Self-adaptive hot-line work attitude control method, system, medium and equipment for intelligent power robot with body

Abstract

The invention belongs to the technical field of intelligent power robots, and provides a self-adaptive hot-line work attitude control method and system for an intelligent power robot with a body, and the technical scheme is as follows: obtaining the working state and working environment data of the power robot in the hot-line work process; performing feature extraction on the working state and working environment data, and fusing the extracted working state features and working environment features to obtain a comprehensive feature vector; and performing operation dynamic attitude adjustment of the power robot by combining the comprehensive feature vector and a reinforcement learning algorithm, in the adjustment process, introducing an isolated forest algorithm to dynamically monitor the operation state of the power robot, triggering a feedback mechanism according to the monitored abnormal state, and generating a self-adaptive hot-line operation attitude control strategy of the power robot. The self-adaptive response to a complex working environment is realized, and the safety and efficiency of hot-line work are improved.

Description

Technical Field

[0001] This invention belongs to the field of embodied intelligent robot technology, and particularly relates to the adaptive live-line working posture control method, system, medium and equipment of embodied intelligent power robot. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the continuous development of power systems, especially the complexity of distribution networks and the widespread use of high-voltage lines, traditional live-line working methods are facing increasing challenges. High-voltage lines are usually located at high altitudes, and the working environment is complex and variable. Traditional manual operations not only pose high safety risks, but also cannot meet the needs of modern power systems in terms of efficiency and accuracy.

[0004] Although some power robots have been applied to live-line work, most still rely on preset work paths and fixed working postures, lacking the ability to perceive and respond to real-time environmental changes. For example, robots may encounter problems such as equipment obstruction, changes in ambient lighting, and electromagnetic interference during operation. These factors may cause the robot to fail to complete the task accurately or even cause safety accidents. In addition, existing posture control systems mostly use traditional control algorithms, which are difficult to achieve high-precision posture adjustment in complex and dynamic working environments. For example, when performing live-line work, improper posture adjustment may lead to collisions with electrical equipment, increasing the risk of the operation.

[0005] During live-line work by power robots, real-time monitoring of their status and timely handling of anomalies are crucial for ensuring operational safety and efficiency. However, existing anomaly detection methods largely rely on static threshold settings and lack the ability to deeply fuse and dynamically analyze multimodal sensor data, making them ill-suited for complex and ever-changing operating environments. Summary of the Invention

[0006] To address at least one of the technical problems mentioned above, this invention provides an adaptive live-line working posture control method and system for an embodied intelligent electric robot. This method and system can sense changes in the complex working environment in real time and dynamically adjust the robot's working posture based on feedback information, thereby achieving adaptive response to the complex working environment and improving the safety and efficiency of live-line work.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of the present invention provides an adaptive live-line working posture control method for an embodied intelligent electric robot, comprising the following steps: Acquire data on the working status and working environment of the power robot during live-line operations; Feature extraction is performed on the work status and work environment data, and the extracted work status features and work environment features are fused to obtain a comprehensive feature vector; This paper combines comprehensive feature vectors and reinforcement learning algorithms to adjust the dynamic posture of electric robots during operation. In the adjustment process, the isolated forest algorithm is introduced to dynamically monitor the operation status of the electric robot. Based on the abnormal state detected, a feedback mechanism is triggered to generate an adaptive live operation posture control strategy for the electric robot.

[0008] Furthermore, in the process of adjusting the dynamic posture of the power robot, an improved isolated forest algorithm is introduced to dynamically monitor the working state of the power robot. Based on the detected abnormal state, a feedback mechanism is triggered to generate an adaptive live-line working posture control strategy for the power robot, including: The working status of the electric robot is dynamically detected. The acquired multimodal feature stream is constructed into a time embedding according to the sliding window and fed into a multi-path isolated forest to obtain the branch anomaly score. Then, through risk-adaptive dynamic threshold and time consistency statistics, an adaptive live-line working attitude control strategy for the electric robot is generated.

