Method and system for multi-objective dynamic optimal control of excavator bucket trajectory

Abstract

The application discloses a kind of excavator bucket trajectory multi-objective dynamic optimal control method and system, belong to open-pit mine mining equipment intelligent control field.It includes: obtaining operating state parameters and environmental parameters, extracting working condition characteristic vector and obtaining bucket tooth tip real-time pose based on kinematic model;Build time, energy consumption, smoothness and tracking accuracy multi-objective optimization model, solve optimal control point and smooth interpolation generation reference trajectory;Using digging resistance model to calculate actual resistance, compare with preset resistance, deviation is over threshold value when local re-planning and smooth splicing;With correction trajectory as target, generate drive signal using feedforward and feedback compound control;Store job data, every preset cycle uses reinforcement learning to update target weight, resistance coefficient and controller gain and feedback to each link.The application realizes multi-objective comprehensive optimization, dynamic self-adaption, high-precision tracking and self-learning optimization.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control technology for open-pit mining equipment, and specifically relates to a multi-objective dynamic optimal control method and system for the trajectory of an excavator bucket. Background Technology

[0002] Currently, excavator bucket trajectory control technology has made some progress, which can be mainly divided into four types: kinematic models, single-objective optimization, trajectory tracking, and intelligent optimization algorithms. The first method is trajectory planning based on kinematic models. Existing technologies plan the desired trajectory of the bucket teeth by establishing a kinematic model of the excavator's working device (e.g., using the DH method to establish forward kinematics solutions). For example, by obtaining the preset starting and ending poses of the bucket coordinate system relative to the base coordinate system, inverse kinematics calculations are used to obtain the trajectory of each joint angle. This type of method can achieve basic trajectory planning, but it usually only considers geometric constraints and fails to fully take into account the multi-objective optimization requirements such as energy consumption, time, and ride comfort.

[0003] The second type is trajectory planning methods based on single-objective optimization. Some studies have proposed excavation trajectory planning methods with the goal of minimizing energy consumption, constructing excavation resistance prediction models by studying the resistance characteristics during the interaction between the bucket and the material. Other studies have proposed trajectory planning methods that achieve a comprehensive optimization of joint running time and joint pulsation, using quintic non-uniform B-spline curves to interpolate discrete sequence points. There are also time-optimal trajectory planning methods based on reinforcement learning. However, most of these methods only optimize for a single objective (one of time, energy consumption, or smoothness), making it difficult to achieve dynamic balance among multiple mutually constraining optimization objectives.

[0004] The third type of control method is based on trajectory tracking and correction. Existing patents have proposed intelligent automatic control methods for the bucket trajectory of mechanical hydraulic excavators. By acquiring target trajectory data and excavation condition data, these methods calculate the desired force and trajectory correction compensation, achieving real-time trajectory correction. While these methods focus on trajectory tracking accuracy, they lack a systematic consideration of multi-objective optimality during the trajectory generation stage.

[0005] The fourth type is trajectory planning methods based on intelligent optimization algorithms. In recent years, researchers have attempted to use artificial fish swarm algorithms and improved RRT algorithms for path planning. For example, a quintic NURBS curve is used to plan the movement path of the bucket tooth tip, while simultaneously optimizing the operation time and movement smoothness. However, most existing intelligent optimization methods perform trajectory planning offline, making it difficult to dynamically adjust according to real-time changing working conditions (such as changes in material pile angle and fluctuations in soil resistance).

[0006] The aforementioned traditional methods have the following defects or shortcomings: Traditional excavator bucket trajectory control methods mainly suffer from difficulties in multi-objective optimization, poor dynamic adaptability, insufficient smoothness, poor matching of dynamic characteristics of the hydraulic system, and insufficient modeling accuracy. Most existing methods only optimize trajectories for a single objective, such as optimal time, optimal energy consumption, or optimal smoothness, lacking the ability to comprehensively optimize multiple objectives simultaneously, including operational efficiency, energy consumption, trajectory smoothness, and hydraulic system impact. In actual excavation operations, these objectives are often interdependent, and single-objective optimization is insufficient to meet the comprehensive performance requirements under complex working conditions.

