AGV robot control method and system, medium and program product

Abstract

The invention discloses an AGV robot control method and system, a medium and a program product, and relates to the field of control systems, and the method comprises the steps: obtaining the historical operation data of a plurality of AGV robots; the historical operation data comprises working state data, environment sensing data, task execution data and energy consumption data; based on the performance evaluation model, extracting control parameters and environment characteristics corresponding to the performance mutation points from the historical operation data, and generating a training sample with a performance label; constructing a mapping relation based on the training sample, and generating an operation optimization model; receiving current environment characteristic data and operation state data sent by the to-be-optimized robot; inputting the current environment characteristic data and the operation state data to the operation optimization model to obtain a target control parameter; and sending a control instruction corresponding to the target control parameter to the to-be-optimized robot so as to control the to-be-optimized robot to adjust the running state. According to the invention, the environment adaptability of the AGV robot can be improved.

Description

Technical Field

[0001] This application relates to the field of control systems, and in particular, to a control method, system, medium, and program product for an AGV robot. Background Art

[0002] With the rapid development of intelligent manufacturing and logistics automation, the AGV (Automated Guided Vehicle) robot, as a core component of the intelligent logistics system, has been widely used in industrial production, warehouse management, and intelligent logistics. With its flexibility, reliability, and high efficiency, the AGV robot can achieve intelligent transportation and precise distribution of materials, greatly improving production efficiency and operation safety.

[0003] In related technologies, the AGV robot control technology is mainly designed based on a single configuration type, and navigation and task execution in a specific environment are achieved through preset control algorithms and parameter settings. These control systems usually perform one-time parameter optimization according to a specific environment and task type, and engineers manually adjust and solidify the parameters according to the initial installation environment.

[0004] However, the factory environment usually changes dynamically, such as changes in the ground friction coefficient, an increase in the personnel density during the production peak, and diversification of cargo types. Traditional AGV control systems mostly use preset PID parameters or static rule tables, and the adaptability of AGV robot operation is poor. Summary of the Invention

[0005] This application provides a control method, system, medium, and program product for an AGV robot to improve the environmental adaptability of the AGV robot.

[0006] In a first aspect, this application provides a control method for an AGV robot, which is applied to a control system. The method includes: obtaining historical operation data of multiple AGV robots; the historical operation data includes working state data, environment perception data, task execution data, and energy consumption data; based on a performance evaluation model, extracting control parameters and environmental characteristics corresponding to performance mutation points from the historical operation data to generate training samples with performance annotations; the performance mutation point is the time point when the change rate of the performance index generated by the performance evaluation model exceeds a preset stable threshold; constructing a mapping relationship between environmental characteristics and optimal control parameters based on the training samples to generate an operation optimization model; receiving current environmental characteristic data and operation state data sent by the robot to be optimized; inputting the current environmental characteristic data and operation state data into the operation optimization model to obtain target control parameters; sending a control instruction corresponding to the target control parameters to the robot to be optimized to control the robot to be optimized to adjust its operation state.

[0007] In the above embodiments, the control system extracts control parameters and environmental characteristics based on the performance mutation points in the historical operation data, establishes the mapping relationship between the environmental characteristics and the optimal control parameters, realizes the dynamic optimization of the control parameters of the AGV robot, and the control system can adaptively adjust the control parameters according to the current environmental characteristics, significantly improving the adaptability and operation efficiency of the AGV robot in different environments, and solving the problem that the traditional fixed parameter configuration cannot cope with environmental changes.

[0008] Combined with some embodiments of the first aspect, in some embodiments, the step of extracting the control parameters and environmental characteristics corresponding to the performance mutation points from the historical operation data based on the performance evaluation model to generate training samples with performance annotations specifically includes: segmenting the historical operation data according to the time series, and calculating the comprehensive performance index of the AGV robot in each time period based on the performance evaluation formula; the comprehensive performance index includes task completion efficiency, navigation accuracy, energy utilization rate and safety factor; identifying the time point when the change rate of the comprehensive performance index exceeds the preset stability threshold as the performance mutation point; extracting the environmental characteristic data and control parameter data within the preset time window before and after the performance mutation point, and calculating the performance difference before and after the performance mutation point; using the performance difference as the performance annotation, and integrating the environmental characteristics and control parameters corresponding to the performance annotation to generate training samples.

[0009] In the above embodiments, the control system constructs a comprehensive performance evaluation system including task completion efficiency, navigation accuracy, energy utilization rate and safety factor, uses the performance evaluation formula to segment and evaluate the historical data, and extracts the differences in environmental characteristics and control parameters before and after the performance mutation points, realizing the accurate quantification of the operation performance of the AGV robot and the identification of key influencing factors.

[0010] In combination with some embodiments of the first aspect, in some embodiments, the performance evaluation formula is: P = W1·Pt + W2·Pn + W3·Pe + W4·Ps; Pt is the task execution performance index; Pt = (T0 / T)·[1 + ∫(dv / dt)²dt / D]·(1 - θ / θmax); T0 is the standard task time, T is the actual execution time; D is the task distance; θ is the path curvature, θmax is the allowable maximum curvature; dv / dt is the rate of change of speed, used to characterize the smoothness of movement; Pn is the navigation accuracy index; Pn = (1 - |ΔD| / D0)·exp(-σ² / σ0²)·(1 + d²Δp / dt²)^(-1); ΔD is the actual trajectory deviation; D0 is the allowable maximum deviation; σ is the standard deviation of the positioning error; σ0 is the reference error value; d²Δp / dt² is the second derivative of the position, used to characterize the movement jitter; Pe is the energy efficiency index; Pe = (E0 / E)·[1 - ∑(Pi·ti) / (Pmax·T)]·(1 - ∂E / ∂t); E0 is the standard energy consumption, E is the actual energy consumption; Pi is the instantaneous power, Pmax is the rated power; ti is the duration of each power segment; ∂E / ∂t is the rate of change of energy consumption; Ps is the safety index; Ps = (dmin / d0)·exp(-v / v0)·[1 - ∑(1 / ri²)] / n; dmin is the minimum distance from the obstacle, d0 is the safety distance threshold; v is the current speed, v0 is the rated speed; ri is the distance from the i-th obstacle; n is the number of detected obstacles; the weight coefficients W1, W2, W3, and W4 satisfy: W1 + W2 + W3 + W4 = 1.

[0011] In the above embodiments, the control system will comprehensively and quantitatively evaluate the performance of the AGV robot from four dimensions: task execution, navigation accuracy, energy efficiency, and safety through a detailed performance evaluation formula system. The calculation of each index fully considers the key features in actual operation, ensuring the scientificity and accuracy of the performance evaluation.

[0012] In combination with some embodiments of the first aspect, in some embodiments, before the step of receiving the current environmental feature data and operating state data sent by the robot to be optimized, the method further includes: real-time monitoring the performance parameters of multiple AGV robots in the operating state, determining multiple target robots to be optimized based on the performance parameters; obtaining the real-time operating parameters and real-time environmental data of the target robots; when it is determined that the target robots do not meet the preset optimization conditions according to the real-time operating parameters and real-time environmental data, marking the target robots as optimization waiting robots to perform optimization determination again after a preset time.

[0013] In the above embodiments, the control system monitors the running state of the AGV robot in real time, screens the targets to be optimized based on performance parameters, evaluates the optimization conditions in combination with real-time environmental data, and establishes an active optimization trigger mechanism to ensure that the system can timely identify and optimize the AGV robots with abnormal performance, improving the timeliness and pertinence of optimization.

[0014] In combination with some embodiments of the first aspect, in some embodiments, after the step of obtaining the real-time running parameters and real-time environmental data of the target robot, the method further includes: constructing an optimization condition evaluation model based on historical optimization records; calculating an environmental complexity score according to the real-time running parameters and calculating a task urgency score according to the real-time environmental data; inputting the environmental complexity score and the task urgency score into the optimization condition evaluation model to obtain an optimization adaptability score; determining preset optimization conditions based on the optimization adaptability score.

