Intelligent control method and system based on electro-hydraulic linkage

Abstract

The invention provides an intelligent control method and system based on electro-hydraulic linkage. The method comprises the steps that multi-mode environment parameters are collected, a digital twin environment model is constructed, and data are synchronized with an electro-hydraulic servo system in real time. And establishing a nonlinear model, identifying parameters by using a least square method and a recursive least square method, verifying the precision through a step response test, and updating. And constructing an adaptive LSTM neural network inverse model to output a control voltage signal. And the signal is subjected to weighted fusion with the output of a PID controller, and the PID controller dynamically adjusts a gain parameter according to an error integral term to generate a control instruction to drive the system. According to the invention, the LSTM inverse model provides feed-forward compensation, the PID controller realizes feedback correction, and the LSTM inverse model and the PID controller are complementary. Meanwhile, based on a related nonlinear model and a parameter self-adaptive updating mechanism, the influence of working condition change is accurately described, and high-precision control of the electro-hydraulic servo system under the complex working condition is achieved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to an intelligent control method and system based on electro-hydraulic linkage. Background Technology

[0002] Electro-hydraulic servo systems, as core execution units in industrial automation, are widely used in CNC machine tools, construction machinery, aerospace, and other scenarios, undertaking high-precision displacement / force control tasks. Currently, the mainstream control method is classical PID control, which, due to its simple structure and ease of engineering implementation, can meet basic control requirements under stable operating conditions. With the development of sensor technology, the acquisition of multimodal environmental parameters (such as temperature, pressure, and vibration) has become increasingly widespread. Some advanced systems are beginning to explore combining adaptive control, fuzzy control, and other technologies to address the challenges of nonlinear and time-varying operating conditions. At the modeling level, traditional methods often employ linearized models or piecewise fitting strategies, combined with offline methods such as step response testing for parameter calibration, initially achieving an approximate description of key physical quantities such as the flow-pressure characteristics and friction effects of hydraulic pumps.

[0003] However, traditional PID control parameters are fixed, and under operating conditions such as load fluctuations and oil temperature changes, it is easy to overshoot or steady-state error, making it difficult to maintain high-precision control. Summary of the Invention

[0004] This invention aims to at least solve the technical problems existing in the prior art, and in particular, it innovatively proposes an intelligent control method based on electro-hydraulic linkage.

[0005] To achieve the above-mentioned objectives of this invention, this invention provides an intelligent control method based on electro-hydraulic linkage, the method comprising: S1. Collect multimodal environmental parameters and perform data fusion processing through Kalman filtering to obtain fused environmental state information; S2. Based on the environmental state information, a target detection algorithm is used to identify dynamic obstacles and construct a digital twin environment model, and real-time data synchronization is performed with the electro-hydraulic servo system; S3. Based on the historical operating data and environmental status information of the electro-hydraulic servo system, a nonlinear model including the hydraulic pump flow-pressure characteristic curve, the Stribeck friction model and the valve port flow formula is established. The least squares method and the recursive least squares method are used to identify parameters, and the accuracy of the model is verified by step response test. The parameters are then updated adaptively. S4. Based on the nonlinear model, the state variables of the electro-hydraulic servo system are estimated by a full-dimensional state observer, and the state variables are used to construct an adaptive LSTM neural network inverse model based on the LSTM network to output the control voltage signal. S5. The control voltage signal and the output of the PID controller are weighted and fused. The PID controller dynamically adjusts the proportional-integral-derivative gain parameters by calculating the error integral term between the output of the adaptive LSTM neural network inverse model and the desired trajectory in real time, and finally generates control commands to drive the electro-hydraulic servo system.

[0006] On the other hand, the present invention also provides an intelligent control system based on electro-hydraulic linkage, the system including a processor and a memory for storing processor-executable instructions; The processor is configured to implement an intelligent control method based on electro-hydraulic linkage when executing the executable instructions.

[0007] The beneficial effects of this invention are as follows: This invention effectively solves the problems of overshoot, steady-state error, and poor adaptability caused by fixed parameters in traditional PID control by dynamically weighting and fusing the inverse model of an adaptive LSTM neural network with the output of a PID controller, and by adjusting the PID gain parameters in real time with the error integral term. Specifically, the LSTM inverse model provides feedforward compensation to capture the nonlinear characteristics of the system, and the PID controller achieves feedback correction through dynamic gain adjustment (proportional gain to suppress overshoot, integral gain to eliminate steady-state error, and derivative gain to suppress oscillation). The two are then weighted and fused to form a complementary control strategy. Simultaneously, based on the nonlinear model of the hydraulic pump flow-pressure characteristics, the Stribeck friction model, and the valve orifice flow formula, combined with the parameter adaptive update mechanism of least squares and recursive least squares methods, it can accurately characterize the impact of changes in operating conditions such as oil temperature and load on the system dynamics, ensuring model accuracy and control robustness, and ultimately achieving high-precision, adaptive, and long-cycle stable control of the electro-hydraulic servo system under complex operating conditions.

[0008] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0009] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart of an intelligent control method based on electro-hydraulic linkage according to the present invention. Detailed Implementation

[0010] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0011] Example 1 like Figure 1 As shown, an intelligent control method based on electro-hydraulic linkage is disclosed, the method comprising: S1. Collect multimodal environmental parameters and perform data fusion processing through Kalman filtering to obtain fused environmental state information; In step S1, the multimodal environmental parameters include hydraulic oil temperature, hydraulic pump outlet pressure, hydraulic cylinder piston displacement, piston speed, proportional valve spool opening, and external load force. These parameters are acquired in real time through corresponding sensors to ensure data accuracy and real-time performance. Sensor placement must consider the system's dynamic characteristics to avoid signal interference and delays. For example, temperature sensors are installed near the hydraulic pump outlet pipeline for rapid response to oil temperature changes, and pressure sensors are placed before and after the valve ports for accurate pressure differential measurement. The acquired raw data is preprocessed (e.g., filtered, normalized) and then input into a Kalman filter. This filter uses a predict-update two-step cyclic framework, combined with the system's dynamic model and measurement noise characteristics, to dynamically adjust the reliability of the measurement data, ultimately outputting fused environmental state information.

[0012] S2. Based on the environmental state information, a target detection algorithm is used to identify dynamic obstacles and construct a digital twin environment model, and real-time data synchronization is performed with the electro-hydraulic servo system; S3. Based on the historical operating data and environmental status information of the electro-hydraulic servo system, a nonlinear model including the hydraulic pump flow-pressure characteristic curve, the Stribeck friction model and the valve port flow formula is established. The least squares method and the recursive least squares method are used to identify parameters, and the accuracy of the model is verified by step response test. The parameters are then updated adaptively. S4. Based on the nonlinear model, the state variables of the electro-hydraulic servo system are estimated by a full-dimensional state observer, and the state variables are used to construct an adaptive LSTM neural network inverse model based on the LSTM network to output the control voltage signal. S5. The control voltage signal and the output of the PID controller are weighted and fused. The PID controller dynamically adjusts the proportional-integral-derivative gain parameters by calculating the error integral term between the output of the adaptive LSTM neural network inverse model and the desired trajectory in real time, and finally generates control commands to drive the electro-hydraulic servo system.

[0013] The principle of the intelligent control method based on electro-hydraulic linkage in this embodiment is as follows: First, multimodal environmental parameters are collected by sensors and fused using Kalman filtering to construct a digital twin environmental model to map the system dynamics in real time. Second, a nonlinear model is established based on historical operating data and environmental state information. This model integrates the hydraulic pump flow-pressure characteristics, the Stribeck friction model, and the valve flow formula. Parameters are adaptively updated using least squares and recursive least squares methods to ensure that the model accuracy is dynamically adjusted according to changes in operating conditions. Further, a full-dimensional state observer is used to estimate the system state variables, and an adaptive LSTM neural network inverse model is constructed. Its input layer receives the concatenated vector of the state estimate and the desired trajectory, and the output layer generates a control voltage signal through a fully connected layer. An online gradient descent mechanism is introduced to dynamically adjust the network weights to capture the nonlinear characteristics of the system. Finally, the control voltage signal output from the LSTM inverse model is weighted and fused with the output of the PID controller. The PID controller adjusts the proportional-integral-derivative gain parameters in real time according to the error integral term, forming a complementary control strategy of feedforward compensation and feedback correction to drive the electro-hydraulic servo system to achieve high-precision motion control. This embodiment effectively solves the problems of overshoot, steady-state error and poor adaptability of traditional PID control under complex working conditions through dynamic weighted fusion mechanism and adaptive parameter update, and significantly improves the control robustness and working condition adaptability of electro-hydraulic servo system.

