Hybrid reality-based power equipment intelligent commissioning system and method

Abstract

The application discloses a mixed reality-based power equipment intelligent debugging system and method, and particularly relates to the field of equipment debugging, and comprises seven modules of setting space electromagnetic field reconstruction, equipment state field estimation, electric field potential energy mapping, multi-source data assimilation, dynamic debugging strategy generation, feedback check self-adaptation and distributed online learning; through collecting on-site space and electromagnetic data, three-dimensional electromagnetic fields are reconstructed and dynamically corrected, equipment operation states are accurately estimated, multi-physical field coupling linkage is realized, and a work safety area is demarcated relying on electromagnetic potential energy; global state perception is completed by fusing heterogeneous operation data, and optimal debugging actions are automatically selected in combination with debugging targets and safety constraints; model parameters are self-adaptively corrected through operation feedback comparison, forming a closed-loop debugging regulation; and multi-site model collaborative optimization is completed by using federated learning; the application realizes intelligent management and control of the whole process of power equipment debugging, and is suitable for intelligent debugging scenes of substations.

Description

Technical Field

[0001] This invention relates to the field of equipment commissioning technology, and more specifically, to a system and method for intelligent commissioning of power equipment based on mixed reality. Background Technology

[0002] As the scale of power system construction continues to expand, substations, as the core nodes of power transmission and control, have increasingly diverse types of internal power equipment and more complex layouts. Daily operation and maintenance and on-site commissioning of equipment have become key links to ensure the stable operation of the power grid.

[0003] During the commissioning of power equipment, it is necessary to collect various types of data, such as on-site spatial location, electromagnetic signals, and equipment operating parameters, and carry out relevant operations in conjunction with the real-time operating conditions of the power grid. This involves multiple technical aspects, including electromagnetic field distribution sensing, equipment operating status monitoring, multi-source data integration, and commissioning strategy formulation.

[0004] Currently, the application of mixed reality technology in the industrial field is gradually expanding, providing a visual and immersive interactive method for on-site operations; at the same time, the rapid development of technologies such as multi-source data fusion, machine learning, and federated learning has also provided technical support for the intelligent upgrading of power equipment commissioning.

[0005] The operating status of various power equipment is closely related to the power flow and electromagnetic environment of the power grid. The commissioning scenarios of different substations are different. On-site commissioning work needs to rely on multi-dimensional data perception and flexible control methods to adapt to the complexity and diversity of power grid operation.

[0006] However, it still has some drawbacks in practical use, such as:

[0007] 1. Insufficient on-site electromagnetic environment perception capability: Traditional power equipment commissioning lacks precise reconstruction methods for full-domain three-dimensional electromagnetic fields, making it impossible to simultaneously acquire spatial electromagnetic distribution and equipment area outlines. Relying solely on basic electrical monitoring methods to collect data makes it difficult to simultaneously grasp real-time electromagnetic changes on-site. It cannot combine power grid flow and equipment on / off status to complete dynamic field updates. The electromagnetic environment data acquisition dimension is singular, failing to provide refined environmental foundation support for on-site operations.

[0008] 2. Poor correlation between equipment status assessment and field: Existing commissioning technology can only independently monitor the operating parameters on the surface of the equipment. It has not established a standardized system for assessing the evolution of equipment status, and cannot accurately predict the actual operating conditions of the equipment based on sensor data. At the same time, there is no two-way linkage adjustment mechanism between equipment status and electromagnetic field parameters. The physical fields are independent of each other and it is difficult to match the changes in operating conditions caused by equipment aging and load changes.

[0009] 3. Low level of data integration and rigid commissioning strategies: Existing technologies have weak integration capabilities for electromagnetic data, equipment status data, power grid electrical data, and transient characteristic data. Unified and standardized integration processing has not been completed, resulting in low data utilization. Commissioning strategies are mostly preset fixed modes, unable to flexibly adjust the action range based on real-time safety conditions and on-site operating conditions. Strategy formulation lacks scientific and quantitative evaluation criteria.

[0010] 4. Lack of closed-loop optimization and global iteration in debugging and control: Traditional debugging processes only complete the issuance of instructions and simple data feedback, without building a complete closed-loop verification and control system. It is impossible to compare the deviation between predicted operating conditions and actual operating conditions in a timely manner, and it is difficult to independently complete the correction of model parameters. At the same time, the debugging data of each site are isolated from each other, and there is no way to share experience across sites or coordinate model optimization. The technical system cannot be upgraded independently for a long time. Summary of the Invention

[0011] In order to overcome the above-mentioned defects of the prior art, the present invention provides a power equipment intelligent commissioning system and method based on mixed reality, which solves the problems mentioned in the background art through the following scheme.

