Intelligent Control Method and System for Power Inspection Robots Based on Digital Twin

By calculating control residuals and twin residuals, and combining environmental safety levels and safety constraints, target control parameters are generated, solving the problem of unstable matching between control parameters and actual dynamic characteristics of power inspection robots, and realizing safe adaptive control of power inspection robots.

CN123113349APending Publication Date: 2026-11-17HAINAN VOCATIONAL COLLEGE OF IND & COMMERCE
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611335551.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-31
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

The existing control parameters of power inspection robots are difficult to continuously reflect the changes in the actual dynamic characteristics of the robot, resulting in reduced stability in the matching between the control parameters and the robot's current operating characteristics, which affects the continuity and safety of the inspection task.

Method used

By acquiring the operational status of the entity and the predicted status of the digital twin model, the control residuals and twin residuals are calculated, candidate control parameters are generated and twin shadow calculations are performed. The target control parameters are then selected by combining the environmental safety level and safety constraints, thereby achieving incremental deployment and entity verification.

Benefits of technology

It effectively distinguishes between control parameter mismatch and external interference, ensuring the reliability and safety of parameter tuning, balancing motion control accuracy and inspection data acquisition quality, and achieving adaptive matching between control parameters and the dynamic characteristics of the entity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123113349A_ABST
    Figure CN123113349A_ABST
Patent Text Reader

Abstract

This invention discloses an intelligent control method and system for a power inspection robot based on digital twins, belonging to the field of digital twin technology. The invention synchronously collects data by running a digital twin model alongside the physical power inspection robot. It calculates control residuals and twin residuals separately for each motion stage. Parameter mismatch is confirmed only when both continuously exceed limits under reliable data. The mismatch type is identified based on residual fingerprints to limit adjustable parameters. Candidate parameters are then generated and tested in the twin model, and target parameters are selected. The fusion ratio is calculated based on the environmental safety level, and the parameters are progressively issued and corresponding decisions are made. Finally, the twin model is updated to form a closed-loop adaptive control. This achieves more reliable identification of mismatches between control parameters and the robot's actual operating characteristics, ensuring the stability of robot motion control and solving the problem of reduced stability in matching control parameters with the robot's current operating characteristics in existing technologies.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, and in particular to an intelligent control method and system for a power inspection robot based on digital twins. Background Technology

[0002] With the continuous expansion of power system operation and the gradual promotion of unmanned and minimally staffed operation and maintenance models, the use of inspection robots for status monitoring of power equipment has become an important technical means to improve inspection efficiency and reduce the risks of human operation. Power inspection robots typically need to travel along a preset route to a designated inspection location and acquire equipment operating information through visible light imaging, infrared thermography, and sound acquisition. To achieve remote monitoring, status mapping, and decision support during the inspection process, existing technologies also apply digital twin technology to power inspection scenarios. By establishing virtual models of the site environment, power equipment, and inspection robots, the robot's position, operating status, and inspection results can be synchronously displayed and managed in virtual space.

[0003] Existing intelligent power robot inspection systems typically generate inspection tasks and travel paths based on a scene map, inspection points, and the robot's initial state. The robot controller then controls the robot's linear velocity, angular velocity, drive wheel speed, acceleration / deceleration, and stopping processes according to preset motion control parameters. During robot operation, encoders, inertial sensors, positioning devices, and motor status detection devices continuously collect motion data. A digital twin platform predicts or presents the robot's operating state based on the robot model and control commands. Once the robot reaches the inspection location, the system controls it to stop moving or perform minor position adjustments. After the robot stabilizes, inspection data collection is completed, and the results are correlated with the corresponding power equipment.

[0004] However, during long-term operation, the load state, wheel-ground contact state, drive mechanism performance, and braking response characteristics of inspection robots may change, causing the robot's actual motion characteristics to gradually deviate from the initial calibration state. Existing technologies typically use control parameters determined according to an initial model or preset working conditions. While some systems can switch control parameters based on road type or operating mode, parameter adjustments still primarily rely on pre-set working condition categories, making it difficult to continuously reflect changes in the robot's actual dynamic characteristics. If existing technologies directly adjust control parameters based solely on trajectory or speed errors, it is difficult to accurately determine whether motion deviations are indeed caused by control parameter mismatch, easily leading to unnecessary or inaccurate parameter adjustments. If the adjusted parameters are directly applied to a specific robot without effective verification, it may further increase control fluctuations, affecting the continuity and safety of inspection tasks. Summary of the Invention

[0005] To address the technical problem of reduced stability in existing technologies due to mismatch between control parameters and the robot's current operating characteristics, this invention provides an intelligent control method and system for a power inspection robot based on digital twins. The technical solution is as follows:

[0006] On the one hand, a digital twin-based intelligent control method for power inspection robots is provided, which includes:

[0007] The system acquires inspection task instructions and obtains the physical operating status of the specified robot and the twin prediction status output by the digital twin model based on the same time reference. It extracts the current motion stage features of the specified robot based on the control instructions in the inspection task instructions, and calculates the control residual and twin residual by combining the physical operating status and the twin prediction status. When the control residual and twin residual meet the preset parameter mismatch conditions, candidate control parameters are generated and input into the digital twin model for trial calculation, outputting the predicted trajectory. The predicted trajectory is filtered based on preset safety constraints, and the target control parameters are determined from the candidate control parameters. The system acquires the environmental safety level of the specified robot's environment, calculates the fusion ratio based on the environmental safety level, and fuses the current operating control parameters and the target control parameters according to the fusion ratio to generate hybrid control parameters. The hybrid control parameters are sent to the specified robot for physical verification operation, and the system decides whether to solidify the target control parameters or revert to the previous control parameters based on the acquired verification feedback data.

[0008] On the other hand, a digital twin-based intelligent control system for power inspection robots is provided, including:

[0009] The inspection data synchronization acquisition module is used to acquire inspection task instructions and, based on the same time reference, acquire the physical running status of the specified robot and the twin prediction status output by the digital twin model. The dual residual calculation module is used to extract the current motion stage features of the specified robot according to the control instructions in the inspection task instructions, and calculate the control residual and twin residual respectively by combining the physical running status and the twin prediction status. The twin trial calculation prediction module is used to generate candidate control parameters when the control residual and the twin residual meet the preset parameter mismatch conditions, and input the candidate control parameters into the digital twin model for trial calculation and output the predicted trajectory. The target parameter determination module is used to filter the predicted trajectory based on preset safety constraints and determine the target control parameters from the candidate control parameters. The progressive deployment verification module is used to acquire the environmental safety level of the environment in which the specified robot is located, calculate the fusion ratio based on the environmental safety level, and fuse the current running control parameters and the target control parameters according to the fusion ratio to generate hybrid control parameters. The hybrid control parameters are sent to the specified robot for physical verification operation, and the target control parameters are fixed or rolled back to the control parameters before the update are determined based on the acquired verification feedback data.