[0014] As an optional embodiment of the present invention, optionally, step S2, which involves using a target detection algorithm to identify dynamic obstacles and construct a digital twin environment model based on the environmental state information, includes: S201. Perform spatiotemporal alignment and noise removal on the environmental state information; In step S201, it is necessary to explain in detail that spatiotemporal alignment is mainly achieved through timestamp synchronization and coordinate system transformation. First, the data collected by each sensor needs to be appended with a high-precision timestamp. A time synchronization algorithm (such as the least squares time synchronization method) is used to eliminate the time deviation of multi-source data, ensuring that the environmental state information at the same moment is aligned in the time dimension. Second, for the spatial coordinate differences caused by the different sensor installation positions, a coordinate system transformation matrix is ​​used to unify mechanical quantities such as hydraulic cylinder piston displacement and valve core opening with physical quantities such as pressure and temperature to the same spatial coordinate system. For example, the valve core opening data is mapped to the coordinate system of the hydraulic cylinder piston movement trajectory, providing a spatial reference for subsequent obstacle localization. The noise removal stage uses the existing sliding window filtering algorithm. The window length and filtering threshold are set according to the dynamic characteristics of the system to smooth the instantaneous changes in parameters such as hydraulic oil temperature and pressure, while retaining the data fluctuations caused by the actual working condition changes, avoiding the loss of key information due to over-filtering. For example, for the pulse fluctuations of hydraulic pump outlet pressure, the filter cutoff frequency is set by analyzing the fluctuation frequency of historical data, effectively filtering out high-frequency measurement noise while retaining low-frequency working condition change signals. After spatiotemporal alignment and noise removal, the spatiotemporal consistency and data purity of environmental status information are significantly improved.

[0015] S202. The target detection algorithm is used to perform cross-scale feature fusion detection on the environmental state information to obtain the dynamic obstacle detection box coordinates, category confidence and motion trajectory prediction based on Kalman filtering. In step S202, it is necessary to explain in detail that the target detection algorithm adopts the existing YOLOv5 model. This model enhances the detection capability of small-sized obstacles by introducing a Cross-Scale Feature Fusion Module (CSFFM). Specifically, CSFFM performs bidirectional feature fusion between shallow feature maps (high resolution, low semantic information) and deep feature maps (low resolution, high semantic information). The shallow features are concatenated with the deep features after adjusting the number of channels through a 1×1 convolution, and then a fused feature map is generated through a 3×3 convolution, which preserves spatial detail information and enhances semantic expressiveness. For dynamic obstacle detection, the model output includes the coordinates of the detection box (center point coordinates, width and height), class confidence (e.g., fixed obstacle, moving obstacle), and the predicted motion trajectory based on Kalman filtering. The Kalman filter uses the center point coordinates of the detection box as the observation value, combines it with the system state transition model (assuming uniform motion) to predict the position at the next moment, and corrects the prediction bias through observation updates, finally outputting a smooth motion trajectory. For example, when a moving obstacle is detected, the model not only outputs its current position and category, but also predicts its movement path within the next 3 seconds. In addition, to improve detection robustness, data augmentation strategies (such as random scaling, rotation, and adding Gaussian noise) are used during model training to simulate complex working conditions, and the model is jointly trained on public datasets (such as COCO and KITTI) and custom industrial scene datasets to ensure the model's generalization ability to obstacles of different sizes and shapes.

[0016] S203. Based on the coordinates of the dynamic obstacle detection box, the category confidence level, and the motion trajectory prediction based on Kalman filtering, a three-dimensional digital twin model is constructed using the Unity engine integrated with the PhysX physics engine, and the dynamic parameters of the hydraulic system are embedded to achieve dynamic mapping between the virtual and real systems.

[0017] In step S203, it is necessary to explain in detail that the Unity engine reads the coordinates (center point coordinates, width and height) and category confidence of the dynamic obstacle detection box, and maps the detection results to a solid model in three-dimensional space. For example, fixed obstacles are rendered as immovable static rigid bodies (such as walls and equipment bases), while moving obstacles are generated as dynamic rigid bodies that can be acted upon by the physics engine (such as transport vehicles and personnel models). For the dynamic parameters of the hydraulic system, the physical constraints of the hydraulic cylinder piston movement and valve core opening are simulated through the joint components (such as hinge joints and sliding joints) of the PhysX engine, mapping parameters such as hydraulic pump outlet pressure and hydraulic oil temperature to physical properties such as the mass and damping coefficient of the rigid body. For example, the hydraulic cylinder piston displacement data drives the linear movement of the piston in the virtual model, and the pressure value is converted into a force on the virtual load through the force feedback module, achieving synchronization of the virtual and real system movements. Simultaneously, using Unity's Timeline function and PhysX's collision detection mechanism, the obstacle position and system state are updated in real time. For example, when a virtual obstacle collides with an actuator model, a safety stop logic is triggered and fed back to the actual control system. To improve the accuracy of dynamic mapping, the model employs a double-buffering mechanism: one frame is used for rendering and display, while the other is used for data synchronization and physical calculations, ensuring the real-time nature of virtual-real interaction. Furthermore, data interaction is achieved between the model and a hydraulic system digital model built with MATLAB / Simulink via Unity's C# script interface. Parameters such as pressure and flow rate from the virtual environment are fed back to the nonlinear model, forming a closed-loop verification system to further guarantee the consistency of virtual-real mapping.

[0018] As an optional embodiment of the present invention, the adaptive parameter update in step S3 may include: S301. Preprocess the historical operating data and environmental status information of the electro-hydraulic servo system and extract key feature variables; In step S301, it is necessary to explain in detail that the preprocessing stage mainly addresses the correction of outliers, missing values, and data scale inconsistencies in historical operating data and environmental status information. First, an outlier detection algorithm based on the 3σ principle is used to statistically filter continuous variables such as hydraulic pump outlet pressure and hydraulic oil temperature. If data at a certain moment deviates from the mean by more than three times the standard deviation, it is identified as an outlier and replaced with the average value of the preceding and following moments. For example, when the hydraulic pump outlet pressure suddenly jumps to twice the normal value due to a sensor malfunction, the algorithm automatically identifies and corrects it to a value within a reasonable range. Second, regarding missing values, different filling strategies are selected based on the proportion of missing data: if the missing proportion is less than 5%, linear interpolation is used to fill in data from adjacent moments; if the missing proportion is between 5% and 15%, a regression model is constructed based on historical data from the same period (such as the pressure-flow curve under the same operating condition) for predictive filling; if the missing proportion exceeds 15%, the data for that period is directly removed and marked as invalid data segment. Finally, to eliminate the problem of significant numerical scale differences caused by the dimensional differences of different physical quantities (such as pressure, temperature, and displacement), the Min-Max normalization method is used to map all data to the [0,1] interval. For example, the original value range of hydraulic oil temperature is 20-80℃, which becomes 0.25-1.0 after normalization, and is on the same order of magnitude as the pressure value (such as 0-1MPa, which is normalized to 0-1).

