High-precision position keeping method and device for multi-dimensional motor cooperative control and storage medium

Abstract

The invention discloses a high-precision position keeping method and device for cooperative control of a multi-dimensional motor and a storage medium. The high-precision position keeping method and device are used for keeping the high-precision position of the multi-dimensional motor. The method comprises the following steps: constructing a multi-dimensional data set; performing standardization processing on the multi-dimensional data set to obtain standard data; calculating a covariance matrix based on the standard data and performing eigenvalue decomposition; determining a plurality of principal component feature vectors according to the cumulative variance contribution rate to form a principal component space; projecting the standard data into the principal component space to form a dimensionality reduction data set; inputting the dimension reduction data set into a pre-trained deep belief network, and outputting a comprehensive feature vector through hierarchical features of a restricted Boltzmann machine; inputting the comprehensive feature vector into a depth deterministic strategy gradient algorithm model, and constructing a reward function; outputting a real-time adjustment strategy based on the reward function, and obtaining an optimization control parameter; and realizing high-precision position keeping of the plurality of motors according to the optimized control parameters.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the field of industrial automation control, and particularly relate to a high-precision position keeping method and device for multi-dimensional motor cooperative control and a storage medium. BACKGROUND

[0002] In the field of high-precision control such as industrial automation, robots, aerospace, etc., multi-dimensional motor cooperative systems are widely used in complex motion control scenarios, such as multi-joint robot arms, precision machining platforms, satellite attitude adjustment systems, etc. Such systems usually require multiple motors to maintain micron-level or even sub-micron-level position accuracy under complex conditions such as dynamic load, environmental disturbance, etc., while meeting multiple constraints such as energy optimization, synchronization control, etc. Traditional single motor control methods cannot meet the coupling effect and cooperative requirements between multiple motors, so multi-dimensional motor cooperative control technology has become the core direction of current research.

[0003] Currently, in multi-dimensional motor cooperative control, only limited dimensions of data such as position, speed and current are collected. These data are input into a pre-set motor mathematical model or a fixed parameter PID control algorithm after linear dimension reduction processing such as traditional low-pass filtering or principal component analysis, to generate control instructions. The control algorithm based on the pre-set model lacks adaptive ability, and when facing complex working conditions such as dynamic load changes and environmental parameter fluctuations, manual parameter adjustment is required, and the current loop, speed loop and position loop parameters cannot be optimized autonomously through real-time data-driven methods, ultimately leading to gradual accumulation of position errors during long-term system operation, making it difficult to meet the stringent requirements of high-end equipment for high-precision position keeping.

[0004] In addition, ignoring the influence of temperature and humidity changes on motor winding resistance, as well as the mechanical resonance risk reflected by vibration sensor data, can easily lead to the inability of the control strategy to comprehensively perceive the system state, and linear dimension reduction methods are difficult to effectively process nonlinear characteristics in motor operation data. SUMMARY

[0005] The present application discloses a high-precision position keeping method, device and storage medium for multi-dimensional motor cooperative control, for keeping the high-precision position of multiple motors.

[0006] The first aspect of the present application discloses a high-precision position keeping method for multi-dimensional motor cooperative control, comprising: real-time collection of position data, speed data, torque data, environmental data, vibration data and current data of multiple motors; aligning the position data, speed data, torque data, environmental data, vibration data and current data based on timestamps to construct a multi-dimensional data set; standardizing the multi-dimensional data set to obtain standard data; calculating a covariance matrix based on the standard data and performing eigenvalue decomposition; determining a plurality of principal component feature vectors according to the cumulative variance contribution rate to form a principal component space; projecting the standard data into the principal component space to form a reduced dimension data set; inputting the reduced dimension data set into a pre-trained deep belief network, outputting a comprehensive feature vector through the hierarchical features of the restricted Boltzmann machine, and the comprehensive feature vector is used to represent the position error dynamic characteristics, load torque coupling relationship, environmental disturbance response characteristics and dynamic compensation parameters required for coordinated control of the plurality of motors; inputting the comprehensive feature vector into a deep deterministic policy gradient algorithm model to construct a reward function; outputting a real-time adjustment strategy of current loop parameters, speed loop compensation coefficients and position loop parameters of the plurality of motors based on the reward function to obtain optimized control parameters; According to the optimized control parameters, the current loop parameters, the speed loop compensation coefficients and the position loop parameters of the plurality of motors are adjusted in real time to realize high-precision position keeping of the plurality of motors.

[0007] Optionally, after outputting a real-time adjustment strategy of current loop parameters, speed loop compensation coefficients and position loop parameters of the plurality of motors based on the reward function to obtain optimized control parameters, and before adjusting the current loop parameters, the speed loop compensation coefficients and the position loop parameters of the plurality of motors in real time to realize high-precision position keeping of the plurality of motors, the method further comprises: calculating the deviation value of the actual position and the target position of the plurality of motors to form a position deviation sequence; calculating a comprehensive error index based on the position deviation sequence; if the comprehensive error index exceeds a preset threshold, triggering a reward function correction mechanism; inputting the current multi-dimensional data set into the deep belief network to output a current comprehensive feature vector, combining a preset fault feature library, and analyzing the error source; According to the error source, dynamically correcting the reward function in the deep deterministic policy gradient algorithm model; inputting the corrected reward function into the deep deterministic policy gradient algorithm model to regenerate the adjustment strategy of the plurality of motors and obtain new optimized control parameters.

[0008] Optionally, the inputting the comprehensive feature vector into the deep deterministic policy gradient algorithm model to construct a reward function comprises: input the comprehensive feature vector into a deep deterministic policy gradient algorithm model to generate initial control parameters, the initial control parameters including current loop parameters, speed loop compensation coefficients and position loop parameters of the motor; According to the initial control parameters, control multiple motors to run, and collect actual position data of multiple motors, synchronization data between multiple motors and energy consumption data of multiple motors in real time to generate actual control parameters; Based on the actual control parameters, calculate position error, synchronization error and energy consumption respectively, and construct a reward function containing position error, synchronization error and energy consumption optimization.

