A forklift operation control optimization method and system based on disturbance observation technology

Abstract

This invention discloses a forklift operation control optimization method and system based on disturbance observation technology, belonging to the field of electric forklift control technology. The method includes: real-time acquisition of forklift operation data; analysis of forklift motion state variables through multi-source sensor information fusion to establish dynamic parameter mapping relationships; construction of a high-order sliding mode observer using measured values ​​and predicted values ​​from the forklift's full-condition dynamics model as inputs to evaluate disturbance source interference and generate motion compensation corrections; construction of a feedforward-feedback fusion composite controller to generate yaw moment pre-compensation commands based on disturbance source interference evaluation results, dynamically correcting forklift trajectory tracking errors; and real-time monitoring of the distance between the forklift's operating point and the instability boundary, triggering a torque redistribution strategy when approaching the instability boundary. This invention significantly improves forklift operation performance under complex conditions by deeply integrating real-time sensor data, dynamic modeling, and intelligent control algorithms.

Description

Technical Field

[0001] This invention relates to the field of electric forklift control technology, and in particular to a forklift operation control optimization method and system based on disturbance observation technology. Background Technology

[0002] Electric forklifts, as an important type of mobile material handling vehicle, are widely used in modern logistics, warehouses, cargo terminals, and logistics factories due to their versatility and flexibility. They play a crucial role in enabling intelligent logistics operations, reducing worker labor intensity, and improving operational efficiency, providing enterprises with efficient and environmentally friendly material handling solutions. As the logistics industry continues to upgrade towards automation and intelligence, the application areas and operating scenarios of electric forklifts are gradually expanding, bringing significant benefits and competitive advantages to modern logistics systems.

[0003] The e-commerce logistics industry is constantly demanding higher handling efficiency, requiring forklifts to maintain millimeter-level trajectory tracking accuracy and vehicle stability even under extreme conditions such as high-speed driving, high-level lifting, and emergency stops to avoid obstacles. However, in actual operation, electric forklifts face severe challenges from multi-dimensional dynamic disturbance coupling, manifested in the following aspects: 1. During fork lifting operations, the shift in the center of gravity of the goods due to changes in height can cause sudden changes in inertial torque, leading to vehicle yaw instability; 2. High-frequency disturbances such as road unevenness, tire slippage, and hydraulic system pressure fluctuations can disrupt the torque balance of the drive system, resulting in accumulated trajectory tracking deviations; 3. Heterogeneous data from multiple sources of sensors (such as lidar, IMU, and wheel speed sensors) have differences in sampling frequency and misalignment of spatiotemporal references, leading to lag or distortion in state observation. The time-varying coupling characteristics of these disturbance sources make it difficult for traditional control methods to decouple and compensate in real time, easily causing vehicle sideslip, cargo overturning, or even vehicle rollover accidents.

[0004] Current forklift control technologies primarily rely on rule-based feedforward compensation and fixed-parameter PID feedback regulation. However, these technologies suffer from insufficient disturbance observation capabilities. Traditional linear observers can only estimate low-frequency disturbances in a single direction, failing to model the coupling effects of multi-source disturbances under conditions such as fork lifting and rapid acceleration. Furthermore, the control parameters are fixed, and existing methods do not consider the dynamic impact of center-of-gravity shift on dynamic parameters such as tire sideslip stiffness and yaw inertia, leading to a phase deviation between the compensation torque and the actual disturbance. In addition, current technologies often use static threshold methods to monitor the center-of-gravity sideslip angle, without constructing a phase plane model of the dynamic stability boundary. This can easily result in misjudgments or delayed responses in scenarios such as ramp turns and sudden load changes, leading to a rigid stability judgment mechanism. These problems collectively cause the trajectory tracking error of existing forklift control systems to exceed the safety threshold under complex operating conditions, making it difficult to meet the requirements of scenarios requiring coordinated control of high-precision trajectory tracking and dynamic stability. Summary of the Invention

[0005] Therefore, it is necessary to provide a forklift operation control optimization method and system based on disturbance observation technology to address the above-mentioned technical problems.

[0006] In a first aspect, the present invention provides a forklift operation control optimization method based on disturbance observation technology, comprising:

[0007] S1. Real-time acquisition of forklift operation data, analysis of forklift motion state variables through multi-source sensor information fusion, and establishment of dynamic parameter mapping relationship;

[0008] S2. Using the measured values ​​and the predicted values ​​of the forklift's full-condition dynamics model as inputs, a high-order sliding mode observer is constructed to evaluate the disturbance source interference and generate motion compensation corrections.

[0009] S3. Construct a feedforward-feedback fusion composite controller, generate yaw moment pre-compensation command based on disturbance source interference assessment results, and dynamically correct forklift trajectory tracking error;

[0010] S4. Real-time monitoring of the distance between the forklift's operating point and the instability boundary. When it approaches the instability boundary, a torque redistribution strategy is triggered, and a self-learning mechanism is introduced to achieve adaptive adjustment of control parameters.

[0011] Furthermore, real-time acquisition of forklift operation data, through multi-source sensor information fusion, analysis of forklift motion state variables, and establishment of dynamic parameter mapping relationships include:

[0012] S11. The multi-source sensors configured in the integrated forklift establish a cross-domain perception network to collect real-time operation data of the forklift operation process and align the sampling timestamps to a unified time reference. The multi-source sensors include lidar, inertial measurement unit, wheel speed sensor and hydraulic system pressure sensor.

[0013] S12. Construct a dynamic parameter mapping model based on the kinematic equations of the forklift, establish the mathematical relationship between the lateral force, longitudinal force and slip ratio of the tires, and calculate the state variables of the forklift motion process. The state variables include the center of gravity sideslip angle calculated by integrating the longitudinal and lateral velocities, and the center of gravity horizontal offset calculated based on the hydraulic system pressure signal and the fork lifting height.

[0014] Furthermore, by using the measured values ​​and the predicted values ​​from the forklift's full-condition dynamics model as input, a high-order sliding mode observer is constructed to evaluate disturbance source interference and generate motion compensation corrections, including:

[0015] S21. Based on the mechanical structure and motion characteristics of the forklift, input the measured value of the current motion state of the forklift, construct a full-condition dynamic model including tire force, inertial force and external disturbance source, decompose the external disturbance source into longitudinal disturbance, lateral disturbance and yaw disturbance, and output the predicted state value of the forklift.

[0016] S22. Construct a nonlinear disturbance observer, input the real-time status and control commands of the forklift, and output the low-frequency disturbance estimate.

[0017] S23. Construct a variable gain high-order sliding mode observer, input the state prediction value and the uncompensated residual disturbance, and output the high-frequency disturbance estimate.

[0018] S24. The low-frequency disturbance estimate and the high-frequency disturbance estimate after integrating the interference from external disturbance sources are used as the motion compensation correction amount for the active control of the forklift.