[0010] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following:

[0011] This technical solution provides an intelligent control method and system for power inspection robots based on digital twins. A digital twin model runs in the background of the physical robot, synchronously acquiring task instructions, physical operating status, and twin prediction status at the same time reference. This provides time-aligned, source-based comparison data for parameter identification, avoiding false deviations caused by time misalignment. Furthermore, control residuals and twin residuals are calculated separately for each motion stage, and mismatch types are identified under confidence gating. This effectively distinguishes between control parameter mismatch and external interference, ensuring the reliability of parameter tuning triggers. Subsequently, candidate parameters are generated within the mismatch type limit, and target parameters are determined through twin shadow calculation, safety constraints, and joint screening based on inspection quality, thus balancing motion control accuracy and inspection data acquisition quality. Finally, deployment is progressively based on environmental safety levels, and physical verification determines whether to solidify or roll back. This achieves safe adaptive matching of control parameters and physical dynamic characteristics without altering the original controller structure.

[0012] The primary distinguishing technical feature of this solution lies in its dual residual collaborative judgment and phased residual fingerprint recognition. It simultaneously constructs control residuals and twin residuals, confirming mismatch only when both synchronously and continuously exceed limits and change in the same direction across multiple motion stages. This differs from existing technologies that trigger parameter tuning based solely on a single deviation, effectively differentiating control parameter mismatch from positioning anomalies and external disturbances. Furthermore, it limits the adjustable parameter range based on the mismatch type identified by residual fingerprint recognition, thereby avoiding parameters with incorrect physical meanings due to mutual compensation among multiple parameters and helping to reduce the probability of mis-tuning.

[0013] Another distinguishing technical feature of this solution lies in its shadow simulation and environmentally adaptive, progressive deployment. Candidate parameters are initially predicted only for subsequent local path segments in twin shadow mode, reducing computational load while completing a safe pre-simulation without affecting actual operation. Target parameters are then determined through a combination of hard screening based on safety constraints and a joint evaluation of inspection data quality loss, ensuring that the selected parameters simultaneously meet both safety and data acquisition quality requirements. Finally, the fusion ratio is dynamically calculated and gradually distributed based on the environmental safety level and parameter variation range, thus balancing parameter tuning efficiency and operational safety even in high-risk environments such as those with energized equipment. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1A flowchart illustrating the intelligent control method for a power inspection robot based on digital twins provided in this application embodiment;

[0016] Figure 2 A schematic diagram of the structure of the intelligent control system for power inspection robots based on digital twins provided in this application embodiment. Detailed Implementation

[0017] Embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present disclosure are shown in the drawings, it should be understood that embodiments of the present disclosure may be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure.

[0018] It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure. In the description of the embodiments of this disclosure, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "this embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects.

[0019] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0020] See Figure 1 The flowchart illustrates an intelligent control method for a power inspection robot based on digital twins, provided in this application embodiment. This method is applied to a control terminal that communicates with a physical inspection robot and includes the following steps:

[0021] Before performing the following steps, this embodiment pre-establishes a digital twin model corresponding to the physical inspection robot. This digital twin model includes at least the robot's geometric model, kinematic model, dynamic model, left and right drive wheel models, motor response model, braking response model, robot load parameters, wheel-ground adhesion or slip parameters, vehicle vibration response model, and inspection road and obstacle model. Its initial parameters are obtained through factory calibration data and offline identification. A set of verified initial stability control parameters and basic parameter sets corresponding to standard load, heavy load, smooth road surface, low-adhesion road surface, slope, and precise parking conditions are pre-stored. Among them, the preset mismatch type feature library is pre-established by collecting historical operating data or simulating and injecting various typical mismatch conditions into the digital twin model, extracting and labeling their residual fingerprint features; the preset inspection quality assessment algorithm is pre-obtained by regression training on historical vehicle vibration, parking error and corresponding inspection image clarity data; each preset threshold is pre-calibrated and determined by the initial stability parameter library and the basic parameter set of working conditions in combination with calibration data.

[0022] S101. Obtain the inspection task instruction, and obtain the physical running status of the specified robot and the twin prediction status output by the digital twin model based on the same time reference.

[0023] Specifically, this step provides the data input foundation for the entire method, establishing the conditions for comparison between the physical robot and the digital twin model. The inspection task instructions include the target inspection point, path point sequence, target linear velocity, target angular velocity, parking position, and arrival posture. The physical robot's operating status is collected and uploaded in real time by corresponding sensors on the robot, including actual linear velocity, actual angular velocity, actual position and heading angle, actual rotational speed of the left and right wheels, current of the left and right motors, and acceleration and angular velocity collected by the inertial measurement unit. The twin prediction status is the predicted linear velocity, predicted angular velocity, predicted position and heading angle, predicted wheel speed, and predicted current output by the digital twin model after receiving the same control instructions as the physical robot. This step acquires the above data based on the same time reference, aligning the three types of data at the same moment, thus ensuring that the subsequent residuals reflect differences in control characteristics rather than spurious deviations caused by time misalignment.

[0024] S102. Extract the current motion stage features of the specified robot based on the control instructions in the inspection task instructions, and calculate the control residual and twin residual respectively by combining the physical running state and the twin prediction state.

[0025] Specifically, this step more accurately characterizes control parameter mismatch through motion phase segmentation and parallel computation of dual residuals. On the one hand, extracting features of the current motion phase avoids misjudging normal characteristics of different phases as mismatches, such as the inherent speed build-up lag in the acceleration phase. On the other hand, control residuals characterize the difference between the control objective and the entity's execution result, while twin residuals characterize the difference between the digital twin prediction result and the entity's execution result. Since a single control residual cannot distinguish the source of deviation, the introduction of twin residuals ensures that only when the two types of residuals appear simultaneously and have a consistent trend are they more likely to correspond to a true physical characteristic mismatch, thus effectively distinguishing parameter mismatch from external disturbances.

[0026] S103. Extract the desired velocity change data and target path curvature from the control command.

[0027] Furthermore, this step determines the judgment quantities for motion phase division. Its function is to extract the expected speed change data and target path curvature directly from the control command without relying on additional calibration actions: the former reflects whether the target speed is increasing, maintaining or decreasing, and the latter reflects the curvature of the current path segment, thus providing input for subsequent determination of motion phase based on speed trend and path curvature.

[0028] When the expected velocity change data indicates that the target velocity is increasing and the curvature of the target path is not greater than a preset curvature threshold, the current motion phase characteristics representing the acceleration phase are generated.