[0019] The key feature variable extraction stage combines the physical characteristics and control requirements of the electro-hydraulic servo system to screen out features that significantly affect the system dynamics from the preprocessed data. Specifically, the feature variables are divided into three categories: first, system state features, including hydraulic cylinder piston displacement, piston speed, and proportional valve spool opening, which directly reflect the actuator's motion state; second, environmental disturbance features, covering hydraulic oil temperature, hydraulic pump outlet pressure, and external load force, which characterize the impact of changes in operating conditions on the system; and third, control performance features, selecting the error integral term of the PID controller (a weighted sum of proportional-integral-derivative errors) and the control voltage signal output by the LSTM inverse model, which are used to evaluate the effectiveness of the control strategy. To further reduce data dimensionality and improve model training efficiency, principal component analysis (PCA) is used to reduce the dimensionality of the above features: first, the covariance matrix of all features is calculated, then the eigenvalues ​​and eigenvectors of the matrix are solved, and the principal components with a cumulative contribution rate exceeding 95% are selected as the final feature variables. For example, if the original feature dimension is 10, after PCA dimensionality reduction, the first 4 principal components (cumulative contribution rate of 96%) may be retained. This not only retains 96% of the information in the original data, but also reduces the feature dimension to 4, which significantly reduces the computational burden of subsequent nonlinear models and LSTM inverse models.

[0020] All the methods for extracting key feature variables mentioned above are based on existing technologies, as are the methods for calculating the covariance matrix of all features. Therefore, this embodiment will not elaborate further.

[0021] S302. Based on the key characteristic variables and the physical mechanism of the hydraulic system, a piecewise nonlinear model structure is constructed. The flow-pressure characteristics of the hydraulic pump are fitted with a quadratic polynomial. The Stribeck friction model describes the static-dynamic friction transition characteristics through an exponential function. The valve orifice flow formula combines the thin-walled orifice flow theory with the Bernoulli equation to perform dynamic pressure-flow mapping. The method for constructing the piecewise nonlinear model structure in step S302 is as follows: First, considering the flow-pressure characteristics of the hydraulic pump, since it exhibits a complex nonlinear relationship under different operating conditions, a quadratic polynomial fitting can effectively capture this characteristic. Specifically, with the hydraulic pump outlet pressure as the independent variable and the flow rate as the dependent variable, a large amount of historical data is collected, and the coefficients of each term of the quadratic polynomial are determined using the least squares method, thereby obtaining the hydraulic pump flow-pressure characteristic curve equation. For example, in practical applications, the quadratic polynomial equation obtained through data fitting can accurately describe the flow rate variation of the hydraulic pump under different pressures.

[0022] The Stribeck friction model focuses on describing the static-dynamic friction transition characteristics. Static friction is high when an object is stationary, decreases rapidly as the object begins to move, and then gradually stabilizes with increasing speed. Using an exponential function to describe this process accurately depicts this trend of frictional change. By setting appropriate parameters for the exponential function, such as the coefficients and constants of the exponent, and combining this with actual measured friction and speed data, a Stribeck friction model that conforms to the actual conditions of an electro-hydraulic servo system can be constructed. For example, during the movement of the valve core in an electro-hydraulic servo valve, this model can accurately predict the frictional force experienced by the valve core at different speeds, providing accurate frictional compensation for system control.

[0023] Regarding the valve orifice flow formula, a dynamic pressure-flow mapping is achieved by combining the thin-walled orifice flow theory and Bernoulli's equation. The thin-walled orifice flow theory describes the relationship between flow rate and pressure difference when a liquid passes through a thin-walled orifice, while Bernoulli's equation reflects the principle of energy conservation during fluid flow. Combining these two theories and considering the flow characteristics of hydraulic oil at the valve orifice, such as velocity and pressure, a dynamic relationship formula between valve orifice flow and pressure is derived. In practical applications, the parameters in the formula are determined based on the geometric parameters of the valve orifice (such as orifice diameter and length) and the physical properties of the hydraulic oil (such as density and viscosity), thus obtaining an accurate valve orifice flow formula. For example, in an electro-hydraulic servo system, this formula can accurately calculate the flow rate through the valve orifice based on the pressure difference before and after the valve, providing a basis for system flow control. Through the construction of these three models, a complete piecewise nonlinear model structure is formed, which can comprehensively and accurately describe the dynamic characteristics of the electro-hydraulic servo system.

[0024] S303. Based on the piecewise nonlinear model structure, the least squares method is used to batch process the historical running data to identify the initial parameters offline. Then, the model parameters of the piecewise nonlinear model structure are updated online by combining the recursive least squares method with real-time environmental status information. S304. Apply a standard step input signal to the electro-hydraulic servo system based on the updated model parameters, collect the actual output response of the electro-hydraulic servo system, compare the actual output response with the prediction result of the piecewise nonlinear model structure, verify the model accuracy of the piecewise nonlinear model structure, and trigger the parameter re-identification process when the error of the model accuracy exceeds the threshold.

[0025] When implementing step S304, the evaluation criteria for model accuracy must first be clarified, typically using root mean square error (RMSE) or mean absolute error (MAE) as quantitative indicators. Specifically, the actual output response of the electro-hydraulic servo system under a standard step input signal (such as hydraulic cylinder piston displacement, valve opening, etc.) is compared point-by-point with the prediction results of the piecewise nonlinear model structure, calculating the error value at each sampling moment. For example, if the actual piston displacement is 5mm and the model prediction is 4.8mm, then the absolute error at that moment is 0.2mm. Subsequently, the error values ​​at all sampling moments are statistically processed: if using the RMSE indicator, the square root of the average of the squared errors is calculated; if using the MAE indicator, the average of the absolute values ​​of the errors is directly calculated. The final RMSE or MAE value is the quantitative representation of the model accuracy.

[0026] To set a reasonable error threshold, it is necessary to consider the actual control requirements and safety specifications of the electro-hydraulic servo system. For example, if the system requires piston displacement control accuracy within ±0.5mm, the error threshold can be set to the RMSE or MAE value corresponding to 0.5mm. When the quantification result of the model accuracy exceeds this threshold, the system automatically triggers the parameter re-identification process. At this time, it is necessary to re-collect the historical operating data and environmental status information of the electro-hydraulic servo system, and perform key feature variable extraction, piecewise nonlinear model structure construction, and parameter updates according to steps S301 to S303. It is worth noting that the parameter re-identification process can adopt an incremental update strategy, that is, only local parameter adjustments are made to the model segment with large errors (such as the hydraulic pump flow-pressure characteristic segment or the Stribeck friction model segment), rather than global re-identification, thereby improving the re-identification efficiency.

[0027] Furthermore, to avoid system oscillations caused by frequent parameter re-identification, a hysteresis threshold and time window mechanism can be introduced into the triggering conditions. For example, the re-identification process should only be formally triggered when the model accuracy exceeds the threshold for three consecutive sampling periods, or when the number of times the threshold is exceeded reaches five times within 10 minutes. Simultaneously, during parameter re-identification, the basic control functions of the electro-hydraulic servo system must be maintained. This can be ensured by switching to a backup control strategy (such as simplified PID control) or reducing the system operating speed (such as reducing the maximum piston speed from 100 mm / s to 50 mm / s). After parameter re-identification is completed, model accuracy verification must be performed again until the error threshold requirements are met before resuming normal control mode. Through these measures, the long-term stability and adaptability of the piecewise nonlinear model structure can be effectively guaranteed.

[0028] As an optional embodiment of the present invention, optionally, the expression for the flow-pressure characteristic of the hydraulic pump in the piecewise nonlinear model structure in step S302 is: in, Indicates the output flow rate of the hydraulic pump. , , and The coefficients of the quadratic / linear terms in the temperature-pressure coupling are represented and identified offline using the least squares method to characterize oil temperature. The influence of nonlinear characteristics on the flow-pressure curve. Indicates the outlet pressure of the hydraulic pump. Indicates the real-time temperature of the hydraulic oil. This represents the temperature compensation base coefficient. Represents the temperature decay rate coefficient. Indicates reference temperature; The expression for the Stribeck friction model in the piecewise nonlinear model structure is: in, This represents the total frictional force. Represents Coulomb friction. This represents the maximum static friction force. Indicates the transition speed threshold. Indicates the speed of moving parts. This represents the Stribeck effect index. Represents a symbolic function. Indicates the coefficient of viscous friction; The expression for the valve orifice flow rate formula in the piecewise nonlinear model structure is as follows: in, Indicates the valve outlet flow rate. Indicates the flow coefficient. Indicates the effective flow area of ​​the valve port. Indicates the valve core opening degree. Indicates the density of hydraulic oil. Indicates the compressibility coefficient of the oil. This indicates the pressure difference across the valve port.