[0019] Furthermore, the expression for the full-condition dynamic model is as follows:

[0020]

[0021] In the formula, m represents the total mass of the forklift; v x Indicates longitudinal velocity; v y R represents lateral velocity; r represents yaw rate; F represents lateral velocity. xf F represents the longitudinal force on the front wheel. xr F represents the lateral force on the rear wheel. yf F represents the lateral force on the front wheel. yr F represents the lateral force on the rear wheel. roll Indicates rolling resistance; I z The moment of inertia about the vertical axis; ΔM hyd Indicates hydraulic additional torque; δ represents steering angle; K p τ represents the steering pressure gain; θ represents the hydraulic delay time; hyd F represents the time constant of the hydraulic system. dist,1 Indicates longitudinal disturbance; F dist,2 Indicates lateral disturbance; M dist,3 The yaw disturbance torque is represented by t; t represents the current time. This represents the predicted longitudinal acceleration value; This represents the predicted lateral acceleration value; This represents the predicted yaw acceleration; a and b both represent length coefficients.

[0022] Furthermore, a variable-gain high-order sliding mode observer is constructed, taking the state prediction and uncompensated residual perturbation as inputs, and outputting high-frequency perturbation estimates including:

[0023] S231. By calculating the deviation between the measured value and the predicted value of the forklift's motion state, a sliding surface is constructed to characterize the influence of external disturbance sources.

[0024] S232. By combining nonlinear functions and integral terms, the update rules for the disturbance estimator are set, and an adaptive mechanism is introduced to dynamically adjust the gain to form a high-order sliding mode observer. When the sliding surface deviation is greater than the set threshold, the gain is automatically increased to accelerate disturbance tracking; when the deviation is less than the set threshold, the gain is reduced to reduce high-frequency chattering of the control signal.

[0025] S233. Inject different disturbance signals into the simulation environment to test the tracking accuracy and response speed of the high-order sliding mode observer. Based on the test results, adjust the integral weight coefficient and the initial gain parameter.

[0026] S234. Using the tested high-order sliding mode observer, receive the state prediction values ​​output from the full-condition dynamic model, and combine them with the residual disturbances that are not completely canceled to output high-frequency disturbance estimates.

[0027] Furthermore, a feedforward-feedback fusion composite controller is constructed. Based on the disturbance source evaluation results, a yaw moment pre-compensation command is generated to dynamically correct the forklift trajectory tracking error, including:

[0028] S31. Based on the motion compensation correction amount of the forklift active control, combined with the real-time trajectory tracking error, yaw moment and compensation command, feedforward compensation moment and feedback correction moment are generated sequentially.

[0029] S32. Construct a reinforcement learning policy network based on the near-end policy optimization algorithm, establish a multi-dimensional state vector set, establish a reward function with the objectives of minimizing tracking error, minimizing energy consumption and optimizing stability, and dynamically allocate the weights of feedforward compensation torque and feedback correction torque according to the perturbation frequency.

[0030] S33. The feedforward compensation torque and the feedback correction torque are fused according to the disturbance frequency characteristics to generate the total control torque, and the total control torque is limited to the peak range of the motor and hydraulic system.

[0031] Furthermore, based on the motion compensation correction amount of the forklift's active control, combined with the real-time trajectory tracking error, the yaw moment and compensation command, feedforward compensation moment, and feedback correction moment are generated sequentially, including:

[0032] S311. Based on the measured values ​​of the current motion state of the forklift, the estimated value of the low-frequency disturbance is mapped to the feedforward compensation torque, and the timing of the application of the feedforward compensation torque is adjusted using the lead compensation filter.

[0033] The expression for the feedforward compensation torque is:

[0034]

[0035] In the formula, M ff K represents the feedforward compensation torque. model Indicates the inverse solution gain; v x Indicates longitudinal velocity; m load Indicates the load mass; This represents the estimated value of the low-frequency disturbance;

[0036] S312. Based on the high-frequency disturbance estimate, the real-time trajectory error of the forklift is calculated by combining the lateral deviation and the yaw angle deviation, and the feedback correction torque is generated by the adaptive proportional-integral-derivative controller.

[0037] The feedback correction torque is expressed as:

[0038]

[0039] In the formula, M fb Indicates feedback correction torque; K p K i K d This indicates the dynamically adjusted PID gain; e y Indicates lateral deviation; This represents the estimated value of the high-frequency disturbance.

[0040] Furthermore, the expression for the reward function is:

[0041]

[0042] In the formula, R t Represents the reward function; e y Indicates lateral deviation; e Ψ Indicates yaw rate deviation; I motor P represents the motor current; hyd Indicates the hydraulic power of the hydraulic system; I max P represents the maximum current of the motor. max Indicates the maximum hydraulic power of the hydraulic system; I stable Indicates the stability flag;

[0043] The formula for calculating the weight of the feedforward compensation torque is:

[0044]

[0045] In the formula, w ff The weight of the feedforward compensation torque is represented by f; the perturbation frequency is represented by f0; and the corner frequency is represented by f0.

[0046] Furthermore, the distance between the forklift's operating point and the instability boundary is monitored in real time. When the forklift approaches the instability boundary, a torque redistribution strategy is triggered, and a self-learning mechanism is introduced to achieve adaptive adjustment of control parameters, including:

[0047] S41. Establish a phase plane analysis model of the center of gravity sideslip angle-yaw rate, draw dynamic stability boundary curves under different load conditions, and monitor the relative position of the forklift's operating point and the instability boundary in real time.

[0048] S42. When the distance between the forklift's operating point and the instability boundary is less than a preset threshold, the torque vector redistribution strategy is triggered to reduce the driving force of the outer wheels and enhance the energy recovery mechanism to suppress the sideslip trend.

[0049] S43. Record the control parameter adjustment records and effect evaluation in historical operations, and optimize the approach rate of the sliding mode observer and the bandwidth parameters of the composite controller through the gradient descent algorithm to achieve adaptive adjustment.

[0050] Secondly, the present invention also provides a forklift operation control optimization system based on disturbance observation technology, the system comprising:

[0051] The data acquisition module is used to collect real-time operating data of the forklift during operation. Through the fusion of information from multiple sources, it analyzes the state variables of the forklift's movement and establishes dynamic parameter mapping relationships.

[0052] The disturbance observation module is used to take the measured values ​​and the predicted values ​​of the forklift's full-condition dynamic model as inputs to build a high-order sliding mode observer, evaluate the disturbance source interference, and generate motion compensation corrections.

[0053] The composite control module is used to construct a feedforward-feedback fusion composite controller. Based on the disturbance source interference assessment results, it generates yaw moment pre-compensation commands to dynamically correct the forklift trajectory tracking error.

[0054] The monitoring and adjustment module is used to monitor the distance between the forklift's operating point and the instability boundary in real time. When it approaches the instability boundary, it triggers a torque redistribution strategy and introduces a self-learning mechanism to achieve adaptive adjustment of control parameters.

