, For activation function, Normalized exponential function; residual mapping function The specific structure is defined as follows:

[0089]

[0090] in, For weight parameters, These are bias parameters; These are the weight matrix and bias vector of the output layer, respectively;

[0091] Based on the interference weights, the state-aware entropy is calculated. In this embodiment, the theoretical entropy value from Embodiment 1 is transformed into an engineering discrete calculation model suitable for graph neural networks. The calculation model is as follows:

[0092]

[0093] in, The source is calculated by the state-aware entropy calculation module, the physical meaning is state-aware entropy, and the unit is dimensionless. The source is preset according to the device type; the physical meaning is the total number of discrete intervals in the physical state space, in units of ; where, the discrete intervals of the physical state space According to the safety level of power grid equipment, it is divided into: normal operation zone, performance degradation zone, early warning deviation zone and cognitive failure zone. Each zone corresponds to a probability dimension output by the causal reasoning model. The source is the residual network output of the causal inference model, and its physical meaning is that the current monitoring data cannot explain the device's current state. The probability distribution of a physical state, in dimensionless units;

[0094] The source is a preset constant, the physical meaning is the noise amplification factor, and the unit is dimensionless. The source is the real-time calculation of the attention mechanism, the physical meaning is the maximum value of the interference weight, and the unit is dimensionless;

[0095] This formula design indicates that when the data contains an uninterpretable probability distribution... The more uniform the entropy, the less the model knows the state of the device. The higher the level, the better in response to the detection of strong external interference. When the entropy value is large, it will be further amplified. This design enables the system to be sensitive to cognitive ambiguity caused by data noise and avoid misjudging sensor failure as stable equipment operation.

[0096] Example 3:

[0097] The proactive verification decision module is used to generate proactive calibration verification instructions in response to the state perception entropy exceeding the deviation critical threshold. This includes: identifying the current operating load rate of the target power grid equipment and preset power supply continuity indicators; searching the preset intervention strategy library when the state perception entropy exceeds the deviation critical threshold; selecting physical actions from the intervention strategy library that can reduce the state perception entropy, including unplanned power outage maintenance or preventive component replacement; and encapsulating the selected physical actions into proactive calibration verification instructions, which are configured to sacrifice short-term operating efficiency in exchange for improved model cognitive accuracy.

[0098] This embodiment details the internal logic of the active verification decision module in generating instructions, adhering to the principle of minimizing the impact on power supply continuity; the system identifies the current operating load rate of the target power grid equipment. and preset power supply continuity indicators; in response to The module retrieves a library of preset intervention strategies containing various levels of physical intervention methods; the module then adjusts the strategy based on the current... Filter from the strategy library to find strategies that can effectively reduce The physical action; responding to the high-frequency oscillation characteristics in Example 2. Dominant, that is and ,in The system determines that the entropy increase originates from external electrical shocks based on the preset characteristic amplitude threshold. At this time, the physical actions selected are unplanned power outage maintenance, filter installation, or isolation testing, because such actions can physically cut off the oscillation source.

[0099] Response to zero-point drift characteristics Dominant, that is and The system determines that the entropy increase originates from the measurement system itself. At this point, the physical action to be selected is the preventive replacement of components, because the maintenance of the main electrical circuit cannot reduce the entropy value caused by sensor aging.

[0100] These actions include, but are not limited to: unplanned power outages for maintenance that force equipment to disconnect for offline testing during low load periods to obtain the highest fidelity data at the expense of power supply reliability; or preventative replacements of critical sensors or vulnerable components without obvious signs of failure, sacrificing maintenance costs for data accuracy; the system encapsulates the selected physical actions as... And send it to the dispatch center or maintenance team;

[0101] This embodiment transforms the verification process from passive data analysis to proactive physical intervention. By pre-setting actions with different costs in the strategy library, the system can intelligently select the physical means with the lowest cost to rebuild the credibility of the digital twin in critical moments of cognitive collapse, realizing dynamic optimization of operation and maintenance decisions and ensuring system controllability under extreme uncertainty.

[0102] Example 4:

[0103] The model calibration module is used to correct the internal parameters of the digital twin model based on calibration feedback data. This includes: receiving physical disassembly data or offline test data generated after executing the active calibration verification command and using it as calibration feedback data; calculating the residual vector between the calibration feedback data and the predicted value of the digital twin model; and using the backpropagation algorithm to adjust the weight parameters of the digital twin model according to the residual vector until the model output prediction state and the fitting degree of the calibration feedback data reach the preset standard.