[0009] Furthermore, the formula for calculating the risk-adaptive dynamic threshold is as follows: , in, Indicates a dynamic threshold, in t The dynamic threshold used to determine anomalies at any given time is adaptively adjusted according to factors such as the working environment and sensor fluctuations. For robust position and scaling estimation of fractions, This represents the risk adaptive coefficient; a larger value loosens the threshold, while a smaller value tightens it. This indicates the upper limit of the risk adjustment coefficient. This represents the lower limit of the risk adjustment coefficient. As an environmental risk vector, It is Sigmoid. It is a linear combination of risk factors.

[0010] Furthermore, during the timing consistency statistics, time series analysis is incorporated into the attitude control loop. A sliding window is used to perform consistency detection on the sensor fractional sequences, and a change point statistic is constructed. , is represented as: , in, and express t Time and t The change point statistics at time -1 For the abnormal strength after robust standardization, For baseline drift compensation, This indicates reinforcement learning. Indicates a dynamic threshold. for t Parameters that can adapt to risk at any time.

[0011] Furthermore, the dynamic posture adjustment of the robot during operation by combining comprehensive feature vectors and reinforcement learning algorithms includes: By jointly modeling the integrated feature vector and the robot posture matrix, a unified state representation is obtained and used as input for reinforcement learning; The unified state representation is used as input to the reinforcement learning network to generate actions, while the Critic network calculates the value of the actions based on the current state s and the action taken. The system calculates the corresponding reward value r, continuously updates the Q-value function, and generates an action selection strategy based on the Q-value function to adjust the dynamic posture of the power robot during operation.

[0012] Furthermore, feature extraction is performed on the aforementioned working status and work environment data, including: Visual feature data is obtained by feature extraction based on multi-dimensional visual modalities; Based on the acquired force and tactile modal data, feature extraction is performed to obtain force feature data and tactile feature data; Environmental feature data is extracted from the acquired environmental sensor data; A comprehensive feature vector is obtained by fusing visual feature data, force feature data, tactile feature data, and environmental feature data.

[0013] Furthermore, the method also includes preprocessing the acquired robot working status and operating environment data, and fusing the data processed by each module using Kalman filtering.

[0014] A second aspect of the present invention provides an adaptive live-line working posture control system for an embodied intelligent electric robot, comprising: The data acquisition module is used to acquire data on the working status and operating environment of the electric robot. The feature extraction module is used to extract features from the work status and work environment data, and to fuse the extracted work status features and work environment features to obtain a comprehensive feature vector. The attitude control module is used to dynamically adjust the working attitude of the electric robot by combining comprehensive feature vectors and reinforcement learning algorithms. During the adjustment process, the isolated forest algorithm is introduced to dynamically monitor the working status of the electric robot. Based on the abnormal state detected, a feedback mechanism is triggered to generate an adaptive live-line working attitude control strategy for the electric robot.

[0015] Furthermore, during the dynamic posture adjustment process of the electric robot, an improved isolated forest algorithm is introduced to dynamically monitor the robot's operating state. Based on the detected abnormal states, a feedback mechanism is triggered to generate an adaptive live-line working posture control strategy for the electric robot, including: The working status of the electric robot is dynamically detected. The acquired multimodal feature stream is constructed into a time embedding according to the sliding window and fed into a multi-path isolated forest to obtain the branch anomaly score. Then, through risk-adaptive dynamic threshold and time consistency statistics, an adaptive live-line working attitude control strategy for the electric robot is generated.

[0016] Furthermore, the formula for calculating the risk-adaptive dynamic threshold is as follows: , in, Indicates a dynamic threshold, in t The dynamic threshold used to determine anomalies at any given time is adaptively adjusted according to factors such as the working environment and sensor fluctuations. For robust position and scaling estimation of fractions, This represents the risk adaptive coefficient; a larger value loosens the threshold, while a smaller value tightens it. This indicates the upper limit of the risk adjustment coefficient. This represents the lower limit of the risk adjustment coefficient. As an environmental risk vector, It is Sigmoid. It is a linear combination of risk factors.