[0015] In the above embodiments, the control system constructs an optimization condition evaluation model, comprehensively considers the environmental complexity and task urgency, realizes intelligent judgment of the optimization timing, avoids interference with the normal operation of the AGV robot caused by improper optimization, and improves the accuracy of the optimization decision.

[0016] In combination with some embodiments of the first aspect, in some embodiments, after the step of sending a control instruction corresponding to the target control parameter to the robot to be optimized to control the robot to be optimized to adjust its running state, the method further includes: recording the optimized performance data of the robot to be optimized after executing the target control parameter; calculating the performance improvement amplitude of the optimized performance data compared with the benchmark performance before optimization, and when the performance improvement amplitude exceeds a preset benchmark threshold, extracting the current environmental characteristics and the corresponding target control parameters as incremental training samples; performing online optimization on the running optimization model based on the incremental training samples.

[0017] In the above embodiments, the control system records the optimized performance data, extracts samples with significant performance improvement for incremental training, realizes the continuous evolution of the optimization model, ensures the adaptability of the model to new environmental characteristics, and improves the generalization performance of the system.

[0018] In combination with some embodiments of the first aspect, in some embodiments, after the step of online optimizing the running optimization model based on incremental training samples, the method further includes: constructing an evaluation function for the running optimization model; the evaluation function is: F = (ΔP / P0)·(1 - σp / σ0)·(1 - |dw / dt| / λ)·(S / S0); ΔP is the amount of performance improvement, P0 is the benchmark performance value; σp is the standard deviation of performance fluctuation, σ0 is the allowable fluctuation threshold; dw / dt is the change rate of model parameters, λ is the upper limit of the learning rate; S is the size of the current sample space, S0 is the size of the initial sample space; when the calculation result of the evaluation function is lower than the preset evaluation threshold, constructing an adversarial sample set based on historical abnormal working condition data, and retraining the running optimization model using the adversarial sample set; adjusting the sampling period and weight distribution of the incremental training samples according to the performance improvement amplitude after retraining.

[0019] In the above embodiments, the control system constructs an evaluation function for the optimization model, comprehensively considers factors such as performance improvement, stability, and learning efficiency, and ensures the effectiveness and robustness of model optimization through adversarial sample retraining and dynamic adjustment of the sampling strategy.

[0020] In a second aspect, an embodiment of the present application provides a control system, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to enable the control system to execute the method described in the first aspect and any possible implementation manner in the first aspect.

[0021] In a third aspect, an embodiment of the present application provides a computer program product containing instructions, when the above computer program product runs on a control system, enabling the above control system to execute the method described in the first aspect and any possible implementation manner in the first aspect.

[0022] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, including instructions, when the above instructions run on a control system, enabling the above control system to execute the method described in the first aspect and any possible implementation manner in the first aspect.

[0023] It can be understood that the control system provided in the second aspect above, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the method provided in the embodiments of the present application. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding method, which will not be elaborated here.

[0024] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. Due to the adoption of the method of extracting performance mutation points based on historical operation data and establishing the mapping relationship between environmental characteristics and optimal control parameters, analyzing the working status, environmental perception, task execution and energy consumption data of multiple AGV robots, identifying the critical moments with significant performance changes and their corresponding environmental and control characteristics, and constructing a dynamic optimization model, the system can accurately capture the influence law of environmental changes on the performance of AGV robots, establish the corresponding relationship between environmental characteristics and optimal control parameters, realize the adaptive adjustment of control parameters, effectively solve the problems in the prior art that the fixed parameter configuration cannot adapt to environmental changes and the performance of AGV robots is easily affected by the environment and fluctuates, and further realize the adaptive ability of the AGV robot control system to different environments, and improve the operation stability and task execution efficiency of AGV robots in complex and changeable environments.

[0025] 2. Due to the adoption of the method of segmenting time series to calculate comprehensive performance indicators and extracting performance mutation points, quantitatively evaluating multi-dimensional indicators such as task completion efficiency, navigation accuracy, energy utilization rate and safety factor of AGV robots, identifying the time points with significant performance changes, and extracting the corresponding environmental characteristics and control parameters to generate training samples, the system can comprehensively evaluate the operation status of AGV robots, accurately locate the critical moments of performance fluctuations, establish the correlation between environmental changes and performance changes, effectively solve the problems in the prior art that there is a lack of scientific performance evaluation criteria and it is difficult to accurately identify the key factors of performance changes, and further realize the precise quantification of the operation performance of AGV robots and the systematic analysis of influencing factors, providing a reliable data basis for optimizing control parameters.

[0026] 3. Due to the adoption of the method of real-time monitoring the performance parameters of AGV robots and determining the optimization timing based on the optimization condition evaluation model, evaluating the necessity and feasibility of optimization by analyzing real-time operation parameters and environmental data, and marking the robots to be optimized according to preset conditions, the system can actively discover AGV robots with abnormal performance, accurately judge the optimization timing, avoid the interference of unnecessary optimization operations on normal operation, effectively solve the problems in the prior art that the optimization trigger mechanism is passive and the improper selection of optimization timing leads to poor optimization effects, and further realize the initiative and accuracy of AGV robot performance optimization, and improve the efficiency and reliability of optimization operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a schematic flow chart of an AGV robot control method in an embodiment of the present application; Figure 2 is another schematic flow chart of an AGV robot control method in an embodiment of the present application; Figure 3 is a schematic structural diagram of an entity device of a control system in an embodiment of the present application. Detailed implementation manners

[0028] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification of the present application, the singular forms "a", "an", "the above", "the", and "this" are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in the present application refers to any or all possible combinations including one or more of the listed items.

[0029] Hereinafter, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as implying or suggesting relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of the present application, unless otherwise specified, the meaning of "a plurality" is two or more.

[0030] For ease of understanding, the application scenarios of the embodiments of the present application are introduced below.

[0031] In a large logistics and warehousing center, there are more than 100 AGV robots working collaboratively. These AGV robots need to complete various handling tasks in a complex and ever-changing environment, including narrow passages between shelves, frequent human-machine interactions in loading and unloading areas, and areas with different road surface conditions. Due to the dynamic changes in environmental conditions, such as the degree of ground slipperiness, lighting conditions, and obstacle distribution, the pre-set control parameters cannot adapt to all working conditions. For example, when a certain AGV robot passes through a wet area, due to the reduced friction coefficient, the pre-set acceleration parameter is too large, resulting in slipping, which affects the positioning accuracy and motion stability. At the same time, there are hardware differences and different degrees of aging among different AGV robots, and the same control parameters perform differently on different AGV robots. These factors together lead to low overall operating efficiency and unstable task completion quality of the AGV robot group.

[0032] In related technologies, the AGV robot control system usually adopts a pre-set parameter configuration table or a parameter adjustment method based on simple rules. This method requires manual experience for parameter configuration and cannot adapt to dynamically changing environmental conditions, resulting in low operating efficiency and unstable performance of AGV robots. The scenario of using the AGV robot control method in related technologies is introduced below.

[0033] A certain factory adopts the traditional AGV robot control method, using a fixed parameter configuration table to adapt to different scenarios. Managers divide the working area into multiple sub-areas according to experience and set specific control parameter groups for each area. When an AGV robot enters a new area, the corresponding parameter configuration is automatically switched. However, this method has obvious defects: First, the parameter configuration table is set based on offline statistics and experience and cannot reflect the real-time changes of the environment; Second, the area division is relatively rough and the local differences within the same area cannot be reflected; Third, the parameter switching is discrete and it is easy to cause discontinuous movement at the area boundary. For example, at the junction of the shelf area and the open area, the AGV robot frequently switches parameters, resulting in movement jitter and affecting the stability of the transported items. In addition, when new AGV robots are added or the environment changes, it is necessary to manually recalibrate and configure the parameters, with high maintenance costs and slow response speed.

[0034] However, by adopting the AGV robot control method in the embodiment of the present application, by establishing the mapping relationship between the environmental characteristics and the optimal control parameters, the real-time optimization of the control parameters is realized, which not only improves the operation efficiency of the AGV robot, but also realizes the automatic accumulation and group sharing of parameter optimization experience. The following describes the scenario using the AGV robot control method in the present application.