[0104] This embodiment describes the closed-loop process of the model calibration module performing corrections, which is essentially a supervised learning process based on real physical feedback; the module receives when... The physical disassembly data or offline test data generated after execution are marked as... Calibration feedback data; the module calculates the difference between the calibration feedback data and the predicted value of the digital twin model at the same time, i.e., the residual vector, and introduces a dimensional normalization factor to ensure the consistency of physical meaning. The calculation model is as follows:

[0105]

[0106] in, The physical meaning is corrected to the standard deviation reference value for measuring noise, and its unit is the same as that of the physical quantity. To maintain consistency, by taking the square. , making the first item This is transformed into a dimensionless Mahalanobis distance form, thus relating it to the second dimensionless regularization term. Maintain consistency in dimensions to ensure that the optimization process is not affected by the magnitude of physical units;

[0107] The source is calculated by the model calibration module, and its physical meaning is the residual scalar containing the regularization term, with the unit being dimensionless; The source is a preset system reference value, and its physical meaning is a normalization factor. Its unit is the same as that of the physical quantity. Maintain consistency, i.e., volts or amperes, used to eliminate the physical dimensions of the residual terms; The source is actual measurement after physical intervention, the physical meaning is calibration feedback data, and the unit depends on the specific physical quantity; The source is the model output, the physical meaning is the predicted value of the digital twin model, and the unit depends on the specific physical quantity; The source is determined according to the L-curve method, its physical meaning is the regularization coefficient, and its unit is dimensionless.

[0108] In specific settings, an L-shaped curve of residual norm and solution norm is plotted, and the value corresponding to the point of maximum curvature is selected as... To balance the fit and model complexity; The source is the current state of the model, the physical meaning is the weight parameters of the digital twin model, and the unit is dimensionless;

[0109] The system utilizes the backpropagation algorithm, based on Update using gradient descent This iterative process continues until the model output matches... When the fit reaches the preset standard, the system determines that the model has recovered its vision and... Force reset to 0;

[0110] This embodiment ensures that the digital twin model has self-healing capabilities. By introducing high-value samples generated by physical intervention, it solves the drift problem caused by the lack of negative samples in the long-term operation of traditional models, ensuring the long-term effectiveness of full life cycle management and enabling the model to adapt to the dynamic evolution of physical entities.

[0111] Example 5:

[0112] The distributed energy impact characteristics acquired in the data acquisition module include: voltage fluctuation frequency, reverse power flow amplitude, and harmonic distortion rate generated by photovoltaic or wind power access points; the state-aware entropy calculation module is further used to establish a dynamic coupling model between the distributed energy impact characteristics and the equipment insulation aging rate, so as to remove data noise generated by external impacts.

[0113] This embodiment specifically defines the distributed energy impact characteristics acquired by the data acquisition module and constructs a dynamic coupling model; the data acquisition module captures specific power quality indicators generated by new energy grid connection points through high-frequency sampling, specifically including the voltage fluctuation frequency caused by photovoltaic inverters. The amplitude of reverse power flow from the user side to the grid side And total harmonic distortion, i.e., harmonic distortion rate Based on this, the state-aware entropy calculation module establishes a dynamic coupling model to isolate data noise generated by external shocks. The dynamic coupling model for insulation aging rate is as follows:

[0114]

[0115] in, To correct the aging rate, Based on aging rate, For frequency sensitivity coefficient, For voltage fluctuation frequency, The amplitude coupling coefficient is... This represents the amplitude of the reverse current flow. Harmonic distortion rate;

[0116] The source is calculated, and the physical meaning is the corrected equipment insulation aging rate, expressed as a percentage per year.

[0117] The data is derived from traditional model calculations and its physical meaning is the basic thermal aging rate calculated based on Arrhenius's law, expressed as a percentage per year.

[0118] The data source is historical fault data regression analysis; its physical meaning is frequency sensitivity coefficient, and the unit is seconds. , used to balance the dimensions of frequency;

[0119] The data is acquired by the data acquisition module, and its physical meaning is the voltage fluctuation frequency, measured in Hertz.

[0120] The source is historical fault data regression analysis, the physical meaning is amplitude coupling coefficient, the unit is dimensionless, and it is used to match the per-unit value dimension;

[0121] The data is obtained from the data acquisition module, and its physical meaning is the reverse current amplitude, with the unit being per unit value. The data source is obtained from the data acquisition module, and its physical meaning is harmonic distortion rate, expressed as a percentage.