[0012] To achieve the above objectives, the present invention provides the following technical solution:

[0013] A mixed reality-based intelligent commissioning system for power equipment includes:

[0014] Space electromagnetic field reconstruction module: Collects on-site space and electromagnetic related data, completes the reconstruction of three-dimensional space electromagnetic field distribution and calibration of space equipment area outline, combines system operation data to realize dynamic correction of electromagnetic field, and outputs electromagnetic field distribution data, space equipment area outline and related uncertain information;

[0015] Equipment State Field Estimation Module: Delineates the corresponding spatial range of the equipment based on the spatial equipment area outline, establishes the equipment state evolution relationship, estimates the equipment operating state by combining sensor monitoring data, and adjusts the electromagnetic field related parameters in reverse according to the equipment state;

[0016] Electric field potential energy mapping module: Receives electromagnetic field distribution data, calculates spatial electromagnetic potential energy and gradient characteristics, delineates hazardous areas after data smoothing, forms unified safety assessment indicators, and outputs them outwards;

[0017] Multi-source data assimilation module: It receives information from the spatial electromagnetic field reconstruction module and information generated by the equipment state field estimation module, and simultaneously receives power grid operation data and equipment transient characteristic data. It completes data standardization processing and fusion calculation, and outputs the overall system operation status results.

[0018] Dynamic debugging strategy generator module: Combining the overall system operating status, preset debugging targets and safety evaluation indicators, it simplifies the system status dimensions, divides the adjustable action range of the equipment, and determines the optimal equipment debugging action.

[0019] Feedback verification and adaptive module: Issues optimal equipment debugging actions and collects actual equipment operation feedback data, compares the difference between the actual operating state and the predicted state, determines the model adaptability and automatically corrects the model parameters, and formulates a new debugging strategy based on the actual working conditions to form a closed-loop control.

[0020] Distributed online learning module: It aggregates debugging and running data from multiple sites through all the aforementioned modules, and uses a federated learning approach to complete the global control model iterative optimization. Each local terminal adapts and adjusts the optimized model accordingly.

[0021] Intelligent commissioning methods for power equipment based on mixed reality include:

[0022] S1. Spatial electromagnetic field reconstruction: Collect multi-source electromagnetic signals and spatial geometric data on site, combine on-site electromagnetic interference factors to construct an electromagnetic field solution model to complete the reconstruction of the three-dimensional spatial electromagnetic field distribution in the whole domain, and combine real-time operation data of the power system to dynamically and linearly correct the electromagnetic field distribution obtained by reconstruction.

[0023] S2. Equipment State Field Estimation: Based on the electromagnetic field reconstruction results, the corresponding spatial regions of each power equipment are divided. The lumped parameter model is used to establish the internal state evolution relationship of the equipment. Combined with the equipment surface temperature data collected by infrared thermography, the equipment operating state is estimated. At the same time, the equipment state data is fed back to the electromagnetic field reconstruction link to realize the bidirectional coupling and linkage of multiple physical fields.

[0024] S3, Electric Field Potential Energy Mapping: Based on the corrected global electromagnetic field data, spatial electromagnetic potential energy and potential energy gradient information are calculated. After smoothing and noise reduction of the gradient data, regional safety assessment indicators are generated. Based on the safety indicators, high-risk areas for power equipment operation are divided and safety constraints are formed.

[0025] S4. Multi-source data assimilation: Integrates electromagnetic field data, equipment status data, real-time electrical operation data of power system and transient fault characteristic data, performs unified and standardized processing on various heterogeneous monitoring data, and completes multi-source data assimilation processing by Bayesian filtering fusion method, outputting the fused overall system operation status information.

[0026] S5. Dynamic debugging strategy generation: The assimilated system operating state is simplified and dimensionality is reduced. The adjustable action space of the equipment is defined in combination with the preset debugging target. A comprehensive cost evaluation system including state deviation and adjustment cost is built based on the heuristic greedy search algorithm. At the same time, regional safety constraints and electrical parameter limit constraints are incorporated to select the optimal power equipment debugging action.

[0027] S6. Feedback Verification and Adaptive Adjustment: The determined optimal debugging actions are sent to the power equipment to complete parameter control, the actual operating response data of the equipment is read back and a unified dimension mapping is completed, the deviation between the actual operating state and the predicted operating state of the equipment is compared by constructing a stability judgment function, the system state inference model is adaptively corrected online based on the deviation results, and the debugging strategy planning is re-completed based on the actual operating state.

[0028] S7. Distributed online learning optimization: It aggregates the running data and model optimization data generated during the intelligent debugging process at multiple sites, and adopts a federated learning mode to complete the global model aggregation optimization in the cloud by uploading only the parameter update. It also improves the global model performance by combining the stochastic gradient dynamics algorithm. Each deployment site receives the optimized global model and completes personalized fine-tuning by combining it with local running data, iteratively optimizing the overall debugging method and running performance.

[0029] The technical effects and advantages of this invention are as follows:

[0030] Achieving comprehensive and accurate perception of the electromagnetic environment across the entire area: This solution relies on mixed reality terminals to complete multi-dimensional data acquisition on-site, and can quickly complete the reconstruction of the three-dimensional electromagnetic field of the entire substation and the calibration of equipment contours. Combined with real-time power grid operation data, it can complete the dynamic correction of the electromagnetic field and output complete electromagnetic field distribution data and spatial uncertainty information in real time, so as to fully grasp the real-time changes of the on-site electromagnetic environment and lay a solid foundation for accurate environmental perception for on-site commissioning.

[0031] 2. Achieve accurate equipment status assessment and multi-field collaborative linkage: This solution builds a standardized state evolution model based on the spatial range of the equipment, and accurately estimates the real-time operating status of the equipment by combining multiple types of sensor data. It can adjust the relevant parameters of the electromagnetic field medium in reverse according to the actual operating status of the equipment, realize the bidirectional coupling and linkage between the equipment state field and the electromagnetic field, conform to the actual aging law of the equipment, and improve the accuracy of equipment condition assessment.