[0035] After adopting this solution, the AGV robot group realizes adaptive parameter optimization. The control system monitors the performance indicators of each AGV robot in real time. When it detects that the performance of a certain AGV robot decreases, it immediately starts the optimization process. For example, when an AGV robot is performing tasks in the loading and unloading area, the system finds that its positioning accuracy decreases, and immediately analyzes the current environmental characteristics (high personnel density, large light changes) and operating status, and generates more suitable control parameters (reducing the maximum speed, increasing the obstacle avoidance distance) through the optimization model. After optimization, the positioning accuracy of the AGV robot is significantly improved, and at the same time, the smoothness of the movement is maintained. More importantly, the system uses this optimization experience as a training sample to continuously improve the optimization model. When other AGV robots encounter similar situations, they can directly use the optimized parameters, realizing the rapid transfer of experience and group learning.

[0036] It can be seen that by adopting the AGV robot control method in the embodiment of the present application, while realizing the adaptive adjustment of the AGV robot control parameters, it can also effectively solve the problem of performance fluctuations caused by dynamic environmental changes, and then realize the intelligent control and continuous optimization of the AGV robot group.

[0037] For easy understanding, the following combines the above scenarios to describe the process of the method provided in this embodiment. Please refer to Figure 1 , which is a schematic flowchart of the AGV robot control method in the embodiment of the present application.

[0038] S101. Obtain the historical operation data of multiple AGV robots.

[0039] Among them, the historical operation data represents various data records generated during the past operation of the AGV robot, including working status data, environmental perception data, task execution data, and energy consumption data. The working status data refers to the motion parameters of the AGV robot such as speed, acceleration, steering angle, etc., and the device parameters such as battery power and component status. The environmental perception data is used to represent the environmental information obtained by the AGV robot through various sensors, such as lidar scan data, visual image data, obstacle distribution data, etc. The task execution data represents the performance indicators related to the task, such as task completion time, path planning, positioning accuracy, etc. The energy consumption data refers to the energy consumption situation of the AGV robot during operation, including parameters such as instantaneous power, average power, and energy utilization efficiency.

[0040] When initializing the operation optimization model, the control system needs to obtain historical data as the training basis. Specifically, the control system first establishes a data connection with the AGV robot management platform and extracts the historical operation data within a specified time range from the database. The control system preprocesses the data, including operations such as data cleaning, outlier processing, and time alignment, to ensure the quality and consistency of the data. At the same time, the control system will perform structured processing on the data, organizing different types of data according to a predefined format for subsequent analysis and use. In addition, the control system will also perform preliminary statistical analysis on the data, calculating the basic statistical characteristics of each indicator to provide a reference benchmark for subsequent performance evaluation.

[0041] In some embodiments, the acquisition and preprocessing of historical data can be achieved in multiple ways: Optionally, the control system can establish a real-time data pipeline, write the data generated by the operation of the AGV robot into a distributed database in real time, and periodically execute data cleaning and feature extraction tasks to form a standardized historical data set; during the data cleaning process, a sliding window median filter is used to remove outliers, linear interpolation is used to supplement missing data, and finally, principal component analysis is used for dimensionality reduction to extract key features. Optionally, the control system can also adopt a batch processing method, periodically collect the original data from the local storage of the AGV robot, and perform data transformation and loading through an ETL tool; wavelet transform is used for denoising in the data preprocessing stage, time series analysis methods are used to detect and correct abnormal data, and finally, a structured data table is constructed through feature engineering. It can be understood that other data collection and preprocessing methods can also be used to achieve the acquisition and standardization of historical data, which is not limited here.

[0042] S102. Based on the performance evaluation model, extract the control parameters and environmental characteristics corresponding to the performance mutation points from the historical operation data, and generate training samples with performance annotations.

[0043] Among them, the performance evaluation model represents a mathematical model for evaluating the comprehensive operation performance of an AGV robot, including calculation formulas and weight coefficients of multiple performance indicators. The performance mutation point is the time point at which the change rate of the performance indicators generated by the performance evaluation model exceeds a preset stable threshold, referring to the time node at which the performance indicators of the AGV robot change significantly. The control parameters are used to represent various adjustable parameters that affect the operation of the AGV robot, such as speed limit, acceleration limit, obstacle avoidance distance, etc. The environmental characteristics represent various characteristic quantities that describe the operation environment of the AGV robot, such as ground conditions, obstacle density, lighting conditions, etc. The performance annotation refers to the quantitative description of the degree of performance change.

[0044] After obtaining the historical data, the control system needs to identify the critical performance change moments and extract the corresponding feature information. Specifically, the control system first segments the historical data according to the time series, and applies the performance evaluation model to calculate the comprehensive performance indicators for the data in each time period. The control system identifies the performance mutation points by analyzing the change trend of the performance indicators. For each performance mutation point, the control system extracts the environmental feature data and control parameter data within a certain time window before and after it, and calculates the amplitude of the performance change as the performance annotation. Finally, the control system integrates the extracted feature data and performance annotation into training samples in a standard format.

[0045] It should be noted that the training of the performance evaluation model is based on a large amount of historical operation data, comprehensively considering multi-dimensional data such as working status, environmental perception, task execution, and energy consumption. The model establishes evaluation criteria for each indicator through statistical analysis of historical data, and determines the importance of different indicators through weight optimization. During the training process, the weight coefficients are adjusted to make the evaluation results of the model consistent with the actual operation effects. The model structure of the performance evaluation model includes four core evaluation dimensions: task execution performance (Pt) considering the execution time ratio and motion smoothness; navigation accuracy (Pn) evaluating the trajectory deviation and positioning error; energy efficiency index (Pe) calculating the energy utilization efficiency; safety index (Ps) measuring the safety distance from obstacles. These indicators are quantitatively calculated through carefully designed mathematical formulas, and finally the comprehensive performance score is obtained through weighted summation. For example, when the AGV robot performs a precision part transportation task, the model will give a higher weight to the navigation accuracy. During operation, the model continuously receives the real-time operation data of the AGV robot, calculates various performance indicators, and outputs the comprehensive performance score reflecting the current operation state. This score is used to identify the performance mutation points and also serves as the evaluation basis for the optimization effect. By continuously monitoring the change trend of the performance score, the system can timely detect performance anomalies and trigger the optimization process.

[0046] In some embodiments, the identification and feature extraction of performance mutation points can be achieved in various ways: Optionally, the control system can use the sliding window method to calculate the short-term and long-term means of performance metrics, and mark it as a mutation point when the change rate of the short-term mean relative to the long-term mean exceeds the threshold; then use the principal component analysis method to extract the main features from the environmental data, record the differences in control parameters before and after the mutation at the same time, and finally set the sample weights based on the mutation amplitude. Optionally, the control system can also adopt the method based on variational mode decomposition to decompose the performance metric sequence into components of different scales, identify the mutation points by analyzing the energy changes of the high-frequency components; then use the autoencoder to extract the low-dimensional representation of the environmental features from the original data, and generate training samples by combining the correlation analysis of control parameters. It can be understood that other data analysis methods can also be used to achieve the detection and feature extraction of performance mutation points, which are not limited here.

[0047] S103. Construct a mapping relationship between environmental features and optimal control parameters based on the training samples to generate an operation optimization model.

[0048] Among them, the mapping relationship represents the mathematical correspondence between environmental features and optimal control parameters, and is used to predict the optimal control strategy in a specific environment. The operation optimization model refers to a model trained by machine learning methods, which can automatically generate optimal control parameters according to the input environmental features. The optimal control parameters represent the parameter combination that can make the AGV robot achieve the best performance in a specific environment. The training samples refer to the data set containing environmental features, control parameters and performance annotations.

[0049] After obtaining the training samples with performance annotations, the control system needs to construct a mapping model between environmental features and control parameters. Specifically, the control system first performs data standardization processing on the training samples to convert features of different dimensions to the same scale range. Then, the control system uses a deep neural network to construct a basic model structure, including a feature extraction layer, a mapping layer and a parameter generation layer. The control system uses a loss function with performance weights for model training and optimizes the model parameters through the backpropagation algorithm. During the training process, the control system adopts the cross-validation method to evaluate the model performance and improves the generalization ability of the model by adjusting the network structure and hyperparameters.