[0029] Specifically, this step identifies the speed increase process under straight or near-straight-line operating conditions, thereby distinguishing the acceleration process from the turning process. The acceleration phase is mainly used to identify load changes, drive gain changes, and motor response speed; separating it separately avoids misjudging speed build-up lag as mismatch. The smaller the preset curvature threshold, the stricter the requirements for straightness.

[0030] When the curvature of the target path is greater than a preset curvature threshold and the expected speed change data does not indicate a decrease in the target speed, the current motion phase feature representing the turning phase is generated.

[0031] Specifically, the condition of not indicating a decrease in target speed is added here to distinguish it from the deceleration phase, avoiding conflicts caused by the deceleration section before the turn being simultaneously judged as the turning phase. The turning phase is mainly used to identify turning overshoot, inconsistent output between the left and right wheels, changes in moment of inertia, and mismatch in angular velocity control gain.

[0032] When the expected velocity change data indicates a decrease in the target velocity, the current motion phase characteristics representing the deceleration phase are generated.

[0033] Specifically, this step identifies the deceleration process of the robot before entering inspection points, obstacles, or turns. The deceleration phase is mainly used to identify braking response delays, insufficient deceleration, and increased braking distance, which helps to determine whether there is a mismatch in braking response parameters during multiple consecutive deceleration processes.

[0034] When a specified robot simultaneously meets multiple motion stage judgment conditions, the current motion stage characteristics are determined according to the preset stage priority.

[0035] Specifically, this step ensures the uniqueness of stage identification, resolving stage assignment conflicts when multiple judgment conditions are simultaneously met. In actual inspections, the robot may simultaneously be in sections with overlapping working conditions, such as a path curvature and target speed decrease before a turn, where the conditions for the turning stage and deceleration stage may be met at the same time. By introducing a preset stage priority, the current motion stage characteristics are uniquely determined according to priority when multiple conditions are met simultaneously. For example, setting the priority of the deceleration stage higher than that of the turning stage ensures that the deceleration section before the turn is preferentially identified as the deceleration stage, thereby ensuring that the subsequent residual statistics use matching statistical rules. Alternatively, when multiple conditions are met simultaneously, the current motion stage characteristics can also be determined by assigning membership degrees to each motion stage and taking the one with the largest membership degree.

[0036] S104. Obtain the hardware operating status of the positioning sensor and communication module to calculate the overall reliability.

[0037] Furthermore, this step establishes a data reliability gating for residual calculation, a crucial design to prevent parameter mistuning. Its purpose is to assess data reliability before residual calculation. Specifically, the hardware operating status of the positioning sensor includes the confidence level of LiDAR or visual positioning and whether discontinuous jumps occur in the positioning results. The hardware operating status of the communication module includes data packet loss and time synchronization errors. These statuses are read in real-time by the self-test interface of the corresponding module. The overall reliability is obtained by weighted summation of the components, expressed as:

[0038] ;

[0039] In the formula, To ensure overall reliability, the values ​​are normalized to [0, 1]. To determine the reliability of the positioning information, it reflects the degree of reliability of the positioning sensor output; Communication reliability reflects the data integrity and time synchronization quality of the communication link; , For the corresponding weight coefficients, and Alternatively, the overall credibility can be further incorporated by including components such as perceived credibility and twin model credibility.

[0040] When the overall confidence level exceeds the preset confidence level threshold, the residual calculation process is triggered.

[0041] Specifically, when the overall credibility exceeds a preset credibility threshold, it indicates that the current positioning and communication data is reliable, and the residual calculation process is triggered to continue subsequent judgments; otherwise, it indicates that there are problems such as positioning jumps, communication packet loss, or excessive time synchronization errors, and the residual calculation and parameter identification are suspended, and the current data window does not participate in parameter tuning. The higher the preset credibility threshold, the stricter the requirements for data quality and the less likely it is to trigger parameter tuning erroneously, effectively preventing positioning anomalies and communication problems from being mistaken for control parameter mismatches.

[0042] Extract the actual motion execution data from the entity's operating state and the expected target data from the inspection task instructions, calculate the deviation between the two, and obtain the control residual.

[0043] Specifically, this step clarifies the data source and calculation method for the control residual. Actual motion execution data is taken from the entity's operational status, while the desired target data is taken from the inspection task instructions. The deviation between the two yields the control residual, reflecting the difference between the control target and the entity's execution result. Its residual vector can be constructed as follows:

[0044] ;

[0045] In the formula, To control the residual vector; For the desired linear velocity, This is the actual linear velocity; For the desired angular velocity, This is the actual angular velocity; This refers to lateral trajectory deviation; This refers to the deviation in heading angle; This is due to a deviation in the parking position.

[0046] Extract the actual motion execution data from the entity's running state and the predicted execution data from the twin prediction state, calculate the deviation between the two, and obtain the twin residual.

[0047] Specifically, the twin residual is obtained by comparing the actual motion execution data with the predicted execution data in the twin prediction state. The difference between the two is calculated, which reflects the difference between the digital twin prediction result and the entity execution result. Its residual vector can be constructed as follows:

[0048] ;

[0049] In the formula, For twin residual vectors; , These are the predicted linear velocity and angular velocity from twins, respectively. , Predict the location coordinates for twins. , These are the coordinates of the entity's position. To predict acceleration for twins, This represents the entity's acceleration. This step is executed in parallel with the previous control residual calculation step, and together they constitute a double residual.

[0050] According to the preset statistical rules corresponding to the characteristics of the current motion stage, the control residuals and twin residuals are statistically analyzed within the target time window to obtain the statistical characteristics of the stage residuals.

[0051] Specifically, this step enables phased mismatch judgment, transforming instantaneous residuals into statistically significant phase residual characteristics, thus preventing single-moment residual fluctuations from directly triggering parameter tuning. Different motion phases correspond to different preset statistical rules, with statistics performed within the target time window. Phase residual statistical characteristics include residual mean, residual peak value, residual change rate, residual duration, symmetry of left and right wheel residuals, and consistency of the change direction between the control residual and the twin residual. The consistency of the change direction is used to determine whether the two types of residuals appear synchronously and have a consistent trend; only when the directions are consistent is it more likely to correspond to a true mismatch. This step effectively suppresses the impact of instantaneous disturbances such as occasional road impacts. Alternatively, the target time window can be a fixed-duration sliding time window or an event-triggered time window corresponding to the start and end of the motion phase.

[0052] S105. When the control residual and the twin residual meet the preset parameter mismatch conditions, candidate control parameters are generated, and the candidate control parameters are input into the digital twin model for trial calculation, and the predicted trajectory is output.