[0017] Furthermore, during the timing consistency statistics, time series analysis is incorporated into the attitude control loop. A sliding window is used to perform consistency detection on the sensor fractional sequences, and a change point statistic is constructed. , is represented as: , in, and express t Time and t The change point statistics at time -1 For the abnormal strength after robust standardization, For baseline drift compensation, This indicates reinforcement learning. Indicates a dynamic threshold. for t Parameters that can adapt to risk at any time.

[0018] Furthermore, the dynamic posture adjustment of the electric robot during operation by combining comprehensive feature vectors and reinforcement learning algorithms includes: By jointly modeling the integrated feature vector and the robot posture matrix, a unified state representation is obtained and used as input for reinforcement learning; The unified state representation is used as input to the reinforcement learning network to generate actions, while the Critic network calculates the value of the actions based on the current state s and the action taken. The system calculates the corresponding reward value r, continuously updates the Q-value function, and generates an action selection strategy based on the Q-value function to adjust the dynamic posture of the power robot during operation.

[0019] Furthermore, feature extraction is performed on the aforementioned working status and work environment data, including: Visual feature data is obtained by feature extraction based on multi-dimensional visual modalities; Based on the acquired force and tactile modal data, feature extraction is performed to obtain force feature data and tactile feature data; Environmental feature data is extracted from the acquired environmental sensor data; A comprehensive feature vector is obtained by fusing visual feature data, force feature data, tactile feature data, and environmental feature data.

[0020] Furthermore, the method also includes preprocessing the acquired robot working status and operating environment data, and fusing the data processed by each module using Kalman filtering.

[0021] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the adaptive live-line working posture control method for an embodied intelligent electric robot as described above.

[0022] A fourth aspect of the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the adaptive live-line working posture control method for an embodied intelligent electric robot as described above.

[0023] A fifth aspect of the present invention provides a program product, which is a computer program product including a computer program that, when executed by a processor, implements the steps in the above-described adaptive live-line working posture control method for an embodied intelligent electric robot.

[0024] Compared with the prior art, the beneficial effects of the present invention are: This invention innovatively proposes an adaptive live-line working posture control method for embodied intelligent electric robots. It can acquire data of the working environment in real time through multimodal sensors, extract features from the multimodal data and fuse the data, and use embodied intelligent reinforcement learning to dynamically adjust the robot's working posture based on feedback information. This enables adaptive response to complex working environments, improves the safety and efficiency of live-line working, and solves the problem that relying on static threshold settings makes it difficult to adapt to complex and ever-changing working environments.

[0025] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0026] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0027] Figure 1 This is a flowchart of the adaptive live-line working posture control method for an embodied intelligent electric robot provided in an embodiment of the present invention. Detailed Implementation

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0029] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0030] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0031] Example 1 like Figure 1 As shown, this embodiment provides an adaptive live-line working posture control method for an embodied intelligent power robot, including the following steps: Step 1: Acquire data on the working status and working environment of the power robot during live-line operation; In this embodiment, the working environment status is perceived in real time through sensors such as vision, force, and touch. For example, vision sensors acquire image information of the equipment, force sensors measure the contact force between the robot and the equipment, and touch sensors perceive the contact point between the robot and the equipment. The data provided by each sensor is converted into a unified perception state space.

[0032] Step 2: Preprocess the acquired power robot's working status and operating environment data to obtain preprocessed power robot working status and operating environment data; First, the acquired multi-sensor data is preprocessed, including: Visual data: Images / point clouds acquired via high-resolution cameras or depth cameras. Preprocessing includes denoising (Gaussian / Median filtering), histogram equalization, and ROI cropping.

[0033] Force and tactile data: time-series signals from sensors, preprocessing includes low-pass filtering to eliminate high-frequency noise and normalization to standard force / pressure units.

[0034] Environmental sensing: Normalization / moving average is used to eliminate fluctuations.

[0035] Then, a Kalman filter is used to fuse data from multiple sensors to reduce sensor noise and improve the overall sensing accuracy.