[0050] It should be noted that the training process of the operation optimization model takes the operation data of multiple AGV robots in different environments as input, including environmental feature data (such as ground conditions, obstacle distribution) and corresponding control parameter data, and uses the performance difference as the optimization goal for training. The model adopts a deep neural network structure, optimizes the network parameters through the backpropagation algorithm, and uses a weighted loss function to ensure that the model can accurately predict the optimal control parameters. During the training process, the model performance is continuously evaluated and adjusted through cross-validation to ensure that the model has good generalization ability. The model architecture of the operation optimization model adopts a multi-layer neural network structure, including a feature extraction layer for processing high-dimensional features of environmental data, a mapping layer for establishing a non-linear relationship between features and parameters, and a parameter generation layer for outputting the optimal control parameters. An attention mechanism is introduced into the model, which can adaptively focus on important environmental features and improve the response ability to key factors. For example, in a complex environment, the model will pay more attention to the obstacle distribution features; when running at high speed, it will pay more attention to the road surface condition features. In practical applications, the model receives the environmental feature data and operation status data of the current AGV robot as input, and after feature extraction and multi-layer mapping, outputs the control parameter combination most suitable for the current scenario. These parameters are then used to adjust the operation status of the AGV robot to achieve real-time adaptive control optimization. For example, when the AGV robot enters a slippery road surface, the model will automatically reduce the speed and acceleration parameters based on the environmental features to improve safety.

[0051] S104. Receive the current environmental feature data and operation status data sent by the robot to be optimized.

[0052] Among them, the robot to be optimized refers to the AGV robot device that currently needs to optimize the control parameters. The current environmental feature data refers to the environmental information sensed by the AGV robot in real time, including features such as the distribution of surrounding obstacles and ground conditions. The operation status data represents the current operation parameters of the AGV robot, such as speed, position, power and other status information.

[0053] The control system needs to receive and process the data uploaded by the AGV robot in real time. Specifically, the control system establishes a real-time communication connection with the robot to be optimized and receives the data packets regularly sent by the AGV robot. The control system parses and validates the received data to ensure the integrity and validity of the data. At the same time, the control system preprocesses the environmental feature data and operation status data respectively, including operations such as data format conversion, outlier detection, and feature extraction. In addition, the control system also maintains a short-term data cache for storing recent data history for trend analysis.

[0054] In some embodiments, data reception and processing can be achieved in various ways: Optionally, the control system can use a communication architecture based on message queues to establish a highly reliable data transmission channel, optimize the transmission efficiency through message serialization and compression, perform data recombination and verification at the receiving end, and finally store the processed data in a real-time database. Optionally, the control system can also adopt a real-time communication protocol based on WebSocket to achieve two-way data streams, use a heartbeat mechanism to maintain the connection status, ensure data synchronization through data frame analysis, and finally perform data conversion and feature extraction. It can be understood that other communication methods can also be used to achieve data reception and processing, which are not limited herein.

[0055] S105. Input the current environmental characteristic data and operating status data into the operation optimization model to obtain target control parameters.

[0056] Among them, the target control parameters represent the optimal parameter combination that is calculated by the operation optimization model and suitable for the current environment. The operation optimization model refers to a trained machine learning model used to generate control parameters. The current environmental characteristic data and operating status data refer to the preprocessed real-time data.

[0057] The control system needs to use the optimization model to generate control parameters suitable for the current environment. Specifically, the control system first organizes the preprocessed environmental characteristic data and operating status data in the format required by the model. Then, the control system calls the trained operation optimization model, inputs the data into the model for forward calculation. The model performs feature transformation and mapping through a multi-layer network structure according to the input features, and finally outputs the predicted target control parameters. The control system performs post-processing on the parameters output by the model, including operations such as parameter range constraint and smoothing processing, to ensure the executability of the parameters.

[0058] In some embodiments, the generation of target control parameters can be achieved in various ways: Optionally, the control system can adopt an integrated inference method, use multiple trained models for prediction simultaneously, fuse the output results of multiple models through a weighted average or voting mechanism, and finally perform parameter constraint and smoothing processing. Optionally, the control system can also use a method based on reinforcement learning, dynamically adjust the parameter generation strategy according to the historical optimization effect, evaluate the potential effect of the parameters through a value network, and finally select the target parameter combination. It can be understood that other inference methods can also be used to achieve the generation of target control parameters, which are not limited herein.

[0059] S106. Send the control instruction corresponding to the target control parameter to the robot to be optimized to control the robot to be optimized to adjust its operating status.

[0060] Among them, the control instruction represents an operation command containing target control parameters, which is used to guide the AGV robot to adjust its control parameters. The running state adjustment refers to the process in which the AGV robot modifies its running parameters according to the received control instruction.

[0061] The control system needs to safely and reliably send the optimized control parameters to the AGV robot. Specifically, the control system first converts the target control parameters into a standard control instruction format, including information such as parameter values, execution timings, and safety verifications. Then, the control system sends the control instruction to the robot to be optimized through a reliable communication channel. The control system will monitor the sending status of the instruction to ensure that the instruction is correctly received by the AGV robot. At the same time, the control system will record the history of parameter distribution for subsequent effect evaluation. In addition, the control system will also monitor the state changes of the AGV robot after the parameter adjustment to ensure the safety of the adjustment process.

[0062] In some embodiments, the issuance and execution of control instructions can be implemented in various ways: Optionally, the control system can adopt a step-by-step parameter adjustment strategy, decompose the parameter change amount into multiple small steps, and gradually adjust to the target value. Safety verification is performed for each step to ensure the smooth and controllable adjustment process. Optionally, the control system can also use a parameter distribution scheme with a rollback mechanism to monitor the state of the AGV robot in real time during the parameter adjustment process and quickly restore to the previous parameter configuration if an abnormality occurs. It can be understood that other control strategies can also be adopted to achieve the safe distribution and adjustment of parameters, which are not limited here.

[0063] In the above embodiments, the optimization model is mainly optimized based on the performance improvement of a single AGV robot. In actual applications, the group collaboration effect can also be considered. The optimization experiences of multiple AGV robots are comprehensively analyzed, common features are extracted, and the continuous evolution of swarm intelligence is achieved. The following supplements the scenario of this embodiment.

[0064] The system continuously accumulates optimization experiences during long-term operation and forms a more intelligent control strategy. For example, the system discovers that in the morning, due to the low temperature, the battery performance of the AGV robot is limited, and automatically adjusts the energy distribution strategy; in rainy days, the system predicts the slippery ground areas based on historical data and adjusts the control parameters for the AGV robots passing through these areas in advance. Further, the system begins to learn optimization strategies at the task level. For example, for the task of transporting precision parts, more conservative control parameters are automatically selected to ensure stability; for the task of emergency material replenishment, the speed parameters are increased on the premise of ensuring safety. The system can also customize personalized parameter optimization strategies for each AGV robot based on long-term data such as battery life prediction and component wear status to achieve the optimization of the overall performance of the group.

[0065] After combining the above scenarios, the following further describes the more specific process of the method provided in this embodiment. Please refer to Figure 2 which is another process schematic diagram of the AGV robot control method in the embodiment of the present application.

[0066] S201. Obtain the historical operation data of multiple AGV robots.

[0067] Referring to step S101, the control system will obtain the historical operation data.

[0068] S202. Segment the historical operation data according to the time series, and calculate the comprehensive performance index of the AGV robot in each time period based on the performance evaluation formula.

[0069] Among them, the time series segmentation refers to dividing the continuous historical data into multiple discrete data segments according to the time interval. The performance evaluation formula represents a mathematical model for calculating the comprehensive performance of the AGV robot, including the calculation formulas and weight coefficients of multiple performance indicators. The comprehensive performance index is a quantitative index reflecting the overall operation status of the AGV robot, which is weighted and composed of multiple sub-indicators such as task completion efficiency, navigation accuracy, energy utilization rate, and safety factor. The data segment refers to the set of historical operation data within a specific time window.