[0029] As an optional embodiment of the present invention, the online updating of the model parameters of the piecewise nonlinear model structure in step S303 may include: S3031. Based on the key feature variables, a linear regression matrix is ​​constructed using the hydraulic pump flow-pressure characteristic curve, Stribeck friction model, and valve orifice flow formula in the piecewise nonlinear model structure. The method for constructing the linear regression matrix in step S3031 is as follows: First, the basic form of the linear regression matrix is ​​defined as Y = Xβ + ε, where Y is the dependent variable vector, X is the independent variable matrix, β is the parameter vector to be estimated, and ε is the error vector. For the electro-hydraulic servo system in this embodiment, taking the hydraulic pump flow-pressure characteristic curve as an example, the hydraulic pump outlet pressure is included as one of the independent variables in the X matrix, and the corresponding flow rate is part of the Y vector. Simultaneously, considering the influence of hydraulic oil temperature on flow rate, the real-time temperature is also included as an independent variable in the X matrix, and corresponding temperature-pressure coupling terms (such as the product of temperature and pressure) are set to capture nonlinear effects. For the Stribeck friction model, the velocity of the moving parts is used as the independent variable, and the total friction force is used as the dependent variable to construct a regression relationship. The square term and absolute value term of the velocity are added to the X matrix to describe the static-dynamic friction transition characteristics. For the valve orifice flow rate formula, the valve core opening and the pressure difference before and after the valve orifice are used as independent variables, and the valve orifice output flow rate is used as the dependent variable. Simultaneously considering the influence of physical parameters such as hydraulic oil density and oil compressibility coefficient on flow rate, these parameters are combined with the independent variables to construct an extended X matrix. Specifically, each row of the X matrix corresponds to data at a sampling time, and each column corresponds to a feature variable (such as pressure, temperature, velocity, opening degree, etc.) or its derivatives (such as square terms, product terms, etc.); each element of the Y vector corresponds to the dependent variable (such as flow rate, friction force, etc.) at the same time. By collecting a large amount of historical operating data, the X matrix and Y vector are filled, and then the least squares method is used to solve for the β vector, thus obtaining the parameter estimates for each model segment.

[0030] S3032. Based on the linear regression matrix, the least squares method is used for batch offline identification to solve for the initial parameter set; The method for solving the initial parameter set in step S3031 is as follows: After constructing the linear regression matrix X and the dependent variable vector Y, the parameter vector β is solved mathematically using the least squares method. Its core principle is to minimize the sum of squared residuals, that is, to minimize the sum of squared errors between the actual observed value Y and the model predicted value Xβ. The specific solution process can be expressed as follows: First, construct the objective function J(β) = (Y - Xβ)ᵀ(Y - Xβ), where the superscript ᵀ denotes matrix transpose; then, take the partial derivative of J(β) with respect to β and set it equal to zero to obtain the normal equation XᵀXβ = XᵀY; finally, solve for β = (XᵀX) through matrix operations. -1 XᵀY, where (XᵀX) -1Let X be the inverse matrix of X. For example, when the X matrix contains hydraulic pump pressure, temperature, and their coupling terms, and the Y vector represents the corresponding flow rate, the elements in the solved β vector correspond to the coefficients of each characteristic variable, including the coefficients of the quadratic and linear terms of the temperature-pressure coupling, thus forming a complete initial parameter set for the hydraulic pump flow-pressure characteristics. For the Stribeck friction model, the X matrix contains velocity, velocity square terms, and absolute value terms, and the Y vector represents the total friction force. The elements in the solved β vector correspond to parameters such as Coulomb friction, maximum static friction, and transition speed threshold, constituting the initial parameter set for the friction model. For the valve orifice flow formula, the X matrix contains valve spool opening, pressure difference, hydraulic oil density, and oil compressibility coefficient, and the Y vector represents the valve orifice output flow rate. The elements in the solved β vector correspond to parameters such as the flow coefficient and effective flow area, forming the initial parameter set for the valve orifice flow model. Through the above methods, the initial parameter sets for the hydraulic pump flow-pressure characteristics, the Stribeck friction model, and the valve orifice flow formula can be obtained respectively.

[0031] S3033. Initialize the recursive least squares method based on the initial parameter set, construct an online update framework by combining real-time environmental state information, and perform state estimation through a Kalman filter to filter out measurement noise. In step S3033, it is important to explain in detail that the initialization process of the recursive least squares method relies on the initial parameter set obtained in step S3032. This parameter set provides an initial benchmark for online updates. Specifically, the initial parameter set is used as the initial estimate for the recursive least squares method, and an appropriate covariance matrix is ​​set to characterize the uncertainty of the parameter estimates.

[0032] The method for constructing an online update framework by combining real-time environmental state information (such as hydraulic oil temperature, pressure, valve spool opening, etc.) is as follows: First, the real-time collected environmental state information is introduced as a new input variable into the iterative process of the recursive least squares method. For example, when the hydraulic oil temperature changes, this temperature value is fed back to the online update framework in real time and used as one of the bases for parameter correction. Simultaneously, a Kalman filter is used to filter out measurement noise. Its core principle is to optimize the state estimation through two steps: prediction and update. In the prediction step, the state at the current moment is predicted based on the state estimate value at the previous moment and the system dynamic model. In the update step, the state estimate is adjusted by combining the current measured value (such as pressure and flow data collected by sensors) and the predicted value, thereby obtaining the optimal state estimate after noise filtering. Specifically, for the hydraulic pump flow-pressure characteristic model, the Kalman filter will correct the temperature-pressure coupling coefficient in real time; for the Stribeck friction model, it will dynamically adjust the maximum static friction force and the transition speed threshold; for the valve orifice flow formula, it will optimize parameters such as the flow coefficient and effective flow area. This online update framework allows model parameters to adapt to changes in system operating status in real time, ensuring the accuracy and robustness of the piecewise nonlinear model structure.

[0033] S3034. Based on the aforementioned online update framework, the model parameter vector of the piecewise nonlinear model structure is updated in real time using the recursive least squares method, including calculating the gain matrix, updating the parameter vector estimates, and updating the covariance matrix. In step S3034, it is necessary to explain in detail that the real-time update process of the recursive least squares method is a cyclical process of dynamically adjusting the parameter vector. First, based on the optimal state estimate provided by the Kalman filter, the gain matrix at the current time is calculated. The calculation of the gain matrix depends on the covariance matrix and the statistical characteristics of the measurement noise, and its role is to determine the correction strength of the parameter estimate by the new measurement information. For example, when the hydraulic oil temperature fluctuates greatly, the gain matrix will increase the adjustment weight of the corresponding temperature-related parameters, thereby accelerating the model's response to environmental changes.

[0034] Subsequently, the parameter vector estimates are updated using the gain matrix. Specifically, newly acquired environmental state information (such as real-time pressure and valve opening) is compared with the model predictions, and the parameter vector is corrected by weighting it using the gain matrix. For example, if there is a deviation between the actual valve flow rate and the model prediction, the system will adjust the estimated values ​​of the flow coefficient and effective flow area according to the gain matrix, making the model output closer to the actual system behavior. Simultaneously, the covariance matrix is ​​also updated to reflect changes in the uncertainty of the parameter estimates. When the measurement noise is low, the covariance matrix shrinks, indicating an increase in the confidence of the parameter estimates; conversely, it expands.