[0055] The beneficial effects of this invention are as follows:

[0056] 1. By deeply integrating real-time sensor data, dynamic modeling, and intelligent control algorithms, the forklift's operational performance under complex conditions is significantly improved. A feedforward-feedback composite control architecture is constructed, centered on disturbance source assessment, to achieve precise separation and dynamic compensation of low-frequency modelable disturbances and high-frequency random disturbances. Simultaneously, reinforcement learning and self-learning mechanisms are introduced, enabling the control system to possess parameter adaptive optimization capabilities, balancing trajectory tracking accuracy, energy efficiency, and operational stability. Through multi-source information fusion and closed-loop disturbance rejection control, the forklift's comprehensive performance under dynamic loads, sudden road surface changes, and external disturbance scenarios achieves a qualitative leap, providing technical support for logistics automation and industrial safety.

[0057] 2. Disturbance observation technology enables real-time identification and quantification of external disturbances during forklift operation (such as hydraulic lag, cargo deviation, and road impact). Feedforward compensation torque preemptively offsets modelable disturbances, while feedback correction torque rapidly suppresses random disturbances, significantly reducing trajectory tracking errors. In typical scenarios such as sharp turns, ramp driving, and high-level lifting, forklift lateral path deviation and yaw angle fluctuations are significantly reduced, improving operational accuracy to industry-leading levels. Simultaneously, a stability monitoring module based on phase plane analysis assesses the distance between the operating state and the instability boundary in real time. Through a torque vector redistribution strategy, it proactively intervenes in sideslip trends, ensuring dynamic stability under extreme conditions and effectively preventing safety hazards such as rollovers and cargo tipping.

[0058] 3. By introducing reinforcement learning policy networks and gradient descent optimization algorithms, the control system can dynamically adjust parameters based on historical operational data and real-time operating conditions. For example, when the load changes, the feedforward weights and PID gain automatically adapt to the load characteristics, reducing reliance on manual parameter tuning; when road surface adhesion conditions change abruptly, the disturbance frequency analysis drives the torque distribution strategy to dynamically switch, balancing control response speed and actuator load. Furthermore, the energy recovery mechanism converts kinetic energy during braking into hydraulic or electrical energy storage, reducing peak motor power consumption and extending the service life of critical components. Adaptability not only improves system robustness but also achieves synergistic optimization of operational efficiency and energy utilization, meeting the core requirements of green industry and sustainable development. Attached Figure Description

[0059] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0060] Figure 1 This is a flowchart of a forklift operation control optimization method based on disturbance observation technology according to an embodiment of the present invention;

[0061] Figure 2 This is a system principle block diagram of a forklift operation control optimization system based on disturbance observation technology according to an embodiment of the present invention.

[0062] Reference numerals: 1. Data acquisition module; 2. Disturbance observation module; 3. Composite control module; 4. Monitoring and adjustment module. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0064] Please see Figure 1 This invention provides a forklift operation control optimization method based on disturbance observation technology, the method comprising:

[0065] S1. Real-time acquisition of forklift operation data, analysis of forklift motion state variables through multi-source sensor information fusion, and establishment of dynamic parameter mapping relationship.

[0066] In the description of this invention, the real-time acquisition of forklift operation data, the analysis of forklift motion state variables through multi-source sensor information fusion, and the establishment of dynamic parameter mapping relationships include:

[0067] S11 integrates multiple source sensors configured within the forklift to build a cross-domain perception network, collects real-time operating data of the forklift operation process, and aligns the sampling timestamps to a unified time reference. The multiple source sensors include lidar, inertial measurement unit, wheel speed sensor, and hydraulic system pressure sensor.

[0068] Specifically, a multi-source sensor network is constructed by integrating the internal sensors of the electric forklift to collect operational data in real time during forklift operation. The multi-source sensors include LiDAR, inertial measurement unit (IMU), wheel speed sensors, and hydraulic system pressure sensors, which are used to acquire information on environmental obstacles, vehicle attitude (three-axis acceleration and angular velocity), wheel speed, and hydraulic system load, respectively. By aligning the sampling timestamps of all sensors using a unified time reference, inconsistencies in data timing caused by sensor hardware delays or asynchronous communication are eliminated, ensuring the spatiotemporal consistency of multi-source data and providing reliable input for subsequent fusion analysis.

[0069] In terms of hardware deployment, the LiDAR is installed on the top of the forklift, with a horizontal scanning angle of 270° and a vertical resolution of 0.1°, for real-time detection of 3D point clouds of shelves, goods and obstacles; the IMU is fixed at the center of gravity of the forklift, and collects the dynamic attitude of the vehicle body through a three-axis accelerometer and gyroscope; the wheel speed sensor is integrated into the drive wheel axle to measure the rotation speed of the left and right wheels in real time; the hydraulic pressure sensor is embedded in the hydraulic cylinder oil inlet line to monitor the lifting load of the forks.

[0070] Time synchronization adopts the Precision Time Protocol (PTP), using the on-board industrial control computer as the clock source and broadcasting synchronization signals via Ethernet to control the timestamp deviation of each sensor within the microsecond level.

[0071] In the data preprocessing stage, the lidar point cloud is denoised by voxel filtering, the IMU data is converted into Euler angles by quaternion calculation, the wheel speed signal is smoothed by Kalman filtering, and the hydraulic pressure signal is filtered by moving average to eliminate high-frequency noise.

[0072] S12. Construct a dynamic parameter mapping model based on the kinematic equations of the forklift, establish the mathematical relationship between the lateral force, longitudinal force and slip ratio of the tires, and calculate the state variables of the forklift motion process. The state variables include the center of gravity sideslip angle calculated by integrating the longitudinal and lateral velocities, and the center of gravity horizontal offset calculated based on the hydraulic system pressure signal and the fork lifting height.

[0073] Specifically, key state variables during the forklift's movement are calculated in real time, including the sideslip angle, longitudinal velocity, lateral velocity, and horizontal offset of the load's center of gravity.

[0074] The center of gravity sideslip angle is calculated using the integral method combined with lateral and longitudinal velocities, reflecting the forklift's steering stability; the load center of gravity offset is derived from the functional relationship between the hydraulic system pressure signal and the fork lifting height, and is used to correct the influence of the forklift load distribution on the motion.

[0075] The tire force model uses Pacejka's Magic Formula to describe the nonlinear mechanical properties of the tire under slip conditions. Longitudinal force F x and lateral force F y The expression is:

[0076] F x =D x sin(C x arctan(B x λ-E x (B x λ-arctan(B x λ))));

[0077] F y =D y sin(C y arctan(B y α-E y (B y α-arctan(B y α))));

[0078] In the formula, B x C x D x E x All represent tire characteristic parameters of longitudinal force, in the following order: stiffness factor, shape factor, peak factor, and curvature factor; B y C y D y E y All represent tire characteristic parameters of lateral force; λ represents longitudinal slip ratio; α represents sideslip angle.