[0007] Existing trajectory planning methods are mostly offline, and once the trajectory is determined, it is difficult to dynamically adjust it according to real-time changing working conditions such as changes in material pile angle, soil hardness fluctuations, and bucket load changes. Open-pit mining conditions are complex and variable, and offline trajectory planning cannot guarantee optimality under various working conditions.

[0008] Traditional trajectory planning methods often result in sudden acceleration changes or large pulsations, causing significant impacts on various structures of the excavator. This not only affects operational stability but also shortens the equipment's lifespan.

[0009] Existing methods rarely consider the dynamic response characteristics, flow limitations, and pressure constraints of hydraulic systems during the trajectory planning stage, leading to tracking deviations during actual trajectory execution.

[0010] The interaction force between the bucket and the material is affected by multiple factors such as the weight of the material, friction, digging speed, and the material on both sides of the bucket. Existing methods lack the accuracy and real-time performance to predict this type of dynamic resistance, which affects the accuracy of optimal trajectory control. Summary of the Invention

[0011] To address the shortcomings of existing technologies, the purpose of this invention is to provide an automatic optimal control method for the trajectory of an excavator bucket.

[0012] The technical solution adopted in this invention is: an automatic optimal control method for the trajectory of an excavator bucket, the key technical points of which include: Obtain the operating status parameters and environmental parameters of the excavator's working device; Based on the obtained operating status parameters and environmental parameters, the current working condition feature vector is extracted, and the real-time pose of the bucket tooth tip is obtained based on the pre-established kinematic model. Based on the extracted working condition feature vectors, a multi-objective optimization model is constructed that comprehensively considers working time, energy consumption, trajectory smoothness and tracking accuracy. Under the conditions of satisfying kinematic constraints and hydraulic system constraints, the optimal control point is solved, and the optimal control point is smoothly interpolated to generate a reference trajectory. Based on the real-time acquired operating status parameters and environmental parameters, the current actual excavation resistance is calculated using the excavation resistance model. The actual excavation resistance is compared with the preset resistance in the planning stage to calculate the resistance deviation. When the resistance deviation exceeds a preset threshold, the unexecuted trajectory segment is extracted for local replanning, and the replanned trajectory segment is smoothly spliced ​​with the completed trajectory segment to obtain the corrected trajectory. Using the corrected trajectory as the desired trajectory, a combined feedforward and feedback control strategy is adopted, and a drive control signal is generated by combining the dynamic response characteristics of the hydraulic system to control the excavator to perform digging actions. The system stores the operation data for each mining cycle, performs parameter updates every preset number of mining cycles, and uses reinforcement learning algorithms to iteratively update the target weights, resistance model coefficients, and controller gain parameters of the multi-objective optimization model. The updated parameters are then fed back to the trajectory planning, resistance correction, and tracking control stages to achieve continuous optimization of control performance. The above-mentioned links sequentially constitute the perception input layer, global decision layer, intermediate dynamic adjustment layer, execution drive layer, and iterative optimization layer, forming a five-layer progressive hierarchical control architecture. The overall structure is a dual-closed-loop coupling system consisting of a millisecond-level real-time operation inner loop and a long-cycle iterative optimization outer loop.

[0013] In the above scheme, the construction of a multi-objective optimization model that comprehensively considers operation time, energy consumption, trajectory smoothness, and tracking accuracy specifically includes: The optimization objective is to minimize the comprehensive evaluation index. The comprehensive evaluation index is composed of the weighted sum of four sub-objectives: total excavation time, total excavation energy consumption, joint impact index, and trajectory tracking error. Each sub-objective is assigned a dynamic weight coefficient, and the sum of all dynamic weight coefficients is always equal to 1. The kinematic constraints require that the angles of each joint be limited to a preset lower limit and an upper limit. The hydraulic system is constrained to ensure that the system pressure does not exceed a preset pressure limit and the main pump flow does not exceed a preset flow limit.

[0014] In the above scheme, the step of calculating the current actual excavation resistance using the excavation resistance model and comparing the actual excavation resistance with the preset resistance in the planning stage to calculate the resistance deviation specifically includes: The total excavation resistance is decomposed into four resistance components: material self-weight resistance, bucket friction resistance, material lateral extrusion resistance, and velocity-added resistance, and then superimposed for calculation. The difference between the pre-set resistance during the planning stage and the actual excavation resistance is taken as the resistance deviation value; When the absolute value of the resistance deviation exceeds a preset deviation threshold, the trajectory correction procedure is triggered.

