A Dock Control Method for Electric Loaders During Loading Operations Based on Load Fluctuation Identification

Abstract

This invention discloses a docking control method for electric loaders based on load fluctuation identification, comprising the following steps: By collecting multi-source signals such as drive motor torque, hydraulic cylinder pressure, vehicle posture, and axle load changes, a system state capable of characterizing the changes in material pile stiffness and reaction force is constructed; an online estimation of different pile stiffness and load fluctuations is achieved using a dynamic sparse neural network with embedded physical constraints; the vehicle's longitudinal speed, yaw response, and pitch and roll attitude are incorporated into a unified dynamic model to establish a three-dimensional coupling relationship between fluctuating loads and vehicle posture changes; the drive torque and deceleration process are adjusted based on the predicted parking distance and posture deviation, and the bucket slewing angle is compensated to automatically offset the effects of vehicle pitch and roll on the bucket posture; this invention solves the problems of identifying pile stiffness and load fluctuations in electric loader loading operations, as well as the difficulties in coordinated control caused by the longitudinal, transverse, and vertical coupling of load and vehicle.

Description

Technical Field

[0001] This invention relates to the technical field of electric drive and dynamic control of engineering machinery, and in particular to a method for stopping and controlling the loading operation of an electric loader based on load fluctuation identification. Background Technology

[0002] The parking operation of electric loaders is the most crucial and frequent part of their workflow. With the increasing demand for operational efficiency and precise quantitative loading in infrastructure construction, precise parking control of electric loader operations has become a current technological bottleneck.

[0003] Existing publicly available solutions mostly employ fixed threshold strategies based on sensor position limits or recording-playback. Some studies utilize visual perception or parameters such as current and hydraulic pressure to improve the repeatability of bucket trajectories and the efficiency of unloading and return. High-precision GNSS positioning and navigation technologies for unmanned mining trucks have also emerged. For example: Chinese invention patent application number CN202510307103.0, entitled "Automatic Loading Control Method, Training Method, Device, Equipment and Medium"; Chinese invention patent application number CN202410823414.8, entitled "Automatic Loading Loader Vehicle Control Method"; Chinese invention patent application number CN202311558172.6, entitled "An Automatic Loading Control Method, Device and Loader"; Chinese invention patent application number CN202111454513.6, entitled "Automatic Loading Control Method and Electric Loader"; Chinese invention patent application number CN202011431315.3, entitled "A Control Method and Loader for Automatic Loading Operation of a Loader". The aforementioned technologies are mostly for automated operations under ideal working conditions or control optimization in independent domains. They are difficult to directly transfer to the precise docking operations of electric loaders in unstructured work sites and in front of complex material piles, and to achieve synchronous and precise control of the docking position and bucket posture under the influence of load fluctuations and attitude coupling.

[0004] Current research faces two key challenges: First, existing loading and stopping control systems mostly adopt fixed threshold strategies, which do not consider the impact of different material stiffness and load fluctuations on vehicle deceleration and attitude. They lack real-time perception and adjustment of dynamic changes in system status, resulting in unstable loading depth and stopping position, making it difficult to achieve precise stopping.

[0005] Second, existing loader docking controls are mostly independent controls of driving and bucket hydraulics, ignoring the influence of load movement in space during loader loading operations. They lack prediction and compensation for the load-vehicle longitudinal, transverse and vertical coupling characteristics, and cannot achieve precise docking in front of complex terrain and complex material piles.

[0006] Therefore, it is necessary to propose a precise docking control method for electric loaders based on load fluctuation identification. By using a dynamic sparse neural network with embedded physical constraints to predict and control the dynamic model of the loader under fluctuating loads, the method can predict and compensate for the load-vehicle longitudinal, transverse and vertical coupling characteristics under the condition of identifying the stiffness of the stockpile and load fluctuations, thereby achieving precise docking. Summary of the Invention

[0007] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0008] In view of the problems existing in the above-mentioned electric loader loading operation docking control method based on load fluctuation identification, the present invention is proposed.

[0009] Therefore, the purpose of this invention is to provide a docking control method for electric loaders in loading operations based on load fluctuation identification, which solves the problems of material stacking stiffness and load fluctuation identification in electric loader loading operations, as well as the difficulties in coordinated control caused by the longitudinal, transverse and vertical coupling of load and vehicle.