[0070] After obtaining the preprocessed historical data, the control system needs to perform time segmentation and performance calculation. Specifically, the control system first determines the appropriate time window size, and sets the segmentation interval based on data characteristics and task characteristics. For the data in each time period, the control system calculates the task completion efficiency index, navigation accuracy index, energy utilization rate index, and safety factor index respectively. Then, the control system combines the indicators into a comprehensive performance index according to the preset weight coefficients. During the calculation process, the control system will consider the time series characteristics of the data, use the sliding window method for continuous calculation, and record the performance index values and their constituent elements at each time point.

[0071] In some embodiments, time series segmentation and performance calculation can be implemented in various ways: Optionally, the control system adopts an adaptive segmentation algorithm, dynamically adjusts the time window size according to the data change characteristics, uses a multi-scale analysis method to extract the characteristic statistics of each time period, and fuses multiple performance indicators through a fuzzy comprehensive evaluation method; Optionally, the control system uses an event-based segmentation method, divides the data segments with the completion of tasks or state transitions as the boundaries, determines the index weights by the analytic hierarchy process, and calculates the comprehensive performance score through grey relational analysis. It can be understood that other data analysis methods can also be used to implement time series segmentation and performance evaluation, which are not limited herein.

[0072] In some embodiments, in the control system, the performance evaluation formula is: P = W1·Pt + W2·Pn + W3·Pe + W4·Ps; Pt is the task execution performance index; Pt = (T0 / T)·[1 + ∫(dv / dt)²dt / D]·(1 - θ / θmax); T0 is the standard task time, T is the actual execution time; D is the task distance; θ is the path curvature, θmax is the allowable maximum curvature; dv / dt is the rate of change of velocity, used to characterize the motion smoothness; Pn is the navigation accuracy index; Pn = (1 - |ΔD| / D0)·exp(-σ² / σ0²)·(1 + d²Δp / dt²)^(-1); ΔD is the actual trajectory deviation; D0 is the allowable maximum deviation; σ is the standard deviation of the positioning error; σ0 is the reference error value; d²Δp / dt² is the second derivative of position, used to characterize the motion jitter; Pe is the energy efficiency index; Pe = (E0 / E)·[1 - ∑(Pi·ti) / (Pmax·T)]·(1 - ∂E / ∂t); E0 is the standard energy consumption, E is the actual energy consumption; Pi is the instantaneous power, Pmax is the rated power; ti is the duration of each power segment; ∂E / ∂t is the rate of change of energy consumption; Ps is the safety index; Ps = (dmin / d0)·exp(-v / v0)·[1 - ∑(1 / ri²)] / n; dmin is the minimum distance from the obstacle, d0 is the safety distance threshold; v is the current speed, v0 is the rated speed; ri is the distance from the i-th obstacle; n is the number of detected obstacles; The weight coefficients W1, W2, W3, and W4 satisfy: W1 + W2 + W3 + W4 = 1.

[0073] Among them, the performance evaluation formula represents a mathematical model for quantitatively calculating the comprehensive operation performance of the AGV robot. The task execution performance index Pt refers to a comprehensive index for measuring the task completion efficiency of the AGV robot, and is evaluated by the standard time ratio, motion smoothness, and path rationality. The navigation accuracy index Pn is used to represent the evaluation index of the motion accuracy of the AGV robot, and is calculated by combining the trajectory deviation, positioning error, and motion stability. The energy efficiency index Pe represents the evaluation index of the energy utilization efficiency of the AGV robot, and is measured by the energy consumption ratio, power utilization rate, and energy consumption change trend. The safety index Ps refers to the evaluation index reflecting the running safety degree of the AGV robot, and is evaluated by considering the obstacle distance, motion speed, and surrounding environment complexity. The weight coefficients W1, W2, W3, and W4 are used to represent the importance of each index in the comprehensive performance.

[0074] When conducting performance evaluation, the control system needs to comprehensively calculate performance indicators in multiple dimensions. Specifically, the control system first collects the task execution data of the AGV robot, including parameters such as actual execution time, speed curve, and path curvature, and calculates the task execution performance indicator Pt. Then, based on the positioning system and sensor data, the control system calculates the trajectory deviation and positioning error, and combines the motion stability to evaluate the navigation accuracy indicator Pn. At the same time, the control system calculates the energy consumption ratio and power distribution by collecting the data of the battery management system to evaluate the energy efficiency indicator Pe. In addition, based on the obstacle detection data, the control system calculates the minimum safety distance and environmental complexity to evaluate the safety indicator Ps. Finally, the control system weights and sums up the various indicators according to the weights to obtain the comprehensive performance evaluation value P. During the calculation process, the control system will monitor the changes of various parameters in real time to ensure the accuracy and timeliness of the calculation results.

[0075] In some embodiments, the calculation of the performance evaluation formula can be implemented in multiple ways: Optionally, the control system uses an adaptive weight adjustment method to establish a weight optimization model based on historical performance data, dynamically adjusts the weight coefficients through the gradient descent algorithm, introduces index correlation analysis to avoid repeated calculations, and finally uses a sliding window method to achieve continuous evaluation; Optionally, the control system adopts a fuzzy comprehensive evaluation method, converts each indicator into a membership function, calculates the comprehensive score through fuzzy inference rules, combines time series analysis to identify the performance change trend, and introduces confidence interval analysis to evaluate the reliability of the results. It can be understood that other mathematical methods can also be used to implement the calculation and comprehensive evaluation of performance indicators, which are not limited here.

[0076] S203. Identify the time point when the change rate of the comprehensive performance indicator exceeds the preset stability threshold as the performance mutation point.

[0077] Among them, the change rate represents the change amplitude between adjacent time points of the comprehensive performance indicator. The preset stability threshold refers to the reference value used to judge the significance of performance changes, which is determined by the system according to the statistical characteristics of historical data. The performance mutation point is used to represent the time node when the performance of the AGV robot changes significantly. The time point refers to the specific sampling moment in the data sequence.

[0078] After obtaining the comprehensive performance indicators for each time period, the control system needs to identify the critical moments of performance change. Specifically, the control system first calculates the rate of change of the performance indicators between adjacent time points, and uses the difference or derivative method to quantify the change trend. Then, the control system compares the calculated rate of change with a preset stability threshold. When the rate of change at a certain time point exceeds the threshold, the control system marks this point as a candidate mutation point. For the candidate mutation point, the control system further analyzes the performance change trend within a certain time range before and after it, and confirms the real performance mutation point through pattern recognition methods. At the same time, the control system also records the characteristic information such as the type and intensity of the mutation point.

[0079] In some embodiments, the identification of the performance mutation point can be achieved in various ways: Optionally, the control system adopts a mutation detection method based on wavelet transform, performs multi-scale decomposition on the performance indicator sequence, analyzes the energy distribution of different frequency components, determines the mutation position through threshold decision, and filters out pseudo-mutation points by combining morphological features; Optionally, the control system uses a method based on statistical tests to construct a probability distribution model of the performance indicators, judges the mutation points of the data distribution through hypothesis testing, and analyzes the state transition characteristics using Markov chains. It can be understood that other data analysis methods can also be used to detect and confirm the performance mutation points, which are not limited here.

[0080] S204. Extract the environmental feature data and control parameter data within a preset time window before and after the performance mutation point, and calculate the performance difference before and after the performance mutation point.

[0081] Among them, the preset time window refers to a period of time centered on the performance mutation point. The environmental feature data represents various characteristic quantities describing the operating environment of the AGV robot. The control parameter data refers to various adjustable parameters that affect the operation of the AGV robot. The performance difference is used to represent the change range of the performance level before and after the mutation point.

[0082] After determining the performance mutation point, the control system needs to analyze the relevant environmental and control factors. Specifically, the control system first determines the appropriate size of the time window and extracts the data within the corresponding time range before and after the mutation point. For the environmental feature data, the control system performs feature extraction and dimensionality reduction processing to obtain the key environmental description factors. For the control parameter data, the control system analyzes the parameter change situation and identifies the key parameters related to the performance change. Then, the control system calculates the average performance level before and after the mutation point to obtain the performance difference. At the same time, the control system also analyzes the direction and rate of the performance change to provide a basis for subsequent optimization.