[0079] The formula for calculating the centroid sideslip angle β is:

[0080]

[0081] The formula for calculating the load centroid offset Δx is:

[0082]

[0083] In the formula, p hyd Indicates hydraulic pressure; h fork The value represents the lifting height of the forks; k represents the stiffness coefficient; and A represents the working area of ​​the hydraulic cylinder.

[0084] S2. Using the measured values ​​and the predicted values ​​of the forklift's full-condition dynamics model as inputs, a high-order sliding mode observer is constructed to evaluate the disturbance source interference and generate motion compensation corrections.

[0085] In the description of this invention, measured values ​​and predicted values ​​from the forklift's full-condition dynamics model are used as inputs to construct a high-order sliding mode observer, evaluate disturbance source interference, and generate motion compensation corrections, including:

[0086] S21. Based on the mechanical structure and motion characteristics of the forklift, input the measured values ​​of the current motion state of the forklift, construct a full-condition dynamic model including tire force, inertial force and external disturbance source, decompose the external disturbance source into longitudinal disturbance, lateral disturbance and yaw disturbance, and output the predicted state value of the forklift.

[0087] Specifically, based on the mechanical structural characteristics and kinematic relationships of the forklift, a full-condition dynamic model incorporating tire forces, inertial forces, and external disturbances is constructed. The model inputs are the forklift's real-time state (longitudinal velocity, lateral velocity, yaw rate) and control commands (steering angle, motor torque). The multi-degree-of-freedom motion of the forklift is described using Newton-Euler equations, and external disturbances are decomposed into longitudinal disturbances (such as changes in road slope and wheel hub friction fluctuations), lateral disturbances (such as cargo center of gravity shift and crosswind interference), and yaw disturbances (such as hydraulic time-delay torque and sudden changes in road adhesion). The model outputs predicted forklift state values ​​(predicted longitudinal acceleration, lateral acceleration, and yaw rate).

[0088] In the description of this invention, the expression for the full-condition dynamic model is as follows:

[0089]

[0090] In the formula, m represents the total mass of the forklift; v x Indicates longitudinal velocity; v y R represents lateral velocity; r represents yaw rate; F represents lateral velocity. xf F represents the longitudinal force on the front wheel. xr F represents the lateral force on the rear wheel. yf F represents the lateral force on the front wheel. yr F represents the lateral force on the rear wheel.roll Indicates rolling resistance; I z The moment of inertia about the vertical axis; ΔM hyd Indicates hydraulic additional torque; δ represents steering angle; K p τ represents the steering pressure gain; θ represents the hydraulic delay time; hyd F represents the time constant of the hydraulic system. dist,1 Indicates longitudinal disturbance; F dist,2 Indicates lateral disturbance; M dist,3 Indicates yaw disturbance; t represents the current time; This represents the predicted longitudinal acceleration value; This represents the predicted lateral acceleration value; This represents the predicted yaw acceleration; a and b both represent length coefficients (in meters).

[0091] S22. Construct a nonlinear disturbance observer, input the real-time status and control commands of the forklift, and output the low-frequency disturbance estimate.

[0092] Specifically, for modelable low-frequency disturbances (such as hydraulic time delay and slow load change), a nonlinear disturbance observer (NDOB) is constructed. The real-time status of the forklift (longitudinal speed and yaw rate) and control commands (steering angle and motor torque) are input, and the low-frequency disturbance estimate is output.

[0093] NDOB estimates the disturbance term by inversely estimating the deviation between the model-predicted state and the actual state, thus compensating for low-frequency interference not covered by the model.

[0094] Observer equation: Define auxiliary variable z i The disturbance is estimated using dynamic equations, and its expression is:

[0095]

[0096] In the formula, l i Indicates the observer gain; x i Represents the state variable; φ i (x,u) represents the known terms of the model; This represents the estimated value of low-frequency disturbances.

[0097] S23. Construct a variable gain high-order sliding mode observer, input the state prediction value and the uncompensated residual disturbance, and output the high-frequency disturbance estimate value.

[0098] In the description of this invention, a variable-gain high-order sliding mode observer is constructed, which takes the state prediction value and the uncompensated residual disturbance as inputs, and outputs a high-frequency disturbance estimate including:

[0099] S231. By calculating the deviation between the measured value and the predicted value of the forklift's motion state, a sliding surface is constructed to characterize the influence of external disturbance sources.

[0100] The expression for the sliding surface si is:

[0101] s i =x i,meas -x i,model +λ i ∫(x i,meas -x i,model )dt;

[0102] In the formula, x i,meas Indicates the measured value of the motion state; x i,model λ represents the predicted state value. i This represents the integral weighting coefficient.

[0103] S232. By combining nonlinear functions and integral terms, update rules for disturbance estimators are set, and an adaptive mechanism is introduced to dynamically adjust the gain to form a high-order sliding mode observer. When the sliding surface deviation is greater than the set threshold, the gain is automatically increased to accelerate disturbance tracking; when the deviation is less than the set threshold, the gain is reduced to reduce high-frequency chattering of the control signal.

[0104] S233. Inject different disturbance signals into the simulation environment to test the tracking accuracy and response speed of the high-order sliding mode observer. Based on the test results, adjust the integral weight coefficient and the initial gain parameter.

[0105] S234. Using the tested high-order sliding mode observer, receive the state prediction values ​​output from the full-condition dynamic model, and combine them with the residual disturbances that are not completely canceled to output high-frequency disturbance estimates.

[0106] Among them, the high-frequency disturbance estimate The calculation formula is:

[0107]

[0108] In the formula, k 1,i and k 2,i Both represent gain coefficients.

[0109] S24. The low-frequency disturbance estimate and the high-frequency disturbance estimate after integrating the interference from external disturbance sources are used as the motion compensation correction amount for the active control of the forklift.

[0110] S3. Construct a feedforward-feedback fusion composite controller, generate yaw moment pre-compensation command based on disturbance source evaluation results, and dynamically correct forklift trajectory tracking error.

[0111] In the description of this invention, a feedforward-feedback fusion composite controller is constructed. Based on the disturbance source evaluation results, a yaw moment pre-compensation command is generated to dynamically correct the forklift trajectory tracking error, including:

[0112] S31. Based on the motion compensation correction amount of the forklift active control, combined with the real-time trajectory tracking error, the yaw moment and compensation command, feedforward compensation moment and feedback correction moment are generated sequentially.

[0113] In the description of this invention, based on the motion compensation correction amount of the forklift active control and combined with the real-time trajectory tracking error, the yaw moment and compensation command, feedforward compensation moment and feedback correction moment are generated sequentially, including:

[0114] S311. Based on the measured values ​​of the current motion state of the forklift, the estimated low-frequency disturbance value is mapped to the feedforward compensation torque, and the timing of the application of the feedforward compensation torque is adjusted using the lead compensation filter.