[0032] 3. Achieve deep fusion of multi-source data and intelligent autonomous decision-making: This solution can access various heterogeneous operating data and complete unified and standardized fusion processing, accurately judge the abnormal operating status of equipment, define the scope of compliant actions based on dimensionality reduction processing and debugging objectives, and select the optimal debugging actions based on a scientific evaluation system. It takes into account electrical limits and on-site operation safety conditions throughout the process, making debugging decisions more scientific, flexible and in line with the actual situation on site.

[0033] 4. Constructing a closed-loop debugging system and a global intelligent iteration system: This solution forms a complete closed-loop control process of action execution, data feedback, deviation verification, and model correction. It can autonomously complete model adaptation and optimization, ensuring the stability of debugging and operation. At the same time, relying on federated learning, it realizes multi-site parameter sharing and iteration, completes global model optimization under the premise of ensuring data security, and each site can be locally adapted and adjusted, continuously improving the overall level of intelligent debugging. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the overall system structure of the present invention.

[0035] Figure 2 This is a schematic diagram of the multi-source data assimilation module of the present invention.

[0036] Figure 3 This is a schematic diagram of the dynamic debugging strategy generator of the present invention.

[0037] Figure 4 This is a schematic diagram of the overall structure of the method of the present invention. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] refer to Figure 1 - Figure 3 The intelligent commissioning system for power equipment based on mixed reality shown includes:

[0040] Space electromagnetic field reconstruction module:

[0041] By deploying on-site spatial positioning and multi-source signal acquisition terminals (such as wearable multi-channel RF detectors, capable of measuring RF signal strength and phase, and synchronized with the spatial positioning function of commercial mixed reality headsets), RF signal strength, phase, and spatial geometry data are collected from multiple viewpoints. By solving Maxwell's equations with random source terms, the electromagnetic field distribution of the entire three-dimensional space is reconstructed. This module performs a complete reconstruction before the commissioning task begins (taking approximately 10 minutes). During subsequent commissioning, real-time SCADA data is used to linearly correct the field distribution to reflect changes in equipment status; the output electric field strength... With magnetic field strength This provides foundational data for subsequent security assessments, and its spatial uncertainty boundary will also be passed to the multi-source data assimilation module; specifically as follows:

[0042] Data acquisition and spatial coordinate alignment:

[0043] The terminal moves along a preset path using spatial positioning and signal acquisition, and the terminal... Automatic acquisition from different viewpoints:

[0044] Radio frequency signal complex amplitude matrix (M frequency points, K antenna elements), including amplitude and phase information;

[0045] Depth image With visible light images , where u is the pixel coordinate.

[0046] Using the terminal's built-in Simultaneous Localization and Mapping (SLAM) algorithm, all viewpoint data are registered to the global coordinate system. In the middle, the position of each voxel is obtained. Observational information.

[0047] Modeling of stochastic Maxwell's equations:

[0048] Considering the presence of unmodeled electromagnetic interference in the field, a random current source term is introduced into Ampere's law. (Gaussian white noise process, unit) The time-domain Maxwell's equations are as follows:

[0049] ;

[0050] in Permeability (unit: H / m) Electrical conductivity (unit: S / m) Let H be the dielectric constant (unit: F / m). Eliminating H yields the wave equation for the electric field E:

[0051] ;

[0052] Stochastic finite element method:

[0053] Discretize the spatial domain into a tetrahedral mesh, with the number of nodes... Taking the weak form of the equation, we approximate the solution using polynomial chaotic expansion (PCE). At each time step, we calculate the expected electric field. and its variance The electric field vector distribution on the boundary is reconstructed using multi-view RF complex amplitude observations (including phase) as the boundary condition (the specific inverse scattering algorithm uses existing technology). Depth images are used to determine the location of the metal casing, and ideal conductor boundary conditions (zero tangential electric field) are applied. The initial electric field distribution is output. and magnetic field distribution .

[0054] Linear correction based on SCADA data:

[0055] During commissioning, the operating status of the equipment will change the actual electromagnetic field distribution. A correction factor is defined, considering that the electric field strength is proportional to the current (approximately in the near-field region), and the current is proportional to the square root of the active power: ;in It is a collection of devices with significant changes in active power. Real-time active power for SCADA (in W). For the initial moment, The contribution weight of the equipment to the electromagnetic field (set according to the distance between the equipment and the key measurement point) , (Take the coordinates of the preset key operation points). The magnetic field correction coefficient is the same as that of the electric field. The corrected electromagnetic field is then: This correction is calculated every 5 seconds and synchronized with subsequent decision-making and control cycles.