[0083] In some embodiments, feature extraction and difference calculation can be achieved in various ways: Optionally, the control system uses an adaptive sampling strategy with a sliding window, dynamically adjusts the window size according to the data change characteristics, extracts the main components of the environmental features through principal component analysis, calculates the correlation matrix of the control parameters at the same time, and finally calculates the performance difference using the weighted average method; Optionally, the control system adopts a deep learning-based feature extraction method, uses an autoencoder to perform feature learning on the environmental data, identifies important control parameters through an attention mechanism, and combines time series analysis to calculate the performance change trend. It can be understood that other data analysis methods can also be used to achieve feature extraction and performance difference calculation, which are not limited here.

[0084] S205. Use the performance difference as the performance annotation, integrate the environmental features and control parameters corresponding to the performance annotation, and generate training samples.

[0085] Among them, the performance annotation represents a quantitative description of the degree of performance change. A training sample refers to a structured data record containing environmental features, control parameters, and performance annotations. The integration process refers to the process of organizing multi-dimensional data into a standard format.

[0086] After obtaining the performance difference and related features, the control system needs to construct standardized training samples. Specifically, the control system first normalizes the performance difference and converts it into a standard scoring interval. Then, the control system organizes the environmental feature data and control parameter data according to a predefined data structure to establish a feature vector and a parameter vector. For each group of data, the control system associates the corresponding performance annotation to form a complete training sample. At the same time, the control system will perform a quality assessment on the samples, including feature integrity checks, numerical range verification, etc. In addition, the control system will also establish a sample index to facilitate quick access to data during subsequent training processes.

[0087] In some embodiments, the generation of training samples can be achieved in various ways: Optionally, the control system adopts a multi-level feature organizational structure, classifies and stores environmental features according to physical meanings, represents control parameters using sparse coding, performs quantile normalization on performance annotations, and finally constructs a sample description file containing metadata; Optionally, the control system uses a tensor representation method, converts multi-dimensional feature data into a unified tensor format, uses attention weights to annotate the importance of features, and expands sample diversity through data augmentation techniques. It can be understood that other data organization methods can also be used to achieve the generation and management of training samples, which are not limited here.

[0088] S206. Based on the training samples, construct a mapping relationship between environmental features and optimal control parameters to generate an operation optimization model.

[0089] Referring to step S103, the control system will construct an operation optimization model.

[0090] S207. Monitor the performance parameters of multiple AGV robots in the running state in real time, and determine multiple target robots to be optimized based on the performance parameters.

[0091] Among them, the performance parameters represent various indicators that reflect the running status of the AGV robot in real time. The target robot to be optimized refers to the AGV robot device whose current performance needs to be optimized. The real-time monitoring process refers to the process of continuously collecting and analyzing the running data of the AGV robot.

[0092] The control system needs to continuously monitor the running state of the AGV robot and identify the optimization requirements. Specifically, the control system establishes a real-time data acquisition channel and continuously receives the performance parameter data uploaded by each AGV robot. For each AGV robot, the control system calculates its current comprehensive performance index and compares it with the historical benchmark value. When it is found that the performance index of a certain AGV robot is continuously lower than the expected value or shows obvious fluctuations, the control system marks it as a target to be optimized. At the same time, the control system will sort the priorities of multiple targets to be optimized to ensure the reasonable allocation of optimization resources.

[0093] In some embodiments, performance monitoring and target recognition can be achieved in various ways: Optionally, the control system adopts a monitoring framework based on a state machine, judges the running state of the AGV robot through multi-dimensional thresholds of performance indicators, uses a sliding time window to analyze the performance trend, and combines the task urgency to evaluate the optimization priority; Optionally, the control system uses an anomaly detection algorithm, establishes a normal fluctuation range model of performance indicators, identifies abnormal patterns through statistical analysis methods, and considers the impact of environmental factors on performance at the same time. It can be understood that other monitoring methods can also be used to evaluate the performance of the AGV robot and determine the optimization target, which is not limited here.

[0094] S208. Obtain the real-time running parameters and real-time environmental data of the target robot.

[0095] Among them, the real-time running parameters represent the current running state information of the AGV robot, including dynamic parameters such as speed, acceleration, position, and battery level. The real-time environmental data refers to the environmental information collected in real time through sensors, such as obstacle distribution and road surface conditions. The target robot refers to the AGV robot device that has been determined to need optimization.

[0096] After determining the optimization objectives, the control system needs to obtain detailed status information. Specifically, the control system establishes a dedicated data communication channel with the target robot, sets a high data sampling frequency, and receives operation parameters and environmental data in real time. The control system processes the received data in real time, including operations such as data parsing, format conversion, and anomaly detection. At the same time, the control system maintains a short-term data cache to store historical data for a recent period of time, which is used to analyze the parameter change trend. In addition, the control system also preprocesses the data to ensure that the data quality meets the optimization requirements.

[0097] In some embodiments, real-time data acquisition and processing can be achieved in multiple ways: Optionally, the control system adopts a real-time communication architecture based on message queues, improves the transmission efficiency through data compression and priority mechanisms, uses multi-level caches to manage recent data, and performs real-time data quality evaluation and anomaly handling simultaneously; Optionally, the control system uses a distributed data acquisition framework to perform preliminary data preprocessing and feature extraction at the AGV robot end, transmits data changes through incremental updates, and combines data synchronization mechanisms to ensure information consistency. It can be understood that other communication methods can also be used to achieve real-time data acquisition and processing, which are not limited here.

[0098] In some embodiments, the control system constructs an optimization condition evaluation model based on historical optimization records; calculates the environmental complexity score according to real-time operation parameters and calculates the task urgency score according to real-time environmental data; inputs the environmental complexity score and the task urgency score into the optimization condition evaluation model to obtain an optimization adaptability score; determines preset optimization conditions based on the optimization adaptability score.

[0099] Among them, the optimization condition evaluation model refers to a mathematical model used to judge whether the AGV robot is suitable for parameter optimization. The environmental complexity score is an index that quantitatively describes the difficulty level of the current environment, including factors such as obstacle density and road surface conditions. The task urgency score is used to represent the time pressure degree of the current task, considering factors such as the task deadline and the remaining workload. The optimization adaptability score represents the degree to which the current state of the AGV robot is suitable for parameter optimization. The preset optimization condition is the standard threshold for triggering parameter optimization.

[0100] Before performing parameter optimization, the control system needs to evaluate the suitability of the optimization timing. Specifically, the control system first establishes an evaluation model based on historical optimization records, including a feature extraction layer, a scoring calculation layer, and a decision output layer. Then, the control system collects environmental data and task data in real time, and calculates the complexity score of the current environment and the urgency score of the task. The control system inputs these scores into the evaluation model to obtain an adaptability score reflecting the suitability of the current state for optimization. Based on the adaptability score, the control system dynamically adjusts the optimization condition threshold to ensure that the optimization process is triggered at the appropriate time. Throughout the process, the control system continuously updates the evaluation model to continuously improve the accuracy of the evaluation.

[0101] It should be noted that the model training process of the optimization condition evaluation model is based on historical optimization records, and analyzes the impact of environmental complexity and task urgency on the optimization effect. By statistically analyzing the characteristics of successful optimization cases, the correlation between environmental conditions, task characteristics, and optimization adaptability is established. During the training process, by continuously adjusting the evaluation parameters, the judgment accuracy of the model for the optimization timing is improved. The model adopts a multi-layer structure, including a feature extraction layer for processing environmental and task data, a scoring calculation layer for comprehensively evaluating the optimization conditions, and a decision output layer for giving the final judgment. The internal of the model uses a fuzzy inference system, and makes decisions through a combination of expert rules and data-driven methods. For example, in a crowded area, the model will comprehensively consider the current task urgency and environmental complexity to determine whether it is suitable for parameter optimization. When in use, the model receives the environmental complexity score and task urgency score calculated in real time, and outputs the optimization adaptability score through comprehensive evaluation. This score is used to judge whether it is suitable for parameter optimization currently, avoiding optimization operations at inappropriate times, and ensuring the safety and effectiveness of the optimization process.