[0115] The expression for the feedforward compensation torque is:

[0116]

[0117] In the formula, M ff K represents the feedforward compensation torque. model Indicates the inverse solution gain; v x Indicates longitudinal velocity; m load Indicates the load mass; This represents the estimated value of the low-frequency disturbance.

[0118] S312. Based on the high-frequency disturbance estimate, the real-time trajectory error of the forklift is calculated by combining the lateral deviation and yaw angle deviation, and the feedback correction torque is generated by the adaptive proportional-integral-derivative controller.

[0119] The feedback correction torque is expressed as:

[0120]

[0121] In the formula, M fb Indicates feedback correction torque; K p K i K d This indicates the dynamically adjusted PID gain; e y Indicates lateral deviation; This represents the estimated value of the high-frequency disturbance.

[0122] S32. Construct a reinforcement learning policy network based on the near-end policy optimization algorithm, establish a multi-dimensional state vector set, establish a reward function with the objectives of minimizing tracking error, minimizing energy consumption, and optimizing stability, and dynamically allocate the weights of feedforward compensation torque and feedback correction torque according to the perturbation frequency.

[0123] Specifically, the Proximal Policy Optimization (PPO) algorithm is a reinforcement learning method based on policy gradients. Its core objective is to enable an agent (in this invention, a forklift control system) to make adaptive decisions in dynamic environments. In this method, the PPO policy network undertakes the following key functions:

[0124] 1. Dynamic parameter optimization: Based on real-time environmental conditions (such as trajectory error, disturbance frequency, actuator load), dynamically adjust the weight distribution of feedforward compensation torque and feedback correction torque, as well as the gain parameters (proportional, integral, and derivative terms) of the PID controller.

[0125] 2. Multi-objective collaboration: With minimizing trajectory tracking error, minimizing energy consumption, and optimizing stability as the optimization directions, the priority of different objectives is balanced through reward function design. For example, under sudden disturbances, error suppression is prioritized, while energy consumption is reduced in steady state.

[0126] 3. Disturbance Frequency Adaptation: The weight ratio of feedforward and feedback is dynamically adjusted according to the characteristics of the disturbance spectrum (low frequency or high frequency dominance). For example, the feedforward weight is increased to 80% for hydraulic time lag (low frequency) and the feedback weight is increased to 70% for road impact (high frequency).

[0127] In this invention, the PPO policy network is adapted to the forklift control system in the following ways:

[0128] 1. State Awareness Fusion: Input a multi-dimensional state vector, including sensor measured data (lateral deviation, yaw rate), disturbance estimation results (feedforward compensation, feedback correction), and actuator states (motor current, hydraulic power). These data collectively characterize the system's dynamic characteristics and the impact of external disturbances.

[0129] 2. Action space mapping: Output parameter adjustment instructions, such as feedforward weight increment, PID gain adjustment, etc., directly affect the torque fusion ratio and control response speed.

[0130] 3. Closed-loop learning mechanism: Feedback on actual control effects (such as residual error and energy consumption) to the reward function drives the policy network to fine-tune parameters online, forming a closed-loop learning link of "perception-decision-execution-optimization".

[0131] In addition, the implementation and operation process of the PPO policy network include the following aspects:

[0132] I. State Space Construction and Data Processing

[0133] The state vector consists of data such as forklift lateral deviation, yaw angle deviation, center of gravity sideslip angle, feedforward and feedback disturbance compensation, motor current, and hydraulic power, which are then normalized and input into the strategy network. Lateral deviation is calculated by comparing the path planning module with LiDAR data; the center of gravity sideslip angle is achieved by fusing IMU and wheel speed signals through Kalman filtering.

[0134] Data real-time performance guarantee: Ensure the consistency of timestamps of multi-source sensor data through time synchronization protocols (such as PTP) to avoid asynchronous errors.

[0135] II. Reward Function Design and Multi-Objective Trade-offs

[0136] Tracking error penalty: The square of the lateral deviation and the yaw angle deviation is used as the main penalty term, which forces the policy network to prioritize reducing path deviation.

[0137] Energy consumption penalty: The sum of the squares of motor current and hydraulic power is used as an energy consumption indicator to suppress excessive load on the actuator.

[0138] Stability bonus: A positive bonus is given when the sideslip angle and yaw rate of the center of mass do not exceed the limits to encourage stable operation.

[0139] III. Motion Space Definition and Parameter Adjustment

[0140] Feedforward weight adjustment: Based on the disturbance spectrum analysis results (low frequency or high frequency dominance), output the increment of the feedforward weight coefficient. For example, when hydraulic time delay disturbance is detected, the feedforward ratio is increased.

[0141] PID gain dynamic adjustment: When the error is large, the proportional gain is automatically increased to accelerate the response, and the integral term is enhanced in steady state to eliminate residual deviation.

[0142] IV. Training and Online Fine-tuning

[0143] Offline pre-training: Simulate various disturbance scenarios (such as cargo displacement and slippery road surface) in a simulation environment, and train the policy network through millions of interactions to initially learn parameter adjustment rules.

[0144] Online Adaptive: During actual vehicle operation, the network parameters are continuously fine-tuned based on real-time data, and the learning rate is set to a low value to ensure stability and avoid oscillations caused by parameter mutations.

[0145] V. Safety Constraints and Cross-Step Collaboration

[0146] Actuator saturation protection: The total output torque is limited to the physical limits of the motor and hydraulic system. When the limits are exceeded, the feedforward and feedback components are compressed proportionally.

[0147] Stability fallback mechanism: In conjunction with step S4 (torque redistribution), when approaching the instability threshold, forced intervention control is implemented, and the update of strategy network parameters is frozen.

[0148] In the description of this invention, the expression for the reward function is:

[0149]

[0150] In the formula, R t Represents the reward function; e y Indicates lateral deviation (dimensionless value); e Ψ Indicates the yaw rate deviation (dimensionless value); I motor P represents the motor current; hyd Indicates the hydraulic power of the hydraulic system; I max P represents the maximum current of the motor. max Indicates the maximum hydraulic power of the hydraulic system; I stable This indicates a stability flag (value is 0 or 1; it is set to 1 when the forklift oscillates slightly near the stability boundary, and is used to provide a fixed reward).

[0151] The formula for calculating the weight of the feedforward compensation torque is:

[0152]

[0153] In the formula, w ff The weight of the feedforward compensation torque is represented by f; the perturbation frequency is represented by f0; and the corner frequency is represented by f0.

[0154] S33. The feedforward compensation torque and the feedback correction torque are fused according to the disturbance frequency characteristics to generate the total control torque, and the total control torque is limited to the peak range of the motor and hydraulic system.