[0036] The update formula for Kalman filtering is: , , in, For the estimated state, This refers to the state at the previous moment. For the observation matrix, The error covariance of the previous time step. For error covariance, For Kalman gain, These are actual measured values. This is the observation matrix.

[0037] Step 3: Based on the preprocessed robot working status and working environment data, feature extraction is performed to obtain the robot's working characteristics and working environment characteristics during live-line operation; In this embodiment, wavelet transform is used to perform multi-scale analysis on the signal to extract key features. These features include the location of the contact point, the magnitude of the contact force, and information about obstacles in the working environment. Wavelet transform can effectively process signals of different frequency bands, helping to extract high-frequency and low-frequency information, thereby providing more accurate data for attitude adjustment.

[0038] Specifically, the steps include the following: Step 301: Extract visual feature data based on multi-dimensional visual modalities. ; In this embodiment, the multi-dimensional visual modal data includes image sequences, depth maps, and point cloud data; Specifically, image sequences are processed through convolutional neural networks. Perform convolution operations to obtain representational features such as edges, shapes, and textures; Extracting the three-dimensional geometric information of an object from a depth map or point cloud, such as its position coordinates, surface normal vectors, and obstacle distribution, enables the robot to accurately perceive the surface of the equipment and the structure of the work space. These visual features ultimately constitute a key description of the device's surface location, the spatial distribution of obstacles, and the ROI of the target area.

[0039] Step 302: Extract force feature data based on the acquired force and tactile modal data. and tactile feature data ; In this embodiment, in the data processing of force and tactile modes, the main frequency components of the force signal are identified from the acquired time-series signals of force and tactile modes, and multi-scale decomposition is performed on the acquired time-series signals of force and tactile modes to obtain multi-scale features. Specifically, when identifying the main frequency components of the force signal from the acquired force and tactile modal time-series signals, a Fast Fourier Transform (FFT) is used to transform the contact force signal. Convert to frequency domain ,get: , in, Represents angular frequency. It is a complex exponential function, thus identifying the main frequency components of the force signal.

[0040] Specifically, wavelet transform is used to decompose the signal at multiple scales, separating the instantaneous impact force from the stable contact force, and extracting the contact point location, the magnitude of the force, and the trend of force change.

[0041] Ultimately, these features provide fine-grained tactile feedback when the robot performs operations.

[0042] Step 303: Extract environmental feature data based on the acquired environmental sensor data. ; For environmental sensing modalities, the system primarily extracts statistical and temporal features. By calculating the mean, variance, and skewness of signals such as temperature and electromagnetic fields, it obtains the stability characteristics of the environment. Furthermore, by combining a sliding window or Long Short-Term Memory (LSTM) network to analyze the temporal trends of the signals, it identifies potential risk factors such as electromagnetic disturbances or temperature fluctuations. These features help the robot make adaptive adjustments in safety-sensitive scenarios.

[0043] Step 304: Integrate visual feature data, force feature data, tactile feature data, and environmental feature data to obtain a comprehensive feature vector, represented as: ; The fused integrated feature vector is input to the task decision network, enabling adaptive attitude adjustment and task planning during live-line work. This multimodal feature extraction and fusion process empowers the robot with comprehensive perception and intelligent decision-making capabilities in complex power environments.

[0044] Step 4: Combine comprehensive feature vectors and reinforcement learning algorithms to adjust the dynamic posture of the electric robot during operation. During the adjustment process, an improved isolated forest algorithm is introduced to dynamically monitor the working state of the electric robot. Based on the detected abnormal state, a feedback mechanism is triggered to generate an adaptive live-line working posture control strategy for the electric robot. Specifically, the steps include the following: Step 401: Jointly model the integrated feature vector and the robot posture matrix to obtain a unified state representation as input for reinforcement learning; In this embodiment, the comprehensive feature vector will be Mapped into the robot posture control model, this is represented by feature vectors. Together with the attitude matrix R(q), they construct the state space. , as input for reinforcement learning.

[0045] in It includes visual features, force and tactile features, and environmental features. These features constitute the "external perception layer," which provides the robot with environmental information.