[0010] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for stopping and controlling the loading operation of an electric loader based on load fluctuation identification, comprising the following steps: 1) By collecting multi-source signals such as drive motor torque, hydraulic cylinder pressure, vehicle body posture and axle load changes, a system state that can characterize the changes in material pile stiffness and reaction force is constructed. A dynamic sparse neural network with embedded physical constraints is used to realize online estimation of different material pile stiffness and load fluctuations. 11) Construct and train dynamic sparse neural networks with embedded physical constraints offline; 12) Collect signals such as drive motor torque, hydraulic cylinder pressure, vehicle attitude and axle load from vehicle-mounted sensors, and perform time synchronization and preprocessing on these data; construct a real-time feature vector that can directly characterize the stiffness characteristics and reaction force of the material pile by calculating the rate of change, peak value and correlation of the signal. 13) The constructed feature vector is input into the trained PDSNN for real-time calculation; real-time estimation is performed on continuous loading resistance and discrete stockpile stiffness classification, thereby realizing online estimation of different stockpile stiffness and load fluctuations. 2) Incorporate vehicle longitudinal speed, yaw response and pitch and roll attitude into a unified dynamic model, establish a three-dimensional coupling relationship between fluctuating load and vehicle attitude change, calculate driving force and future attitude deviation in real time based on estimated load characteristics, and realize feedforward compensation of docking point and bucket end attitude. 21) Construct a dynamic model of the loader under fluctuating load, which consists of a six-degree-of-freedom dynamic model of the whole vehicle, a tire-ground model, and hydraulic and electric drive models; 22) Based on the real-time estimated load fluctuation, the dynamic model of the loader under fluctuating load is used to make predictions, calculate the remaining parking distance and the possible future pitch and roll attitude deviations of the vehicle during the parking process in real time, and achieve accurate prediction of the vehicle's final state. 23) Calculate the required feedforward compensation based on the prediction results, and control the drive system and hydraulic system; 3) Adjust the drive torque and deceleration process based on the predicted parking distance and attitude deviation, and compensate for the bucket slewing angle so that the bucket attitude can automatically counteract the effects of vehicle body pitch and tilt, and achieve synchronous and precise control of the final parking position and bucket attitude. 31) Based on the prediction results in step 22) and the feedforward compensation calculated in step 23), calculate the drive torque required by the electric loader in real time; 32) Based on the prediction results in step 22) and the feedforward compensation calculated in step 23), calculate the required hydraulic pressure for bucket slewing angle compensation in real time; 33) Based on the feedforward compensation calculated in step 23), perform coordinated control compensation of the motor and bucket hydraulics to achieve synchronous and precise control of the final docking position and bucket posture.

[0011] As a preferred embodiment of the electric loader loading operation docking control method based on load fluctuation identification described in this invention, wherein: in step 11), the constructed dynamic sparse neural network with embedded physical constraints introduces a physical prior mask matrix in the input layer:

[0012] In the formula, For the output vector, For the input vector, It is a non-linear activation function. This is the weight matrix. This is a binary physical prior mask matrix. It is the bias vector; Among them, the binary physical prior mask matrix Used to determine input features With hidden layer features Physical correlations reduce invalid computations and prevent the network from learning false correlations; To constrain the inference logic of the neural network to conform to the basic laws of shovel dynamics, a loss function with physical constraints is constructed:

[0013] In the formula, These are the task loss, physical consistency loss, and sparsity regularization term, respectively. For hyperparameter weights, These are the feed resistance and stack stiffness predicted by the network, respectively. These are actual feed resistance labels and actual stacking stiffness labels, respectively. For hydraulic pressure, This indicates the direction of movement of the hydraulic cylinder.

[0014] As a preferred embodiment of the electric loader loading operation docking control method based on load fluctuation identification described in this invention, wherein: in step 12), time synchronization and preprocessing specifically involve: performing linear interpolation processing on the data collected by the sensor at a fixed time step to achieve time synchronization of the original data; and filtering the original data to eliminate random noise and reflect the true trend of the physical quantity.

[0015] As a preferred embodiment of the electric loader loading operation docking control method based on load fluctuation identification described in this invention, wherein: in step 21), the six-degree-of-freedom dynamic model of the vehicle under load includes force balance equations and moment balance equations, wherein the force balance equations are:

[0016] In the formula, For the curb weight of the loader, For load mass, , The loader's speed and acceleration in the longitudinal, transverse, and vertical directions. The pitch, roll, and yaw rates of the loader. These are the pitch angle and roll angle of the loader, respectively. For the driving force of the loader, These are the longitudinal and lateral components of the loading resistance identified in step 13). These are the longitudinal and lateral forces on the tire, respectively. This is the vertical component of the lift force. The vertical force of the tire is calculated in real time by the onboard recognition unit; The torque balance equation is:

[0017] In the formula, They are respectively around xyz The time-varying moment of inertia of the shaft. These are the loader's pitch, roll, yaw angular velocities and angular accelerations, respectively. These are the yaw moments generated by the ground and the loads, respectively. These are the axle load transfer moment caused by the longitudinal force of the tire and the gravitational moment of the load, respectively. For hydraulic reaction torque, This is the tire return torque. , These are the load inertia yaw moment and the loading resistance yaw moment, respectively; Among them, in the above equation It is obtained from the ground-tire model based on the magic formula, which includes the longitudinal tire force model and the lateral tire force model; The longitudinal tire force model is as follows:

[0018] In the formula, As the peak factor, Stiffness factor For shape factor, These parameters, representing curvature factors, were obtained through offline testing and fitting. The longitudinal slip ratio is calculated in real time by combining wheel speed and motor torque estimation units. The lateral tire force model is as follows:

[0019] In the formula, This refers to the tire slip angle; The remaining torques are obtained from the tire force and distance from the ground-tire model, and are calculated from the lever arm length of the center of gravity position identified in step 13). Among them, the vertical component of the lifting force in the above model and hydraulic reaction torque The hydraulic model is derived from the hydraulic model itself.

[0020] In the formula, These are the total thrust of the boom hydraulic cylinder and the total thrust of the rocker arm hydraulic cylinder, respectively. These are the thrust angles of the boom and rocker arm, respectively. These are the equivalent boom and the rocker arm, respectively, and these parameters are obtained in real time by the hydraulic and displacement recognition unit. The driving force of the loader in the above model The electric drive model is derived from the electric drive model, which is as follows:

[0021] In the formula, The output torque of each motor is obtained by the wheel speed and motor torque estimation unit. For the gearbox reduction ratio, For transmission efficiency, The radius of the tire; At this point, the dynamic model of the loader under fluctuating load has been established.

[0022] As a preferred embodiment of the electric loader loading operation docking control method based on load fluctuation identification described in this invention, wherein: in step 22), model predictive control is performed based on the loader dynamics model under fluctuating load established in step 21). First, the fluctuating load dynamic equation established in step 21) is discretized to construct a nonlinear state-space predictive model:

[0023]

[0024]

[0025]

[0026] In the formula, Let be the system state vector. For the system input vector, For the load disturbance vector, Sampling time, The dynamic model of the loader under fluctuating load established in step 21);

[0027] Furthermore, a multi-objective optimization objective function is constructed:

[0028] In the formula, For the output error term, These are the prediction time domain and the control time domain, respectively. These are the target position of the loader and the target angle of the bucket, respectively. To control the increment, This is the weight matrix. This represents the remaining parking distance under the current input. Furthermore, establish the constraint equations:

[0029] In the formula, These are the minimum and maximum values ​​of the control quantity, respectively. These are the minimum and maximum values ​​of the rate of change of the control quantity, respectively. These are the minimum and maximum values ​​of the bucket rotation angle, respectively. This represents the maximum value of the longitudinal force of the tire. Furthermore, a rolling optimization solution is performed:

[0030] In the formula, This is the calculated sequence of optimal control commands for the future. For actual control instructions, only the first value in the sequence is taken; Furthermore, this step outputs actual control commands. Predicted state and remaining parking distance .

[0031] As a preferred embodiment of the electric loader loading operation docking control method based on load fluctuation identification described in this invention, wherein: in step 23), based on the output result of step 22), feedforward compensation is performed on the electric drive torque and loading hydraulic pressure; Furthermore, regarding feedforward compensation for electric drive torque:

[0032] In the formula, These are the longitudinal endpoint velocities predicted in step 22). These are the stiffness gain, damping gain, and error sensitivity factor, obtained through offline testing. Furthermore, regarding feedforward compensation for loading hydraulics:

[0033] In the formula, Let $\mathbf{a}$ be the moment of inertia of the bucket. For hydraulic stiffness gain, This is the initial pitch angle of the loader.

[0034] As a preferred embodiment of the electric loader loading operation docking control method based on load fluctuation identification described in this invention, wherein: in step 31), the control quantity of the motor driving force is:

[0035] Furthermore, in step 32), the control quantity of the loading hydraulic pressure is:

[0036] Furthermore, in step 33), coordinated control compensation of the motor and bucket hydraulics is performed, and a dynamic coupling matrix is ​​introduced:

[0037] In the formula, This represents the gain adjustment coefficient of the boom for the drive. To drive the anti-pitch coefficient of the boom, The feedforward coefficient for the bucket's resistance to the drive is denoted as . To drive the attitude fidelity coefficient of the bucket; The final output control quantity is: .