[0102] S209. When it is determined that the target robot does not meet the preset optimization conditions according to the real-time operation parameters and real-time environmental data, mark the target robot as an optimization waiting robot to perform the optimization determination again after a preset time.

[0103] Among them, the preset optimization conditions represent the environmental and state requirements for the AGV robot to perform parameter optimization. The optimization waiting robot refers to the AGV robot device that is temporarily not suitable for parameter optimization. The marking process refers to the operation of updating the AGV robot state to waiting for optimization.

[0104] The control system needs to evaluate whether the AGV robot meets the optimization conditions. Specifically, the control system first checks whether the real-time operating parameters are within the safe range, including key indicators such as speed, acceleration, and battery level. Then, the control system analyzes the real-time environmental data to evaluate whether the current environment is suitable for parameter adjustment. When it is found that the AGV robot is in a state or environment that is not suitable for optimization, the control system updates its state to optimization waiting. At the same time, the control system will record the specific reasons for the waiting and set corresponding trigger conditions to automatically resume the optimization process when the conditions are met.

[0105] In some embodiments, the evaluation of optimization conditions and state management can be achieved in various ways: Optionally, the control system uses a multi-dimensional condition evaluation matrix, comprehensively considering operating safety, environmental complexity, and task urgency, determines the optimization timing through fuzzy logic reasoning, and establishes a state transition trigger mechanism; Optionally, the control system adopts a rule-based judgment system, sets up a multi-level condition check process, dynamically adjusts the judgment threshold in combination with historical optimization experience, and simultaneously maintains the priority sorting of the optimization waiting queue. It can be understood that other evaluation methods can also be used to achieve the judgment of optimization conditions and state management, which are not limited here.

[0106] S210. Receive the current environmental feature data and operating status data sent by the robot to be optimized.

[0107] Referring to step S104, the control system will receive the data information sent by the robot to be optimized.

[0108] S211. Input the current environmental feature data and operating status data into the operation optimization model to obtain the target control parameters.

[0109] Referring to step S105, the control system will determine the target control parameters.

[0110] S212. Send the control instruction corresponding to the target control parameter to the robot to be optimized to control the robot to be optimized to adjust its operating state.

[0111] Referring to step S106, the control system will control the robot to be optimized to adjust its operation.

[0112] In some embodiments, the control system will record the optimization performance data of the robot to be optimized after executing the target control parameter; calculate the performance improvement amplitude of the optimization performance data compared with the benchmark performance before optimization, and when the performance improvement amplitude exceeds the preset benchmark threshold, extract the current environmental features and the corresponding target control parameters as incremental training samples; perform online optimization on the operation optimization model based on the incremental training samples.

[0113] Among them, the optimized performance data represents the operation performance record of the AGV robot after adopting the target control parameters. The performance improvement rate refers to the degree of improvement relative to the baseline performance after optimization. The baseline performance is used to represent the performance level before optimization. The incremental training samples represent the new data used for continuous optimization of the model. The preset baseline threshold is the standard value for judging the significance of the optimization effect. Online optimization represents the real-time update process of the model parameters.

[0114] After the control system completes the parameter optimization, it needs to evaluate the optimization effect and continuously improve the model. Specifically, the control system first records the operation data of the AGV robot after adopting the target control parameters, including performance indicators, environmental characteristics, and control parameters. Then, the control system calculates the performance difference before and after optimization and evaluates the significance of the optimization effect. When the performance improvement exceeds the preset threshold, the control system extracts the feature data and corresponding target parameters of the current scenario and constructs new training samples. The control system uses these incremental samples to update the optimization model, continuously improving the adaptability and accuracy of the model. At the same time, the control system will maintain the diversity of the sample library to ensure the generalization ability of the model.

[0115] In some embodiments, the incremental optimization of the model can be achieved in various ways: Optionally, the control system adopts an online learning algorithm, updates the model parameters through the stochastic gradient descent method, uses the experience replay mechanism to process historical samples, introduces a regularization term to prevent overfitting, and finally evaluates the model performance through cross-validation; Optionally, the control system uses the transfer learning method, maintains the underlying feature extraction ability, only fine-tunes the high-level mapping parameters, improves the feature representation ability through contrastive learning, and finally adopts an integration strategy to fuse the old and new models. It can be understood that other learning methods can also be used to achieve the continuous optimization of the model, which is not limited here.

[0116] In some embodiments, the control system will construct an evaluation function for the operation optimization model; the evaluation function is: F=(ΔP / P0)·(1-σp / σ0)·(1-|dw / dt| / λ)·(S / S0); ΔP is the performance improvement amount, P0 is the baseline performance value; σp is the standard deviation of performance fluctuation, σ0 is the allowable fluctuation threshold; dw / dt is the model parameter change rate, λ is the upper limit of the learning rate; S is the current sample space size, S0 is the initial sample space size; when the calculation result of the evaluation function is lower than the preset evaluation threshold, an adversarial sample set is constructed based on the historical abnormal working condition data, and the operation optimization model is retrained using the adversarial sample set; according to the performance improvement rate after retraining, the sampling period and weight distribution of the incremental training samples are adjusted.

[0117] Among them, the evaluation function represents a mathematical formula used to quantitatively evaluate the performance of the optimization model. The performance improvement amount refers to the value of the performance improvement brought about by model optimization. The standard deviation of performance fluctuation is used to represent the stability of the optimization effect. The model parameter change rate represents the speed of model update. The sample space size refers to the amount of data used for training. The adversarial sample set is used to represent the set of training data under special working conditions.

[0118] The control system needs to continuously evaluate the effect of the operation optimization model. Specifically, the control system first calculates the evaluation metrics of the optimization model in real time, including the performance improvement effect, optimization stability, learning rate, and data coverage. When the evaluation result is lower than the preset threshold, the control system filters out abnormal working conditions from the historical data and constructs an adversarial sample set. Then, the control system uses these samples to train the model specifically to improve the performance of the model in special scenarios. Based on the effect of retraining, the control system dynamically adjusts the subsequent incremental training strategies, including the sampling frequency and sample weight allocation. At the same time, the control system continuously monitors the generalization ability of the model to ensure the reliability of the optimization process.

[0119] In some embodiments, the evaluation and optimization of the model can be achieved in various ways: Optionally, the control system uses a multi-objective optimization method, considering performance improvement, stability, and generalization simultaneously, finding the optimal balance point through Pareto front analysis, adopting an adaptive weight adjustment sampling strategy, and finally ensuring the optimization effect through cross-validation; Optionally, the control system adopts an active learning method based on uncertainty, identifies the regions with low prediction confidence of the model, focuses on collecting training samples in these regions, adjusts the model hyperparameters through Bayesian optimization, and finally evaluates the robustness of the model in different scenarios. It can be understood that other evaluation methods can also be used to achieve the quantitative evaluation and optimization of the model performance, which is not limited here.

[0120] In the embodiments of the present application, since the mapping relationship between environmental features and control parameters based on deep learning is established and combined with a multi-dimensional performance evaluation mechanism, it is possible to identify performance mutation points in real time and generate optimal control parameters, and at the same time continuously optimize the model performance through incremental learning. It effectively solves the problems such as fixed parameter configuration, lagging optimization response, and difficulty in accumulating experience in traditional AGV robot control systems, and thus realizes the adaptive control optimization of AGV robot groups. Through real-time performance evaluation and parameter optimization, the adaptability of AGV robots in dynamic environments is improved; through the incremental training mechanism, automatic accumulation and group sharing of optimization experience are realized; through multi-dimensional performance evaluation, the safety and stability of the optimization process are ensured; finally, through continuous optimization of the model, the intelligent evolution of the control system is realized.

[0121] The control system in the embodiments of the present invention application will be described from the perspective of hardware processing. Please refer to Figure 3, which is a schematic structural diagram of an entity device of the control system in the embodiments of the present application.