[0046] The attitude control of electric robots employs quaternion representation to effectively avoid the singularity problem in traditional Euler angle models. Quaternions are mathematical tools that can represent rotations in three-dimensional space, providing higher computational accuracy and more stable control. The basic form of attitude control can be described by the following relationship between quaternions and rotation matrices: , in, This is a scalar component, related to the rotation angle of the electric robot, and is equivalent to the rotation "amplitude" factor; The vector part, related to the rotation axis, describes the direction in which the rotation occurs.

[0047] Quaternions directly correspond to the robot's orientation in three-dimensional space. Compared to Euler angles (roll, pitch, yaw), quaternions can continuously and smoothly represent large-angle rotations, avoiding discontinuities or singularities during pose calculation. With rotation matrix R The relationship is as follows: , The rotation matrix R This paper describes the rotational transformations of an electric robot in three-dimensional space, enabling real-time adjustment of the robot's posture to adapt to environmental changes during operation. This is achieved using quaternions and rotation matrices. R The conversion can seamlessly transfer the posture of the robot's joints / body to the controller for kinematic and dynamic calculations; The rotation matrix R(q) maps the robot's end effector posture to the global coordinate system, providing a mathematical description for the interaction between the robot's actions and the external task environment. This is equivalent to an "internal state layer," ensuring the accuracy of the robot's kinematics and dynamics posture.

[0048] External environmental characteristics By jointly modeling with the attitude matrix R(q), a unified state representation is obtained: , in, This is the feature fusion function, which can be a simple concatenation, a non-linear mapping, or a deep neural network. This is how it's constructed. It includes both "external environment information" and "robot internal posture," fully depicting the robot's operational status at the current moment.

[0049] Step 402: Use the unified state representation as input for reinforcement learning, and generate an initial action selection strategy for robot posture adjustment through reinforcement learning; In this embodiment, Q-learning or deep reinforcement learning (such as DDPG) is used at the decision-making level to achieve action selection.

[0050] Will As input for reinforcement learning, it is passed to the Actor network to generate actions. The Critic network calculates the value of actions. In this process, the robot can dynamically adjust its posture and task execution strategy based on the fused state representation, thereby achieving adaptive response to complex power operation environments. This is based on the current state s and the actions taken. To calculate the corresponding reward value r, the system continuously updates the Q-value function: , in, It represents the expected cumulative reward that can be obtained after taking action a in the current state s, and is a value estimation function. The estimated value that can be obtained after taking action a' in the next state s'. For learning rate, As a discount factor, In order to implement the reward, For the next state, This prepares for the next action. Through continuous learning, the robot can adaptively adjust its posture based on real-time perception data, optimizing work efficiency and safety. In the decision-making process, probability is considered. Choose the action with the highest Q value, based on probability. Randomly select an action to ensure a balance between exploring new actions and utilizing existing knowledge.

[0051] Step 403: Based on the initial action selection strategy, the working state of the electric robot is dynamically detected. The acquired multimodal feature stream is constructed into a time embedding according to the sliding window and sent into a multi-path isolated forest to obtain the branch anomaly score. Then, through risk-adaptive dynamic threshold and time consistency statistics, the adaptive live-line working posture control strategy of the robot is generated. In this embodiment, during the process of adjusting the robot's dynamic posture, an improved isolated forest algorithm is introduced to dynamically monitor the working state of the electric robot. Based on the detected abnormal state, a feedback mechanism is triggered to generate an adaptive live-line working posture control strategy for the robot. Outliers in the Isolation Forest concept typically have two characteristics: low number and sparse distribution. Therefore, they are more likely to be isolated than normal points during data partitioning. The Isolation Forest algorithm constructs a series of random trees to divide the samples into increasingly smaller subspaces and calculates the path length required to isolate each point; the shorter the path, the more likely the point is to be an outlier.