[0053] Specifically, the above steps complete the transition from identifying mismatches to generating and rehearsing solutions, which is crucial for ensuring the safety of parameter tuning. On the one hand, candidate control parameters are only generated when the control residuals and twin residuals both meet the preset parameter mismatch conditions, avoiding unnecessary changes caused by a single fluctuation. On the other hand, the generated candidate control parameters are not directly issued to the entity, but are first input into the digital twin model for trial calculation, outputting the predicted trajectory, thereby completing parameter rehearsal without affecting the normal inspection of the entity.

[0054] When the multi-dimensional stage residual statistical characteristics continuously exceed the preset residual threshold and reach the preset duration, it is confirmed that the preset parameter mismatch condition is met.

[0055] Furthermore, this step uses persistence and multi-dimensionality as mismatch confirmation conditions to avoid accidental disturbances triggering parameter tuning. Its purpose is to avoid relying on a single dimension or a single moment exceeding the limit, but rather requires that the statistical characteristics of the stage residuals in multiple dimensions simultaneously and continuously exceed the preset residual thresholds for a preset duration. This confirmation condition can be formally expressed as:

[0056] ;

[0057] In the formula, For the first Statistical characteristics of residuals in the dimensional stage To correspond to the preset residual threshold, The set of feature dimensions used in the judgment; For the duration of exceeding the limit, The preset duration; To assess overall credibility, A preset credibility threshold is set. This represents a logical AND operation. A higher preset residual threshold results in more conservative parameter tuning, while a longer preset duration effectively filters out instantaneous impacts. This constraint effectively prevents parameter updates from being triggered by a single road impact.

[0058] The statistical features of stage residuals within different time windows are combined to generate residual fingerprint features.

[0059] Specifically, this step is the core of identifying mismatch types. Its function is to aggregate the residual statistical features scattered across different time windows and different motion stages into residual fingerprint features that can characterize the mismatch pattern. The residual within a single time window can only reflect local deviations, but by combining the features within the time windows corresponding to stages such as acceleration, constant speed, turning, and deceleration in a certain structure, the resulting residual fingerprint features can depict the overall performance pattern of mismatch under different operating conditions. For example, the load-increasing type is characterized by a decrease in actual acceleration and a synchronous increase in the current of the left and right motors during the acceleration stage, while the single-sided drive attenuation type is characterized by asymmetric response of the left and right wheels and continuous deviation to the same side during straight-line driving.

[0060] It should be added that the residual fingerprint feature can be constructed as a feature vector composed of the statistical features of the stage residuals within the corresponding time windows of each motion stage. As an example, for the braking hysteresis type, the feature vector can be composed of components such as the mean of the linear velocity residual during the deceleration stage, the mean of the parking position deviation, and the mean of the difference between the physical acceleration and the twin predicted acceleration in the twin residuals; for the load increase type, the feature vector can be composed of components such as the actual acceleration residual during the acceleration stage, the increase in current of the left and right motors, and the symmetry of the left and right wheel speeds. During comparison, the feature distance is calculated between the current residual fingerprint feature vector and the feature patterns of each type in the preset mismatch type feature library, such as calculating the Euclidean distance or Mahalanobis distance. The type with the smallest feature distance that is less than the preset matching threshold is taken as the parameter mismatch type identified.

[0061] The residual fingerprint features are compared with a preset mismatch type feature library to identify and output the corresponding parameter mismatch type.

[0062] Specifically, the current residual fingerprint features are compared one by one with each type of feature in the preset mismatch type feature library. The matching type is taken as the output parameter mismatch type. The parameter mismatch types include load increase type, low adhesion or slip type, single-sided drive attenuation type, rotational response mismatch type, braking hysteresis type, and positioning anomaly type. It should be noted that for the case where the identification result is a positioning anomaly type, the encoder, inertial measurement unit, and motor response are basically normal, and only the external positioning result changes abruptly. In this case, no control parameter correction is performed, but the process is switched to positioning recovery or repositioning, thereby avoiding mistaking positioning anomaly for control parameter mismatch. Alternatively, the comparison between the residual fingerprint features and the mismatch type feature library can use nearest neighbor matching based on feature distance, or a pre-trained classification model can be used for identification.

[0063] Based on the parameter mismatch type, the corresponding adjustable parameter range constraint data is extracted from the initial stable parameter library, and multiple sets of candidate control parameters are generated within the adjustable parameter range constraint data.

[0064] Specifically, by establishing a correspondence between parameter mismatch types and adjustable parameters—for example, increasing load type corresponds to adjusting acceleration gain, velocity feedforward, and braking advance; low adhesion type corresponds to adjusting the upper limit of acceleration, torque change rate, and turning speed; single-side drive attenuation type corresponds to adjusting the compensation coefficients of the left and right wheels; rotational response mismatch type corresponds to adjusting angular velocity gain and steering feedforward; and braking hysteresis type corresponds to adjusting deceleration start distance, deceleration gain, and stopping gain—the generated candidate control parameters are all within a preset safety range, with single changes not exceeding the allowable range, and parameters unrelated to the current mismatch type are not adjusted. Furthermore, the original controller structure and upper-level control command format are not altered, thus limiting the search space to a finite and safe range and reducing the computational load of subsequent trial calculations. Alternatively, the adjustable parameter range constraint data can be extracted from the initial stable parameter library or jointly determined by combining the basic parameter sets corresponding to different operating conditions.

[0065] S106. Based on preset safety constraints, the predicted trajectory is screened, and the target control parameter is determined from the candidate control parameters.

[0066] Specifically, when there are no candidate control parameters that satisfy all safety constraints, the current operating control parameters are maintained and the system switches to deceleration operation, rather than forcibly selecting suboptimal parameters.

[0067] Switch the digital twin model to shadow trial mode and input the current pose state data and subsequent local path segment data of the specified robot into the digital twin model.

[0068] Furthermore, the so-called shadow simulation mode refers to a scenario where the digital twin model and the physical robot receive the same current pose state data and subsequent local path segment data. However, the output of candidate control parameters in this mode is only used for simulation prediction and is not directly sent to the physical robot. Here, only the subsequent local path segment data is input, rather than the entire inspection path, significantly reducing the computational load compared to resimulating the entire inspection task, thereby improving the real-time performance of the simulation. Alternatively, the length of the subsequent local path segment can be set to a path segment corresponding to a fixed distance or a fixed time window.

[0069] For each set of candidate control parameters, simulations are performed in shadow trial mode, and the predicted trajectory is output, including the maximum lateral deviation, the predicted collision distance, and the predicted motor current.