[0009] Optionally, the reward function outputs real-time adjustment strategies of current loop parameters, speed loop compensation coefficients and position loop parameters of multiple motors based on the reward function, to obtain optimized control parameters, including: Based on the reward function, evaluate the running state of multiple motors under the actual control parameters, and output evaluation results; According to the evaluation results, generate optimized control parameters based on the adjustment strategy of the deep deterministic policy gradient algorithm.

[0010] Optionally, based on the actual control parameters, calculate position error, synchronization error and energy consumption respectively, and construct a reward function containing position error, synchronization error and energy consumption optimization, including: The penalty term of the position error is designed by using an exponential decay function, and the formula is:

[0011] wherein, is a position error sensitivity coefficient, is the actual position of the motor, is the target position of the motor; The penalty term of the synchronization error is designed by using a Gaussian kernel function, and the formula is:

[0012] wherein, M is the number of motors, is a synchronization error sensitivity coefficient, and is the actual position of the motor i and j; The optimization term of the energy consumption is designed by using a current flat integral form, and the formula is:

[0013] wherein, is an energy consumption weight coefficient, is the motor current.

[0014] Optionally, after aligning the position data, rotational speed data, torque data, environmental data, and current data based on timestamps to construct a multidimensional dataset, and before standardizing the multidimensional dataset to obtain standard data, the method further includes: Wavelet transform is used to perform multi-scale decomposition on the multidimensional dataset to remove high-frequency noise.

[0015] Optionally, the step of performing multi-scale decomposition of the multidimensional dataset using wavelet transform to remove high-frequency noise includes: The multidimensional data is decomposed into different decomposition layers by using wavelet basis functions at multiple scales. An adaptive threshold is used to reduce noise in the different decomposition layers; The high-frequency coefficients in the different decomposition layers are processed by a soft thresholding function to obtain the denoised coefficients. The noise-reduced coefficients are reconstructed with the low-frequency approximation coefficients to complete the high-frequency noise removal.

[0016] A second aspect of this application provides a high-precision position holding device for multi-dimensional motor cooperative control, comprising: The data acquisition unit is used to collect position data, speed data, torque data, environmental data, vibration data, and current data of multiple motors in real time. An alignment unit is used to align the position data, rotational speed data, torque data, environmental data, vibration data, and current data based on timestamps to construct a multidimensional dataset. The processing unit is used to standardize the multidimensional dataset to obtain standard data. The decomposition unit is used to calculate the covariance matrix and perform eigenvalue decomposition based on the standard data. Forming units are used to determine multiple principal component eigenvectors based on the cumulative variance contribution rate, forming the principal component space; A projection unit is used to project the standard data into the principal component space to form a dimension-reduced dataset. The output unit is used to input the dimensionality reduction dataset into a pre-trained deep belief network and output a comprehensive feature vector through the hierarchical features of the restricted Boltzmann machine. The comprehensive feature vector is used to characterize the dynamic characteristics of the position error of multiple motors, the load torque coupling relationship, the environmental disturbance response characteristics, and the dynamic compensation parameters required for the coordinated control of multiple motors. The construction unit is used to input the comprehensive feature vector into the deep deterministic policy gradient algorithm model to construct the reward function; The acquisition unit is used to output a real-time adjustment strategy for the current loop parameters, speed loop compensation coefficients and position loop parameters of multiple motors based on the reward function, so as to obtain optimized control parameters. A holding unit is configured to adjust the current loop parameters, the speed loop compensation coefficients and the position loop parameters of the plurality of motors in real time according to the optimized control parameters, so as to achieve high-precision position holding of the plurality of motors.

[0017] The third aspect of the present application provides a high-precision position holding device for multi-dimensional motor cooperative control, comprising: a processor, a memory, an input / output unit and a bus; The processor is connected with the memory, the input / output unit and the bus; The memory stores a program, and the processor invokes the program to execute the method of the first aspect and any optional method of the first aspect.

[0018] The fourth aspect of the present application provides a computer readable storage medium, and the computer readable storage medium stores a program, and the program executes the method of the first aspect and any optional method of the first aspect when executed on a computer.

[0019] From the above technical solutions, the embodiments of the present application have the following advantages: Firstly, the present application constructs a full-dimensional data containing position, speed, torque, environmental parameters and current, which is more accurate than the traditional method relying only on basic data such as position and current. The newly added torque data can accurately reflect the load change, and the temperature and humidity and vibration sensor data can capture the environmental interference in real time. A complete data set containing motor operating state, load characteristics and environmental interference is constructed, the current loop parameters are adjusted in real time through temperature data, the error is reduced, and the precision is improved.

[0020] Secondly, the fusion mechanism of principal component analysis dimension reduction + deep belief network hierarchical feature extraction breaks through the bottleneck of traditional linear dimension reduction method in processing nonlinear characteristics. Principal component analysis removes data redundancy, and then performs nonlinear transformation through restricted Boltzmann machine to mine deep coupling features between multi-source data, effectively processing nonlinear laws in motor operation.