[0056] Equipment State Field Estimation Module:

[0057] Building upon the spatial profile of the equipment provided by the spatial electromagnetic field reconstruction module, this module focuses on the continuous state estimation within each power device. A lumped parameter model is used to describe the aging or temperature state of each device to reduce computational complexity and ensure observability. The state of each device is represented by a single scalar. The normalized aging factor or normalized temperature is represented, and its evolution satisfies an ordinary differential equation. Kalman filtering correction is performed using the external surface temperature measured by an infrared thermal imager. This module updates on a minute-level timescale (sampling period of 60 seconds), but to match second-level control, an extrapolation method is used to provide state estimates between adjacent update points; the output... The uncertainty of this data is fed into the multi-source data assimilation module, and simultaneously used to correct the dielectric constant in the electromagnetic field reconstruction module, forming a bidirectional coupling of multi-physics fields; specifically as follows:

[0058] Equipment area division: Based on the 3D contours of the equipment reconstructed by the electromagnetic field reconstruction module, the area division of each piece of equipment is determined. Corresponding spatial region Regional boundaries The device casing is accurately identified by the data acquisition terminal.

[0059] Lumped parameter state equations: defining device states (0 represents new / low temperature, 1 represents end of life / high temperature). Its evolution satisfies: ;in, The time constant (in seconds). The load influence coefficient (dimensionless) is used during the initial system commissioning phase. Initial values ​​are estimated online using recursive least squares estimation based on the load change and temperature response curves recorded by temperature sensors. The real-time active power of the equipment (in W). The rated power of the equipment (in watts); for equipment without power measurement (such as switches that only measure current), the normalized squared value of the current can be used instead. .

[0060] Observation model: Average surface temperature of infrared thermal imaging sensor measuring equipment (Unit: K). Assume a linear relationship between state and temperature: ,in, Ambient temperature (in K). The maximum allowable temperature of the equipment (in K).

[0061] The observation equation is: , For observation noise (Gaussian white noise, variance) ).

[0062] Kalman filtering and extrapolation: Discretizing the state equation, sampling period Seconds. The infrared thermography data is fused using a standard Kalman filter to obtain an estimate for each sampling time. and its variance Between two filtering moments (second-level control cycles), zero-order hold extrapolation is used: for any The variance remains constant, where, For the k-th Kalman filter sampling time, For equipment At sampling time The actual state value.

[0063] Feedback to the electromagnetic field reconstruction module: Update the average dielectric constant within the device area based on the state estimate. ,in is the vacuum permittivity (unit: F / m). This is the sensitivity coefficient (dimensionless) to polarizability during aging; this value is passed back to the space electromagnetic field reconstruction module for dielectric constant adjustment in electromagnetic field correction.

[0064] Electric potential energy mapping module:

[0065] Receive the corrected electromagnetic field output from the space electromagnetic field reconstruction module Calculate the electromagnetic potential energy density at each point in space. and its gradient The output potential energy gradient, smoothed by an exponentially weighted moving average, serves as an objective indicator of the safety of the quantitative debugging operation and will be used as a constraint in subsequent decision-making modules. This module executes every 5 seconds, synchronized with electromagnetic field correction, to ensure the real-time nature of safety indications; details are as follows:

[0066] Potential energy density calculation: For any point x in space and time t, the electromagnetic potential energy density is: ;in Vacuum permeability (unit: H / m).

[0067] Potential energy gradient calculation: Calculate the gradient with respect to spatial coordinates: ;

[0068] The central difference method is used to calculate on the discrete grid output by the electromagnetic field reconstruction module. For a typical operating point on each device, the gradient magnitude of that point is extracted. .

[0069] Safety indicator smoothing: The gradient magnitude is smoothed using an exponentially weighted moving average (EWMA) to eliminate spikes caused by switching transients. ;in Let be the smoothing factor. Define the reference gradient. (Maximum value at initial moment), then the safety index is: ;when At that time, the area where device i is located is marked as a "high-risk area". This indicator will be directly sent to the subsequent decision-making module to adjust the set of allowed actions.

[0070] Multi-source data assimilation module:

[0071] The system receives electromagnetic field characteristics from the space electromagnetic field reconstruction module, equipment states from the equipment state field estimation module, real-time electrical quantities from the SCADA system, and transient characteristics from the protection recording device. To meet real-time requirements, a simplified Kalman filter-based Bayesian update is employed, centered on the current observation vector and considering the observation noise covariance, to assimilate these heterogeneous data sources into a unified state vector. and its covariance This module runs in 5-second cycles, and its output is directly fed into the dynamic debugging strategy generator to provide an estimate of the current optimal state for decision-making; specifically as follows:

[0072] Input feature extraction and standardization:

[0073] From the spatial electromagnetic field reconstruction module: select the electric field amplitude of L=50 preset key points. and its variance .

[0074] From the device state field estimation module: each device and its variance .

[0075] From SCADA: Voltage of each device Current Meritorious No merit and switch position (Binary).

[0076] From the protection waveform recording: the power frequency amplitude within the most recent second. Attenuation DC time constant .

[0077] Stack all features into the original observation vector Dimension Standardize each dimension independently: ;in, These are the historical mean and standard deviation for this dimension (calculated from offline data during system initialization); the standardized observation vector is denoted as... .

[0078] Observation model and prior distribution settings: Assume the observation noise is Gaussian white noise, and the covariance matrix... ,in, (Based on sensor noise level settings). Therefore, the likelihood distribution of the current observation is: ;

[0079] System state prior distribution The state transition is obtained from the posterior of the previous time step via state transition. The state transition employs a reduced-order model in the dynamic debugging strategy generator. The linear approximation is used, but since the model is nonlinear, in practice it can be simplified to directly use the posterior mean of the previous time step as the current prior mean, while the covariance is propagated through a constant decay factor: ,in, , Let be the process noise covariance matrix.