[0035] This process iterates continuously using a recursive formula: for each new environmental state data point acquired, the system completes a "measurement-correction-update" cycle. For example, in updating the flow-pressure characteristic model of a hydraulic pump, if the real-time temperature rises from 40℃ to 50℃, the system will prioritize adjusting the temperature-pressure coupling coefficient through the gain matrix, while the covariance matrix records the impact of this adjustment on parameter uncertainty. Through this mechanism, the model parameters can dynamically track changes in the system's operating state, avoiding the computational burden of global re-identification and ensuring the long-term adaptability of the piecewise nonlinear model structure under complex working conditions.

[0036] S3035. When the real-time observation error exceeds the preset dynamic threshold, the model parameter vector is updated, and the update condition is stopped by judging the convergence.

[0037] In step S3035, it is important to explain in detail that the determination of real-time observation error relies on a preset dynamic threshold, which is dynamically adjusted according to the actual operating conditions and accuracy requirements of the electro-hydraulic servo system. For example, when the deviation between the measured value of the hydraulic pump output flow or valve port flow and the model prediction value exceeds the threshold range (e.g., ±5% of rated flow), or when the error between the total friction force calculated by the Stribeck friction model and the sensor feedback value exceeds a set proportion (e.g., ±10% of rated friction force), the system will automatically trigger the model parameter vector update process. After triggering, the recursive least squares method corrects the parameter vector through the gain matrix, while the Kalman filter continuously optimizes the state estimation until the observation error converges to the threshold range. The convergence judgment adopts a dual criterion: first, whether the absolute value of the error is continuously lower than the threshold for N sampling periods (e.g., N=5); second, whether the rate of change of the parameter vector update amplitude is less than the set value (e.g., Δβ / Δt<0.01). When the above conditions are met simultaneously, the system determines that the parameter update is complete, stops the iteration, and returns to the normal control mode. For example, in updating the flow-pressure characteristic model of a hydraulic pump, if the temperature suddenly rises from 50℃ to 60℃, causing the flow error to exceed the limit, the system will prioritize adjusting the temperature-pressure coupling coefficient and dynamically adjust the uncertainty of parameter estimation through the covariance matrix until the error is below the threshold for five consecutive cycles and the parameter change rate approaches zero. At this point, the model parameter update will terminate, ensuring that the system can respond quickly to environmental changes under complex working conditions while avoiding oscillations caused by overcorrection.

[0038] As an optional embodiment of the present invention, optionally, the output of the control voltage signal in step S4 includes: S401. Using real-time sensor data and the hydraulic pump flow-pressure characteristic curve, Stribeck friction model and valve orifice flow formula in the nonlinear model, construct a full-dimensional state observer and use the full-dimensional state observer to output real-time state estimates. S402. Based on the real-time state estimate and the desired trajectory signal, construct an adaptive LSTM neural network inverse model using an LSTM network; The method for constructing the adaptive LSTM neural network inverse model in step S402 is as follows: First, the structural characteristics of LSTM (Long Short-Term Memory) are clarified, which includes an input gate, a forget gate, and an output gate, and can effectively handle long-term dependencies in sequence data. For the control requirements of the electro-hydraulic servo system, real-time state estimates (such as hydraulic pump flow rate, pressure, and moving part speed) and desired trajectory signals (such as target flow rate and target position) are used as input sequences to construct a two-layer LSTM network architecture. The first layer of LSTM is used to extract the temporal features of the input sequence, dynamically filtering historical information through the forget gate, and updating the key features of the current moment through the input gate; the second layer of LSTM further integrates the temporal features and system dynamic characteristics, outputting the predictive control quantity of the inverse model. To enhance the model's adaptability, a dynamic weight adjustment mechanism is introduced during training: when the real-time observation error (such as flow tracking error and position deviation) exceeds a threshold, the weight of error-related features is increased, making the network more focused on the correction needs of the current operating condition; simultaneously, the network's dependence on historical information is adjusted in real time by updating the cell state and hidden state of the LSTM online. For example, in a hydraulic pump flow control scenario, if the real-time flow rate is lower than the expected value, the system will increase the weight of the flow error feature in the input gate to accelerate the adjustment of the predicted value of the flow coefficient parameter. If the speed of moving parts fluctuates, the system will suppress noise interference by dynamically adjusting the retention ratio of historical speed data in the forget gate. Furthermore, to improve the model's generalization ability, data augmentation techniques are used to expand the training set: Gaussian noise is added to simulate sensor measurement errors, and time windows are used to generate sequence samples of different durations, covering the entire operating range of the system from steady state to transient state. During training, a mean squared error loss function combined with an L2 regularization term is used to prevent overfitting, and the learning rate is dynamically adjusted using the Adam optimizer to accelerate model convergence. The final adaptive LSTM neural network inverse model can output a control voltage signal that matches the desired trajectory in real time, achieving high-precision tracking control of the electro-hydraulic servo system.

[0039] S403. The input layer of the adaptive LSTM neural network inverse model receives the concatenated vector of the state estimate and the desired trajectory. The output layer generates a control voltage signal through a fully connected layer and introduces an adaptive mechanism to adjust the weights of the adaptive LSTM neural network inverse model through online gradient descent.

[0040] The expression for the output layer of the inverse model of the adaptive LSTM neural network is: in, express Control voltage signal at any time, Indicates trainable weights. This represents the hidden state vector of the LSTM layer. Indicates a trainable bias; The specific implementation of the adaptive mechanism in step S403 is as follows: First, a loss function is defined to quantify the error between the control voltage signal and the actual system response. For example, the mean square error is used as the objective function, and the model performance is evaluated by calculating the sum of squares of the deviations between the predicted control voltage and the actual required voltage. Then, an online gradient descent algorithm (such as stochastic gradient descent or its variant Adam optimizer) is used to minimize the loss function. Its core steps include: calculating the gradient of the loss function with respect to the trainable weights and biases; adjusting the weight parameters according to the gradient direction (e.g., scaling the gradient by the learning rate η and backpropagating to update), so that the subsequently predicted control voltage is closer to the actual requirement. For example, when the hydraulic pump flow tracking of the desired trajectory deviates, the system will calculate the gradient of the flow error with respect to the weights in real time, dynamically correct the weight matrix of the LSTM output layer through gradient descent, thereby adjusting the amplitude or phase of the control voltage signal until the error converges to within a preset threshold. Simultaneously, to balance convergence speed and stability, an adaptive learning rate adjustment strategy is introduced: when the error rate of change is large (e.g., the error has not decreased for N consecutive periods), the learning rate is increased to accelerate correction; when the error approaches the threshold, the learning rate is decreased to avoid oscillation. Furthermore, to prevent overfitting during weight updates, an L2 regularization term is added to the loss function to limit model complexity by constraining the norm of the weight vector. For example, in a valve flow control scenario, if real-time pressure fluctuations cause flow overshoot, the system will prioritize adjusting pressure-related weight parameters through gradient descent, while simultaneously suppressing the weight growth of irrelevant features through the regularization term, ensuring the model's generalization ability under complex operating conditions. Through this adaptive mechanism, the LSTM neural network inverse model can optimize weight parameters in real time, enabling the control voltage signal to dynamically match the nonlinear characteristics of the electro-hydraulic servo system, ultimately achieving high-precision trajectory tracking control.