[0021] Finally, with the help of deep deterministic policy gradient algorithm, an intelligent decision-making mechanism for multi-objective optimization is constructed, taking position error, synchronization error and energy consumption optimization as joint optimization objectives, to realize autonomous dynamic adjustment of control parameters. The deep deterministic policy gradient algorithm automatically learns the optimal control strategy under different working conditions through real-time interaction with the motor operating environment, without human intervention to cope with complex scenarios such as load mutation and environmental fluctuations. In addition, through the closed-loop control of "collection-decision-execution-feedback", the position deviation is continuously monitored and the error source is analyzed combined with the data fusion model, the reward function is corrected accordingly, and the error accumulation effect is suppressed, so that the system maintains position stability in long-term operation, meeting the stringent requirements of high-end equipment for high precision and high reliability. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0023] Figure 1 An embodiment schematic diagram of a high-precision position holding method for multi-dimensional motor cooperative control of the present application; Figure 2 An embodiment schematic diagram of a real-time error dynamic optimization method for adjustment strategy of the present application; Figure 3 An embodiment schematic diagram of a reward function construction method of the present application; Figure 4 An embodiment schematic diagram of an optimized control parameter obtaining method of the present application; Figure 5 An embodiment schematic diagram of a multi-scale decomposition method for multi-dimensional data set by using wavelet transform to remove high-frequency noise of the present application; Figure 6 An embodiment schematic diagram of a high-precision position holding device for multi-dimensional motor cooperative control of the present application; Figure 7 An embodiment schematic diagram of a high-precision position holding device for multi-dimensional motor cooperative control of the present application. DETAILED DESCRIPTION

[0024] In the following description, specific details are set forth in order to provide a thorough understanding of embodiments of the present application. However, persons of ordinary skill in the art will readily appreciate that embodiments of the present application can be practiced without these specific details. In other instances, well-known structures, devices, circuits, and methods have not been described in detail in order to avoid obscuring the present application.

[0025] It should be understood that the term "comprising" as used in the specification and in the claims indicates the presence of the recited features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0026] It should also be understood that the term "and / or" as used in the specification and in the claims indicates any combination of one or more of the associated listed items and all possible combinations of those items.

[0027] As used in the specification and the appended claims, the term “if’ can be interpreted as meaning “when” or “upon” or “in response to determining” or “in response to detecting” depending on the context. Similarly, the phrase “if it is determined” or “if [the described condition or event] is detected” can be interpreted as meaning “upon determining” or “in response to determining” or “upon detecting [the described condition or event]” or “in response to detecting [the described condition or event]” depending on the context.

[0028] In addition, in the description of the specification and the appended claims, the terms “first”, “second”, “third”, etc. are only used to distinguish descriptions and cannot be understood as indicating or implying relative importance.

[0029] In the specification of the present application, the reference “one embodiment” or “some embodiments” and the like means that the specific features, structures or characteristics described in connection with the embodiment are included in one or more embodiments of the present application. Therefore, the statements “in one embodiment”, “in some embodiments”, “in other some embodiments”, “in further some embodiments” and the like appearing in different places in the specification are not necessarily all referring to the same embodiment, but mean “one or more but not all embodiments”, unless otherwise specifically emphasized. The terms “include”, “contain”, “have” and their variants mean “include but not limited to”, unless otherwise specifically emphasized.

[0030] Based on this, the present application discloses a multi-dimensional motor cooperative control high-precision position keeping method, device and storage medium, which is used for keeping the high-precision position of the multi-dimensional motor.

[0031] The technical solutions in the present application will be described clearly and completely in the specification of the present application combined with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0032] The method of the present application can be applied to a server, a device, a terminal or other devices with logical processing capability, and the present application is not limited thereto. For the convenience of description, the following will be described taking the system as an example.

[0033] Please refer to Figure 1 The present application provides one embodiment of a multi-dimensional motor cooperative control high-precision position keeping method, which comprises: 101, collecting position data, speed data, torque data, environmental data, vibration data and current data of a plurality of motors in real time; 102. Aligning the position data, the rotation speed data, the torque data, the environmental data, the vibration data and the current data based on timestamps, and constructing a multi-dimensional data set; In step 101, various sensors are installed on the motor equipment to collect key data during the operation of the motor in real time. The position data and the rotation speed data of the motor are accurately obtained by using the encoder, the torque data output by the motor is monitored by relying on the torque sensor, the environmental data of the motor is collected by deploying environmental sensors such as temperature and humidity, air pressure, etc., the vibration of the motor shell is captured by means of the vibration sensor, and the current data is recorded by using the current detection module in the motor driving circuit. These sensors and detection modules cooperate with each other to continuously and uninterruptedly collect data, ensuring that the multi-aspect information of the motor operation can be obtained completely and timely.

[0034] Since the time and frequency of data collection by each sensor in step 101 may be different, the collected position data, rotation speed data, torque data, environmental data, vibration data and current data are aligned based on timestamps in step 102. With a unified time reference, different types of data at the same time are integrated together to ensure the consistency and correlation of the data in the time dimension. After completing the data alignment, the data of different dimensions are combined to form a multi-dimensional data set. The multi-dimensional data set covers multi-aspect information such as the motor operating state and the working environment.

[0035] 103. Standardizing the multi-dimensional data set to obtain standard data; 104. Calculating the covariance matrix based on the standard data and performing eigenvalue decomposition; 105. Determining a plurality of principal component feature vectors according to the cumulative variance contribution rate to form a principal component space; 106. Projecting the standard data into the principal component space to form a reduced dimension data set; In step 103, the multi-dimensional data set is standardized to eliminate the differences in dimensions and orders of magnitude of different types of data. Specifically, for the data of each dimension, the mean and standard deviation are first calculated, then the original data is subtracted from the mean and divided by the standard deviation to obtain the standard data.

[0036] In step 104, the covariance matrix is calculated based on the standard data, and the covariance matrix mainly reflects the correlation between the data of each dimension. Eigenvalue decomposition of the covariance matrix can obtain a series of eigenvalues and corresponding eigenvectors. The size of the eigenvalue indicates the variance of the data in the corresponding eigenvector direction, and the larger the variance, the more data information contained in the direction.

[0037] The formula for calculating the covariance matrix is:

[0038] in For standard sample vectors, The mean of the sample; The eigenvalues ​​of the covariance matrix obtained by the QR iterative algorithm are: The corresponding feature vector is: .

[0039] The formula for calculating the cumulative variance contribution rate in step 105 is as follows:

[0040] Dynamic selection to satisfy The values ​​are used as principal component eigenvectors to form the principal component space.