[0070] Specifically, the predicted trajectory includes at least the maximum lateral deviation, the predicted collision distance, and the predicted motor current. It may also include the maximum heading angle deviation, turning overshoot, parking position error, remaining speed upon reaching the inspection point, the speed difference between the left and right wheels, vehicle vibration, and expected energy consumption. Among these, the maximum lateral deviation represents the trajectory tracking accuracy, the predicted collision distance represents the safety margin for obstacles and electrical equipment, and the predicted motor current represents the load on the drive system.

[0071] The predicted trajectory is compared with each of the boundary thresholds in the preset safety constraints.

[0072] Specifically, each predicted quantity is compared with the corresponding boundary threshold in the preset safety constraints. Each boundary threshold includes the maximum allowable trajectory error, the minimum safe distance from obstacles and energized equipment, the upper limit of the allowable motor current, the mechanical limits of wheel speed and acceleration, and the allowable parking area. Among these, the stricter the boundary threshold setting, the fewer candidate parameters will pass the screening, and the greater the safety margin.

[0073] Candidate control parameters that fail the comparison are removed, and those that meet the various boundary thresholds are retained as preliminary candidate parameters.

[0074] Specifically, this step uses safety constraints as a hard condition to filter candidate control parameters. Candidate control parameters must meet all boundary thresholds to be retained; any parameter that fails to meet any threshold is eliminated. This eliminates candidate parameters with potential collision risks, current overruns, boundary violations, or excessive trajectory errors at the source. This screening method, which uses all safety constraints as hard thresholds, differs from conventional parameter selection based solely on performance indicators. It ensures that every set of parameters entering subsequent processes is safe and feasible, thereby improving the safety assurance capability of the parameter tuning process for high-risk objects such as energized equipment.

[0075] The predicted vehicle vibration data and predicted parking position error are obtained by reading the preliminary candidate parameters through a digital twin model.

[0076] Furthermore, this step extracts the necessary inputs for the inspection data quality assessment, aiming to further evaluate the quality of inspection data collection on a safe and feasible basis. Among them, the predicted vehicle vibration data reflects the degree of vehicle shaking during robot operation and parking, and the predicted parking position error reflects the positioning accuracy of the robot when it stops at the inspection point. Both are predicted by the digital twin model in the shadow trial calculation for each initially screened candidate parameter.

[0077] The predicted vehicle vibration data and the predicted parking position error are input into the preset inspection quality assessment algorithm to calculate the predicted inspection data quality loss data.

[0078] Specifically, a pre-defined inspection quality assessment algorithm maps physical quantities at the motion control level to data acquisition quality indicators at the inspection task level. Since vehicle vibration and parking position errors directly affect acquisition quality—for example, excessive vibration can lead to image blurring, and excessive parking deviation can cause the target device to deviate from the camera's field of view—a mapping relationship can be established from control energy to inspection data quality loss, expressed as:

[0079] ;

[0080] In the formula, This is a predicted data point for the quality loss of inspection data; the larger the value, the worse the data collection quality. To predict vehicle body vibration; To predict the settling time, which is the time required for the vehicle to stabilize after parking; This is the error in parking position; This refers to the vehicle's orientation error. This is the mapping function corresponding to the preset inspection quality assessment algorithm; to These are non-negative weighting coefficients, predetermined by historical data regression training.

[0081] The maximum lateral deviation in the predicted trajectory and the predicted quality loss of inspection data are weighted and calculated to output comprehensive evaluation data.

[0082] Specifically, this step integrates control performance indicators and inspection quality indicators into unified comprehensive evaluation data, achieving multi-objective joint evaluation. The comprehensive evaluation data is represented as follows:

[0083] ;

[0084] In the formula, To comprehensively evaluate the data, the smaller the value, the better the overall performance; The prediction trajectory error corresponds to the maximum lateral deviation in the predicted trajectory; For predicting the quality loss of inspection data; , The corresponding weighting coefficients are used to adjust the relative importance of trajectory accuracy and inspection quality in the comprehensive evaluation. Alternatively, the comprehensive evaluation data can be further weighted by incorporating predicted parking errors, predicted vibrations, estimated energy consumption, and changes relative to current operating control parameters.

[0085] Extract the candidate parameter with the smallest value from the comprehensive evaluation data and mark it as the target control parameter.

[0086] Specifically, since the comprehensive evaluation data is obtained by weighting loss indicators, the smaller the value, the better the overall performance. Therefore, the preliminary candidate parameter with the smallest comprehensive evaluation data is extracted and marked as the target control parameter. This constitutes a two-level parameter determination mechanism: first, hard filtering by safety constraints, and then selection by comprehensive evaluation. This ensures that the target control parameter satisfies all safety constraints while achieving optimal overall performance in terms of trajectory accuracy and inspection data acquisition quality. Alternatively, when multiple sets of candidate parameters with similar comprehensive evaluation data exist, the one with the smallest change relative to the current operating control parameter can be further selected to reduce the parameter variation.

[0087] S107. Obtain the environmental safety level of the environment where the specified robot is located, calculate the fusion ratio based on the environmental safety level, and fuse the current operating control parameters and target control parameters according to the fusion ratio to generate hybrid control parameters.

[0088] Specifically, the preset rating rules can set threshold levels for the distance to surrounding obstacles and the calculated road smoothness, and take the higher risk level as the environmental safety level. As an example, when the distance to surrounding obstacles is less than a first distance threshold, or the calculated road smoothness indicates that the road surface undulation exceeds a preset undulation threshold, it is rated as a high-risk, non-deployable area; when the distance to surrounding obstacles is greater than a second distance threshold and the road surface undulation is less than the preset undulation threshold, it is rated as a high-safety level within the permitted deployment area; and when it falls between these two, it is rated as a general safety level within the permitted deployment area. Correspondingly, the adjustment factor can take discrete values ​​according to the environmental safety level, for example, a larger value for high-safety levels, a smaller value for general safety levels, and zero for high-risk, non-deployable areas, thus increasing the fusion ratio as the environmental safety level increases.

[0089] Specifically, the deployment and quantity of the robot are linked to the real-time risks of its environment, enabling a gradual deployment of parameters rather than a one-time replacement. Even if the target control parameters have passed trial calculations and safety screening, their actual effectiveness must be determined based on the environmental safety level. The fusion process is represented as follows:

[0090] ;

[0091] In the formula, These are the current operating control parameters; These are the hybrid control parameters generated after this fusion. For target control parameters; The value range for this fusion ratio is [value range missing]. .in, The smaller the value, the closer the mixed control parameters are to the current operating control parameters, and the more conservative the parameter changes. The larger the value, the closer it is to the target control parameter, and the faster the benefits of parameter tuning become apparent.

[0092] Obtain data on the distance to surrounding obstacles and the smoothness of the running road at the current location of the specified robot.