[0080] Kalman filter update: Let the mean of the prior state be... The covariance is Observation vector The dimension is the same as that of the state vector (both are...). ), observation matrix (That is, the state is directly observable, but contains noise). Then the Kalman gain is: Posterior mean and covariance: , Take the assimilated state estimate covariance .

[0081] Anomaly detection: For each observed component, calculate the absolute value of the deviation between its standardized value and the predicted value (the component corresponding to the prior mean). If the deviation exceeds 3 times the standard deviation (from...), anomaly detection is performed. If the square root of the diagonal element is obtained, the sensor may be faulty, and an alarm will be issued on the human-machine interface.

[0082] Dynamic debugging strategy generator module: Receives the current state estimate output by the multi-source data assimilation module. Covariance And the debugging goals set by the maintenance personnel. The sequence planning problem is transformed into a mixed-integer decision problem. To achieve good control performance within a finite time, a greedy search-based heuristic decision-making approach is adopted: only the current action is optimized, and the state after the action is evaluated through a heuristic future cost to approximate the long-term effect. The control period is uniformly set to 5 seconds; after discretizing the action space, an enumeration method is used to select the action that minimizes the immediate cost plus the future heuristic; the optimal action is output. (Equipment number and discretized parameter values); Safety constraint checks are achieved through key physical quantities output by the reduced-order model; details are as follows:

[0083] State dimensionality reduction: using the principal component projection matrix calculated offline during system initialization. , the original state Projected onto the principal component space, denoted as Debugging target Convert to target principal component Using a deterministic model, the state transition is as follows: ;in For the reduced-order dynamic model of the power system (neural network), the step size is... Seconds. This reduced-order model, in addition to outputting the principal components of the next time step, also... In addition, it outputs key physical quantities (such as the amplitude of key bus voltage and the amplitude of key line current), denoted as... This is used for subsequent safety constraint checks.

[0084] Action space discretization: Action Equipment number These are discrete variables.

[0085] Continuous parameters (For example, protection settings) Discretize by equipment type: If the equipment parameter itself is discrete (such as transformer tap positions), directly use the original tap position; if it is a continuously adjustable parameter, use equal-interval discretization, with the interval taken as the minimum allowable adjustment step size of the equipment (such as 0.01A or 0.1s), and the number of taps. The size of the action space after discretization is obtained by dividing the parameter range by the step size. Typical value On average per device The total number of actions is approximately 100.

[0086] Cost function and constraints: The immediate cost consists of state deviation and parameter adjustment. Due to the state... For standardized dimensionless quantities, and parameters Having physical dimensions, the parameter adjustment values ​​need to be normalized: ;in This is the state deviation weight matrix. (Dimensionless coefficient) These are the rated or default values ​​for the device parameters. This refers to the adjustable range of the parameter (e.g., the maximum set value minus the minimum set value).

[0087] The cost of future heuristics: Pick It is an identity matrix.

[0088] Safety constraints consist of two parts:

[0089] High-risk zone constraint from the electric potential energy mapping module: If device k is marked as a high-risk zone, then no modifications to the device are allowed (i.e., all actions of the device are excluded from the action space).

[0090] Physical safety constraints such as voltage and current: key physical quantities output from the reduced-order model (e.g., voltage amplitude, current amplitude) directly perform threshold checks. If any physical quantity exceeds the limit (e.g., voltage below 0.95pu or above 1.05pu, current exceeding the thermal stability limit), then an extremely high cost is imposed on the action, preventing it from being selected.

[0091] Greedy decision-making process:

[0092] At each decision time t: enumerate all allowed actions. (Filtered by high-risk area);

[0093] For each action, the next state is calculated using a reduced-order model. and key physical quantities ;

[0094] like Skip the action if any safety constraints are violated;

[0095] Calculate the total cost ;

[0096] Select The smallest action as .

[0097] The complexity of this process ,in The time for a single neural network feedforward is approximately 0.1ms, and the total time is less than 10ms, which fully meets the 5-second cycle requirement.

[0098] Output optimal action and the predicted next state (For use in subsequent verification modules).

[0099] Feedback verification and adaptive module: issues actions in the dynamic debugging strategy generator. Subsequently, device settings are written via the IEC 61850 protocol, and then the actual state changes are read back from SCADA. First, the original read-back state is standardized and PCA projected using the same methods as the multi-source data assimilation module to ensure spatial consistency. Then, a Lyapunov function is constructed on the dimensionality-reduced state space, and the difference between the actual response and the predicted response is compared. When the residual exceeds a threshold, the parameters of the reduced-order model are corrected online. The dynamic debugging strategy generator is then re-triggered based on the current actual state for planning; this module runs in 5-second cycles, with time progressing naturally to ensure the gradual stability of the closed-loop debugging process; details are as follows:

[0100] Storage of projection matrix and standardized parameters: During system initialization, the standardized parameters, namely the mean vector, are calculated offline. Standard deviation vector ; and principal component projection matrix These parameters are stored in the system.