[0052] To address the limitations of traditional isolated forest algorithms in high-dimensional data processing, this invention improves upon them by addressing the limitations of multimodal feature flows. The time embedding is constructed using a sliding window, and then fed into a multi-path isolated forest to obtain the branch anomaly score. The alarm is then determined by a risk-adaptive dynamic threshold and time-series consistency statistics. Finally, a quality-aware integration strategy is used to perform score fusion and self-calibration.

[0053] Specifically, the steps include the following: Step 4031: Obtain the multimodal feature stream Construct time embeddings using a sliding window, and feed them into a multi-path isolated forest to obtain path outlier scores. ; Step 4032: Calculate the risk adaptive threshold based on the multimodal anomaly score; The system uses robust statistical tracking of the center and scale of the score, and directly maps environmental risks (such as electromagnetic disturbances, temperature gradients, visibility, etc.) into threshold tightening / relaxation modulation quantities, forming a discrimination boundary that adapts to the context. , in, Indicates a dynamic threshold, in t The dynamic threshold used to determine anomalies at any given time is adaptively adjusted according to factors such as the working environment and sensor fluctuations. For robust position and scaling estimation of fractions, This represents the risk adaptive coefficient; a larger value loosens the threshold, while a smaller value tightens it. This indicates the upper limit of the risk adjustment coefficient. This represents the lower limit of the risk adjustment coefficient. As an environmental risk vector, It is Sigmoid. It is a linear combination of risk factors.

[0054] Unlike fixed thresholds, this improvement can automatically tighten the judgment in high-risk scenarios and loosen it in low-risk scenarios, significantly improving sensitivity and accuracy.

[0055] Step 4033: Add time series analysis to the attitude control loop, perform consistency detection on the sensor fractional sequences using a sliding window, and construct change point statistics. : , in, and express t Time and t The change point statistics at time -1 for t Parameters that can adapt to risk at any time. This indicates reinforcement learning; For baseline drift compensation, Adaptable to risk The robustly standardized anomaly strength is compared with the dynamic threshold from the previous step. In comparison, if the threshold is exceeded and continues to accumulate over time... Only then will an alarm be triggered. The robot does not adjust its posture based on a single momentary abnormal fluctuation, but rather combines abnormal trends and persistence to determine whether to perform posture correction. Only when this happens will the system determine it as a "continuous anomaly" and trigger attitude replanning, thereby avoiding misoperation caused by noise or momentary contact.

[0056] Step 4033: The ensemble learning mechanism plays a role in the robot's multimodal perception. Visual, force, tactile, and environmental sensors each establish an isolated forest, outputting local scores. The system weights and fuses these parameters based on modal health (such as sensor signal-to-noise ratio or calibration status): , Score of different sensors (vision, force, touch, environment) under multimodal input. The global score will be formed through weighted fusion. Then As standardized score This is then fed into the decision framework of the first two formulas. (Fusing score) The final alarm is determined by the threshold and time series analysis in the above formula, and the embodied controller is linked to perform a graded handling of "deceleration-yield-replanning". The above three improvements introduce risk modulation, time consistency and quality perception integration without changing the core mechanism of the isolated forest, forming a real-time, robust and implementable anomaly detection system for live-line work.

[0057] Example 2 This embodiment provides an adaptive live-line working posture control system for an embodied intelligent power robot, including: The data acquisition module is used to acquire data on the working status and operating environment of the electric robot. The feature extraction module is used to extract features from the work status and work environment data, and to fuse the extracted work status features and work environment features to obtain a comprehensive feature vector. The attitude control module is used to dynamically adjust the working attitude of the electric robot by combining comprehensive feature vectors and reinforcement learning algorithms. During the adjustment process, an improved isolated forest algorithm is introduced to dynamically monitor the working status of the electric robot. Based on the abnormal state detected, a feedback mechanism is triggered to generate an adaptive live-line working attitude control strategy for the electric robot.