[0093] Furthermore, this step collects basic data characterizing the risks of the current deployment environment. The distances to surrounding obstacles are measured by the robot's LiDAR or visual perception module, reflecting the spatial margin between the robot and obstacles and powered equipment. The road surface smoothness is calculated by the inertial measurement unit in conjunction with the vehicle's vibration response, reflecting the current road surface smoothness. The smaller the distance or the more uneven the road surface, the greater the risk posed by parameter adjustments.

[0094] Specifically, the road surface smoothness measurement data can be represented by the statistics of the vehicle's vertical acceleration collected by the inertial measurement unit over a certain driving distance. For example, the root mean square value of the vertical acceleration over that distance can be taken. The larger the root mean square value, the more uneven the road surface. As an alternative, the vehicle vibration response model can also be used to characterize the road surface smoothness by the degree of deviation between the measured vibration and the model's expected vibration.

[0095] An environmental safety level is generated based on data calculated from the distance to surrounding obstacles and the smoothness of the running road.

[0096] Specifically, multiple environmental parameters are integrated into a unified environmental safety level, providing a tiered basis for determining the subsequent fusion ratio. Data on the distance to surrounding obstacles and the smoothness of the operating road are comprehensively evaluated according to preset rating rules to obtain an environmental safety level characterizing the deployment risk at the current location, such as dividing it into high-risk, non-deployable areas and permitted deployment areas.

[0097] When the environmental security level indicates that the current location is a high-risk, non-deployable area, a freeze update command is output, setting the fusion ratio to zero to suspend the generation and distribution of fusion control parameters.

[0098] Specifically, in high-risk environments, parameter updates are forcibly frozen to avoid parameter adjustments in unsuitable road sections. When in a high-risk, undeployable area, a freeze update command is output, and the fusion ratio is set to zero. As shown in the aforementioned fusion formula, when the fusion ratio is zero, the mixed control parameters are equal to the current operating control parameters, meaning no parameter changes occur. High-risk, undeployable areas correspond to situations such as passing through narrow passages, going up or down slopes, crossing thresholds, being too close to energized equipment, having entered the final precise docking stage, or having unstable communication. It should be noted that the freeze update does not terminate the original inspection task; the robot continues to operate using the current operating control parameters until it enters a suitable deployment section.

[0099] When the environmental security level indicates that the current location is a permitted deployment area, the initial weights are calculated based on the environmental security level and the magnitude of parameter changes in the target control parameters, and the fusion ratio is generated.

[0100] Specifically, this step determines the initial value of the fusion ratio within the permitted deployment area based on environmental risks and parameter variation. The fusion ratio for this step can be:

[0101] ;

[0102] In the formula, This represents the integration ratio for this transaction; The upper limit of the preset fusion ratio; For environmental safety level, For the corresponding adjustment factor, the range of values ​​is... The safer the environment, the larger it will be; The magnitude of the change in the target control parameter relative to the current operating control parameter can be taken as . , Represents the vector norm; The fluctuation range suppression coefficient and This formula shows that the safer the environment, the better. The larger the value, the faster the parameter convergence; the greater the range of parameter variation, the faster the convergence. The smaller the change and the more conservative the single change, the safer the deployment, and the faster the deployment. The more dangerous or larger the change, the slower the deployment. This adaptive incremental deployment achieves the goal of being safer and deploying faster when the change is smaller, and deploying slower when the change is larger or more dangerous.

[0103] S108. Send the mixed control parameters to the designated robot for physical verification operation, and decide whether to solidify the target control parameters or revert to the control parameters before the update based on the obtained verification feedback data.

[0104] Specifically, this step constitutes a closed-loop verification link in the entire parameter tuning process. The actual operating effect of the entity is used as the final criterion to decide whether to retain or discard parameters. This avoids relying solely on twin predictions while ignoring the actual response of the entity, thereby preventing erroneous parameters from being permanently adopted due to accidental data or twin model bias.

[0105] Based on the fusion ratio, the current operating control parameters and the target control parameters are weighted and summed to generate and distribute the hybrid control parameters.

[0106] Furthermore, this step clarifies the generation method of the hybrid control parameters, ensuring that the target control parameters do not completely replace the current operating control parameters at once. Instead, the hybrid control parameters are generated by weighting and summing the two parameters according to the aforementioned fusion ratio before being issued. The weighted summation process is the aforementioned fusion formula. This allows for the gradual introduction of parameter changes within a limited scope.

[0107] Within the preset entity verification time window, recalculate the updated residual verification data generated based on the hybrid control parameters.

[0108] Specifically, the actual effects of parameter changes are collected under actual operating conditions. Within a preset entity verification time window, updated residual verification data generated from operation based on hybrid control parameters are recalculated. This updated residual verification data includes updated control residuals and twin residuals, calculated in the same way as the aforementioned residual calculations, thus ensuring the comparability of residuals before and after the update. A longer preset entity verification time window results in more robust verification results, but slower parameter confirmation.

[0109] When the updated residual verification data indicates that the control residual is reduced and the twin residual is not expanded, and the clarity assessment data of the synchronously acquired entity inspection image is not degraded, a continue fusion instruction is generated to increase the fusion ratio without exceeding the preset fusion ratio limit.

[0110] Specifically, taking the dual improvement of control performance and inspection quality as a condition for continuing deployment, a three-condition joint judgment is adopted, which can be formally expressed as:

[0111] ;

[0112] In the formula, These are the control residuals before and after the update, respectively. , These are the twin residuals before and after the update, respectively. , These are the image clarity assessment data for physical inspection before and after the update, respectively. This represents a logical AND operation. A decrease in control residual indicates improved execution accuracy; no expansion of twin residuals indicates that the entity response has not deviated from the model's expectations, ruling out the possibility of spurious improvements in control residuals due to model inaccuracies; and no degradation in image sharpness assessment data indicates that parameter changes have not impaired acquisition quality. A continued fusion instruction is generated only when all three conditions are met. This increases the fusion ratio, within a preset upper limit, further converging the mixed control parameters towards the target control parameters, thus accelerating convergence. The preset upper limit restricts the maximum magnitude of a single advancement, preventing parameter abrupt changes.

[0113] Furthermore, the image sharpness assessment data for physical inspection is used to characterize the sharpness of images acquired by the robot at inspection points, and it can be calculated from the image gradient energy. Specifically, the gradient energy of the acquired inspection images is calculated as the sharpness assessment value:

[0114] ;

[0115] In the formula, This is data for evaluating the clarity of images during physical inspections; the larger the value, the clearer the image. Image in pixel coordinates The grayscale value at that location; The total number of pixels involved in the calculation; summation iterates through each pixel in the image. The more blurred the image, the smoother the grayscale change between adjacent pixels and the smaller the gradient energy. The smaller, therefore when updated Smaller than before the update Immediately, it is determined that the image sharpness assessment data has deteriorated.