[0155] Specifically, the feedforward compensation torque M ff With feedback correction torque M fb Based on the fusion of disturbance frequency characteristics, the total control torque M is generated. total The system safety is ensured through actuator saturation limits and stability monitoring. The total torque is limited by the peak torque of the motor and hydraulic system (≤200 N·m). When the limit is exceeded, the feedforward and feedback components are compressed proportionally. At the same time, the centroid side slip angle β and yaw rate r are monitored in real time. If they exceed the stability threshold (β>4° or |r|>1.2 rad / s), the torque redistribution strategy in step S4 is triggered.

[0156] S4. Real-time monitoring of the distance between the forklift's operating point and the instability boundary. When it approaches the instability boundary, a torque redistribution strategy is triggered, and a self-learning mechanism is introduced to achieve adaptive adjustment of control parameters.

[0157] In the description of this invention, the distance between the forklift's operating point and the instability boundary is monitored in real time. When the forklift approaches the instability boundary, a torque redistribution strategy is triggered, and a self-learning mechanism is introduced to achieve adaptive adjustment of control parameters, including:

[0158] S41. Establish a phase plane analysis model of the center of gravity sideslip angle-yaw rate, plot the dynamic stability boundary curves under different load conditions, and monitor the relative position of the forklift's operating point and the instability boundary in real time.

[0159] Specifically, a phase-plane analysis model for the dynamic stability of forklifts is established through a combination of experiments and theory. First, real-world tests are conducted under different load conditions, collecting the sideslip angle and yaw angle of the center of gravity. Stability boundary curves are then plotted using the dynamic equations. Under no-load conditions, the boundary range is wider, allowing for larger sideslip angles and yaw velocities; under full load, the boundary tightens, reducing the stability tolerance. Real-time monitoring continuously acquires the current sideslip angle and yaw velocities of the center of gravity, calculating their Euclidean distance from the nearest stable boundary. When the distance is less than a preset threshold, the forklift is considered to be nearing instability.

[0160] The electric forklift control system incorporates a dynamic database that stores stability boundary curves under different load and road conditions. During operation, it automatically matches the corresponding curve based on real-time load information and calculates the distance between the current operating point and the boundary. If the distance continues to decrease and falls below the safety threshold, an audible and visual warning is triggered, and stability control is prepared to be executed.

[0161] For example, when a forklift turns fully loaded on a wet and slippery surface, its sideslip angle and yaw rate rapidly approach the stability boundary. The system calculates the distance in real time and triggers an early warning, providing a basis for subsequent control decisions.

[0162] S42. When the distance between the forklift's operating point and the instability boundary is less than a preset threshold, the torque vector redistribution strategy is triggered to reduce the driving force of the outer wheels and enhance the energy recovery mechanism to suppress the sideslip trend.

[0163] Specifically, when the operating point is detected to be approaching the instability boundary, the control system immediately activates the torque redistribution strategy. The driving force of the outer wheels is reduced proportionally to decrease the exacerbating effect of centrifugal force on sideslip; the braking pressure of the inner wheels is increased, generating a counter-yawing torque through differential torque to counteract the sideslip tendency. At the same time, the brake energy recovery system is activated, converting the kinetic energy during deceleration into electrical energy for storage, reducing the overall energy consumption of the system.

[0164] Control commands are sent to the drive motor and hydraulic braking unit via the CAN bus. The torque command of the outer motor decreases linearly, while the braking pressure of the inner motor increases rapidly, ensuring that the torque difference is generated quickly. The energy recovery system synchronously adjusts the inverter parameters to maximize the feedback efficiency.

[0165] For example, when making a sharp turn and the cargo is heavily loaded on one side, the system detects the risk of sideslip, reduces the outer driving force by 30%, increases the inner braking force by 50%, and the difference in yaw moment allows the vehicle's attitude to return to stability within 0.5 seconds.

[0166] S43. Record the control parameter adjustment records and effect evaluation in historical operations, and optimize the approach rate of the sliding mode observer and the bandwidth parameters of the composite controller through the gradient descent algorithm to achieve adaptive adjustment.

[0167] Specifically, the system continuously records detailed data for each stability control operation, including the state parameters at the time of triggering, torque adjustment, and control effects (such as sideslip suppression time and energy recovery). The effectiveness of each control operation is evaluated through a weighted scoring mechanism, with scoring indicators covering stability recovery speed, energy efficiency, and trajectory tracking accuracy.

[0168] Based on historical data, the gradient descent algorithm is used to analyze the correlation between control parameters and performance. For example, if the sliding mode observer's response rate is insufficient under certain operating conditions, leading to control delay, the system automatically increases its gain coefficient; if excessive controller bandwidth causes actuator oscillation, the bandwidth range is dynamically reduced. Every 50 work cycles, the system initiates a round of global parameter optimization, updating the core parameters of the sliding mode observer, torque distribution strategy, and energy recovery logic. The optimization process runs in the background, without affecting real-time control, ensuring continuous improvement in control performance throughout the forklift's lifecycle.

[0169] Please see Figure 2 The present invention also provides a forklift operation control optimization system based on disturbance observation technology, the system comprising:

[0170] Data acquisition module 1 is used to collect real-time operating data of the forklift operation process. Through the perception and fusion of multi-source sensor information, it analyzes the state variables of the forklift movement and establishes dynamic parameter mapping relationships.

[0171] The disturbance observation module 2 is used to construct a high-order sliding mode observer by taking the measured values ​​and the predicted values ​​of the forklift's full-condition dynamic model as inputs, evaluating the disturbance source interference, and generating motion compensation corrections.

[0172] Composite control module 3 is used to construct a feedforward-feedback fusion composite controller. Based on the disturbance source interference evaluation results, it generates yaw moment pre-compensation commands to dynamically correct the forklift trajectory tracking error.

[0173] The monitoring and adjustment module 4 is used to monitor the distance between the forklift's operating point and the instability boundary in real time. When it approaches the instability boundary, it triggers a torque redistribution strategy and introduces a self-learning mechanism to achieve adaptive adjustment of control parameters.

[0174] The present invention will be further illustrated below with specific examples.

[0175] In daily operations at a logistics warehouse, fully loaded electric forklifts need to perform right-angle turns on slippery surfaces. At this time, the forklift forks are raised to a height of 2 meters, causing the center of gravity to shift and the center of mass to move laterally. Simultaneously, water on the road reduces the traction between the tires and the ground. After the forklift starts, its body sensors begin collecting environmental and self-status information in real time: LiDAR scans the shelf outline and obstacles in the path; the inertial measurement unit monitors the yaw rate and roll angle; wheel speed sensors report the difference in speed between the left and right wheels; and hydraulic system pressure sensors detect the load lifted by the forks. After time synchronization and filtering, this multi-source data is input into a full-condition dynamics model. The model, combined with the forklift's current load, steering angle, and speed, predicts the theoretical lateral acceleration and yaw rate under normal operating conditions.