[0041] As an optional embodiment of the present invention, optionally, the output of the real-time state estimate using the full-dimensional state observer in step S401 includes: S4011. Based on the hydraulic pump flow-pressure characteristic curve, Stribeck friction model and valve port flow formula, combined with real-time sensor data, define a state variable vector, and construct a nonlinear state space equation containing temperature-pressure coupling terms based on the state variable vector. The expression for the nonlinear state-space equation is: ; , , ; in, Indicates time Differentiate, Represents the state transition matrix. express The state vector at any given time includes displacement, velocity, pressure, and valve opening. Represents the control input matrix. express The control voltage generated by the LSTM inverse model at time t. Indicates temperature coupling term, integrated hydraulic pump flow rate and friction The nonlinear characteristics, express The displacement of the hydraulic cylinder piston at any given moment. express The piston speed at any given moment. express The hydraulic pump outlet pressure at any given time. express The opening degree of the proportional valve spool at any given time. express Friction at all times Indicates the mass of moving parts. Indicates the compressibility coefficient of the oil. Indicates the hydraulic pump flow rate. express The volume of the hydraulic cylinder at any given time; In step S4011, when constructing the nonlinear state-space equations containing temperature-pressure coupling terms, the complex coupling relationships between various physical quantities in the electro-hydraulic servo system must be fully considered. Temperature changes not only affect the viscosity of hydraulic oil, thus altering the flow output characteristics of the hydraulic pump, but also significantly impact friction. For example, in the Stribeck friction model, temperature fluctuations change the friction coefficients in the boundary lubrication, mixed lubrication, and fluid lubrication regions. Therefore, introducing temperature-pressure coupling terms can more accurately characterize the dynamic behavior of the system under different temperature conditions. Taking the piston motion of a hydraulic cylinder as an example, when constructing the state-space equations, the temperature-pressure coupling terms are considered as one of the key factors affecting piston displacement and velocity. When the temperature increases, leading to an increase in the hydraulic pump flow rate, the pressure inside the hydraulic cylinder also rises accordingly. This pressure change affects the piston's motion state through the coupling terms, causing changes in the piston's displacement and velocity. Simultaneously, the piston's motion, in turn, affects the hydraulic pump's flow rate and pressure, forming a complex closed-loop feedback system. By constructing nonlinear state-space equations containing temperature-pressure coupling terms, the dynamic characteristics of the electro-hydraulic servo system can be described more comprehensively. For example, in practical hydraulic systems, when the ambient temperature changes significantly, traditional linear state-space equations may not be able to accurately predict the system response. However, nonlinear state-space equations that include temperature-pressure coupling terms can more accurately estimate the real-time state of the system, such as piston displacement, velocity, and hydraulic pump outlet pressure, by considering the influence of temperature on various system parameters, thereby improving the control accuracy and stability of the entire electro-hydraulic servo system.

[0042] S4012. Linearize the nonlinear state-space equation using an extended Kalman filter, initialize the state covariance matrix and the process noise covariance matrix, and construct a prediction-update two-step loop framework. The method for constructing the prediction-update two-step cyclic framework in step S4012 is as follows: First, in the prediction step, the extended Kalman filter predicts the state vector at the next moment based on the current state estimate and the system dynamic model (i.e., the nonlinear state-space equation). Specifically, using the state transition matrix and the control input matrix, combined with the state estimate from the previous moment and the control voltage signal, the prior estimate of the state at the next moment is calculated. Simultaneously, based on the characteristics of the system process noise, the prior value of the state covariance matrix is ​​updated to reflect the uncertainty in the prediction process.

[0043] Subsequently, the update step begins. When new sensor measurement data arrives, the extended Kalman filter calculates the residual (i.e., the innovation sequence) between the measured value and the prior state estimate. By combining the residual with the measurement noise covariance matrix, the Kalman gain is calculated. This gain is used to adjust the prior state estimate to obtain the posterior state estimate. This process effectively reduces the uncertainty of the system state estimate by fusing sensor data with model predictions. Simultaneously, based on the Kalman gain and residual, the posterior value of the state covariance matrix is ​​updated, providing more accurate initial conditions for the prediction of the next time step.

[0044] Taking hydraulic pump flow control as an example, in the prediction step, the extended Kalman filter estimates the current hydraulic pump outlet pressure, piston displacement, and velocity, and combines this with the control voltage generated by the LSTM inverse model to predict the changes in these state variables at the next moment. In the update step, when new pressure and displacement sensor data arrive, the filter calculates the residual between the actual measured value and the predicted value, and adjusts the state estimate using the Kalman gain. For example, if the actual pressure is lower than the predicted value, the pressure value in the state estimate is corrected by increasing the pressure-related component in the Kalman gain, making it closer to the true value. Simultaneously, the state covariance matrix is ​​updated to reflect the impact of this correction process on system uncertainty.

[0045] Furthermore, to accommodate the nonlinear characteristics of electro-hydraulic servo systems, the extended Kalman filter performs local linearization of the nonlinear state-space equations in each iteration. This linearization approximation is achieved by calculating the Jacobian matrix of the state transition matrix and the control input matrix at the current state point. This approach allows the filter to more accurately track the nonlinear dynamics of the system while maintaining computational efficiency. For example, during the movement of a hydraulic cylinder piston, when the piston speed changes significantly, the Jacobian matrix can capture the impact of this change on the system state, thereby adjusting the linearized model and ensuring the accuracy of the state estimation.

[0046] By constructing a prediction-update two-step cyclic framework, the extended Kalman filter can estimate the state variables of the electro-hydraulic servo system in real time and accurately, thereby achieving high-precision trajectory tracking control.

[0047] S4013. The real-time measurement data and the state estimate are weighted and fused using the prediction-update two-step loop framework, and the reliability of the real-time measurement data is dynamically adjusted by Kalman gain to filter out measurement noise and correct the state prediction deviation, thereby obtaining the real-time state estimate.

[0048] In step S4013, it is important to explain in detail that the process of weighted fusion and state correction using the prediction-update two-step cyclic framework is highly dynamic and accurate. After acquiring real-time measurement data, the extended Kalman filter first compares the real-time measurement data with the prior state estimate to obtain the difference between the two, i.e., the residual. This residual reflects the degree of inconsistency between the measurement data and the model prediction; its magnitude and direction contain the true information of the system state and any potential measurement noise.

[0049] The Kalman gain is dynamically calculated based on the state covariance matrix and the measurement noise covariance matrix. The magnitude of the Kalman gain determines the weight of real-time measurement data and prior state estimates in the final posterior state estimate. When measurement noise is low, i.e., the measurement data is relatively reliable, the Kalman gain increases, allowing real-time measurement data to play a larger role in the posterior state estimate; conversely, when measurement noise is high and the reliability of the measurement data decreases, the Kalman gain decreases, relying more on the prior state estimate.

[0050] Taking pressure measurement in a hydraulic system as an example, if the pressure sensor is subjected to external interference, causing significant fluctuations in the measurement data, the measurement noise is substantial. An extended Kalman filter, by calculating the Kalman gain, reduces the weight of the measurement data in state estimation, thereby preventing excessive influence of noise on the state estimation. Conversely, if the pressure sensor is functioning correctly and the measurement data is accurate and reliable, the Kalman gain will increase, allowing the measurement data to play a greater role in correcting state prediction biases.

[0051] By dynamically adjusting the reliability of real-time measurement data, the extended Kalman filter can effectively filter out measurement noise. Simultaneously, by adjusting the prior state estimate based on the Kalman gain, a posterior state estimate, or real-time state estimate, is obtained. This real-time state estimate comprehensively considers information from model predictions and actual measurement data, enabling it to more accurately reflect the true state of the electro-hydraulic servo system. For example, during the movement of a hydraulic cylinder piston, by continuously updating the estimated piston displacement, velocity, and other state values, the piston's motion state can be monitored in real time, thereby achieving high-precision trajectory tracking control of the electro-hydraulic servo system and improving the stability and reliability of the entire system.