[0116] When the updated residual verification data indicates that the control residual continues to expand or the physical inspection image sharpness assessment data deteriorates, a parameter rollback instruction is generated.

[0117] Specifically, this step triggers a rollback promptly when parameter changes produce adverse effects, and its judgment condition can be formally expressed as:

[0118] ;

[0119] In the formula, This indicates a logical OR. When the control residual continues to increase or the image clarity assessment data of the physical inspection deteriorates, it indicates that the change has failed to improve control performance or has damaged the acquisition quality. In this case, a parameter rollback command is generated to restore the physical robot to the control parameters before the update. The OR relationship is used here to reflect the dual protection of operational safety and inspection quality. Alternatively, the triggering conditions for the parameter rollback command can further include situations such as increased turning oscillation, increased parking error, motor current exceeding the allowable range, or significantly increased vehicle vibration; after rollback, the original inspection task continues to be executed, and deceleration operation is prioritized.

[0120] When no parameter rollback instruction is received within several consecutive control decision cycles, a parameter fixation instruction is generated.

[0121] Furthermore, using continuous validity over multiple cycles as the solidification trigger condition prevents accidentally valid parameters from being prematurely adopted permanently. Solidification is not based on a single valid verification, but requires that no parameter rollback instruction be received within multiple consecutive control decision cycles before a parameter solidification instruction is generated. The more consecutive control decision cycles required, the more robust the solidification. Alternatively, the solidification condition can further require that the parameter verification results are consistent in at least two movement phases, or that its validity is repeatedly verified when traversing the same road segment multiple times.

[0122] In response to parameter solidification commands, the current robot dynamics attribute data, vehicle load status data, and road type label data are extracted and associated with the target control parameters to generate scene control experience records.

[0123] Specifically, verified and effective target control parameters are bound to their applicable operating conditions, forming reusable structured experience. Since the applicability of the same set of control parameters is closely related to the robot's state and road conditions at the time, after responding to the parameter solidification command, the current robot dynamics attribute data, vehicle load state data, and road type label data are extracted and associated with the target control parameters to generate a scenario control experience record. The robot dynamics attribute data includes identified equivalent mass, moment of inertia, and rolling resistance. This scenario control experience record characterizes the correspondence between road segment attributes, load state, drive state, and stable control parameters, allowing the parameters to be directly invoked when re-entering the same or similar operating conditions. Alternatively, the scenario control experience record can be further bound to the applicable speed range, turning radius range, motor health status, and corresponding inspection data quality.

[0124] The scenario control experience records are written into the stable parameter library file, and the parameter mapping relationship or simulation parameter correction amount of the corresponding working condition in the digital twin model is updated based on the scenario control experience records.

[0125] Specifically, on the one hand, scenario control experience is recorded and written into a stable parameter library file, allowing verified parameters and their applicable operating conditions to be permanently stored for future reuse. On the other hand, based on this record, the parameter mapping relationship or simulation parameter correction amount in the digital twin model corresponding to the operating condition is updated, allowing the digital twin model to be corrected based on this identification and verification, thus improving its prediction accuracy under that operating condition. Since subsequent twin residual calculations and shadow trials both rely on the digital twin model, the improvement in model prediction accuracy will directly improve the accuracy of subsequent mismatch judgments and parameter trials, thereby forming a closed loop of identification, verification, solidification, and model updating. It should be noted that parameters that have not been verified by entities must not be permanently written into the stable parameter library to prevent unverified parameters from contaminating the parameter library and the digital twin model.

[0126] Through the above steps, this embodiment can run in the background without taking over the original task planning and navigation control or changing the upper-level control command format. It can also more accurately distinguish between control parameter mismatch and external interference. While ensuring the safe operation of high-risk environments such as live equipment, it improves the adaptability of control parameters to the dynamic characteristics of the entity and the quality of inspection data collection.

[0127] like Figure 2 The diagram shown is a structural schematic of the intelligent control system for a power inspection robot based on digital twins provided in this application embodiment, including:

[0128] The inspection data synchronization acquisition module is used to acquire inspection task instructions and, based on the same time reference, acquire the physical operating status of the specified robot and the twin prediction status output by the digital twin model.

[0129] The dual residual calculation module is used to extract the current motion stage features of the specified robot based on the control instructions in the inspection task instructions, and calculate the control residual and twin residual respectively by combining the physical running state and the twin prediction state.

[0130] The twin trial calculation prediction module is used to generate candidate control parameters when the control residual and the twin residual meet the preset parameter mismatch conditions, and input the candidate control parameters into the digital twin model for trial calculation, and output the predicted trajectory.

[0131] The target parameter determination module is used to filter the predicted trajectory based on preset safety constraints and determine the target control parameters from the candidate control parameters.

[0132] The progressive deployment verification module is used to obtain the environmental safety level of the environment in which the specified robot is located, calculate the fusion ratio based on the environmental safety level, and fuse the current operating control parameters and target control parameters according to the fusion ratio to generate hybrid control parameters. The hybrid control parameters are then sent to the specified robot for physical verification operation, and the module decides whether to solidify the target control parameters or revert to the control parameters before the update based on the obtained verification feedback data.

[0133] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the above functions can be divided into different functional modules to complete all or part of the functions described above.

[0134] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligent control of a power inspection robot based on digital twins, characterized in that, Includes the following steps: Obtain inspection task instructions, and based on the same time reference, obtain the physical operating status of the specified robot and the twin prediction status output by the digital twin model; The current motion stage features of the designated robot are extracted based on the control instructions in the inspection task instructions, and the control residual and twin residual are calculated by combining the entity's running state and the twin prediction state. When the control residual and the twin residual meet the preset parameter mismatch condition, candidate control parameters are generated, and the candidate control parameters are input into the digital twin model for trial calculation, and the predicted trajectory is output. The predicted trajectory is filtered based on preset safety constraints, and the target control parameter is determined from the candidate control parameters. Obtain the environmental safety level of the environment where the specified robot is located, calculate the fusion ratio based on the environmental safety level, and fuse the current operation control parameters and the target control parameters according to the fusion ratio to generate hybrid control parameters; The hybrid control parameters are sent to the designated robot for physical verification, and the target control parameters are either fixed or rolled back to the control parameters before the update are determined based on the obtained verification feedback data.