[0038] As a preferred embodiment of the electric loader loading operation docking control method based on load fluctuation identification described in this invention, the implementation of this method relies on an electric loader loading system, which includes an axle-mounted identification unit, a wheel speed and motor torque estimation unit, an attitude identification unit, a hydraulic and displacement identification unit, a neural network identification unit, an electric drive hydraulic coordination controller, a distributed electric drive execution unit, and a loading electric hydraulic execution unit. The bridge-mounted identification unit is used to identify the changes in the load on the front and rear axles of the loader; The wheel speed and motor torque estimation unit is used to estimate the wheel speed, vehicle speed, and motor torque of the loader; The attitude recognition unit is used to identify changes in the overall attitude of the loader; The hydraulic and displacement recognition unit is used to identify hydraulic changes in the shovel-loading electro-hydraulic actuator and changes in the bucket's posture. The neural network recognition unit is used to run the trained strategy network and output load fluctuation and material stiffness; The aforementioned electro-hydraulic co-controller is used for decision-making and model predictive control to achieve precise docking control of the electric loader; The distributed electric drive execution unit is used to drive the electric loader and execute commands from the electric drive hydraulic coordinated control module; The loading electro-hydraulic actuator is used to drive the electric loader to perform loading operations and execute commands from the electro-hydraulic coordinated control module.

[0039] The beneficial effects of this invention are: 1. This invention breaks through the limitations of traditional fixed threshold control, significantly improves environmental adaptability and recognition accuracy. By using a dynamic sparse neural network with embedded physical constraints to deeply fuse electromechanical and hydraulic multi-source signals, it realizes real-time quantitative estimation and hierarchical recognition of loading resistance and stacking stiffness in unstructured environments. It effectively solves the problem of misjudgment of working conditions under sensor noise interference and greatly enhances the robustness of the system in complex working environments.

[0040] 2. This invention solves the endpoint control deviation caused by multi-domain coupling of load and vehicle, and achieves synchronous and precise control of docking position and attitude. A dynamic model of the loader under fluctuating load is constructed. By predicting the remaining parking distance and future attitude deviation in real time, an electro-hydraulic collaborative feedforward compensation strategy is adopted to automatically offset the influence of vehicle pitch on bucket attitude, achieving precise control of docking position while ensuring smooth and shock-free deceleration. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a method strategy diagram of the electric loader loading operation docking control method based on load fluctuation identification according to the present invention.

[0042] Figure 2 This is a schematic diagram illustrating the system working principle of the electric loader loading and unloading operation docking control method based on load fluctuation identification according to the present invention. Detailed Implementation

[0043] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0044] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0045] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0046] Secondly, the present invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of the present invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not according to the usual scale. Furthermore, the schematic diagrams are merely examples and should not limit the scope of protection of the present invention. In addition, actual fabrication should include three-dimensional spatial dimensions of length, width, and depth.

[0047] Reference Figures 1-2A method for controlling the docking of electric loaders during loading operations based on load fluctuation identification is provided, including the following steps: 1) By collecting multi-source signals such as drive motor torque, hydraulic cylinder pressure, vehicle body posture and axle load changes, a system state that can characterize the changes in material pile stiffness and reaction force is constructed. A dynamic sparse neural network with embedded physical constraints is used to realize online estimation of different material pile stiffness and load fluctuations. 11) Construct and train dynamic sparse neural networks with embedded physical constraints offline; Specifically, in 11), the constructed dynamic sparse neural network with embedded physical constraints introduces a physical prior mask matrix into the input layer:

[0048] In the formula, For the output vector, For the input vector, It is a non-linear activation function. This is the weight matrix. This is a binary physical prior mask matrix. It is the bias vector; Among them, the binary physical prior mask matrix Used to determine input features With hidden layer features Physical correlations reduce invalid computations and prevent the network from learning false correlations; To constrain the inference logic of the neural network to conform to the basic laws of shovel dynamics, a loss function with physical constraints is constructed:

[0049] In the formula, These are the task loss, physical consistency loss, and sparsity regularization term, respectively. For hyperparameter weights, These are the feed resistance and stack stiffness predicted by the network, respectively. These are actual feed resistance labels and actual stacking stiffness labels, respectively. For hydraulic pressure, The direction of movement of the hydraulic cylinder; 12) Collect signals such as drive motor torque, hydraulic cylinder pressure, vehicle attitude and axle load from vehicle-mounted sensors, and perform time synchronization and preprocessing on these data; construct a real-time feature vector that can directly characterize the stiffness characteristics and reaction force of the material pile by calculating the rate of change, peak value and correlation of the signal. In section 12), time synchronization and preprocessing specifically involve: performing linear interpolation on the data collected by the sensor at a fixed time step to achieve time synchronization of the original data; and filtering the original data to eliminate random noise and reflect the true trend of the physical quantity. 13) The constructed feature vector is input into the trained PDSNN for real-time calculation; real-time estimation is performed on continuous loading resistance and discrete stockpile stiffness classification, thereby realizing online estimation of different stockpile stiffness and load fluctuations. 2) Incorporate vehicle longitudinal speed, yaw response and pitch and roll attitude into a unified dynamic model, establish a three-dimensional coupling relationship between fluctuating load and vehicle attitude change, calculate driving force and future attitude deviation in real time based on estimated load characteristics, and realize feedforward compensation of docking point and bucket end attitude. 21) Construct a dynamic model of the loader under fluctuating load, which consists of a six-degree-of-freedom dynamic model of the whole vehicle, a tire-ground model, and hydraulic and electric drive models; In section 21), the six-degree-of-freedom dynamic model of the vehicle under load includes force balance equations and moment balance equations, wherein the force balance equations are:

[0050] In the formula, For the curb weight of the loader, For load mass, , The loader's speed and acceleration in the longitudinal, transverse, and vertical directions. The pitch, roll, and yaw rates of the loader. These are the pitch angle and roll angle of the loader, respectively. For the driving force of the loader, These are the longitudinal and lateral components of the loading resistance identified in step 13). These are the longitudinal and lateral forces on the tire, respectively. This is the vertical component of the lift force. The vertical force of the tire is calculated in real time by the onboard recognition unit; The torque balance equation is:

[0051] In the formula, They are respectively around xyz The time-varying moment of inertia of the shaft. These are the loader's pitch, roll, yaw angular velocities and angular accelerations, respectively. These are the yaw moments generated by the ground and the loads, respectively. These are the axle load transfer moment caused by the longitudinal force of the tire and the gravitational moment of the load, respectively. For hydraulic reaction torque, This is the tire return torque. , These are the load inertia yaw moment and the loading resistance yaw moment, respectively; Among them, in the above equation It is obtained from the ground-tire model based on the magic formula, which includes the longitudinal tire force model and the lateral tire force model; The longitudinal tire force model is as follows:

[0052] In the formula, As the peak factor, Stiffness factor For shape factor, These parameters, representing curvature factors, were obtained through offline testing and fitting. The longitudinal slip ratio is calculated in real time by combining wheel speed and motor torque estimation units. The lateral tire force model is as follows:

[0053] In the formula, This refers to the tire slip angle; The remaining torques are obtained from the tire force and distance from the ground-tire model, and are calculated from the lever arm length of the center of gravity position identified in step 13). Among them, the vertical component of the lifting force in the above model and hydraulic reaction torque The hydraulic model is derived from the hydraulic model itself.

[0054] In the formula, These are the total thrust of the boom hydraulic cylinder and the total thrust of the rocker arm hydraulic cylinder, respectively. These are the thrust angles of the boom and rocker arm, respectively. These are the equivalent boom and the rocker arm, respectively, and these parameters are obtained in real time by the hydraulic and displacement recognition unit. The driving force of the loader in the above model The electric drive model is derived from the electric drive model, which is as follows:

[0055] In the formula, The output torque of each motor is obtained by the wheel speed and motor torque estimation unit. For the gearbox reduction ratio, For transmission efficiency, The radius of the tire; At this point, the dynamic model of the loader under fluctuating load has been established. 22) Based on the real-time estimated load fluctuation, the dynamic model of the loader under fluctuating load is used to make predictions, calculate the remaining parking distance and the possible future pitch and roll attitude deviations of the vehicle during the parking process in real time, and achieve accurate prediction of the vehicle's final state. In step 22), model predictive control is performed based on the loader dynamics model established in step 21). First, the dynamic equations of the fluctuating load established in step 21) are discretized to construct a nonlinear state-space predictive model.

[0056]

[0057]

[0058]

[0059] In the formula, Let be the system state vector. For the system input vector, For the load disturbance vector, Sampling time, The dynamic model of the loader under fluctuating load established in step 21);

[0060] Furthermore, a multi-objective optimization objective function is constructed:

[0061] In the formula, For the output error term, These are the prediction time domain and the control time domain, respectively. These are the target position of the loader and the target angle of the bucket, respectively. To control the increment, This is the weight matrix. This represents the remaining parking distance under the current input. Furthermore, establish the constraint equations:

[0062] In the formula, These are the minimum and maximum values ​​of the control quantity, respectively. These are the minimum and maximum values ​​of the rate of change of the control quantity, respectively. These are the minimum and maximum values ​​of the bucket rotation angle, respectively. This represents the maximum value of the longitudinal force of the tire. Furthermore, a rolling optimization solution is performed:

[0063] In the formula, This is the calculated sequence of optimal control commands for the future. For actual control instructions, only the first value in the sequence is taken; Furthermore, this step outputs actual control commands. Predicted state and remaining parking distance ; 23) Calculate the required feedforward compensation based on the prediction results, and control the drive system and hydraulic system; Specifically, in step 23), based on the output of step 22), feedforward compensation is performed on the electric drive torque and the loading hydraulic pressure. Furthermore, regarding feedforward compensation for electric drive torque:

[0064] In the formula, These are the longitudinal endpoint velocities predicted in step 22). These are the stiffness gain, damping gain, and error sensitivity factor, obtained through offline testing. Furthermore, regarding feedforward compensation for loading hydraulics:

[0065] In the formula, Let $\mathbf{a}$ be the moment of inertia of the bucket. For hydraulic stiffness gain, This is the initial pitch angle of the loader; 3) Adjust the drive torque and deceleration process based on the predicted parking distance and attitude deviation, and compensate for the bucket slewing angle so that the bucket attitude can automatically counteract the effects of vehicle body pitch and tilt, and achieve synchronous and precise control of the final parking position and bucket attitude. 31) Based on the prediction results in step 22) and the feedforward compensation calculated in step 23), calculate the drive torque required by the electric loader in real time; 32) Based on the prediction results in step 22) and the feedforward compensation calculated in step 23), calculate the required hydraulic pressure for bucket slewing angle compensation in real time; 33) Based on the feedforward compensation calculated in step 23), perform coordinated control compensation of the motor and bucket hydraulics to achieve synchronous and precise control of the final docking position and bucket posture.

[0066] In part 31), the control quantity of the motor driving force is:

[0067] Furthermore, in 32), the control quantity of the loading hydraulic system is:

[0068] Furthermore, in section 33), coordinated control compensation of the motor and bucket hydraulics is performed, and a dynamic coupling matrix is ​​introduced:

[0069] In the formula, This represents the gain adjustment coefficient of the boom for the drive. To drive the anti-pitch coefficient of the boom, The feedforward coefficient for the bucket's resistance to the drive is denoted as . To drive the attitude fidelity coefficient of the bucket; The final output control quantity is: .

[0070] Example 2 Reference Figure 2 It provides an electric loader loading system, which includes an axle-mounted identification unit, a wheel speed and motor torque estimation unit, an attitude recognition unit, a hydraulic and displacement recognition unit, a neural network recognition unit, an electric drive hydraulic coordination controller, a distributed electric drive execution unit, and a loading electric hydraulic execution unit; The bridge-mounted identification unit is used to identify changes in the load on the front and rear axles of the loader; The wheel speed and motor torque estimation unit is used to estimate the wheel speed, vehicle speed, and motor torque of the loader; The attitude recognition unit is used to identify changes in the overall attitude of the loader; The hydraulic and displacement recognition unit is used to identify hydraulic changes in the shovel electro-hydraulic actuator and changes in the bucket's posture. The neural network recognition unit is used to run the trained policy network and output load fluctuation and stack stiffness; The electro-hydraulic co-controller is used for decision-making and model predictive control to achieve precise docking control of the electric loader; A distributed electric drive execution unit is used to drive the electric loader and execute commands from the electric drive hydraulic coordination control module; The shoveling electro-hydraulic actuator is used to drive the electric loader in shoveling operations and execute commands from the electro-hydraulic co-control module.

[0071] The system operates as follows: the bridge-mounted identification unit identifies load signal a, the wheel speed and motor torque estimation unit estimates motor torque signal b, the attitude identification unit identifies attitude signal c, and the hydraulic and displacement identification unit identifies hydraulic signal d and bucket attitude signal h, all transmitted to the neural network identification unit via the vehicle network. Simultaneously, based on these multi-source signals, the neural network identification unit outputs stockpiling stiffness and load fluctuation signals e through a trained identification strategy network. The electric drive co-controller, based on signal e, uses a loader dynamics model prediction method under fluctuating loads for prediction and compensation, outputting torque optimization command f to the distributed electric drive execution unit, and simultaneously outputting hydraulic control command g to the loading electro-hydraulic execution unit. The action results of each execution unit form a closed-loop feedback through the sensor network for feedback correction of the model predictive control. Through this mechanism, the system achieves the identification of stockpiling stiffness and load fluctuations during electric loader loading operations, as well as the coordinated control of travel and bucket hydraulics.