[0052] As an optional embodiment of the present invention, optionally, the final generation of control commands to drive the electro-hydraulic servo system in step S5 includes: S501. Based on the control voltage signal output by the adaptive LSTM neural network inverse model and the PID controller output, the control command is obtained by fusing the dynamic weighting coefficients. The weighting coefficients are adjusted in real time according to the magnitude of the error integral term to balance the control weights of feedforward compensation and feedback correction. In step S501, it is necessary to explain in detail that the adjustment mechanism of the dynamic weighting coefficients is based on intelligently allocating the contribution ratio of feedforward compensation (based on the LSTM inverse model) and feedback correction (based on the PID controller) according to the real-time error state of the system. Specifically, the weighting coefficients are dynamically generated by the error integral term through a nonlinear mapping function, which adopts a piecewise exponential decay model: when the error integral term exceeds a preset threshold, the weighting coefficients are tilted towards the PID controller to strengthen feedback correction and prevent the system from overshooting due to excessive feedforward compensation; when the error integral term is less than the threshold, the weighting coefficients are shifted towards the LSTM inverse model to increase the proportion of feedforward compensation, making full use of its nonlinear mapping capability to achieve accurate trajectory tracking. For example, in the hydraulic cylinder position control scenario, if the position deviation continues to accumulate due to sudden load changes (the error integral term increases), the system will automatically reduce the weight of the LSTM inverse model and enhance the integral action of the PID controller, quickly eliminating steady-state error through feedback correction; conversely, when the system approaches the target position, the high-precision prediction capability of the LSTM inverse model will dominate, reducing the frequent adjustments of the PID controller due to small errors, thereby avoiding system oscillation. Furthermore, to prevent control discontinuities caused by frequent switching of weighting coefficients near the threshold, a hysteresis characteristic is introduced into the mapping function (such as the hyperbolic tangent function). That is, when the error integral term fluctuates near the threshold, the weighting coefficients maintain their current values ​​until the error integral term exceeds the hysteresis bandwidth, ensuring a smooth transition of control commands. Through this dynamic weight allocation mechanism, the system can simultaneously achieve both fast response and steady-state accuracy.

[0053] S502, The PID controller calculates the dynamic gain parameter in real time based on the error integral term, wherein the proportional gain is adaptively adjusted by the error magnitude to suppress overshoot, the integral gain is adjusted by the error accumulation to eliminate steady-state error, and the derivative gain is adjusted by the error change rate to suppress system oscillation. In step S502, it is necessary to explain in detail that the dynamic gain parameter adjustment mechanism of the PID controller is the core of its high-precision control. The adaptive adjustment of the proportional gain (Kp) is based on the magnitude of the real-time error. When the error is large, the Kp value increases accordingly to speed up the system response, but at the same time, a limiting circuit prevents excessive control input from causing overshoot. When the error is close to zero, the Kp value gradually decreases to avoid the system oscillating around the target value. For example, in the position control of a hydraulic cylinder, if the deviation between the actual displacement and the target displacement exceeds the set range, the proportional gain will dynamically increase, allowing the control voltage to correct the piston movement direction more quickly. The integral gain (Ki) is adjusted through the accumulated error (i.e., the integral term of the error), and its function is to eliminate steady-state error. When there is a continuous deviation in the system, Ki will continuously accumulate the error signal and convert it into a control input until the deviation is completely eliminated; however, to prevent integral saturation, the adjustment of Ki will introduce an anti-saturation mechanism, such as pausing integration or using a variable integral limit when the integral term of the error exceeds the upper limit. The differential gain (Kd) is dynamically adjusted based on the error rate of change (i.e., the first derivative of the error), and its core function is to suppress system oscillations. When the error rate of change is too fast (such as overshoot when the system approaches the target value), Kd increases to apply damping in advance and slow down the rate of change of the control variable; conversely, when the error rate of change is small, Kd decreases to avoid over-suppressing the system's dynamic response. For example, in hydraulic pump flow control, if the flow rate of change suddenly increases, the differential gain will suppress flow overshoot by generating a reverse control voltage. Furthermore, the gain adjustment of the PID controller also incorporates a temperature compensation mechanism. Since the viscosity of hydraulic oil changes with temperature, affecting the system's dynamic characteristics, the gain parameter is corrected based on real-time temperature data using a lookup table or fitted curve to ensure optimal control performance under different operating conditions. Through this dynamic gain adjustment strategy, the PID controller can simultaneously meet the requirements of fast response, steady-state accuracy, and anti-interference capability, complementing the feedforward compensation of the LSTM inverse model to jointly achieve high-performance control of the electro-hydraulic servo system.

[0054] S503. After the fused control command is adjusted by the dynamic gain parameter, a control command is generated to drive the proportional valve of the electro-hydraulic servo system and control the hydraulic oil flow and pressure to move the actuator.

[0055] In step S503, it is important to explain in detail that after the fused control command has undergone dynamic gain parameter adjustment, its values ​​and characteristics fully reflect the current state requirements and control objectives of the system. This adjustment process ensures that the control command can respond quickly to system changes while maintaining steady-state accuracy, and effectively suppresses overshoot and oscillation. The adjusted control command is transmitted to the proportional valve drive module of the electro-hydraulic servo system. This module precisely adjusts the opening of the proportional valve according to the voltage or current signal in the command. As a key actuator of the electro-hydraulic servo system, the opening change of the proportional valve directly determines the flow rate and pressure of the hydraulic oil. When the control command requires an increase in hydraulic oil flow, the proportional valve opening increases accordingly, allowing more hydraulic oil to enter the actuator (such as a hydraulic cylinder), thereby driving the actuator to generate a larger displacement or force; conversely, when the control command requires a decrease in flow or pressure, the proportional valve opening decreases, restricting the flow of hydraulic oil and achieving precise control of the actuator's movement. In hydraulic cylinder motion control scenarios, if the control command requires the hydraulic cylinder to extend at a relatively fast speed, the proportional valve opening increases, allowing hydraulic oil to flow rapidly into the rodless chamber of the hydraulic cylinder, pushing the piston outward. Simultaneously, the system monitors the piston position and speed in real time via sensors and transmits feedback signals to the controller, forming a closed-loop control. If the actual movement speed deviates from the commanded speed, the controller dynamically adjusts the control command and the proportional valve opening until the piston movement reaches the expected target.

[0056] As an optional embodiment of the present invention, optionally, the expression for calculating the dynamic gain parameter in step S502 is: , , ; in, express Dynamic proportional gain at any given time. Indicates the base proportional gain. This represents the proportional adjustment coefficient. express Error signal at time, express Dynamic integral gain at time step (t). Indicates the basic integral gain. This represents the integral adjustment factor. Indicates time Error signal at time, express The dynamic differential gain at time step (t) Represents the fundamental differential gain. This represents the differential adjustment coefficient.

[0057] Example 2 An intelligent control system based on electro-hydraulic linkage includes a processor and a memory for storing processor-executable instructions; wherein the processor is configured to implement an intelligent control method based on electro-hydraulic linkage when executing the executable instructions.

[0058] It should be noted that the computer device includes a processor, a memory, and may also include one or more of a multimedia component, an input / output (I / O) interface, and a communication component.

[0059] The processor is used to control the overall operation of the computer device to complete all or part of the steps in the above-mentioned intelligent control method based on electro-hydraulic linkage.

[0060] Memory is used to store various types of data to support the operation of the computer device. This data may include, for example, instructions for any application or method used to operate on the computer device, as well as application-related data. Memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0061] The multimedia component may include a screen and an audio component, wherein the screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals; for example, the audio component may include a microphone for receiving external audio signals, the received audio signals may be further stored in memory or transmitted via a communication component; the audio component may also include at least one speaker for outputting audio signals.

[0062] I / O interfaces provide interfaces between the processor and other interface modules, such as keyboards, mice, buttons, etc.; these buttons can be virtual buttons or physical buttons.

[0063] The communication component is used for wired or wireless communication between the computer device and other devices; wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G or 5G, or one or more combinations thereof, and the corresponding communication component may include: Wi-Fi module, Bluetooth module, NFC module, mobile communication module.

[0064] As a preferred embodiment, the computer device may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to execute the above-described intelligent control method based on electro-hydraulic linkage.