[0041] In step 106, the standard data is projected onto the principal component space determined in step 105. Specifically, this involves performing matrix multiplication between the original data and the principal component eigenvectors, thereby mapping the high-dimensional data to the low-dimensional principal component space, ultimately forming a dimensionality-reduced dataset.

[0042] 107. Input the dimensionality-reduced dataset into the pre-trained deep belief network, and output a comprehensive feature vector through the hierarchical features of the restricted Boltzmann machine. The comprehensive feature vector is used to characterize the dynamic characteristics of the position error of multiple motors, the load torque coupling relationship, the environmental disturbance response characteristics, and the dynamic compensation parameters required for the coordinated control of multiple motors. 108. Input the comprehensive feature vector into the deep deterministic policy gradient algorithm model to construct the reward function; 109. Based on the reward function, a real-time adjustment strategy for the current loop parameters, speed loop compensation coefficients, and position loop parameters of multiple motors is developed to obtain optimized control parameters. 110. Based on optimized control parameters, adjust the current loop parameters, speed loop compensation coefficient, and position loop parameters of multiple motors in real time to achieve high-precision position holding of multiple motors.

[0043] In step 107, the dimensionality-reduced dataset obtained in the previous steps is input into the pre-trained deep belief network. The deep belief network is composed of multiple stacked Restricted Boltzmann Machines (RBMs), each of which can perform layer-by-layer training and feature learning on the input data. Through hierarchical feature extraction, complex nonlinear relationships in multiple motor operating data are automatically captured. The final output comprehensive feature vector fully characterizes the dynamic characteristics of the motor's position error, the coupling relationship between load torque, environmental disturbance response characteristics, and the dynamic compensation parameters required for coordinated control.

[0044] After layer-by-layer training and feature learning on the input data using a Restricted Boltzmann Machine, the formula for outputting the comprehensive feature vector is:

[0045] wherein, is the weight matrix of the first layer, is the output of the previous hidden layer, is the bias vector, is the activation function, and L is the number of network layers.

[0046] In step 108, the comprehensive feature vector obtained in step 107 is input into the deep deterministic policy gradient algorithm model to construct a reward function. Specifically, the reward function is constructed based on the comprehensive feature vector. By reasonably setting the reward function, the multiple goals of motor control can be converted into quantifiable numerical indicators, guiding the algorithm to learn towards the expected control effect.

[0047] The deep deterministic policy gradient algorithm model in step 109 optimizes the control policy with the goal of maximizing the reward function. After training, the model outputs real-time adjustment strategies for the current loop parameters, speed loop compensation coefficients, and position loop parameters of multiple motors. The optimized control parameters are dynamically generated by the algorithm based on the current motor operating state and environmental conditions, fully considering the collaborative relationship between motors and the influence of various disturbance factors. In this way, intelligent mapping from data features to control parameters is achieved, providing precise parameter support for high-precision motor control.

[0048] In step 110, the current loop parameters, speed loop compensation coefficients, and position loop parameters of multiple motors are adjusted in real time based on the obtained optimized control parameters. In actual motor control systems, the current loop, speed loop, and position loop are the core links for precise control. Through real-time adjustment of these parameters, a complete closed-loop control system is formed. The system can continuously adjust the control strategy according to the actual operating state of the motor, timely compensate for various errors and disturbances, and thus achieve high-precision position maintenance of multiple motors, ensuring stable and accurate operation of the motor under complex working conditions.

[0049] In this embodiment, first, a full-dimensional data containing position, speed, torque, environmental parameters, and current is constructed. Compared with the traditional method which only relies on basic data such as position and current, the newly added torque data can accurately reflect load changes, and the temperature and humidity and vibration sensor data can capture environmental disturbances in real time. A complete data set containing motor operating state, load characteristics, and environmental disturbances is constructed, the current loop parameters are adjusted in real time through temperature data, the error is reduced, and the precision is improved.

[0050] Secondly, the fusion mechanism of principal component analysis dimension reduction + deep belief network hierarchical feature extraction is adopted to break through the bottleneck of traditional linear dimension reduction method in processing nonlinear features. Principal component analysis removes data redundancy first, and then excavates deep coupling features between multi-source data through nonlinear transformation of restricted Boltzmann machine, thereby effectively processing nonlinear laws in motor operation.

[0051] Finally, the intelligent decision-making mechanism of multi-objective optimization is constructed by means of deep deterministic policy gradient algorithm, and the position error, synchronization error and energy consumption optimization are taken as joint optimization objectives to realize autonomous dynamic adjustment of control parameters. Through real-time interaction with the motor operating environment, the deep deterministic policy gradient algorithm automatically learns the optimal control strategy under different working conditions, and can cope with complex scenes such as load mutation and environmental fluctuations without human intervention. In addition, through the closed-loop control of “collection-decision-execution-feedback”, the position deviation is continuously monitored, and the error source is analyzed combined with the data fusion model, the reward function is corrected, and the error accumulation effect is suppressed, so that the system can maintain position stability in long-time operation and meet the stringent requirements of high-end equipment for high precision and high reliability.

[0052] Please refer to Figure 2 An embodiment of the real-time error dynamic optimization method for adjusting strategy is provided, which comprises the following steps: 201. Calculate the deviation value of the actual position of the plurality of motors from the target position to form a position deviation sequence; The actual position data of each motor is collected in real time by using a high-precision encoder installed on the motor driving shaft, and compared with the preset target position. Specifically, the position deviation value in each sampling period is obtained by subtraction operation, and the position deviation sequence is recorded in time sequence. The sequence reflects the dynamic error trend of position control of the motor in the running process.

[0053] 202. Calculate a comprehensive error index based on the position deviation sequence; Based on the position deviation sequence obtained in step 201, a statistical analysis combined with weighted calculation is used to calculate a comprehensive error index. The comprehensive error index integrates multiple-dimensional parameters such as mean, variance and maximum peak value of the deviation, and gives different weights to long-time cumulative error and instantaneous severe fluctuation error through weighted summation, so as to comprehensively reflect the stability and accuracy of motor position control. The effectiveness of the current control strategy can be quantitatively evaluated through the comprehensive error index, thereby providing a quantitative basis for system decision-making.