[0122] It should be noted that when substituting into the formula... When performing calculations, It must be converted to per-unit form, that is, if the distortion rate is... Then the value substituted is To prevent the exponent from being incorrectly substituted with percentage values. The result overflows, ensuring the nonlinear correction coefficient. It falls within a physically explainable range. Inside;

[0123] This formula reveals the nonlinear acceleration mechanism of distributed energy on equipment lifespan. By explicitly incorporating these impact characteristics into the aging model, this embodiment can distinguish between normal aging and impact damage, thereby eliminating false abnormal signals caused by external impacts when calculating entropy values, and significantly improving the purity of the system's perception in complex power grid environments.

[0124] Example 6:

[0125] The cascaded fault risk prediction module is used to calculate the potential contribution rate of regional large-scale power outages caused by state deviations based on state perception entropy and the node importance of the target power grid equipment in the power grid topology. The active verification decision module is further used to dynamically adjust the deviation critical threshold according to the potential contribution rate. It is configured to reduce the deviation critical threshold when the potential contribution rate increases in order to improve the system's sensitivity to data distortion.

[0126] This embodiment introduces a cascaded fault risk prediction module, designed to assess the impact of single-point cognitive failure on overall power grid security; this module is based on state-aware entropy. Based on the node importance of the target power grid equipment in the power grid topology, calculate the potential contribution rate of regional large-scale power outages caused by state deviations. ; in order to make the formula and Having computability, this embodiment clarifies its underlying algorithm:

[0127] For node importance The betweenness centrality algorithm based on the entire network topology is used for calculation; this index reflects the importance of the device node as a power transmission bridge, and its calculation formula is as follows:

[0128]

[0129] in, For nodes arrive The total number of shortest paths, For the nodes The number of shortest paths, and For the network excluding the target node Any two distinct nodes other than those mentioned above; for potential contribution rate The system employs static security analysis simulation based on the N-1 criterion; it logically disconnects the target device node in the digital twin space, performs power flow calculations, and statistically analyzes the total network load loss after the disconnection. With total system load The ratio is calculated using the following formula:

[0130]

[0131] in, This represents the total load shedding after a fault. This represents the total system load before the failure.

[0132] The active verification decision module is based on The deviation critical threshold is dynamically adjusted, and the dynamic threshold adjustment model is as follows:

[0133]

[0134] in, The source is calculated, and the physical meaning is the critical threshold of deviation after dynamic adjustment. The unit is dimensionless.

[0135] The source is preset, and the physical meaning is the preset basic deviation critical threshold, which corresponds to the initial setting value in Example 1. The unit is dimensionless.

[0136] The source is preset, the physical meaning is sensitivity adjustment factor, and the unit is dimensionless; The source is the risk prediction module's simulation calculation based on the N-1 criterion, and its physical meaning is the potential contribution rate, expressed as a percentage.

[0137] The source is power grid topology analysis, the physical meaning is node importance, and the unit is dimensionless.

[0138] This formula reflects a strategy of lower tolerance for higher risks. For critical equipment in pivotal positions, the system will automatically and significantly reduce the deviation threshold once the potential risk contribution rate increases. This means that for core equipment, the cloud platform will immediately trigger active verification if there is even the slightest uncertainty, thereby nipping the risk of large-scale power outages in the bud and achieving differentiated risk management.

[0139] Example 7:

[0140] The process of generating active calibration verification instructions by the active verification decision module follows the following optimization objective function: construct a joint optimization function containing a first sub-objective and a second sub-objective; the first sub-objective is to minimize the accumulated value of state-aware entropy; the second sub-objective is to minimize the physical resource consumption cost generated by executing active calibration verification instructions; under the premise that the state confidence is not lower than the preset safety baseline, solve the joint optimization function to determine the optimal active calibration verification instruction.

[0141] In this embodiment, the active verification decision module solves the joint optimization function when generating instructions. The goal is to find the optimal solution between cognitive accuracy and physical cost. The system constructs a joint optimization function that includes a first sub-objective and a second sub-objective, where the first sub-objective is to minimize the accumulated value of state perception entropy, and the second sub-objective is to minimize the physical resource consumption cost generated by executing active calibration verification instructions.