[0015] In the above scheme, the step of extracting the unexecuted trajectory segment for local replanning and smoothly splicing the replanned trajectory segment with the completed trajectory segment specifically includes: During trajectory correction, only the trajectory segments that have not yet been executed are locally replanned, while the completed trajectory segments remain unchanged; By adjusting the control points, the replanned trajectory segment and the completed trajectory segment can be smoothly spliced ​​at the splicing point to meet the conditions of position continuity, velocity continuity, and acceleration continuity.

[0016] In the above scheme, the specific steps of using a combined feedforward and feedback control strategy to generate drive control signals based on the dynamic response characteristics of the hydraulic system include: The total control signal of the servo valve is composed of the superposition of two parts: the feedforward compensation amount and the feedback adjustment amount. The feedforward compensation is generated by taking the desired joint angular velocity, desired joint angular acceleration and desired trajectory resistance as inputs, multiplying each by the corresponding feedforward gain coefficient and summing them to achieve feedforward pre-compensation for the desired trajectory. The feedback adjustment adopts an adaptive fuzzy PID control strategy, which adjusts the PID control parameters online according to the tracking deviation between the actual trajectory and the desired trajectory to achieve closed-loop feedback adjustment.

[0017] In the above scheme, the iterative updating of the objective weights, resistance model coefficients, and controller gain parameters of the multi-objective optimization model using reinforcement learning algorithms specifically includes: With the goal of maximizing the cumulative reward value, the cumulative reward value of the current cycle is calculated and stored in the job database after each mining cycle. The cumulative reward value is obtained by weighted summation of multiple performance reward coefficients. The parameters are updated once every preset number of mining cycles. The updated objects include: the objective weights of the multi-objective optimization model, the resistance coefficients of the mining resistance model, and the control gain of the trajectory tracking controller. The updated target weights are fed back to the trajectory planning stage, the updated drag coefficients are fed back to the drag correction stage, and the updated control gain is fed back to the tracking control stage.

[0018] In the above scheme, in the dual closed-loop coupling system: The perception input layer serves as the sole data source for the entire system, providing standardized multi-source feature data for subsequent trajectory planning, resistance calculation, and tracking control. The global decision layer is used to determine the theoretical performance boundary of the operation; The intermediate dynamic adjustment layer is used to compensate for the mismatch between the static global trajectory and the dynamic working conditions on site; The execution drive layer is used to output valve control commands to drive the execution mechanism; The iterative optimization layer is used to achieve self-optimization throughout the entire lifecycle. The millisecond-level real-time operation inner loop is composed of the perception input layer, the global decision layer, the intermediate dynamic adjustment layer, and the execution drive layer connected in sequence. The long-cycle iterative optimization outer loop is independently composed of the iterative optimization layer and forms a long-cycle optimization loop. The real-time operation inner loop and the iterative optimization outer loop communicate with each other to form a dual closed-loop coupled control system.

[0019] In the above scheme, the forward real-time link realizes dynamic optimal control throughout the entire single mining process, the reverse iterative link continuously optimizes the system control performance over a long period of time, data is exchanged between each link, the hydraulic model and the resistance model are globally coupled, and the process constraints are unified.

[0020] An automatic optimal control system for excavator bucket trajectory, the key technical point of which is to achieve automatic optimal control of excavator bucket trajectory by adopting the aforementioned automatic optimal control method for excavator bucket trajectory.

[0021] An excavator, the key technical feature of which is that it includes the aforementioned automatic optimal control system for the excavator bucket trajectory.

[0022] Compared with the prior art, the present invention has the following beneficial effects: This invention overcomes the limitations of traditional single-objective or finite-objective optimization by establishing a multi-objective optimization model that simultaneously considers operation time, energy consumption, trajectory smoothness, system impact, and tracking accuracy. Through dynamic weighting coefficients, it achieves adaptive allocation of each sub-objective, enabling a dynamic optimal balance among multiple mutually constraining optimization objectives. Compared to existing technologies that only focus on full bucket rate, digging time, and digging energy consumption, this invention adds trajectory tracking accuracy as an independent optimization objective, making the planned trajectory both theoretically optimal and feasible.