[0101] Standardize and project the actual state: execute actions After that, wait 5 seconds (consistent with the decision cycle) and read back the original observation vector from SCADA. (Same dimension as the input of the multi-source data assimilation module); standardization is performed first: ;in Then project it onto the principal component space: ;

[0102] Lyapunov function: Definition of the Lyapunov function after dimension reduction: in To debug the projection of the target onto the principal component space.

[0103] Calculation of actual change: Calculate the actual change: ;in The next state predicted by the dynamic debugging strategy generator;

[0104] Ideally, it should satisfy , =0.1; It can be adjusted based on on-site debugging experience.

[0105] Residual determination and model update:

[0106] Define residual: ;like The model is accurate, accepts the action, and will As the initial state for the next moment Then proceed to the next round of decision-making. If This indicates a model mismatch. At this point, the following steps should be taken:

[0107] For the parameters of the reduced-order model Perform gradient descent updates to minimize the prediction error in one step:

[0108] ;in .

[0109] In actual readback state As the current state, discard the previous predicted trajectory and use the updated model to re-invoke the dynamic debugging policy generator for decision-making (i.e., start from the current actual state and reselect the action). Time has naturally advanced by 5 seconds, so no special time step processing is required.

[0110] Will , The changes and device response curves are stored in a local database as the basis for subsequent distributed learning.

[0111] The distributed online learning module collects data generated by multiple substations (i.e., multiple deployment instances of this system) during real-world commissioning tasks. This data is then federated through the cloud: each substation only uploads updated model parameters (not the original data). The cloud aggregates these parameters to generate a global model, which is then distributed to each substation for personalized fine-tuning. Bayesian learning is implemented using stochastic gradient Langevin dynamics (SGLD) with decaying steps to ensure parameter convergence. Updated parameters are periodically distributed to the dynamic commissioning strategy generator and feedback verification and adaptation modules at each substation, enabling all deployed systems to continuously improve based on each other's commissioning experience. Specifically:

[0112] Local model training and parameter upload: Each substation completes each training session. After the debugging task, use local data to adjust the current model parameters. Perform SGD training for E=5 epochs to obtain the parameter update amount. Only Upload to the cloud, not the original data.

[0113] Cloud Aggregation and SGLD: Cloud Servers Maintain Global Parameters For each batch of updates received (from at least C=3 sites), a federated average is performed: Then, noise is added using an SGLD with a decay step size. ;in , This represents the global iteration count. .

[0114] Parameter delivery and personalized fine-tuning: The cloud will send parameters every 24 hours. Broadcast to all sites. Each site, upon receiving the broadcast, uses data from its most recent 100 tasks for additional analysis. Fine-tuning over several cycles yields personalized parameters. It is used for dynamic debugging strategy generator and feedback verification and adaptive module.

[0115] Based on the above method, this application also discloses a hydrogen-powered unmanned aerial vehicle (UAV) energy monitoring and management system, with reference to... Figure 4 The intelligent commissioning method for power equipment based on mixed reality, as shown, includes:

[0116] S1. Spatial electromagnetic field reconstruction: Relying on a mixed reality terminal equipped with spatial positioning and signal acquisition functions, multi-source electromagnetic signals and spatial geometric data are collected on site. An electromagnetic field solution model is constructed in combination with on-site electromagnetic interference factors to complete the reconstruction of the full-domain three-dimensional spatial electromagnetic field distribution. The reconstructed electromagnetic field distribution is dynamically and linearly corrected in combination with real-time power system operation data.

[0117] S2. Equipment State Field Estimation: Based on the electromagnetic field reconstruction results, the corresponding spatial regions of each power equipment are divided. The lumped parameter model is used to establish the internal state evolution relationship of the equipment. Combined with the equipment surface temperature data collected by infrared thermography, the equipment operating state is estimated. At the same time, the equipment state data is fed back to the electromagnetic field reconstruction link to realize the bidirectional coupling and linkage of multiple physical fields.

[0118] S3, Electric Field Potential Energy Mapping: Based on the corrected global electromagnetic field data, spatial electromagnetic potential energy and potential energy gradient information are calculated. After smoothing and noise reduction of the gradient data, regional safety assessment indicators are generated. Based on the safety indicators, high-risk areas for power equipment operation are divided and safety constraints are formed.

[0119] S4. Multi-source data assimilation: Integrates electromagnetic field data, equipment status data, real-time electrical operation data of power system and transient fault characteristic data, performs unified and standardized processing on various heterogeneous monitoring data, and completes multi-source data assimilation processing by Bayesian filtering fusion method, outputting the fused overall system operation status information.

[0120] S5. Dynamic debugging strategy generation: The assimilated system operating state is simplified and dimensionality is reduced. The adjustable action space of the equipment is defined in combination with the preset debugging target. A comprehensive cost evaluation system including state deviation and adjustment cost is built based on the heuristic greedy search algorithm. At the same time, regional safety constraints and electrical parameter limit constraints are incorporated to select the optimal power equipment debugging action.

[0121] S6. Feedback Verification and Adaptive Adjustment: The determined optimal debugging actions are sent to the power equipment to complete parameter control, the actual operating response data of the equipment is read back and a unified dimension mapping is completed, the deviation between the actual operating state and the predicted operating state of the equipment is compared by constructing a stability judgment function, the system state inference model is adaptively corrected online based on the deviation results, and the debugging strategy planning is re-completed based on the actual operating state.