[0142] Solving the above joint optimization function At that time, the system uses the interior-point method combined with gradient descent sequences, while satisfying the state confidence level. Optimization is performed within the constrained convex space to ensure the real-time performance and convergence of the computation results; while satisfying the state confidence level... Not lower than the preset safety baseline Under the premise that the system solves for the following joint optimization function, the calculation formula is as follows:

[0143]

[0144] in The objective function value, This is the risk-cost conversion coefficient. For state-aware entropy, For the integration time window, These are the weighting coefficients. For physical consumption costs;

[0145] The source is calculated, the physical meaning is the optimized target value, and the unit is monetary unit; The source is dynamically set based on the current power grid operation mode; its physical meaning is a risk cost conversion coefficient, with units of monetary units per second, used to convert time integral units into monetary units; in order to achieve future time windows Inside The integral calculation in this embodiment is based on the current time. Based on the historical entropy value sequence of the previous 5 sampling periods, a first-order Taylor expansion trend prediction model is constructed; the specific prediction formula is as follows:

[0146]

[0147] in, The entropy growth rate at the current moment is calculated using the five-point difference method:

[0148]

[0149] Based on this linear trend model, the above integral term It can be interpreted as:

[0150]

[0151] This analytical expression transforms the integral based on the unknown future state into one based on the current state. and trends The deterministic computation makes the objective function It is solvable;

[0152] in, The source is model prediction, and the physical meaning is the prediction time window. The cumulative state-perception entropy within the region, in dimensionless units; The source is preset according to the power grid dispatch command cycle, and its physical meaning is the integral time window parameter, with the unit being seconds; The source is preset, the physical meaning is weighting coefficient, and the unit is dimensionless; The source is calculated based on the preset cost in the strategy library, and the physical meaning is the cost of physical resources consumed by executing the active calibration verification command, in monetary units;

[0153] This model weighs abstract cognitive risks against specific economic costs within the same mathematical framework, ensuring that the system will not experience frequent power outages for minor improvements in accuracy, nor will it allow cognitive blind spots to expand in order to save money, thus achieving the global optimal solution for full lifecycle management under multi-objective constraints.

[0154] Example 8:

[0155] The state-aware entropy calculation module is also used to: monitor the temporal correlation decay trend of sensor time-series data; when the correlation coefficient between multi-source sensor data is detected to be lower than the preset consistency threshold, it is determined that there is a hidden data consistency failure, and a preset penalty factor is introduced to the state-aware entropy for exponential amplification.

[0156] This embodiment integrates a detection mechanism for sensor covert failure into the state-aware entropy calculation module; the module continuously monitors the temporal correlation decay trend of sensor time-series data; when the correlation coefficient between multi-source sensor data is detected... Below the preset consistency threshold At this point, the system determines that a hidden data consistency failure has occurred; based on this, the module adjusts the state-aware entropy. Introduce a preset penalty factor With exponential amplification, the correlation-penalized entropy model is as follows:

[0157]

[0158] in, The source is calculated, the physical meaning is the corrected final state perception entropy, and the unit is dimensionless.

[0159] The source is derived from the aforementioned steps, and its physical meaning is the basic entropy value calculated based on the data distribution, with the unit being dimensionless.

[0160] The source is a pre-set large positive number, the physical meaning is a penalty factor, and the unit is dimensionless;

[0161] The source is preset, the physical meaning is the consistency threshold, and the unit is dimensionless.

[0162] The data is derived from real-time calculations and its physical meaning is the Pearson correlation coefficient between current multi-source sensor data. The unit is dimensionless.

[0163] The function of this formula is to execute a veto logic. In physical devices, different parameters usually have strong physical correlations. If this correlation suddenly disappears, it often means that the sensor itself has a problem rather than the device itself. By exponentially amplifying the entropy value, the system can quickly identify such data source pollution and force the system to trigger active verification to check for sensor failures, thus preventing model output errors caused by input data source pollution.