[0176] When the forklift enters a curve, the slippery road surface causes slight slippage of the outer drive wheel. Simultaneously, the shift in the center of gravity of the cargo generates additional lateral torque, resulting in a significant deviation between the measured lateral velocity and the model prediction. At this point, the nonlinear disturbance observer detects the response delay of the hydraulic system and the low-frequency disturbance caused by the cargo shift, generating a feedforward compensation command to adjust the output torque of the drive motor in advance to counteract the hydraulic time lag. Simultaneously, the high-order sliding mode observer captures high-frequency disturbance signals caused by road slippage and cargo swaying, dynamically adjusting the gain for a rapid response and generating a feedback correction torque to suppress sudden sideslip. Based on the real-time analysis of the disturbance spectrum characteristics, the system weights and fuses the feedforward compensation and feedback correction, outputting a yaw moment control command. This command moderately reduces the torque of the outer wheels and increases the braking of the inner wheels, creating a reverse yaw moment to balance the lateral force.

[0177] As the forklift approaches the midpoint of the curve, real-time monitoring data of the sideslip angle and yaw rate show that the operating point is nearing the dynamic stability boundary. The stability control function immediately intervenes, further reducing the outer driving force and increasing the inner braking pressure, with differential torque quickly restoring the vehicle's posture to normal. Simultaneously, the energy recovery system converts the kinetic energy generated during braking into electrical energy for storage, reducing the motor load. Throughout the process, the control system continuously records the torque adjustment range, disturbance suppression effect, and energy consumption data, optimizing the sliding mode observer's response rate and control parameters through a self-learning mechanism to ensure further improvement in control efficiency under similar conditions in the future.

[0178] Ultimately, the forklift completed the turn with a stable posture, the cargo remained on track, the path tracking error was significantly reduced, the hydraulic system energy consumption was reduced, and the system entered standby mode to await the next work instruction.

[0179] In summary, by leveraging the technical solutions described above, the forklift's performance under complex working conditions is significantly improved through the deep integration of real-time sensor data, dynamic modeling, and intelligent control algorithms. A feedforward-feedback composite control architecture is constructed with disturbance source assessment as the core, achieving precise separation and dynamic compensation of low-frequency modelable disturbances and high-frequency random disturbances. Simultaneously, reinforcement learning and self-learning mechanisms are introduced, enabling the control system to possess adaptive parameter optimization capabilities, balancing trajectory tracking accuracy, energy efficiency, and operational stability. Through multi-source information fusion and closed-loop disturbance rejection control, the forklift's comprehensive performance under dynamic loads, sudden road surface changes, and external interference scenarios achieves a qualitative leap, providing technical support for logistics automation and industrial safety.

[0180] Perturbation observation technology enables real-time identification and quantification of external disturbances during forklift operation (such as hydraulic lag, cargo deviation, and road impact). Feedforward compensation torque preemptively offsets modelable disturbances, while feedback correction torque rapidly suppresses random disturbances, significantly reducing trajectory tracking errors. In typical scenarios such as sharp turns, ramp driving, and high-lift maneuvers, forklift lateral path deviation and yaw angle fluctuations are significantly reduced, achieving industry-leading operational accuracy. Simultaneously, a stability monitoring module based on phase plane analysis continuously assesses the distance between the operating state and the instability boundary, proactively intervening in sideslip trends through a torque vector redistribution strategy to ensure dynamic stability under extreme conditions, effectively preventing safety hazards such as rollovers and cargo tipping.

[0181] By introducing reinforcement learning policy networks and gradient descent optimization algorithms, the control system can dynamically adjust parameters based on historical operational data and real-time operating conditions. For example, when the load changes, the feedforward weights and PID gain automatically adapt to the load characteristics, reducing reliance on manual parameter tuning. When road surface adhesion conditions change abruptly, disturbance frequency analysis drives the torque distribution strategy to dynamically switch, balancing control response speed and actuator load. Furthermore, an energy recovery mechanism converts kinetic energy during braking into hydraulic or electrical energy for storage, reducing peak motor power consumption and extending the lifespan of critical components. This adaptability not only enhances system robustness but also achieves synergistic optimization of operational efficiency and energy utilization, meeting the core requirements of green industry and sustainable development.

[0182] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

Claims

1. A forklift operation control optimization method based on disturbance observation technology, characterized in that, include: S1. Real-time acquisition of forklift operation data, analysis of forklift motion state variables through multi-source sensor information fusion, and establishment of dynamic parameter mapping relationship; S2. Using the measured values ​​and the predicted values ​​of the forklift's full-condition dynamics model as inputs, a high-order sliding mode observer is constructed to evaluate the disturbance source interference and generate motion compensation corrections. S3. Construct a feedforward-feedback fusion composite controller, generate yaw moment pre-compensation command based on disturbance source interference assessment results, and dynamically correct forklift trajectory tracking error; S4. Real-time monitoring of the distance between the forklift's operating point and the instability boundary. When it approaches the instability boundary, a torque redistribution strategy is triggered, and a self-learning mechanism is introduced to achieve adaptive adjustment of control parameters. The process of using measured values ​​and predicted values ​​from the forklift's full-condition dynamics model as input to construct a high-order sliding mode observer, assess disturbance source interference, and generate motion compensation corrections includes: S21. Based on the mechanical structure and motion characteristics of the forklift, input the measured value of the current motion state of the forklift, construct a full-condition dynamic model including tire force, inertial force and external disturbance source, decompose the external disturbance source into longitudinal disturbance, lateral disturbance and yaw disturbance, and output the predicted state value of the forklift. S22. Construct a nonlinear disturbance observer, input the real-time status and control commands of the forklift, and output the low-frequency disturbance estimate. S23. Construct a variable gain high-order sliding mode observer, input the state prediction value and the uncompensated residual disturbance, and output the high-frequency disturbance estimate. S24. The low-frequency disturbance estimate and the high-frequency disturbance estimate after integrating the interference from external disturbance sources are used as the motion compensation correction amount for the active control of the forklift. The constructed feedforward-feedback fusion composite controller, based on the disturbance source evaluation results, generates yaw moment pre-compensation commands to dynamically correct forklift trajectory tracking errors, including: S31. Based on the motion compensation correction amount of the forklift active control, combined with the real-time trajectory tracking error, yaw moment and compensation command, feedforward compensation moment and feedback correction moment are generated sequentially. S32. Construct a reinforcement learning policy network based on the near-end policy optimization algorithm, establish a multi-dimensional state vector set, establish a reward function with the objectives of minimizing tracking error, minimizing energy consumption and optimizing stability, and dynamically allocate the weights of feedforward compensation torque and feedback correction torque according to the perturbation frequency. S33. The feedforward compensation torque and the feedback correction torque are fused according to the disturbance frequency characteristics to generate the total control torque, and the total control torque is limited to the peak range of the motor and hydraulic system. The motion compensation correction amount based on forklift active control, combined with real-time trajectory tracking error, sequentially generates yaw moment and compensation command, feedforward compensation moment and feedback correction moment, including: S311. Based on the measured values ​​of the current motion state of the forklift, the estimated value of the low-frequency disturbance is mapped to the feedforward compensation torque, and the timing of the application of the feedforward compensation torque is adjusted using the lead compensation filter. S312. Based on the high-frequency disturbance estimate, the real-time trajectory error of the forklift is calculated by combining the lateral deviation and yaw angle deviation, and the feedback correction torque is generated by the adaptive proportional-integral-derivative controller.