[0054] 203. If the comprehensive error index exceeds a preset threshold, a reward function correction mechanism is triggered; The threshold range of the comprehensive error index is preset, and the preset threshold is determined according to the motor control accuracy requirement and the system operation stability requirement. When the comprehensive error index calculated in step 202 exceeds the preset threshold, it indicates that the current control strategy cannot meet the high-precision position keeping requirement, and the system automatically triggers the reward function correction mechanism to start the subsequent dynamic optimization process.

[0055] 204, input the current multidimensional data set into the deep belief network, output the current comprehensive feature vector, and analyze the error source in combination with the preset fault feature library; The current collected multidimensional data set is input into the pre-trained deep belief network. The deep belief network extracts features from the multidimensional data set through multiple layers of restricted Boltzmann machines, and outputs a comprehensive feature vector that can represent the operating state of the motor. The comprehensive feature vector is matched and compared with the preset fault feature library, which stores feature patterns corresponding to different types of errors (such as mechanical wear, electromagnetic interference, and parameter mismatch). Through similarity calculation and pattern recognition algorithm, the specific reason for the error exceeding the standard is quickly located, and the source of the error is analyzed.

[0056] 205, dynamically correct the reward function in the deep deterministic policy gradient algorithm model according to the error source; According to the error source determined in step 204, the reward function in the deep deterministic policy gradient algorithm model is dynamically adjusted. For example, when the error is caused by torque fluctuation due to sudden change of motor load, the weight of torque stability related parameters in the reward function is increased; when the motor performance is affected by environmental temperature change, the environmental parameters are included in the consideration range of the reward function. Through dynamic adjustment of the reward function, the reward function can reflect the control target in real time under the current working condition, and guide the deep deterministic policy gradient algorithm to generate a control strategy that better meets the actual demand.

[0057] 206, input the corrected reward function into the deep deterministic policy gradient algorithm model, regenerate the adjustment strategy of multiple motors, obtain new optimized control parameters, and form a closed-loop control.

[0058] The corrected reward function is re-input into the deep deterministic policy gradient algorithm model. Based on the principle of reinforcement learning, through continuous trial and error and policy iteration, the adjustment strategy of current loop parameters, speed loop compensation coefficients and position loop parameters for multiple motors is output, and new optimized control parameters are obtained. These parameters will directly act on the closed-loop control system of the motor driver, realize real-time adjustment of motor current, speed and position, and thus build a complete closed-loop control system of “data acquisition-error evaluation-strategy correction-parameter adjustment”, to ensure that the motor can continuously maintain high-precision position synchronous operation under complex working conditions.

[0059] In this embodiment, by calculating the position deviation in real time and dynamically adjusting the control strategy, the system can automatically compensate for the position error caused by factors such as load change, mechanical wear, environmental disturbance, etc., so that multiple motors can always maintain high-precision synchronous operation and reduce position tracking error; the combination of deep belief network and fault feature library realizes intelligent identification of error sources, shortens fault troubleshooting time and improves production efficiency compared with manual inspection; adjusting the weight of the reward function according to the real-time error source makes the reinforcement learning algorithm quickly adapt to different working conditions; The design of the comprehensive error index takes into account the mean, variance and other multi-dimensional parameters of the position deviation, effectively suppressing system jitter and overshoot while ensuring position accuracy, achieving a cooperative optimization of stability and speed, and improving dynamic response speed; the entire correction mechanism forms a complete closed loop, and the system can continuously operate without human intervention, and has strong adaptability to unknown disturbances and time-varying parameters.

[0060] Please refer to Figure 3 The present application provides an embodiment of a method for constructing a reward function, comprising: 301, input the comprehensive feature vector into the deep deterministic policy gradient algorithm model to generate initial control parameters, the initial control parameters including current loop parameters, speed loop compensation coefficients and position loop parameters of the motor; 302, control the multiple motors to operate according to the initial control parameters, and real-time collect actual position data of the multiple motors, synchronization data between the multiple motors and energy consumption data of the multiple motors to generate actual control parameters; 303, based on the actual control parameters, calculate the position error, synchronization error and energy consumption respectively, and construct a reward function containing position error, synchronization error and energy consumption optimization.

[0061] This embodiment is based on a deep deterministic policy gradient algorithm model to construct a reward function containing position error, synchronization error and energy consumption optimization. Specifically, an exponential decay function is used to design the penalty term of the position error, and the formula is:

[0062] wherein, is the position error sensitivity coefficient, is the actual position of the motor, is the target position of the motor; A Gaussian kernel function is used to design the penalty term of the synchronization error, and the formula is:

[0063] wherein, M is the number of motors, is the synchronization error sensitivity coefficient, and is the actual position of the motor i and j; The optimization item of energy consumption is designed in the form of current flat integral, and the formula is:

[0064] wherein, is the energy consumption weight coefficient, is the motor current.

[0065] The position error, synchronization error and energy consumption are balanced by dynamic weight, and the formula is: wherein, the constraint condition is .

[0066] In the embodiment, the initial control parameters are generated by inputting the comprehensive feature vector into the algorithm model, which changes the traditional parameter setting mode relying on artificial experience, so that the initial values of the current loop, speed loop and position loop parameters are more suitable for the actual state of motor operation, effectively improving the parameter setting efficiency and accuracy. In the actual control process, the actual control parameters are generated by real-time collection of multi-dimensional operation data, the motor operation information is fully captured, and the precise control is ensured. In the reward function construction link, the position error, synchronization error and energy consumption are optimized, so that the algorithm considers the accuracy, synchronization and energy consumption in operation, avoids the performance imbalance caused by single target optimization, and improves the comprehensive performance and applicability of the motor cooperative control system.