[0023] Meanwhile, this invention employs a closed-loop mechanism of "real-time working condition perception, dynamic resistance prediction, and local trajectory replanning" to enable the planned trajectory to be dynamically adjusted based on real-time working conditions such as changes in material pile angle, soil hardness fluctuations, and bucket load changes. Unlike existing technologies that only passively compensate after offline planning, this invention adopts an active correction strategy of "triggering local replanning when deviation exceeds a threshold." This achieves trajectory retention while optimizing unexecuted segments online, balancing trajectory stability and working condition adaptability. It overcomes the poor adaptability of traditional offline planning methods and is suitable for the complex and ever-changing operating environment of open-pit mines.

[0024] This invention employs quintic NURBS curves for smooth interpolation of control points and introduces joint impact indices as optimization targets. By constraining the smoothness of acceleration during the trajectory planning stage, it effectively reduces abrupt acceleration changes at each joint. Simultaneously, in the trajectory correction stage, the invention uses triple continuity constraints on position, velocity, and acceleration at splicing points to ensure a smooth transition between the corrected and altered trajectories, reducing the impact generated by various structures of the excavator during operation and extending the equipment's service life.

[0025] This invention incorporates the dynamic response characteristics of the hydraulic system into the entire process of trajectory planning and tracking control. In the planning stage, the physical feasibility of the trajectory is ensured by hydraulic system constraints (pressure upper limit, flow upper limit). In the control stage, a composite control strategy of feedforward compensation (velocity feedforward, acceleration feedforward and resistance feedforward) and adaptive fuzzy PID feedback is used to achieve high-precision tracking of the desired trajectory.

[0026] This invention also utilizes an online learning mechanism of "operation data storage, reinforcement learning iteration, and three types of parameter feedback" to enable the system to continuously improve itself during use. Parameter updates are performed every preset cycle, achieving autonomous evolution of the control strategy according to changing operating conditions, maintaining optimal control performance over the long term, and overcoming the shortcomings of existing technologies that lack continuous optimization capabilities.

[0027] The five-layer progressive hierarchical control architecture proposed in this invention, consisting of a "sensing input layer, global decision-making layer, intermediate dynamic adjustment layer, execution drive layer, and iterative optimization layer," along with a dual-closed-loop coupling system of "millisecond-level real-time operation inner loop and long-cycle iterative optimization outer loop," is not disclosed in existing technologies. This architecture achieves a system-level integration of dynamic optimization throughout a single excavation process via a forward real-time link and continuous long-term optimization of system control performance via a reverse iterative link. It fundamentally solves the comprehensive technical challenges of multi-objective imbalance, poor adaptability to operating conditions, low tracking accuracy, and lack of autonomous iterative capability in traditional offline single-objective trajectory control, providing a novel systematic solution for intelligent excavator control. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in 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.

[0029] Figure 1 This is a flowchart of the multi-objective dynamic optimal control method for the excavator bucket trajectory in an embodiment of the present invention. Detailed Implementation

[0030] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the following description is provided in conjunction with the accompanying drawings. Figure 1 The present invention will be further described in detail below with reference to specific embodiments.

[0031] This embodiment employs an automatic optimal control method for the trajectory of an excavator bucket, which specifically includes the following steps: Step 1: Obtain the operating status parameters and environmental parameters of the excavator's working device.

[0032] Parameters are acquired in real time using multi-source sensors, specifically including: Joint sensing: Output from boom / stick / bucket tilt sensors Differentiate to get angular velocity .

[0033] Hydraulic sensors: boom cylinder pressure P1, stick cylinder pressure P2, bucket cylinder pressure P3, main pump flow rate Q, servo valve opening u; Environmental perception: The lidar outputs a 3D point cloud of the material pile, and the binocular camera identifies the pile angle β and the material hardness level H; Load sensing: The hydraulic cylinder pressure difference is converted into the real-time load (mL) of the bucket.

[0034] Step 2: Extract the current operating condition feature vector S based on the obtained operating status parameters and environmental parameters.

[0035] In this embodiment, the working condition feature vector ,in H represents the angle of the material pile, and H represents the hardness of the material (1-5 levels, 1 for loose soil and 5 for hard rock). To explore in depth, For bucket load, To determine the excavation feed rate.