[0065] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. An electro-hydraulic linkage based intelligent control method, characterized in that, The method comprises: S1, collecting multi-modal environment parameters, and performing data fusion processing through Kalman filtering to obtain fused environment state information; S2, based on the environment state information, using a target detection algorithm to identify dynamic obstacles and construct a digital twin environment model, and synchronizing real-time data with an electro-hydraulic servo system; S3, based on the historical operation data of the electro-hydraulic servo system and the environment state information, establishing a nonlinear model including a hydraulic pump flow-pressure characteristic curve, a Stribeck friction model and a valve port flow formula, using least squares method and recursive least squares method for parameter identification, and verifying the model accuracy through step response test, and performing parameter adaptive update; S4, based on the nonlinear model, estimating the state variables of the electro-hydraulic servo system through a full-dimensional state observer, using the state variables to construct an adaptive LSTM neural network inverse model based on LSTM network, and outputting a control voltage signal; S5, weighting and fusing the control voltage signal and the output of the PID controller, and the PID controller dynamically adjusts the proportional-integral-derivative gain parameters by real-time calculation of the error integral term of the adaptive LSTM neural network inverse model output and the expected trajectory, and finally generates a control command to drive the electro-hydraulic servo system.

2. The method of claim 1, wherein the method is based on electro-hydraulic linkage. In step S2, based on the environment state information, using a target detection algorithm to identify dynamic obstacles and construct a digital twin environment model comprises: S201, performing spatio-temporal alignment and noise removal on the environment state information; S202, using a target detection algorithm to perform cross-scale feature fusion detection on the environment state information to obtain dynamic obstacle bounding box coordinates, class confidence and Kalman filter-based motion trajectory prediction; S203, based on the dynamic obstacle bounding box coordinates, class confidence and Kalman filter-based motion trajectory prediction, using Unity engine to integrate PhysX physical engine to construct a three-dimensional digital twin model, and embedding hydraulic system dynamics parameters to realize virtual-real dynamic mapping.

3. The method of claim 1, wherein the method is based on electro-hydraulic linkage. In step S3, parameter adaptive update comprises: S301, preprocessing the historical operation data of the electro-hydraulic servo system and the environment state information, and extracting key feature variables; S302, based on the key feature variables and the physical mechanism of the hydraulic system, constructing a segmented nonlinear model structure, wherein the hydraulic pump flow-pressure characteristic is fitted by a quadratic polynomial, the Stribeck friction model describes the static-dynamic friction transition characteristic by an exponential function, and the valve port flow formula combines the thin-walled orifice flow theory and Bernoulli equation for pressure-flow dynamic mapping; S303, based on the segmented nonlinear model structure, using least squares method to batch-process offline identification of initial parameters from the historical operation data, and then using recursive least squares method to update model parameters of the segmented nonlinear model structure online combined with real-time environment state information; S304, based on the updated model parameters, a standard step input signal is applied to the electro-hydraulic servo system, the actual output response of the electro-hydraulic servo system is collected, the actual output response is compared with the prediction result of the segmented nonlinear model structure, the model accuracy of the segmented nonlinear model structure is verified, and when the error of the model accuracy exceeds a threshold value, a parameter re-identification process is triggered.

4. The method of claim 3, wherein the method further comprises: In step S302, the expression of the hydraulic pump flow-pressure characteristic in the segmented nonlinear model structure is: wherein, represents a hydraulic pump output flow rate, , , and represents a temperature-pressure coupling quadratic term / linear term coefficient, represents a hydraulic pump outlet pressure, represents a hydraulic oil real-time temperature, represents a temperature compensation base quantity coefficient, represents a temperature decay rate coefficient, represents a reference temperature; The expression of the Stribeck friction model in the segmented nonlinear model structure is: wherein denotes the total friction force, denotes the Coulomb friction force, denotes the static friction force maximum, denotes the transition velocity threshold, denotes the moving part velocity, denotes the Stribeck effect index, denotes the sign function, denotes the viscous friction coefficient; The expression of the valve port flow formula in the segmented nonlinear model structure is: wherein, represents the valve port output flow rate, represents the flow rate coefficient, represents the valve port effective flow area, represents the spool opening degree, represents the hydraulic oil density, represents the oil compressibility coefficient, represents the valve port pressure difference.

5. The method of claim 3, wherein the method further comprises: In step S303, the model parameters of the segmented nonlinear model structure are updated online, including: S3031, based on the key characteristic variables, a linear regression matrix is constructed using the hydraulic pump flow-pressure characteristic curve, the Stribeck friction model and the valve port flow formula in the segmented nonlinear model structure; S3032, based on the linear regression matrix, a least squares method is used for batch offline identification to solve an initial parameter set; S3033, based on the initial parameter set, a recursive least squares method is initialized, an online update framework is constructed combined with real-time environmental state information, and state estimation is performed through a Kalman filter to filter out measurement noise; S3034, based on the online update framework, a recursive least squares method is used to update the model parameter vector of the segmented nonlinear model structure in real time, including calculating a gain matrix, updating a parameter vector estimate value and updating a covariance matrix, S3035, when the real-time observation error exceeds a preset dynamic threshold value, the model parameter vector update is triggered, and the update condition is stopped through convergence judgment.

6. The electro-hydraulic linkage based intelligent control method of claim 1, wherein, In step S4, the output control voltage signal includes: S401, using real-time sensor data and the hydraulic pump flow-pressure characteristic curve, the Stribeck friction model and the valve port flow formula in the nonlinear model, a full-dimensional state observer is constructed, and a real-time state estimate value is output using the full-dimensional state observer; S402, based on the real-time state estimate value and the expected trajectory signal, an adaptive LSTM neural network inverse model is constructed using an LSTM network; S403, the input layer of the adaptive LSTM neural network inverse model receives a splicing vector of the state estimate value and the expected trajectory, the output layer generates a control voltage signal through a fully connected layer, and an adaptive mechanism is introduced to adjust the adaptive LSTM neural network inverse model weight through online gradient descent.

7. The method of intelligent control based on electro-hydraulic linkage according to claim 6, characterized in that, In step S401, the real-time state estimate value is output using the full-dimensional state observer, including: S4011, based on the hydraulic pump flow-pressure characteristic curve, the Stribeck friction model and the valve port flow formula, a state variable vector is defined combined with real-time sensor data, and a nonlinear state space equation containing temperature-pressure coupling terms is constructed based on the state variable vector; S4012, the nonlinear state space equation is linearized using an extended Kalman filter, a state covariance matrix and a process noise covariance matrix are initialized, and a prediction-update two-step loop framework is constructed; S4013, the real-time measurement data and state estimation value are fused by using the prediction-update double-step cycle framework, and the real-time measurement data credibility is dynamically adjusted by Kalman gain, measurement noise is filtered out, state prediction deviation is corrected, and a real-time state estimation value is obtained.

8. The method of intelligent control based on electro-hydraulic linkage according to claim 1, wherein, The final control instruction is generated in step S5 to drive the electro-hydraulic servo system, comprising: S501, based on the control voltage signal output by the adaptive LSTM neural network inverse model and the PID controller output, the dynamic weighting coefficient is used for fusion processing to obtain the fused control instruction, and the weighting coefficient is adjusted in real time according to the size of the error integral term to balance the control weight of feedforward compensation and feedback correction; S502, the PID controller calculates dynamic gain parameters in real time based on the error integral term, wherein the proportional gain is adaptively adjusted by the error size to suppress overshoot, the integral gain is adjusted by the error accumulation to eliminate steady-state error, and the differential gain is adjusted by the error change rate to suppress system oscillation; S503, the fused control instruction is adjusted by the dynamic gain parameter to generate a control instruction to drive the proportional valve of the electro-hydraulic servo system, control the hydraulic oil flow and pressure, and execute the actuator movement.

9. The electro-hydraulic linkage based intelligent control method of claim 8, wherein, The expression for calculating the dynamic gain parameter in step S502 is: wherein represents a dynamic proportional gain at the time instant represents a base proportional gain, represents a proportional adjustment coefficient, represents an error signal at the time instant represents a dynamic integral gain at the time instant represents a base integral gain, represents an integral adjustment coefficient, represents an error signal at the time instant represents a dynamic derivative gain at the time instant represents a base derivative gain, represents a derivative adjustment coefficient.

10. An electro-hydraulic linkage based intelligent control system, characterized in that, The system includes a processor and a memory for storing processor executable instructions; Wherein, the processor is configured to execute the executable instructions to realize the electro-hydraulic linkage-based intelligent control method in any one of claims 1 to 9.

Images