[0067] Referring to Figure 4 , the application provides an embodiment of a method for obtaining optimized control parameters, which comprises: 401. Based on the reward function, the running state of the plurality of motors under the actual control parameters is evaluated, and the evaluation result is output. The actual control parameters obtained in Figure 3 are substituted into the reward function to evaluate the running state of the plurality of motors. The reward function comprehensively considers the position error, synchronization error and energy consumption and other multi-dimensional indicators. By quantitatively calculating these indicators, a specific evaluation result is obtained. The evaluation result can intuitively reflect the performance of the current motor control system.

[0068] 402. According to the evaluation result, the optimized control parameters are generated based on the adjustment strategy of the deep deterministic policy gradient algorithm.

[0069] According to the evaluation result of the running state, the control strategy is adjusted by the deep deterministic policy gradient algorithm. The deep deterministic policy gradient algorithm has strong adaptive learning ability and can dynamically adjust the control parameters according to the current evaluation result to maximize the reward function. Specifically, the algorithm analyzes the gap between each index in the evaluation result and the expected target, and then optimizes the current loop parameters, speed loop compensation coefficients and position loop parameters. Through continuous iteration and learning, a set of optimal control parameters is found, so that the motor achieves the best balance in position accuracy, synchronization performance and energy consumption. The final optimized control parameters will be applied to the motor control system to achieve precise adjustment of the motor running state.

[0070] In this embodiment, the motor running state is quantitatively evaluated by the reward function, which can convert multi-objective requirements such as position accuracy, synchronization and energy consumption into measurable indicators, avoiding the problem of performance degradation caused by single index optimization, and achieving comprehensive and comprehensive evaluation of the motor running state, providing objective and accurate basis for control strategy optimization. Based on the deep deterministic policy gradient algorithm, the optimized control parameters are dynamically generated, giving the system strong adaptive ability. According to the real-time evaluation result, the current loop, speed loop and position loop parameters are quickly adjusted, so that the motor control system can quickly adapt to complex working conditions such as load changes and environmental disturbances, reduce the control lag or deviation caused by fixed parameters, and improve the dynamic response speed and steady-state accuracy of motor control.

[0071] Please refer to Figure 5 The present application provides an embodiment of a method for removing high-frequency noise by using wavelet transform to perform multi-scale decomposition on a multi-dimensional data set, comprising: 501. Perform multi-scale decomposition on the multi-dimensional data using a wavelet basis function to obtain different decomposition layers; 502. Use an adaptive threshold to denoise the different decomposition layers; 503. Process the high-frequency coefficients in different decomposition layers using a soft threshold function to obtain denoised coefficients; 504. Reconstruct the denoised coefficients with the low-frequency approximation coefficients to complete high-frequency noise removal.

[0072] In step 501, the wavelet basis function is used to process the multi-dimensional data to realize multi-scale decomposition. In the context of motor data, position, current and other signals are decomposed into low-frequency and high-frequency parts. The low-frequency part mainly contains the essential characteristics of the signal, such as the actual running trend of the motor; while the high-frequency part contains noise and some detailed information. Decomposing the multi-dimensional data makes it faster and more efficient to process the data.

[0073] In step 502, the adaptive threshold method is used to process the different decomposition layers obtained by decomposing step 501. The adaptive threshold dynamically determines the appropriate threshold value according to the specific characteristics of each decomposition layer data. For high-frequency layers with more noise, a relatively high threshold value is set to effectively remove noise; while for low-frequency layers containing important signals, a lower threshold value is set to avoid deleting useful information. The adaptive threshold setting method can more accurately identify and remove noise.

[0074] In step 503, the soft threshold function is used to process the high-frequency coefficients in different decomposition layers. High-frequency coefficients contain not only noise components, but also some valuable signal analysis details. Therefore, by processing the high-frequency coefficients in the decomposition layer with the soft threshold function, when the absolute value of the high-frequency coefficient is less than the set threshold value, it is set to zero, thereby removing the noise; when the absolute value of the high-frequency coefficient is greater than the threshold value, it will be shrunk to a certain extent, rather than retaining the original value, which can both denoise and preserve signal details, avoiding signal distortion caused by excessive processing.

[0075] In step 504, the processed noise-reduced coefficients and the low-frequency approximation coefficients are reconstructed, and through the reconstruction operation, an approximate version of the original signal can be recovered, and the approximate signal has effectively removed high-frequency noise. In motor control, the clean signal after reconstruction provides a more reliable basis for subsequent data analysis, fault diagnosis, and control decision-making, which helps to improve the accuracy and stability of the motor control system.

[0076] In this embodiment, the multi-scale decomposition of multi-dimensional data by wavelet basis function can expand the original data at different frequencies and time scales, accurately separate high-frequency noise and low-frequency effective signal, and avoid the loss of signal characteristics caused by traditional filtering methods. Secondly, the adaptive threshold denoising combined with the soft threshold function processing high-frequency coefficients dynamically adjusts the denoising strength according to the data characteristics, which can effectively suppress high-frequency noise caused by current fluctuations, environmental interference, etc., and maximize the preservation of key signal details such as position and speed, thereby improving the signal-to-noise ratio of the data. Finally, the reconstruction of the noise-reduced coefficients and the low-frequency approximation coefficients realizes high-quality data recovery, improves the accuracy and stability of the motor control strategy, reduces the control deviation caused by data noise, and helps to achieve high-precision position maintenance and collaborative control of multiple motors, while improving the adaptability and anti-interference ability of the system to complex working conditions.

[0077] Referring to Figure 6 An embodiment of a high-precision position maintenance device for multi-dimensional motor collaborative control is provided, comprising: The acquisition unit 601 is configured to acquire position data, speed data, torque data, environmental data, vibration data, and current data of multiple motors in real time. An alignment unit 602 is configured to align the position data, the rotation speed data, the torque data, the environment data, the vibration data, and the current data based on timestamps, and construct a multidimensional data set; Optionally, the method further includes a removing unit 603 configured to: performing multi-scale decomposition on the multidimensional data set by using a wavelet transform, and removing high-frequency noise.