[0122] S7. Distributed online learning optimization: It aggregates the running data and model optimization data generated during the intelligent debugging process at multiple sites, and adopts a federated learning mode to complete the global model aggregation optimization in the cloud by uploading only the parameter update. It also improves the global model performance by combining the stochastic gradient dynamics algorithm. Each deployment site receives the optimized global model and completes personalized fine-tuning by combining it with local running data, iteratively optimizing the overall debugging method and running performance.

[0123] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0124] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A power equipment intelligent commissioning system based on mixed reality, characterized in that, include: Space electromagnetic field reconstruction module: Collects on-site space and electromagnetic related data, completes the reconstruction of three-dimensional space electromagnetic field distribution and calibration of space equipment area outline, combines system operation data to realize dynamic correction of electromagnetic field, and outputs electromagnetic field distribution data, space equipment area outline and related uncertain information; Equipment State Field Estimation Module: Delineates the corresponding spatial range of the equipment based on the spatial equipment area outline, establishes the equipment state evolution relationship, estimates the equipment operating state by combining sensor monitoring data, and adjusts the electromagnetic field related parameters in reverse according to the equipment state; Electric field potential energy mapping module: Receives electromagnetic field distribution data, calculates spatial electromagnetic potential energy and gradient characteristics, delineates hazardous areas after data smoothing, forms unified safety assessment indicators, and outputs them outwards; Multi-source data assimilation module: It receives information from the spatial electromagnetic field reconstruction module and information generated by the equipment state field estimation module, and simultaneously receives power grid operation data and equipment transient characteristic data. It completes data standardization processing and fusion calculation, and outputs the overall system operation status results. Dynamic debugging strategy generator module: Combining the overall system operating status, preset debugging targets and safety evaluation indicators, it simplifies the system status dimensions, divides the adjustable action range of the equipment, and determines the optimal equipment debugging action. Feedback verification and adaptive module: Issues optimal equipment debugging actions and collects actual equipment operation feedback data, compares the difference between the actual operating state and the predicted state, determines the model adaptability and automatically corrects the model parameters, and formulates a new debugging strategy based on the actual working conditions to form a closed-loop control. Distributed online learning module: It aggregates debugging and running data from multiple sites through all the aforementioned modules, and uses a federated learning approach to complete the global control model iterative optimization. Each local terminal adapts and adjusts the optimized model accordingly.

2. The intelligent commissioning system for power equipment based on mixed reality according to claim 1, characterized in that: The space electromagnetic field reconstruction module specifically includes: Spatial location information and electromagnetic signal data are collected by the acquisition equipment deployed on site. Based on the acquired data, the overall distribution of the three-dimensional electromagnetic field in the whole domain is constructed. At the same time, the outline of the area where various power equipment is located is uniformly calibrated. Then, the real-time operation data of the power system is connected. The electromagnetic field distribution is dynamically corrected according to the actual operating conditions of the equipment. Finally, the electromagnetic field distribution data, the outline of the spatial equipment area, and the information related to the spatial uncertainty of the electromagnetic field are stably output. The dynamic correction includes adjusting the electromagnetic field distribution based on real-time changes in power grid flow and equipment on / off status, matching the corresponding correction logic to eliminate field distribution errors caused by changes in equipment operating conditions.

3. The intelligent commissioning system for power equipment based on mixed reality according to claim 2, characterized in that: The device state field estimation module includes: Based on the spatial equipment area outline, the dedicated spatial range of each power equipment is delineated, and a state evolution relationship that can reflect the operating law of the equipment is established. The real-time operating status of the equipment is judged by the monitoring data collected by the field-deployed sensing equipment. At the same time, the judged equipment operating status information is transmitted back to the space electromagnetic field reconstruction module. Based on the actual operating conditions of the equipment, the relevant parameters of the electromagnetic field medium properties are changed, thereby completing the reverse adaptation adjustment of the electromagnetic field related parameters.

4. The intelligent commissioning system for power equipment based on mixed reality according to claim 1, characterized in that: The delineation of hazardous work areas, forming a unified safety assessment index, includes: Based on electromagnetic field distribution data, the global electromagnetic potential energy and potential energy gradient data are obtained through calculation. The obtained data is smoothed and noise-reduced to eliminate data deviations caused by instantaneous fluctuations in the working conditions. Combined with on-site operation safety regulations, different levels of work hazard areas are divided according to the potential energy values. Then, the regional hazard level information is integrated to form standardized safety evaluation indicators.

5. The intelligent commissioning system for power equipment based on mixed reality according to claim 1, characterized in that: The process of completing data standardization and fusion calculations, and outputting the overall system operating status results, includes: First, various heterogeneous monitoring data, including electromagnetic field characteristics, equipment operating status characteristics, power grid electrical quantity data, and equipment transient operating characteristics, are extracted and categorized. The raw data of different dimensions and magnitudes are then standardized and normalized to eliminate the impact of differences in data dimensions. Next, the processed and normalized data are fused using a Bayesian filtering fusion algorithm to integrate multi-dimensional operating information and complete a unified simulation of the system's overall operating status. Abnormal monitoring data is identified and fault sensing information is alarmed simultaneously. Finally, the overall operating status of the power grid and equipment, including status confidence information, is output.