[0072] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for stopping and controlling electric loader loading operations based on load fluctuation identification, characterized in that, Includes the following steps: 1) By collecting multi-source signals such as drive motor torque, hydraulic cylinder pressure, vehicle body posture and axle load changes, a system state that can characterize the changes in material pile stiffness and reaction force is constructed. The dynamic sparse neural network PDSNN with embedded physical constraints is used to realize online estimation of different material pile stiffness and load fluctuations. 11) Construct and train dynamic sparse neural networks with embedded physical constraints offline; 12) Collect signals such as drive motor torque, hydraulic cylinder pressure, vehicle attitude and axle load from vehicle-mounted sensors, and perform time synchronization and preprocessing on these data; construct a real-time feature vector that can directly characterize the stiffness characteristics and reaction force of the material pile by calculating the rate of change, peak value and correlation of the signal. 13) The constructed feature vector is input into the trained PDSNN for real-time calculation; real-time estimation is performed on continuous loading resistance and discrete stockpile stiffness classification, thereby realizing online estimation of different stockpile stiffness and load fluctuations. 2) Incorporate vehicle longitudinal speed, yaw response and pitch and roll attitude into a unified dynamic model, establish a three-dimensional coupling relationship between fluctuating load and vehicle attitude change, calculate driving force and future attitude deviation in real time based on estimated load characteristics, and realize feedforward compensation of docking point and bucket end attitude. 21) Construct a dynamic model of the loader under fluctuating load, which consists of a six-degree-of-freedom dynamic model of the whole vehicle, a tire-ground model, and hydraulic and electric drive models; 22) Based on the real-time estimated load fluctuation, the dynamic model of the loader under fluctuating load is used to make predictions, calculate the remaining parking distance and the possible future pitch and roll attitude deviations of the vehicle during the parking process in real time, and achieve accurate prediction of the vehicle's final state. 23) Calculate the required feedforward compensation based on the prediction results, and control the drive system and hydraulic system; 3) Adjust the drive torque and deceleration process based on the predicted parking distance and attitude deviation, and compensate for the bucket slewing angle so that the bucket attitude can automatically counteract the effects of vehicle body pitch and tilt, and achieve synchronous and precise control of the final parking position and bucket attitude. 31) Based on the prediction results in step 22) and the feedforward compensation calculated in step 23), calculate the drive torque required by the electric loader in real time; 32) Based on the prediction results in step 22) and the feedforward compensation calculated in step 23), calculate the required hydraulic pressure for bucket slewing angle compensation in real time; 33) Based on the feedforward compensation calculated in step 23), perform coordinated control compensation of the motor and bucket hydraulics to achieve synchronous and precise control of the final docking position and bucket posture.

2. The electric loader loading operation docking control method based on load fluctuation identification according to claim 1, characterized in that: In step 11), the constructed dynamic sparse neural network with embedded physical constraints introduces a physical prior mask matrix into the input layer: （1）； In the formula, For the output vector, For the input vector, It is a non-linear activation function. This is the weight matrix. This is a binary physical prior mask matrix. It is the bias vector; Among them, the binary physical prior mask matrix Used to determine input features With hidden layer features Physical correlations reduce invalid computations and prevent the network from learning false correlations; To constrain the inference logic of the neural network to conform to the basic laws of shovel dynamics, a loss function with physical constraints is constructed: （2）； In the formula, These are the task loss, physical consistency loss, and sparsity regularization term, respectively. For hyperparameter weights, These are the feed resistance and stack stiffness predicted by the network, respectively. These are actual feed resistance labels and actual stacking stiffness labels, respectively. For hydraulic pressure, This indicates the direction of movement of the hydraulic cylinder.

3. The electric loader loading operation docking control method based on load fluctuation identification according to claim 2, characterized in that: In step 12), time synchronization and preprocessing specifically involve: performing linear interpolation on the data collected by the sensor at a fixed time step to achieve time synchronization of the original data; and filtering the original data to eliminate random noise and reflect the true trend of the physical quantity.

4. The electric loader loading operation docking control method based on load fluctuation identification according to claim 3, characterized in that: In step 21), the six-degree-of-freedom dynamic model of the vehicle under load includes force balance equations and moment balance equations, wherein the force balance equations are: ； In the formula, For the curb weight of the loader, For load mass, , The loader's speed and acceleration in the longitudinal, transverse, and vertical directions. The pitch, roll, and yaw rates of the loader. These are the pitch angle and roll angle of the loader, respectively. For the driving force of the loader, These are the longitudinal and lateral components of the loading resistance identified in step 13). These are the longitudinal and lateral forces on the tire, respectively. This is the vertical component of the lift force. The vertical force of the tire is calculated in real time by the onboard recognition unit; The torque balance equation is: ； In the formula, They are respectively around xyz The time-varying moment of inertia of the shaft. These are the loader's pitch, roll, yaw angular velocities and angular accelerations, respectively. These are the yaw moments generated by the ground and the loads, respectively. These are the axle load transfer moment caused by the longitudinal force of the tire and the gravitational moment of the load, respectively. For hydraulic reaction torque, This is the tire return torque. , These are the load inertia yaw moment and the loading resistance yaw moment, respectively; Among them, in the above equation It is obtained from the ground-tire model based on the magic formula, which includes the longitudinal tire force model and the lateral tire force model; The longitudinal tire force model is as follows: ； In the formula, As the peak factor, Stiffness factor For shape factor, These parameters, representing curvature factors, were obtained through offline testing and fitting. The longitudinal slip ratio is calculated in real time by combining wheel speed and motor torque estimation units. The lateral tire force model is as follows: ； In the formula, This refers to the tire slip angle; The remaining torques are obtained from the tire force and distance from the ground-tire model, and are calculated from the lever arm length of the center of gravity position identified in step 13). Among them, the vertical component of the lifting force in the above model and hydraulic reaction torque The hydraulic model is derived from the hydraulic model itself. ； In the formula, These are the total thrust of the boom hydraulic cylinder and the total thrust of the rocker arm hydraulic cylinder, respectively. These are the thrust angles of the boom and rocker arm, respectively. These are the equivalent boom and the rocker arm, respectively, and these parameters are obtained in real time by the hydraulic and displacement recognition unit. The driving force of the loader in the above model The electric drive model is derived from the electric drive model, which is as follows: ； In the formula, The output torque of each motor is obtained by the wheel speed and motor torque estimation unit. For the gearbox reduction ratio, For transmission efficiency, The radius of the tire; At this point, the dynamic model of the loader under fluctuating load has been established.