[0078] Optionally, the removing unit 603 is further configured to: performing multi-scale decomposition on the multidimensional data by using a wavelet basis function, to obtain different decomposition layers; performing noise reduction on the different decomposition layers by using an adaptive threshold; processing high-frequency coefficients in the different decomposition layers by using a soft threshold function, to obtain reduced noise coefficients; reconstructing the reduced noise coefficients and low-frequency approximation coefficients, to complete removal of the high-frequency noise.

[0079] A processing unit 604 is configured to perform standardization processing on the multidimensional data set, to obtain standard data; A decomposition unit 605 is configured to calculate a covariance matrix based on the standard data, and perform eigenvalue decomposition; A forming unit 606 is configured to determine a plurality of principal component feature vectors according to a cumulative variance contribution rate, and form a principal component space; A projection unit 607 is configured to project the standard data into the principal component space, to form a reduced dimension data set; An output unit 608 is configured to input the reduced dimension data set into a pre-trained deep belief network, output a comprehensive feature vector through a hierarchical feature of a restricted Boltzmann machine, and use the comprehensive feature vector to represent position error dynamic characteristics of the plurality of motors, load torque coupling relationships, environment disturbance response characteristics, and dynamic compensation parameters required for cooperative control of the plurality of motors; A constructing unit 609 is configured to input the comprehensive feature vector into a deep deterministic policy gradient algorithm model, and construct a reward function; Optionally, the constructing unit 609 is further configured to: input the comprehensive feature vector into the deep deterministic policy gradient algorithm model, to generate an initial control parameter, the initial control parameter including a current loop parameter, a speed loop compensation coefficient, and a position loop parameter of the motor; control the plurality of motors to operate according to the initial control parameter, and collect actual position data of the plurality of motors, synchronization data between the plurality of motors, and energy consumption data of the plurality of motors in real time, to generate an actual control parameter; calculate a position error, a synchronization error, and energy consumption based on the actual control parameter, and construct a reward function including optimization of the position error, the synchronization error, and the energy consumption.

[0080] The first obtaining unit 610 is configured to output a real-time adjustment strategy of current loop parameters, speed loop compensation coefficients and position loop parameters of the plurality of motors based on the reward function, and obtain the optimized control parameters. Optionally, the first obtaining unit 610 is further configured to: evaluate the running state of the plurality of motors under the actual control parameters based on the reward function, and output an evaluation result; generate the optimized control parameters based on the adjustment strategy of the deep deterministic policy gradient algorithm according to the evaluation result.

[0081] Optionally, the device further comprises a first calculation unit 611, configured to: calculate a deviation value of the actual positions of the plurality of motors from the target positions, and form a position deviation sequence; Optionally, the device further comprises a second calculation unit 612, configured to: calculate a comprehensive error index based on the position deviation sequence; Optionally, the device further comprises a triggering unit 613, configured to: trigger a reward function correction mechanism if the comprehensive error index exceeds a preset threshold; Optionally, the device further comprises an analysis unit 614, configured to: input a current multidimensional data set into a deep belief network, output a current comprehensive feature vector, combine a preset fault feature library, and analyze an error source; Optionally, the device further comprises a correction unit 615, configured to: dynamically correct the reward function in the deep deterministic policy gradient algorithm model according to the error source; Optionally, the device further comprises a generation unit 616, configured to: input the corrected reward function into the deep deterministic policy gradient algorithm model, regenerate the adjustment strategy of the plurality of motors, and obtain new optimized control parameters.

[0082] The maintaining unit 617 is configured to real-time adjust pulse duty cycles and current vectors of the plurality of motors according to the optimized control parameters, so as to realize high-precision position maintaining of the plurality of motors.

[0083] The specific implementation process is described in the Figures 1 to 5 embodiment, which will not be repeated here.

[0084] Please refer to Figure 7 The application provides a high-precision position maintaining device for coordinated control of a plurality of motors, comprising: a processor 701, a memory 702, an input and output unit 704 and a bus 703.

[0085] The processor 701 is connected with the memory 702, the input and output unit 704 and the bus 703.

[0086] The memory 702 stores a program, and the processor 701 invokes the program to perform the method in any one of Figure 1 , Figure 2 , Figure 3 , Figure 4 and Figure 5 .

[0087] The present application provides a computer readable storage medium, which stores a program, and the program performs the method in any one of Figure 1 , Figure 2 , Figure 3 , Figure 4 and Figure 5 when executed on a computer.

[0088] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0089] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented by other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0090] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment.

[0091] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0092] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

Claims

1. A high-precision position holding method of multi-dimensional motor cooperative control, characterized by, The method comprises the following steps: real-time acquisition of position data, speed data, torque data, environmental data, vibration data and current data of multiple motors; aligning the position data, speed data, torque data, environmental data, vibration data and current data based on timestamps to construct a multidimensional data set; standardizing the multidimensional data set to obtain standard data; calculating the covariance matrix based on the standard data and performing eigenvalue decomposition; determining multiple principal component feature vectors according to the cumulative variance contribution rate to form a principal component space; projecting the standard data into the principal component space to form a reduced dimension data set; inputting the reduced dimension data set into a pre-trained deep belief network to output a comprehensive feature vector through the hierarchical features of the restricted Boltzmann machine, wherein the comprehensive feature vector is used to represent the position error dynamic characteristics, load torque coupling relationship, environmental disturbance response characteristics of multiple motors and the dynamic compensation parameters required for the coordinated control of multiple motors; inputting the comprehensive feature vector into a deep deterministic policy gradient algorithm model to construct a reward function; outputting real-time adjustment strategies of current loop parameters, speed loop compensation coefficients and position loop parameters of multiple motors based on the reward function to obtain optimized control parameters; real-time adjustment of the current loop parameters, speed loop compensation coefficients and position loop parameters of multiple motors according to the optimized control parameters to achieve high-precision position keeping of multiple motors.