[0164] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A cloud platform for the full lifecycle management of power grid equipment, characterized in that, include: The data acquisition module is used to acquire the operation monitoring data stream of the target power grid equipment. The operation monitoring data stream includes: sensor time series data, distributed energy impact characteristics, historical operation and maintenance logs, and environmental interference parameters. The state-aware entropy calculation module is used to identify the nonlinear correlation between data features and physical state based on the operation monitoring data stream using a causal reasoning model, and to calculate the state-aware entropy that characterizes the degree of deviation between the digital twin model and the physical entity state. The confidence assessment module is used to compare the state perception entropy with a preset deviation threshold to determine the state confidence of the current full life cycle management system in the true physical state of the target power grid equipment. An active verification decision module is used to generate an active calibration verification command in response to the state perception entropy exceeding the deviation critical threshold. The active calibration verification command is used to forcibly trigger physical intervention operations on the target power grid equipment in order to obtain high-fidelity calibration feedback data. The model calibration module is used to correct the internal parameters of the digital twin model based on the calibration feedback data, and reset the state-aware entropy to a preset safe range, so as to restore the system's model fitting accuracy to the target power grid equipment. The state-aware entropy calculation module is used to identify the nonlinear correlation between data features and physical states based on the operation monitoring data stream using a causal inference model, and to calculate the state-aware entropy, which characterizes the degree of deviation between the digital twin model and the physical entity state, including: The sensor timing data and the environmental interference parameters are retrieved. Construct a causal topology graph of device status based on a graph neural network, and extract high-frequency oscillation features and zero-point drift features from the operation monitoring data stream; Quantify the interference weights of the high-frequency oscillation characteristics and the zero-point drift characteristics on the equipment insulation status assessment; Based on the aforementioned interference weights, the probability distribution of the evolution of the device's physical state that cannot be explained by the current monitoring data is calculated, and the degree of dispersion of the probability distribution is defined as the state perception entropy; the formula for calculating the state perception entropy is: ； in, State-aware entropy, with units of dimensionless; The total number of discrete intervals in the physical state space, expressed in units of ; The current monitoring data cannot explain the device's current state. The probability distribution of a physical state, in dimensionless units; This is the noise amplification factor, in dimensionless form. This represents the maximum value of the interference weight, in dimensionless form. The active verification decision module is used to generate an active calibration verification instruction in response to the state perception entropy exceeding the deviation critical threshold, including: Identify the current operating load rate and preset power supply continuity indicators of the target power grid equipment; When the state perception entropy exceeds the deviation critical threshold, a preset intervention strategy library is retrieved; Physical actions that can reduce the state perception entropy are selected from the intervention strategy library, including unplanned power outage maintenance or preventive component replacement. The selected physical actions are encapsulated into the active calibration verification instruction, which is configured to sacrifice short-term running efficiency in exchange for improved model cognitive accuracy. The process by which the active verification decision module generates the active calibration verification instruction follows the following optimization objective function: Construct a joint optimization function that includes the first sub-objective and the second sub-objective; the formula for calculating the joint optimization function is as follows: ； in, The objective function value, This is the risk-cost conversion coefficient. At the current sampling time, For state-aware entropy, For the integration time window, These are the weighting coefficients. For physical consumption costs; The first sub-objective is to minimize the cumulative value of the state-aware entropy; The second sub-objective is to minimize the physical resource consumption cost incurred in executing the active calibration verification command; Provided that the state confidence level is not lower than a preset safety threshold, the joint optimization function is solved to determine the optimal active calibration verification command.

2. The cloud platform for full lifecycle management of power grid equipment according to claim 1, characterized in that, The model calibration module is used to correct the internal parameters of the digital twin model based on the calibration feedback data, including: Receive physical disassembly data or offline test data generated after executing the active calibration verification command, and use it as the calibration feedback data; Calculate the residual vector between the calibration feedback data and the predicted value of the digital twin model; Using the backpropagation algorithm, the weight parameters of the digital twin model are adjusted according to the residual vector until the model output prediction state and the fitting degree of the calibration feedback data reach a preset standard.

3. The cloud platform for full lifecycle management of power grid equipment according to claim 1, characterized in that, The distributed energy impact characteristics acquired by the data acquisition module include: Voltage fluctuation frequency, reverse power flow amplitude, and harmonic distortion rate generated by photovoltaic or wind power access points; The state-aware entropy calculation module is further used to establish a dynamic coupling model between the distributed energy impact characteristics and the equipment insulation aging rate, so as to remove data noise generated by external impacts.

4. The cloud platform for full lifecycle management of power grid equipment according to claim 1, characterized in that, Also includes: The cascaded fault risk prediction module is used to calculate the potential contribution rate of regional large-scale power outages caused by state deviations based on the state perception entropy and the node importance of the target power grid equipment in the power grid topology. The active verification decision module is further configured to dynamically adjust the deviation critical threshold based on the potential contribution rate, wherein when the potential contribution rate increases, the deviation critical threshold is reduced to improve the system's sensitivity to data distortion.

5. The cloud platform for full lifecycle management of power grid equipment according to claim 1, characterized in that, The state-aware entropy calculation module is also used for: Monitor the temporal correlation decay trend of the sensor's time-series data; When the correlation coefficient between multi-source sensor data is detected to be lower than a preset consistency threshold, it is determined that a hidden data consistency failure has occurred, and a preset penalty factor is introduced to exponentially amplify the state perception entropy.

Images