[0058] As a further implementation, during the dynamic posture adjustment process of the power robot, an improved isolated forest algorithm is introduced to dynamically monitor the working state of the power robot. Based on the detected abnormal state, a feedback mechanism is triggered to generate an adaptive live-line working posture control strategy for the power robot, including: The working status of the electric robot is dynamically detected. The acquired multimodal feature stream is constructed into a time embedding according to the sliding window and fed into a multi-path isolated forest to obtain the branch anomaly score. Then, through risk-adaptive dynamic threshold and time consistency statistics, an adaptive live-line working attitude control strategy for the electric robot is generated.

[0059] As a further implementation method, the formula for calculating the risk-adaptive dynamic threshold is: , in, Indicates a dynamic threshold, in t The dynamic threshold used to determine anomalies at any given time is adaptively adjusted according to factors such as the working environment and sensor fluctuations. For robust position and scaling estimation of fractions, This represents the risk adaptive coefficient; a larger value loosens the threshold, while a smaller value tightens it. This indicates the upper limit of the risk adjustment coefficient. This represents the lower limit of the risk adjustment coefficient. As an environmental risk vector, It is Sigmoid. It is a linear combination of risk factors.

[0060] As a further implementation, during the timing consistency statistics, time series analysis is incorporated into the attitude control loop. A sliding window is used to detect the consistency of the sensor fractional sequences, and a change-point statistic is constructed. , is represented as: , in, and express t Time and t The change point statistics at time -1 For the abnormal strength after robust standardization, For baseline drift compensation, This indicates reinforcement learning. Indicates a dynamic threshold. for t Parameters that can adapt to risk at any time.

[0061] As a further implementation, the dynamic posture adjustment of the electric robot during operation by combining comprehensive feature vectors and reinforcement learning algorithms includes: By jointly modeling the integrated feature vector and the robot posture matrix, a unified state representation is obtained and used as input for reinforcement learning; The unified state representation is used as input to the reinforcement learning network to generate actions, while the Critic network calculates the value of the actions based on the current state s and the action taken. The system calculates the corresponding reward value r, continuously updates the Q-value function, and generates an action selection strategy based on the Q-value function to adjust the dynamic posture of the power robot during operation.

[0062] As a further implementation, feature extraction is performed on the working status and work environment data, including: Visual feature data is obtained by feature extraction based on multi-dimensional visual modalities; Based on the acquired force and tactile modal data, feature extraction is performed to obtain force feature data and tactile feature data; Environmental feature data is extracted from the acquired environmental sensor data; A comprehensive feature vector is obtained by fusing visual feature data, force feature data, tactile feature data, and environmental feature data.

[0063] As a further implementation, the method also includes preprocessing the acquired robot working status and operating environment data, and fusing the data processed by each module using Kalman filtering.

[0064] It should be noted that the specific implementation of the adaptive live-line working posture control system for the embodied intelligent power robot in this embodiment of the invention is similar to the specific implementation of the adaptive live-line working posture control method for the embodied intelligent power robot in this embodiment of the invention. For details, please refer to the description in the method section. To reduce redundancy, it will not be repeated here.

[0065] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the adaptive live-line working posture control method for an embodied intelligent power robot as described above.

[0066] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the adaptive live-line working posture control method for an embodied intelligent power robot as described above.

[0067] Example 5 This embodiment provides a program product, which is a computer program product, including a computer program. When the computer program is executed by a processor, it implements the steps in the adaptive live-line working posture control method for embodied intelligent electric robots as described above.

[0068] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0069] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0070] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0071] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0072] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0073] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for adaptive live-line working posture control of an embodied intelligent electric robot, characterized in that, Includes the following steps: Acquire data on the working status and working environment of the power robot during live-line operations; Feature extraction is performed on the work status and work environment data, and the extracted work status features and work environment features are fused to obtain a comprehensive feature vector; This paper combines comprehensive feature vectors and reinforcement learning algorithms to adjust the dynamic posture of electric robots during operation. In the adjustment process, the isolated forest algorithm is introduced to dynamically monitor the operation status of the electric robot. Based on the abnormal state detected, a feedback mechanism is triggered to generate an adaptive live operation posture control strategy for the electric robot.