6. The intelligent commissioning system for power equipment based on mixed reality according to claim 5, characterized in that: The simplified system state dimension, which divides the adjustable action range of the equipment and determines the optimal debugging action for the equipment, includes: The overall operating status of the equipment is simplified by reducing its dimensions, and the preset commissioning targets are synchronously converted into target states under the corresponding dimensions. The adjustable action range of various power equipment is defined by combining the equipment's own adjustment authority and safety constraints, and a comprehensive evaluation system is constructed that takes into account both the effect of achieving the commissioning target and the equipment adjustment cost. A heuristic greedy search method is used to screen and compare within the compliant action range, and the action screening and verification are completed by combining electrical operation limits and high-risk operation area restrictions. Finally, the optimal equipment commissioning action is selected.

7. The intelligent commissioning system for power equipment based on mixed reality according to claim 1, characterized in that: The process of determining model suitability and autonomously correcting model parameters, and re-formulating debugging strategies based on actual working conditions to form a closed-loop control includes: The selected optimal commissioning action is issued and executed, and the actual operation feedback data of the equipment is obtained. After the actual operation data is converted into a unified format and dimension, the difference is compared with the predicted operation state in advance by the system. The degree of state deviation is judged by the stability judgment rules, thereby determining the matching and adaptation degree of the operation simulation model. When the deviation exceeds the allowable range, the internal parameters of the model are automatically adjusted by gradient optimization. The commissioning strategy is re-planned based on the actual on-site operation conditions, and the entire process of action execution, state feedback, model correction and strategy reconfiguration is continuously repeated to achieve closed-loop stable control of the entire intelligent commissioning process of power equipment.

8. The intelligent commissioning system for power equipment based on mixed reality according to claim 1, characterized in that: The method of using federated learning to iteratively optimize the global control model includes: Each local deployment site only collects and retains the model parameter update information generated during the debugging process, without uploading the original on-site operating data. Instead, it uploads the parameter update information to the cloud server. The cloud server aggregates the parameter update content uploaded by multiple sites, performs mean aggregation to obtain the basic global model, and then introduces a stochastic gradient dynamics mechanism to optimize and correct the global model parameters, completing the overall performance iteration and upgrade of the global control model. The cloud server then distributes the optimized global model to each local site. Each site, based on its own actual on-site debugging conditions and local historical operating data, makes targeted adjustments and adaptations to the received global model to obtain a dedicated debugging and control model adapted to its local scenario.

9. A method for intelligent commissioning of power equipment based on mixed reality, characterized in that: The method is used in any one of claims 1 to 8 for the intelligent commissioning system of power equipment based on mixed reality, comprising: S1. Spatial electromagnetic field reconstruction: Collect multi-source electromagnetic signals and spatial geometric data on site, combine on-site electromagnetic interference factors to construct an electromagnetic field solution model to complete the reconstruction of the three-dimensional spatial electromagnetic field distribution in the whole domain, and combine real-time operation data of the power system to dynamically and linearly correct the electromagnetic field distribution obtained by reconstruction. S2. Equipment State Field Estimation: Based on the electromagnetic field reconstruction results, the corresponding spatial regions of each power equipment are divided. The lumped parameter model is used to establish the internal state evolution relationship of the equipment. Combined with the equipment surface temperature data collected by infrared thermography, the equipment operating state is estimated. At the same time, the equipment state data is fed back to the electromagnetic field reconstruction link to realize the bidirectional coupling and linkage of multiple physical fields. S3, Electric Field Potential Energy Mapping: Based on the corrected global electromagnetic field data, spatial electromagnetic potential energy and potential energy gradient information are calculated. After smoothing and noise reduction of the gradient data, regional safety assessment indicators are generated. Based on the safety indicators, high-risk areas for power equipment operation are divided and safety constraints are formed. S4. Multi-source data assimilation: Integrates electromagnetic field data, equipment status data, real-time electrical operation data of power system and transient fault characteristic data, performs unified and standardized processing on various heterogeneous monitoring data, and completes multi-source data assimilation processing by Bayesian filtering fusion method, outputting the fused overall system operation status information. S5. Dynamic debugging strategy generation: The assimilated system operating state is simplified and dimensionality is reduced. The adjustable action space of the equipment is defined in combination with the preset debugging target. A comprehensive cost evaluation system including state deviation and adjustment cost is built based on the heuristic greedy search algorithm. At the same time, regional safety constraints and electrical parameter limit constraints are incorporated to select the optimal power equipment debugging action. S6. Feedback Verification and Adaptive Adjustment: The determined optimal debugging actions are sent to the power equipment to complete parameter control, the actual operating response data of the equipment is read back and a unified dimension mapping is completed, the deviation between the actual operating state and the predicted operating state of the equipment is compared by constructing a stability judgment function, the system state inference model is adaptively corrected online based on the deviation results, and the debugging strategy planning is re-completed based on the actual operating state. S7. Distributed online learning optimization: It aggregates the running data and model optimization data generated during the intelligent debugging process at multiple sites, and adopts a federated learning mode to complete the global model aggregation optimization in the cloud by uploading only the parameter update. It also improves the global model performance by combining the stochastic gradient dynamics algorithm. Each deployment site receives the optimized global model and completes personalized fine-tuning by combining it with local running data, iteratively optimizing the overall debugging method and running performance.

Images