[0036] Step 3: Obtain the real-time pose of the bucket tooth tip based on the pre-established kinematic model. The specific method is as follows: (1) Kinematic modeling of the excavator working device DH.

[0037] The excavator's working device includes four links: slewing platform base, boom joint J1, stick joint J2, and bucket joint J3, as well as four sets of DH coordinate systems. The DH parameter table is defined as follows.

[0038]

[0039] Homogeneous transformation matrix:

[0040] Total transformation matrix from base coordinate system to bucket tooth tip:

[0041] tooth tip coordinates in bucket coordinate system Calculation of tooth tip position in base coordinate system:

[0042] in Iterative solution for the real-time three-dimensional coordinates of the bucket tooth tip. Joint angle.

[0043] Step 4: Based on the extracted working condition feature vector, construct a multi-objective optimization model that comprehensively considers working time, energy consumption, trajectory smoothness and tracking accuracy, solve for the optimal control point and smoothly interpolate to generate the reference trajectory.

[0044] Step 4.1 Establish a multi-objective optimization model.

[0045] Using the control points of the bucket tooth trajectory as optimization variables, the weighted comprehensive objective function is constructed as follows:

[0046] In the formula, As a comprehensive evaluation indicator, Dynamic weights for time, energy consumption, smoothness, and tracking accuracy ( ), This represents the total time for a single excavation. To explore total energy consumption, This is an indicator of joint impact. This represents the trajectory tracking error.

[0047] Step 4.2 Determine the constraints of the above multi-objective optimization model.

[0048] Motion constraints are as follows: joint angles are limited to a preset lower limit. With upper limit Between, that is ; The hydraulic constraint is: the system pressure does not exceed the preset pressure upper limit. The main pump flow rate does not exceed the preset flow rate limit, i.e. .

[0049] Step 4.3 Trajectory interpolation solution.

[0050] The optimal control points were obtained using a multi-objective particle swarm optimization algorithm, and smooth interpolation using fifth-order NURBS curves was performed.

[0051] in For trajectory curves, As control points, For control point weights, It is a fifth-order basis function.

[0052] Step 5: Based on the real-time acquired operating status parameters and environmental parameters, calculate the current actual excavation resistance using the excavation resistance model, and compare it with the preset resistance in the planning stage to calculate the resistance deviation. When the deviation exceeds the threshold, extract the unexecuted trajectory segment for local replanning and smooth splicing. In this embodiment, the excavation resistance model decomposes the total excavation resistance into four resistance components for superposition calculation, specifically: Step 5.1 Construct the excavation resistance model.

[0053]

[0054] In the formula, For total excavation resistance, The resistance is due to the material's own weight. For the frictional resistance of the bucket, The lateral extrusion resistance of the material, Add drag to speed.

[0055] Step 5.2 Dynamic trajectory correction.

[0056] The actual excavation resistance is compared with the preset resistance during the planning stage to calculate the resistance deviation. In this embodiment, the resistance deviation is calculated using the following formula:

[0057] In the formula, Pre-set obstacles for the planning phase, if The process involves extracting unexecuted trajectory segments for local replanning, updating NURBS control points, and smoothly stitching them together without altering the completed trajectory. This results in a corrected trajectory, enabling real-time adaptation of the planned trajectory to dynamic on-site conditions.

[0058] Step 6: Using the corrected trajectory as the desired trajectory, a combined feedforward and feedback control strategy is adopted, and a drive control signal is generated by combining the dynamic response characteristics of the hydraulic system to control the excavator to perform digging actions.

[0059] Using the corrected trajectory as the desired trajectory, a feedforward + adaptive fuzzy PID composite control is adopted. The total control signal of the servo valve is composed of the superposition of the feedforward compensation and the feedback adjustment. The total control output is: ,in This is the main control signal for the servo valve. This is the feedforward compensation amount. This is the feedback adjustment amount.

[0060] The feedforward compensation is calculated by multiplying the desired joint angular velocity, desired joint angular acceleration, and desired trajectory drag by their respective inputs, and then summing them. Feedforward compensation

[0061] In the formula, For the feedforward gain of acceleration, velocity, and drag, The desired joint angular velocity and angular acceleration are given. Feedforward compensation enables pre-compensation for the desired trajectory, improving the system's response speed.