2. The forklift operation control optimization method based on disturbance observation technology according to claim 1, characterized in that, The real-time acquisition of forklift operation data, through multi-source sensor information fusion, analysis of forklift motion state variables, and establishment of dynamic parameter mapping relationships include: S11. A cross-domain sensing network is built by integrating multi-source sensors configured in the forklift to collect real-time operation data of the forklift operation process and align the sampling timestamps to a unified time reference. The multi-source sensors include lidar, inertial measurement unit, wheel speed sensor and hydraulic system pressure sensor. S12. Construct a dynamic parameter mapping model based on the kinematic equations of the forklift, establish the mathematical relationship between the lateral force, longitudinal force and slip ratio of the tires, and calculate the state variables of the forklift motion process. The state variables include the center of gravity sideslip angle calculated by integrating the longitudinal and lateral velocities, and the center of gravity horizontal offset calculated based on the hydraulic system pressure signal and the fork lifting height.

3. The forklift operation control optimization method based on disturbance observation technology according to claim 1, characterized in that, The expression for the full-condition dynamic model is: ； In the formula, m represents the total mass of the forklift; v x Indicates longitudinal velocity; v y R represents lateral velocity; r represents yaw rate; F represents lateral velocity. xf F represents the longitudinal force on the front wheel. xr F represents the lateral force on the rear wheel. yf F represents the lateral force on the front wheel. yr F represents the lateral force on the rear wheel. roll Indicates rolling resistance; I z The moment of inertia about the vertical axis; ΔM hyd Indicates hydraulic additional torque; δ represents steering angle; K p τ represents the steering pressure gain; θ represents the hydraulic delay time; hyd F represents the time constant of the hydraulic system. dist,1 Indicates longitudinal disturbance; F dist,2 Indicates lateral disturbance; M dist,3 Indicates the yaw disturbance moment; t represents the current time; This represents the predicted longitudinal acceleration value; This represents the predicted lateral acceleration value; This represents the predicted yaw acceleration; a and b both represent length coefficients.

4. The forklift operation control optimization method based on disturbance observation technology according to claim 3, characterized in that, The variable-gain high-order sliding mode observer is constructed by taking the state prediction value and the uncompensated residual disturbance as inputs, and outputting high-frequency disturbance estimates, including: S231. By calculating the deviation between the measured value and the predicted value of the forklift's motion state, a sliding surface is constructed to characterize the influence of external disturbance sources. S232. By combining nonlinear functions and integral terms, the update rules for the disturbance estimator are set, and an adaptive mechanism is introduced to dynamically adjust the gain to form a high-order sliding mode observer. When the sliding surface deviation is greater than the set threshold, the gain is automatically increased to accelerate disturbance tracking; when the deviation is less than the set threshold, the gain is reduced to reduce high-frequency chattering of the control signal. S233. Inject different disturbance signals into the simulation environment to test the tracking accuracy and response speed of the high-order sliding mode observer. Based on the test results, adjust the integral weight coefficient and the initial gain parameter. S234. Using the tested high-order sliding mode observer, receive the state prediction values ​​output from the full-condition dynamic model, and combine them with the residual disturbances that are not completely canceled to output high-frequency disturbance estimates.

5. The forklift operation control optimization method based on disturbance observation technology according to claim 1, characterized in that, The expression for the feedforward compensation torque is: ； In the formula, M ff K represents the feedforward compensation torque. model Indicates the inverse solution gain; v x Indicates longitudinal velocity; m load Indicates the load mass; This represents the estimated value of the low-frequency disturbance; The feedback correction torque is expressed as follows: ； In the formula, M fb Indicates feedback correction torque; K p K i K d This indicates the dynamically adjusted PID gain; e y Indicates lateral deviation; This represents the estimated value of the high-frequency disturbance.

6. The forklift operation control optimization method based on disturbance observation technology according to claim 5, characterized in that, The expression for the reward function is: ； In the formula, R t Represents the reward function; e y Indicates lateral deviation; e Ψ Indicates yaw rate deviation; I motor P represents the motor current. hyd Indicates the hydraulic power of the hydraulic system; I max P represents the maximum current of the motor. max Indicates the maximum hydraulic power of the hydraulic system; I stable Indicates the stability flag; The formula for calculating the weight of the feedforward compensation torque is as follows: ； In the formula, w ff The weight of the feedforward compensation torque is represented by f; the perturbation frequency is represented by f0; and the corner frequency is represented by f0.

7. The forklift operation control optimization method based on disturbance observation technology according to claim 1, characterized in that, The real-time monitoring of the distance between the forklift's operating point and the instability boundary, triggering a torque redistribution strategy when approaching the instability boundary, and introducing a self-learning mechanism to achieve adaptive adjustment of control parameters includes: S41. Establish a phase plane analysis model of the center of gravity sideslip angle-yaw rate, draw dynamic stability boundary curves under different load conditions, and monitor the relative position of the forklift's operating point and the instability boundary in real time. S42. When the distance between the forklift's operating point and the instability boundary is less than a preset threshold, the torque vector redistribution strategy is triggered to reduce the driving force of the outer wheels and enhance the energy recovery mechanism to suppress the sideslip trend. S43. Record the control parameter adjustment records and effect evaluation in historical operations, and optimize the approach rate of the sliding mode observer and the bandwidth parameters of the composite controller through the gradient descent algorithm to achieve adaptive adjustment.

8. A forklift operation control optimization system based on disturbance observation technology, used to implement the forklift operation control optimization method based on disturbance observation technology as described in any one of claims 1-7, characterized in that, The system includes: The data acquisition module is used to collect real-time operating data of the forklift during operation. Through the fusion of information from multiple sources, it analyzes the state variables of the forklift's movement and establishes dynamic parameter mapping relationships. The disturbance observation module is used to take the measured values ​​and the predicted values ​​of the forklift's full-condition dynamic model as inputs to build a high-order sliding mode observer, evaluate the disturbance source interference, and generate motion compensation corrections. The composite control module is used to construct a feedforward-feedback fusion composite controller. Based on the disturbance source interference assessment results, it generates yaw moment pre-compensation commands to dynamically correct the forklift trajectory tracking error. The monitoring and adjustment module is used to monitor the distance between the forklift's operating point and the instability boundary in real time. When it approaches the instability boundary, it triggers a torque redistribution strategy and introduces a self-learning mechanism to achieve adaptive adjustment of control parameters.

Images