5. The electric loader loading operation docking control method based on load fluctuation identification according to claim 4, characterized in that: In step 22), model predictive control is performed based on the loader dynamics model established in step 21). First, the dynamic equations of the fluctuating load established in step 21) are discretized to construct a nonlinear state-space predictive model: ； ； ； ； In the formula, Let be the system state vector. For the system input vector, For the load disturbance vector, Sampling time, The dynamic model of the loader under fluctuating load established in step 21); ； Furthermore, a multi-objective optimization objective function is constructed: ； In the formula, For the output error term, These are the prediction time domain and the control time domain, respectively. These are the target position of the loader and the target angle of the bucket, respectively. To control the increment, This is the weight matrix. This represents the remaining parking distance under the current input. Furthermore, establish the constraint equations: ； In the formula, These are the minimum and maximum values ​​of the control quantity, respectively. These are the minimum and maximum values ​​of the rate of change of the control quantity, respectively. These are the minimum and maximum values ​​of the bucket rotation angle, respectively. This represents the maximum value of the longitudinal force of the tire. Furthermore, a rolling optimization solution is performed: ； In the formula, This is the calculated sequence of optimal control commands for the future. For actual control instructions, only the first value in the sequence is taken; Furthermore, this step outputs actual control commands. Predicted state and remaining parking distance .

6. The electric loader loading operation docking control method based on load fluctuation identification according to claim 5, characterized in that: In step 23), based on the output of step 22), feedforward compensation is performed on the electric drive torque and loading hydraulic pressure. Furthermore, regarding feedforward compensation for electric drive torque: ； In the formula, These are the longitudinal endpoint velocities predicted in step 22). These are the stiffness gain, damping gain, and error sensitivity factor, obtained through offline testing. Furthermore, regarding feedforward compensation for loading hydraulics: ； In the formula, Let $\mathbf{a}$ be the moment of inertia of the bucket. For hydraulic stiffness gain, This is the initial pitch angle of the loader.

7. The electric loader loading operation docking control method based on load fluctuation identification according to claim 5, characterized in that: In step 31), the control quantity of the motor driving force is: ； Furthermore, in step 32), the control quantity of the loading hydraulic pressure is: ； Furthermore, in step 33), coordinated control compensation of the motor and bucket hydraulics is performed, and a dynamic coupling matrix is ​​introduced: ； In the formula, This represents the gain adjustment coefficient of the boom for the drive. To drive the anti-pitch coefficient of the boom, The feedforward coefficient for the bucket's resistance to the drive. To drive the attitude fidelity coefficient of the bucket; The final output control quantity is: 。 8. The electric loader loading operation docking control method based on load fluctuation identification according to any one of claims 1-7, wherein the implementation of the method depends on an electric loader loading system, the system including an axle-mounted identification unit, a wheel speed and motor torque estimation unit, an attitude identification unit, a hydraulic and displacement identification unit, a neural network identification unit, an electric drive hydraulic cooperative controller, a distributed electric drive execution unit, and a loading electric hydraulic execution unit; The bridge-mounted identification unit is used to identify the changes in the load on the front and rear axles of the loader; The wheel speed and motor torque estimation unit is used to estimate the wheel speed, vehicle speed, and motor torque of the loader; The attitude recognition unit is used to identify changes in the overall attitude of the loader; The hydraulic and displacement recognition unit is used to identify hydraulic changes in the shovel-loading electro-hydraulic actuator and changes in the bucket's posture. The neural network recognition unit is used to run the trained strategy network and output load fluctuation and material stiffness; The aforementioned electro-hydraulic co-controller is used for decision-making and model predictive control to achieve precise docking control of the electric loader; The distributed electric drive execution unit is used to drive the electric loader and execute commands from the electric drive hydraulic coordinated control module; The loading electro-hydraulic actuator is used to drive the electric loader to perform loading operations and execute commands from the electro-hydraulic coordinated control module.

Images