2. The intelligent control method for power inspection robots based on digital twins as described in claim 1, characterized in that: The calculation of the control residuals and twin residuals includes: The hardware operating status of the positioning sensor and communication module is obtained to calculate the overall reliability. When the overall confidence level is greater than a preset confidence threshold, the residual calculation process is triggered; Extract the actual motion execution data from the entity's operating state and the expected target data from the inspection task instruction, calculate the deviation between the two, and obtain the control residual. Extract the actual motion execution data in the entity's running state and the predicted execution data in the twin prediction state, calculate the deviation between the two, and obtain the twin residual; According to the preset statistical rules corresponding to the current motion stage characteristics, the control residual and the twin residual are statistically analyzed within the target time window to obtain the stage residual statistical characteristics.

3. The intelligent control method for power inspection robots based on digital twins as described in claim 1, characterized in that: The step of extracting the current motion stage features of the designated robot based on the control instructions in the inspection task command includes: Extract the desired velocity change data and target path curvature from the control command; When the expected velocity change data indicates that the target velocity is increasing and the curvature of the target path is not greater than a preset curvature threshold, the current motion phase feature representing the acceleration phase is generated; When the curvature of the target path is greater than the preset curvature threshold and the expected speed change data does not indicate a decrease in the target speed, the current motion phase feature representing the turning phase is generated; When the desired velocity change data indicates a decrease in target velocity, the current motion phase feature representing the deceleration phase is generated; When the specified robot simultaneously meets multiple motion stage judgment conditions, the current motion stage characteristics are determined according to the preset stage priority.

4. The intelligent control method for power inspection robots based on digital twins as described in claim 2, characterized in that: When the control residual and the twin residual satisfy a preset parameter mismatch condition, candidate control parameters are generated, including: When the multi-dimensional stage residual statistical characteristics continuously exceed the preset residual threshold and reach the preset duration, it is confirmed that the preset parameter mismatch condition is met. The statistical features of the stage residuals within different time windows are combined to generate residual fingerprint features; The residual fingerprint features are compared with a preset mismatch type feature library to identify and output the corresponding parameter mismatch type; Based on the parameter mismatch type, the corresponding adjustable parameter range constraint data is extracted from the initial stable parameter library, and multiple sets of candidate control parameters are generated within the adjustable parameter range constraint data.

5. The intelligent control method for power inspection robots based on digital twins as described in claim 1, characterized in that: The process of filtering the predicted trajectory based on preset safety constraints includes: Switch the digital twin model to shadow trial mode, and input the current pose state data and subsequent local path segment data of the specified robot into the digital twin model; For each set of candidate control parameters, simulation is performed in the shadow trial mode to output the predicted trajectory, including the maximum lateral deviation, the predicted collision distance, and the predicted motor current. The predicted trajectory is compared item by item with each of the boundary thresholds in the preset safety constraints. Candidate control parameters that fail the comparison are removed, and candidate control parameters that meet the various boundary thresholds are retained as preliminary candidate parameters.

6. The intelligent control method for power inspection robots based on digital twins as described in claim 5, characterized in that: After obtaining the preliminary candidate parameters, the step of determining the target control parameters further includes: The predicted vehicle vibration data and predicted parking position error corresponding to the preliminary candidate parameters are read through the digital twin model. The predicted vehicle vibration data and the predicted parking position error are input into a preset inspection quality assessment algorithm to calculate the predicted inspection data quality loss data. The maximum lateral deviation in the predicted trajectory and the predicted quality loss of the inspection data are weighted together to output comprehensive evaluation data. Extract the preliminary candidate parameter with the smallest value from the comprehensive evaluation data and mark it as the target control parameter.

7. The intelligent control method for power inspection robots based on digital twins as described in claim 1, characterized in that: The step of obtaining the environmental safety level of the environment where the specified robot is located, and calculating the fusion ratio based on the environmental safety level, includes: Obtain the distance to surrounding obstacles and the smoothness of the running road at the current location of the specified robot; The environmental safety level is generated by rating the data based on the distance to surrounding obstacles and the smoothness of the operating road. When the environmental security level indicates that the current location is a high-risk, non-deployable area, a freeze update command is output to set the fusion ratio to zero, thereby pausing the generation and distribution of the fusion control parameters; When the environmental security level indicates that the current location is a permitted deployment area, the initial weight is calculated based on the environmental security level and the parameter variation range of the target control parameters, and the fusion ratio is generated.

8. The intelligent control method for power inspection robots based on digital twins as described in claim 1, characterized in that: The step of sending the hybrid control parameters to the designated robot for entity verification and running, and deciding whether to solidify the target control parameters or revert to the control parameters before the update based on the obtained verification feedback data, includes: Based on the fusion ratio, the current operating control parameters and the target control parameters are weighted and summed to generate and distribute the hybrid control parameters; Within the preset entity verification time window, the updated residual verification data generated based on the hybrid control parameters is recalculated. When the updated residual verification data indicates that the control residual is reduced and the twin residual is not expanded, and the clarity evaluation data of the synchronously acquired entity inspection image is not degraded, a continuing fusion instruction is generated to increase the fusion ratio without exceeding the preset fusion ratio upper limit. When the updated residual verification data indicates that the control residual continues to expand or the entity inspection image clarity assessment data deteriorates, a parameter rollback instruction is generated.

9. The intelligent control method for a power inspection robot based on digital twin as described in claim 8, characterized in that: The method further includes: When no parameter rollback instruction is received within multiple consecutive control determination cycles, a parameter solidification instruction is generated. In response to the parameter solidification command, the current robot dynamics attribute data, vehicle load status data, and road type label data are extracted and associated with the target control parameters to generate a scene control experience record. The scenario control experience records are written into a stable parameter library file, and the parameter mapping relationship or simulation parameter correction amount of the corresponding working condition in the digital twin model is updated based on the scenario control experience records.

10. A digital twin-based intelligent control system for power inspection robots, characterized in that, include: The inspection data synchronization acquisition module is used to acquire inspection task instructions and acquire the physical running status of the specified robot and the twin prediction status output by the digital twin model based on the same time base. The dual residual calculation module is used to extract the current motion stage features of the specified robot according to the control instructions in the inspection task instruction, and calculate the control residual and twin residual respectively by combining the entity running state and the twin prediction state. The twin trial calculation and prediction module is used to generate candidate control parameters when the control residual and the twin residual meet the preset parameter mismatch conditions, and input the candidate control parameters into the digital twin model for trial calculation and output the predicted trajectory; The target parameter determination module is used to filter the predicted trajectory based on preset safety constraints and determine the target control parameter from the candidate control parameters; The progressive deployment verification module is used to obtain the environmental safety level of the environment in which the specified robot is located, calculate the fusion ratio based on the environmental safety level, and fuse the current running control parameters and the target control parameters according to the fusion ratio to generate hybrid control parameters. The hybrid control parameters are then sent to the specified robot for physical verification operation, and the module decides whether to solidify the target control parameters or revert to the control parameters before the update based on the obtained verification feedback data.