[0062] The feedback adjustment employs an adaptive fuzzy PID control strategy. This strategy adjusts the proportional, integral, and derivative parameters of the PID controller online through fuzzy inference based on the tracking deviation and rate of change between the actual and desired trajectories, achieving closed-loop feedback adjustment. The control signal drives the hydraulic cylinders via servo valves, controlling each joint of the excavator to perform digging actions according to the desired trajectory.

[0063] Step 7: After each mining cycle ends, store all types of data from this cycle into the job database.

[0064] The stored data includes: pile angle, material hardness grade, digging depth, real-time bucket load, digging feed speed, actual trajectory and expected trajectory, total digging energy consumption and energy consumption of each hydraulic cylinder, trajectory tracking deviation of each joint, and bonus value data.

[0065] Step 8: Perform parameter updates once every preset number of mining cycles. Use reinforcement learning algorithms to iteratively update the target weights, resistance model coefficients, and controller gain parameters of the multi-objective optimization model, and feed the updated parameters back to the trajectory planning stage, resistance correction stage, and tracking control stage, respectively.

[0066] In this embodiment, parameter updates are performed every preset number of mining cycles, using a reinforcement learning algorithm to iteratively update the three types of parameters, as detailed below: (1) Setting the reward function. The optimization objective is to maximize the cumulative reward value, which is calculated by weighted summation of multiple performance reward coefficients:

[0068] Where R is the reinforcement learning reward value. The parameters are iterated by maximizing R to represent the performance reward coefficients. After each mining cycle, the cumulative reward value R is calculated based on the actual performance of the cycle and stored in the job database.

[0069] (2) Parameter update strategy.

[0070] The parameters are updated every 10 mining cycles. Finally, the three types of parameters are updated in reverse: the objective weights of the multi-objective optimization model, the resistance coefficients of the mining resistance model, and the resistance coefficients of the mining resistance model.

[0071] The updated parameters are fed back to the trajectory planning, drag correction, and tracking control stages to continuously optimize control performance. The updated three types of parameters are fed back to their respective stages: multi-objective weighting. Input into trajectory planning, resistance model coefficients The resistance prediction is fed into the controller, and the controller gain is fed into the tracking control.

[0072] In this embodiment, the above-mentioned links sequentially constitute a perception input layer, a global decision-making layer, an intermediate dynamic adjustment layer, an execution driving layer, and an iterative optimization layer, forming a five-layer progressive hierarchical control architecture. The overall structure is a dual-closed-loop coupling system consisting of a millisecond-level real-time operation inner loop and a long-cycle iterative optimization outer loop. The working condition perception and data acquisition links constitute the perception input layer, which serves as the sole data source for the entire system, providing standardized multi-source feature data for subsequent trajectory planning, resistance calculation, and tracking control.

[0073] The multi-objective optimization trajectory planning stage constitutes the global decision-making layer. Based on the input of the perception layer, it constructs a four-objective optimization model of time, energy consumption, smoothness, and tracking accuracy, and solves it to generate a smooth reference trajectory, thereby determining the theoretical performance boundary of the operation.

[0074] The dynamic resistance prediction and trajectory correction process constitutes an intermediate dynamic adjustment layer. It synchronously reuses real-time sensor data and baseline trajectory, and locally replans and smooths the remaining trajectory based on material resistance fluctuations, thus compensating for the mismatch between the static global trajectory and the dynamic working conditions on site.

[0075] The trajectory tracking and closed-loop control links constitute the execution drive layer. With the corrected trajectory as the target, the composite controller designed in combination with the hydraulic dynamic characteristics outputs valve control commands to drive the excavator to perform digging actions, and continuously transmits back the equipment operating status to refresh the input of the perception layer, forming an uninterrupted real-time operation cycle.

[0076] The online learning and parameter adaptation process constitutes the iterative optimization layer. It collects all operation data from the first four layers across layers, and iteratively updates the multi-objective weights, resistance models, and controller parameters through reinforcement learning, and then sends them back to the planning, correction, and control modules to achieve full lifecycle self-optimization.