2. The high-precision position holding method according to claim 1, characterized by, After outputting real-time adjustment strategies of current loop parameters, speed loop compensation coefficients and position loop parameters of multiple motors based on the reward function to obtain optimized control parameters, and before real-time adjustment of the current loop parameters, speed loop compensation coefficients and position loop parameters of multiple motors according to the optimized control parameters to achieve high-precision position keeping of multiple motors, the method further comprises: calculating the deviation value of the actual position and the target position of multiple motors to form a position deviation sequence; calculating a comprehensive error index based on the position deviation sequence; if the comprehensive error index exceeds a preset threshold, triggering a reward function correction mechanism; inputting the current multidimensional data set into the deep belief network to output a current comprehensive feature vector, combining a preset fault feature library to analyze the error source; dynamically correcting the reward function in the deep deterministic policy gradient algorithm model according to the error source; inputting the corrected reward function into the deep deterministic policy gradient algorithm model to regenerate the adjustment strategies of multiple motors and obtain new optimized control parameters.

3. The high precision position holding method according to claim 1, wherein The method of inputting the comprehensive feature vector into the deep deterministic policy gradient algorithm model to construct a reward function comprises the following steps: inputting the comprehensive feature vector into the deep deterministic policy gradient algorithm model to generate initial control parameters, wherein the initial control parameters include current loop parameters, speed loop compensation coefficients and position loop parameters of the motor; controlling multiple motors to operate according to the initial control parameters, and real-time acquisition of actual position data of multiple motors, synchronization data between multiple motors and energy consumption data of multiple motors to generate actual control parameters; Based on the actual control parameters, position error, synchronization error and energy consumption are calculated respectively, and a reward function containing position error, synchronization error and energy consumption optimization is constructed.

4. The high precision position holding method according to claim 1, characterized by, The reward function outputs real-time adjustment strategies of current loop parameters, speed loop compensation coefficients and position loop parameters of multiple motors, and obtains optimized control parameters, including: Based on the reward function, the running state of multiple motors under the actual control parameters is evaluated, and the evaluation result is output; According to the evaluation result, the adjustment strategy based on the deep deterministic policy gradient algorithm is used to generate the optimized control parameters.

5. The high-precision position holding method according to claim 3, characterized by, Based on the actual control parameters, position error, synchronization error and energy consumption are calculated respectively, and a reward function containing position error, synchronization error and energy consumption optimization is constructed, including: An exponential decay function is used to design the penalty term of the position error, and the formula is: wherein, is a position error sensitivity coefficient, is an actual position of the motor, is a target position of the motor; A Gaussian kernel function is used to design the penalty term of the synchronization error, and the formula is: wherein M is the number of motors, is a synchronization error sensitivity coefficient, and are actual positions of motors i and j; The optimization term of the energy consumption is designed in the form of current flat integral, and the formula is: wherein, is the energy consumption weight coefficient, is the motor current.

6. The high precision position holding method according to claim 1, wherein Before aligning the position data, the speed data, the torque data, the environmental data and the current data based on the time stamp to construct a multi-dimensional data set, and normalizing the multi-dimensional data set to obtain standard data, the method further includes: Wavelet transform is used to perform multi-scale decomposition on the multi-dimensional data set to remove high-frequency noise.

7. The high precision position holding method according to claim 6, wherein The wavelet transform is used to perform multi-scale decomposition on the multi-dimensional data set to remove high-frequency noise, including: The multi-dimensional data is decomposed by a wavelet basis function to obtain different decomposition layers; An adaptive threshold is used to denoise the different decomposition layers; The high-frequency coefficients in the different decomposition layers are processed by a soft threshold function to obtain denoised coefficients; The denoised coefficients and low-frequency approximation coefficients are reconstructed to complete high-frequency noise removal.

8. A high-precision position holding device of multi-dimensional motor cooperative control, characterized by, It includes: The acquisition unit is used to collect position data, speed data, torque data, environmental data, vibration data and current data of multiple motors in real time; The alignment unit is used to align the position data, the speed data, the torque data, the environmental data, the vibration data and the current data based on the time stamp to construct a multi-dimensional data set; The processing unit is used to normalize the multi-dimensional data set to obtain standard data; The decomposition unit is used to calculate the covariance matrix based on the standard data and perform eigenvalue decomposition; The formation unit is used to determine a plurality of principal component feature vectors according to the cumulative variance contribution rate to form a principal component space; The projection unit is used to project the standard data into the principal component space to form a reduced dimension data set; The output unit is used to input the reduced dimension data set into a pre-trained deep belief network, output a comprehensive feature vector through a hierarchical feature of a restricted Boltzmann machine, and the comprehensive feature vector is used to represent position error dynamic characteristics, load torque coupling relationship, environmental disturbance response characteristics and dynamic compensation parameters required for multiple motor cooperative control of multiple motors; The construction unit is used to input the comprehensive feature vector into a deep deterministic policy gradient algorithm model to construct a reward function; An obtaining unit is configured to output a real-time adjustment strategy of current loop parameters, speed loop compensation coefficients and position loop parameters of a plurality of motors based on the reward function, and obtain optimized control parameters. A maintaining unit is configured to adjust the current loop parameters, the speed loop compensation coefficients and the position loop parameters of the plurality of motors in real time according to the optimized control parameters, so as to realize high-precision position maintaining of the plurality of motors.

9. A high-precision position holding device of multi-dimensional motor cooperative control, characterized by, The application relates to a high-precision position maintaining method, comprising the following steps: A processor, a memory, an input / output unit and a bus, wherein the processor is connected with the memory, the input / output unit and the bus, the memory stores a program, and the processor invokes the program to execute the high-precision position maintaining method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a program, and the program is executed on the computer to execute the high-precision position maintaining method according to any one of claims 1 to 7.

Images