2. The adaptive live-line working posture control method for an embodied intelligent power robot as described in claim 1, characterized in that, In the process of adjusting the dynamic posture of the power robot, an improved isolated forest algorithm is introduced to dynamically monitor the robot's operating state. Based on the detected abnormal state, a feedback mechanism is triggered to generate an adaptive live-line working posture control strategy for the power robot, including: The working status of the electric robot is dynamically detected. The acquired multimodal feature stream is constructed into a time embedding according to the sliding window and fed into a multi-path isolated forest to obtain the branch anomaly score. Then, through risk-adaptive dynamic threshold and time consistency statistics, an adaptive live-line working attitude control strategy for the electric robot is generated.

3. The adaptive live-line working posture control method for an embodied intelligent power robot as described in claim 2, characterized in that, The formula for calculating the risk-adaptive dynamic threshold is: ， in, Indicates a dynamic threshold, in t The dynamic threshold used to determine anomalies at any given time is adaptively adjusted according to factors such as the working environment and sensor fluctuations. For robust position and scaling estimation of fractions, This represents the risk adaptive coefficient; a larger value loosens the threshold, while a smaller value tightens it. This indicates the upper limit of the risk adjustment coefficient. This indicates the lower limit of the risk adjustment coefficient. As an environmental risk vector, It is Sigmoid. It is a linear combination of risk factors.

4. The adaptive live-line working posture control method for an embodied intelligent power robot as described in claim 2, characterized in that, When performing time-series consistency statistics, time series analysis is incorporated into the attitude control loop. Consistency detection of sensor fractional sequences is performed using a sliding window to construct change-point statistics. , is represented as: ， in, and express t Time and t The change point statistics at time -1 For the abnormal strength after robust standardization, For baseline drift compensation, This indicates reinforcement learning. Indicates dynamic threshold. for t Parameters that can adapt to risk at any time.

5. The adaptive live-line working posture control method for an embodied intelligent power robot as described in claim 1, characterized in that, The method of combining integrated feature vectors and reinforcement learning algorithms to adjust the dynamic posture of the electric robot includes: By jointly modeling the integrated feature vector and the robot posture matrix, a unified state representation is obtained and used as input for reinforcement learning; The unified state representation is used as input to the reinforcement learning network to generate actions, while the Critic network calculates the value of the actions based on the current state s and the action taken. The system calculates the corresponding reward value r, continuously updates the Q-value function, and generates an action selection strategy based on the Q-value function to adjust the dynamic posture of the power robot during operation.

6. The adaptive live-line working posture control method for an embodied intelligent power robot as described in claim 1, characterized in that, Feature extraction is performed on the aforementioned working status and work environment data, including: Visual feature data is obtained by feature extraction based on multi-dimensional visual modalities; Based on the acquired force and tactile modal data, feature extraction is performed to obtain force feature data and tactile feature data; Environmental feature data is extracted from the acquired environmental sensor data; A comprehensive feature vector is obtained by fusing visual feature data, force feature data, tactile feature data, and environmental feature data.

7. The adaptive live-line working posture control method for an embodied intelligent power robot as described in claim 1, characterized in that, The method also includes preprocessing the acquired robot working status and operating environment data, and fusing the data processed by each module using Kalman filtering.

8. An adaptive live-line working posture control system for an embodied intelligent electric robot, characterized in that, include: The data acquisition module is used to acquire data on the working status and operating environment of the electric robot. The feature extraction module is used to extract features from the work status and work environment data, and to fuse the extracted work status features and work environment features to obtain a comprehensive feature vector. The attitude control module is used to dynamically adjust the working attitude of the electric robot by combining comprehensive feature vectors and reinforcement learning algorithms. During the adjustment process, an improved isolated forest algorithm is introduced to dynamically monitor the working status of the electric robot. Based on the abnormal state detected, a feedback mechanism is triggered to generate an adaptive live-line working attitude control strategy for the electric robot.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the adaptive live-line working posture control method for an embodied intelligent electric robot as described in any one of claims 1-7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the adaptive live-line working posture control method for an embodied intelligent electric robot as described in any one of claims 1-7.

Images