[0077] In this embodiment, the perception input layer, global decision layer, intermediate dynamic adjustment layer, and execution drive layer are sequentially connected to form a millisecond-level real-time operation inner loop, achieving dynamic optimal control throughout a single excavation process. The data flow between each link is as follows: the perception layer provides the decision layer with working condition characteristic data and tooth tip pose data; the decision layer provides the adjustment layer with the baseline trajectory and preset resistance; the adjustment layer provides the execution layer with the corrected trajectory; and the execution layer continuously transmits the equipment operating status back to refresh the perception layer input, forming an uninterrupted real-time operation cycle.

[0078] The iterative optimization layer independently constitutes the long-cycle iterative optimization outer loop, collects all operation data from the first four layers across layers, it iteratively updates the multi-objective weights, resistance models and controller parameters through reinforcement learning, and sends them back to the planning, correction and control modules to continuously optimize the system control performance over the long term.

[0079] Data is exchanged between the real-time operation inner loop and the iterative optimization outer loop, the hydraulic model and the resistance model are globally coupled, and process constraints are unified, forming a dual-closed-loop coupled control system. The forward real-time link realizes dynamic optimal control throughout a single excavation, while the reverse iterative link continuously optimizes the system control performance over the long term, fundamentally solving the shortcomings of traditional offline single-target trajectory control, such as multi-target imbalance, poor adaptability to working conditions, low tracking accuracy, and lack of autonomous iteration capability.

[0080] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for automatic optimal control of excavator bucket trajectory, characterized in that, include: Obtain the operating status parameters and environmental parameters of the excavator's working device; Based on the obtained operating status parameters and environmental parameters, the current working condition feature vector is extracted, and the real-time pose of the bucket tooth tip is obtained based on the pre-established kinematic model. Based on the extracted working condition feature vectors, a multi-objective optimization model is constructed that comprehensively considers working time, energy consumption, trajectory smoothness and tracking accuracy. Under the conditions of satisfying kinematic constraints and hydraulic system constraints, the optimal control point is solved, and the optimal control point is smoothly interpolated to generate a reference trajectory. Based on the real-time acquired operating status parameters and environmental parameters, the current actual excavation resistance is calculated using the excavation resistance model. The actual excavation resistance is compared with the preset resistance in the planning stage to calculate the resistance deviation. When the resistance deviation exceeds a preset threshold, the unexecuted trajectory segment is extracted for local replanning, and the replanned trajectory segment is smoothly spliced ​​with the completed trajectory segment to obtain the corrected trajectory. Using the corrected trajectory as the desired trajectory, a combined feedforward and feedback control strategy is adopted, and a drive control signal is generated by combining the dynamic response characteristics of the hydraulic system to control the excavator to perform digging actions. The system stores the operation data for each mining cycle, performs parameter updates every preset number of mining cycles, and uses reinforcement learning algorithms to iteratively update the target weights, resistance model coefficients, and controller gain parameters of the multi-objective optimization model. The updated parameters are then fed back to the trajectory planning, resistance correction, and tracking control stages to achieve continuous optimization of control performance. The above-mentioned links sequentially constitute the perception input layer, global decision layer, intermediate dynamic adjustment layer, execution drive layer, and iterative optimization layer, forming a five-layer progressive hierarchical control architecture. The overall structure is a dual-closed-loop coupling system consisting of a millisecond-level real-time operation inner loop and a long-cycle iterative optimization outer loop.

2. The automatic optimal control method for excavator bucket trajectory according to claim 1, characterized in that, The construction of the multi-objective optimization model that comprehensively considers operation time, energy consumption, trajectory smoothness, and tracking accuracy specifically includes: The optimization objective is to minimize the comprehensive evaluation index. The comprehensive evaluation index is composed of the weighted sum of four sub-objectives: total excavation time, total excavation energy consumption, joint impact index, and trajectory tracking error. Each sub-objective is assigned a dynamic weight coefficient, and the sum of all dynamic weight coefficients is always equal to 1. The kinematic constraints require that the angles of each joint be limited to a preset lower limit and an upper limit. The hydraulic system is constrained to ensure that the system pressure does not exceed a preset pressure limit and the main pump flow does not exceed a preset flow limit.