[0122] It should be noted that Figure 3 the structure of the control system shown is only an example and should not impose any limitations on the functions and usage scope of the embodiments of the present invention.

[0123] As Figure 3 shown, the control system includes a CPU 301, which can perform various appropriate actions and processes according to the program stored in the ROM 302 or the program loaded into the RAM 303 from the storage section 308, such as executing the method described in the above embodiments. In the RAM 303, various programs and data required for system operation are also stored. The CPU 301, ROM 302, and RAM 303 are connected to each other via a bus 304. The I / O interface 305 is also connected to the bus 304.

[0124] The following components are connected to the I / O interface 305: an input section 306 including an audio input device, a button switch, etc.; an output section 307 including a liquid crystal display (LCD), an audio output device, an indicator light, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A driver 310 is also connected to the I / O interface 305 as needed. A removable medium 311, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the driver 310 as needed so that the computer program read from it can be installed into the storage section 308 as needed.

[0125] Specifically, according to the embodiments of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments of the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication section 309 and / or installed from the removable medium 311. When the computer program is executed by the CPU 301, various functions defined in the present invention are executed.

[0126] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present invention. Among them, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the above-mentioned module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the block may occur in a different order from that marked in the accompanying drawings.

[0127] Specifically, the control system of this embodiment includes a processor and a memory. A computer program is stored on the memory. When the computer program is executed by the processor, it implements the AGV robot control method provided in the above-mentioned embodiment.

[0128] On the other hand, the present invention also provides a computer-readable storage medium. This storage medium may be included in the control system described in the above-mentioned embodiment; or it may exist alone without being assembled into the control system. The above storage medium carries one or more computer programs. When the above one or more computer programs are executed by a processor of the control system, the control system is enabled to implement the AGV robot control method provided in the above-mentioned embodiment.

[0129] As mentioned above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the various embodiments of the present application.

[0130] As used in the above embodiments, depending on the context, the term "when..." may be interpreted to mean "if...", or "after...", or "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if detecting (the stated condition or event)" may be interpreted to mean "if determining...", or "in response to determining...", or "when detecting (the stated condition or event)", or "in response to detecting (the stated condition or event)".

Claims

1. A control method for an AGV robot, characterized in that, Applied to a control system, the method includes: Obtain the historical operation data of multiple AGV robots; the historical operation data includes working state data, environmental perception data, task execution data, and energy consumption data; Based on a performance evaluation model, extract the control parameters and environmental characteristics corresponding to performance mutation points from the historical operation data, and generate training samples with performance annotations; the performance mutation point is the time point when the change rate of the performance index generated by the performance evaluation model exceeds a preset stability threshold; Based on the training samples, construct a mapping relationship between environmental characteristics and optimal control parameters to generate an operation optimization model; Receive the current environmental characteristic data and operation state data sent by the robot to be optimized; Input the current environmental characteristic data and the operation state data into the operation optimization model to obtain target control parameters; Send a control instruction corresponding to the target control parameter to the robot to be optimized to control the robot to be optimized to adjust its operation state.

2. The method according to claim 1, characterized in that, The step of extracting the control parameters and environmental characteristics corresponding to performance mutation points from the historical operation data based on the performance evaluation model to generate training samples with performance annotations specifically includes: Segment the historical operation data according to the time series, and calculate the comprehensive performance index of the AGV robot in each time period based on the performance evaluation formula; the comprehensive performance index includes task completion efficiency, navigation accuracy, energy utilization rate, and safety factor; Identify the time point when the change rate of the comprehensive performance index exceeds the preset stability threshold as the performance mutation point; Extract the environmental characteristic data and control parameter data within a preset time window before and after the performance mutation point, and calculate the performance difference before and after the performance mutation point; Use the performance difference as the performance annotation, and integrate the environmental characteristics and control parameters corresponding to the performance annotation to generate training samples.

3. The method according to claim 2, wherein The performance evaluation formula is: P = W1·Pt + W2·Pn + W3·Pe + W4·Ps; The Pt is the task execution performance index; Pt=(T0 / T)·[1+∫(dv / dt)²dt / D]·(1 - θ / θmax); where T0 is the standard task time, T is the actual execution time; D is the task distance; θ is the path curvature, θmax is the allowable maximum curvature; dv / dt is the speed change rate, which is used to characterize the motion smoothness; The Pn is the navigation accuracy index; Pn=(1 - |ΔD| / D0)·exp(-σ² / σ0²)·(1 + d²Δp / dt²)^(-1); where ΔD is the actual trajectory deviation; D0 is the allowable maximum deviation; σ is the standard deviation of the positioning error; σ0 is the reference error value; d²Δp / dt² is the second derivative of the position, which is used to characterize the motion jitter; The Pe is the energy efficiency index; Pe=(E0 / E)·[1-∑(Pi·ti) / (Pmax·T)]·(1-∂E / ∂t); where E0 is the standard energy consumption, E is the actual energy consumption; Pi is the instantaneous power, Pmax is the rated power; ti is the duration of each power segment; ∂E / ∂t is the energy consumption change rate; Ps is the safety index; Ps=(dmin / d0)·exp(-v / v0)·[1-∑(1 / ri²)] / n; where dmin is the minimum distance to the obstacle, d0 is the safety distance threshold; v is the current speed, v0 is the rated speed; ri is the distance to the i-th obstacle; n is the number of detected obstacles; The weight coefficients W1, W2, W3, W4 satisfy: W1 + W2 + W3 + W4 = 1.

4. The method according to claim 1, wherein Before the step of receiving the current environmental feature data and operation status data sent by the robot to be optimized, the method further includes: Real-time monitoring the performance parameters of multiple AGV robots in the running state, and determining multiple target robots to be optimized based on the performance parameters; Obtaining the real-time operation parameters and real-time environmental data of the target robots; When it is determined according to the real-time operation parameters and the real-time environmental data that the target robots do not meet the preset optimization conditions, marking the target robots as optimization waiting robots to perform the optimization determination again after a preset time.

5. The method according to claim 4, wherein After the step of obtaining the real-time operation parameters and real-time environmental data of the target robots, the method further includes: Constructing an optimization condition evaluation model based on historical optimization records; Calculating the environmental complexity score according to the real-time operation parameters, and calculating the task urgency score according to the real-time environmental data; Inputting the environmental complexity score and the task urgency score into the optimization condition evaluation model to obtain an optimization adaptability score; Determining the preset optimization conditions based on the optimization adaptability score.

6. The method according to claim 1, characterized in that, After the step of sending the control instruction corresponding to the target control parameter to the robot to be optimized to control the robot to be optimized to adjust its operation state, the method further includes: Recording the optimization performance data of the robot to be optimized after executing the target control parameter; Calculating the performance improvement amplitude of the optimization performance data compared with the benchmark performance before optimization, and when the performance improvement amplitude exceeds the preset benchmark threshold, extracting the current environmental features and the corresponding target control parameters as incremental training samples; Performing online optimization on the operation optimization model based on the incremental training samples.

7. The method according to claim 6, wherein After the step of performing online optimization on the operation optimization model based on the incremental training samples, the method further includes: Constructing an evaluation function of the operation optimization model; the evaluation function is: F = (ΔP / P0)·(1 - σp / σ0)·(1 - |dw / dt| / λ)·(S / S0); where ΔP is the performance improvement amount, P0 is the reference performance value; σp is the standard deviation of performance fluctuation, σ0 is the allowable fluctuation threshold; dw / dt is the model parameter change rate, λ is the upper limit of the learning rate; S is the current sample space size, S0 is the initial sample space size; When the calculation result of the evaluation function is lower than the preset evaluation threshold, construct an adversarial sample set based on historical abnormal working condition data, and use the adversarial sample set to retrain the operation optimization model; Adjust the sampling period and weight distribution of the incremental training samples according to the performance improvement amplitude after retraining.

8. A control system, characterized in that, The control system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to enable the control system to execute the method according to any one of claims 1-7.

9. A computer-readable storage medium, comprising instructions, characterized in that, When the instruction runs on the control system, enable the control system to execute the method according to any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product runs on the control system, enable the control system to execute the method according to any one of claims 1-7.

Images