3. The automatic optimal control method for excavator bucket trajectory according to claim 1, characterized in that, The calculation of the current actual excavation resistance using the excavation resistance model, and the comparison of the actual excavation resistance with the preset resistance in the planning stage, specifically includes: The total excavation resistance is decomposed into four resistance components: material self-weight resistance, bucket friction resistance, material lateral extrusion resistance, and velocity-added resistance, and then superimposed for calculation. The difference between the pre-set resistance during the planning stage and the actual excavation resistance is taken as the resistance deviation value; When the absolute value of the resistance deviation exceeds a preset deviation threshold, the trajectory correction procedure is triggered.

4. The automatic optimal control method for excavator bucket trajectory according to claim 3, characterized in that, The step of extracting the unexecuted trajectory segment for local replanning and smoothly splicing the replanned trajectory segment with the completed trajectory segment specifically includes: During trajectory correction, only the trajectory segments that have not yet been executed are locally replanned, while the completed trajectory segments remain unchanged; By adjusting the control points, the replanned trajectory segment and the completed trajectory segment can be smoothly spliced ​​at the splicing point to meet the conditions of position continuity, velocity continuity, and acceleration continuity.

5. The automatic optimal control method for excavator bucket trajectory according to claim 1, characterized in that, The specific steps of employing a combined feedforward and feedback control strategy to generate drive control signals, based on the dynamic response characteristics of the hydraulic system, include: The total control signal of the servo valve is composed of the superposition of two parts: the feedforward compensation amount and the feedback adjustment amount. The feedforward compensation is generated by taking the desired joint angular velocity, desired joint angular acceleration and desired trajectory resistance as inputs, multiplying each by the corresponding feedforward gain coefficient and summing them to achieve feedforward pre-compensation for the desired trajectory. The feedback adjustment adopts an adaptive fuzzy PID control strategy, which adjusts the PID control parameters online according to the tracking deviation between the actual trajectory and the desired trajectory to achieve closed-loop feedback adjustment.

6. The automatic optimal control method for excavator bucket trajectory according to claim 1, characterized in that, The iterative updating of the objective weights, resistance model coefficients, and controller gain parameters of the multi-objective optimization model using reinforcement learning algorithms specifically includes: With the goal of maximizing the cumulative reward value, the cumulative reward value of the current cycle is calculated and stored in the job database after each mining cycle. The cumulative reward value is obtained by weighted summation of multiple performance reward coefficients. The parameters are updated once every preset number of mining cycles. The updated objects include: the objective weights of the multi-objective optimization model, the resistance coefficients of the mining resistance model, and the control gain of the trajectory tracking controller. The updated target weights are fed back to the trajectory planning stage, the updated drag coefficients are fed back to the drag correction stage, and the updated control gain is fed back to the tracking control stage.

7. The automatic optimal control method for excavator bucket trajectory according to claim 1, characterized in that, In the dual closed-loop coupling system: The perception input layer serves as the sole data source for the entire system, providing standardized multi-source feature data for subsequent trajectory planning, resistance calculation, and tracking control. The global decision layer is used to determine the theoretical performance boundary of the operation; The intermediate dynamic adjustment layer is used to compensate for the mismatch between the static global trajectory and the dynamic working conditions on site; The execution drive layer is used to output valve control commands to drive the execution mechanism; The iterative optimization layer is used to achieve self-optimization throughout the entire lifecycle. The millisecond-level real-time operation inner loop is composed of the perception input layer, the global decision layer, the intermediate dynamic adjustment layer, and the execution drive layer connected in sequence. The long-cycle iterative optimization outer loop is independently composed of the iterative optimization layer and forms a long-cycle optimization loop. The real-time operation inner loop and the iterative optimization outer loop communicate with each other to form a dual closed-loop coupled control system.

8. The automatic optimal control method for excavator bucket trajectory according to any one of claims 1 to 7, characterized in that, The forward real-time link enables dynamic optimal control throughout a single mining operation, while the reverse iterative link continuously optimizes the system control performance over the long term. Data is exchanged between each link, the hydraulic model and the resistance model are globally coupled, and process constraints are unified.

9. An automatic optimal control system for the bucket trajectory of an excavator, characterized in that, The method described in any one of claims 1 to 8 can be used to achieve automatic optimal control of the excavator bucket trajectory.

10. An excavator, characterized in that, It includes the automatic optimal control system for excavator bucket trajectory as described in claim 9.

Images