Unmanned excavator energy consumption optimal excavation trajectory planning method based on skilled driver experience
By combining lidar and neural networks with soil failure theory and the Lagrange equation, the adaptive capability and energy consumption of unmanned excavators on complex material surfaces were solved, achieving automatic, energy-saving, and efficient excavation trajectory planning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-08
- Publication Date
- 2026-03-17
AI Technical Summary
Existing unmanned excavators struggle to mimic the efficient operating modes of skilled drivers when faced with complex and variable material surfaces, resulting in sudden changes in digging resistance, high energy consumption, and a lack of adaptability.
By scanning the material surface with lidar, the material morphology features are mapped to the reference trajectory of skilled drivers using a neural network model. A digging resistance model is established by combining soil failure theory, and a multibody dynamics model is constructed using the Lagrange equation. An optimal trajectory planning model is defined, and a sequential quadratic programming algorithm is used to optimize the trajectory to achieve energy-saving and efficient digging.
It enables automatic, energy-saving, and efficient excavation under any material surface, improving the adaptability and operational continuity of unmanned excavators.
Smart Images
Figure CN121680433A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of intelligent control of construction machinery, and more particularly to a method for planning the energy-optimal excavation trajectory of an unmanned excavator based on the experience of a skilled driver. Background Technology
[0002] Excavators, as core construction equipment in earthmoving engineering, are widely used in complex environments such as mining, road construction, and disaster relief. Compared to traditional manual operation, unmanned excavators liberate operators from high-risk and harsh environments, thus greatly improving the safety and continuity of operations. In recent years, to cope with complex and ever-changing working conditions, excavators have been developing towards intelligence and unmanned operation. Among these developments, trajectory planning for excavation on arbitrary material surfaces has become a significant technical challenge in the unmanned development process. Because the shape of material surfaces is arbitrary and unknown in actual operations, it can easily lead to sudden changes in excavation resistance, high energy consumption, and stiff, uneven trajectories. Considering that current mainstream trajectory planning algorithms struggle to effectively embed the expert experience of skilled drivers into the dynamic energy consumption model to adapt to actual working conditions, it is necessary to invent an energy-optimal excavation trajectory planning method for unmanned excavators based on the experience of skilled drivers. This is of great significance for ensuring the adaptive capability of unmanned excavators on arbitrary material surfaces and achieving energy-efficient and autonomous excavation. Summary of the Invention
[0003] To address the aforementioned technical problems in existing unmanned excavator trajectory planning, such as lack of adaptability to material environments, difficulty in mimicking the efficient operating modes of skilled drivers, and high energy consumption, this invention provides an energy-optimal excavation trajectory planning method for unmanned excavators based on the experience of skilled drivers. This invention integrates the operational experience of skilled drivers into the optimization objective of trajectory planning, achieving automatic, efficient, and energy-saving excavation under any material surface.
[0004] The technical means employed in this invention are as follows: A method for planning the energy-optimal excavation trajectory of an unmanned excavator based on the experience of a skilled driver, characterized by the following steps: S1. The lidar scans the surface of the material to be excavated in front of the excavator, and uses a surface fitting algorithm to fit the point cloud data of the material surface to extract the material morphology feature parameters; and collects excavation operation data of skilled drivers under different material surfaces. S2. Establish a neural network model to map the material morphology feature parameters to the reference excavation trajectory parameters of a skilled driver; S3. Based on the soil failure theory, establish a digging resistance model and use the Lagrange equation to establish a multibody dynamics model of the excavator working device that includes the digging resistance term. S4. Define the optimal excavation trajectory planning model, use parametric curves to parametrically describe the speed trajectory of each joint of the excavator, use the speed curve control points and excavation time as optimization variables, and construct an objective function and constraints that comprehensively consider system energy consumption and the similarity of the trajectory of skilled drivers. S5. Using the reference excavation trajectory parameters predicted in step S2, calculate the initial values of NURBS control points and the estimated excavation time in reverse, and initialize the optimization variables. S6. Based on the current optimization variables, reconstruct the velocity, position and acceleration trajectories of the boom, stick and bucket in the entire time domain; S7. Based on the reconstructed trajectory and the excavation resistance and dynamics model in step S3, calculate the driving torque of each joint in real time. S8. Calculate the numerical value of the objective function under the current trajectory and the degree of violation of geometric, kinematic and dynamic constraints; S9. Determine whether the optimization process meets the convergence condition; if it does, output the optimal mining control command; if it does not, proceed to step S10. S10. Update the optimization variables using a nonlinear optimization algorithm and proceed to step S6.
[0005] Further, S1 includes the following steps: S1.1, Data Collection by Skilled Drivers: Skilled drivers are invited to perform excavation operations on material surfaces with different topographical features. Angle sensors installed at the joints of the excavator's boom, stick, and bucket are used to collect data at a sampling frequency... Record Time series data of the first complete mining cycle; The dataset collected in this experiment is denoted as . ,in These are the material surface response surface coefficients corresponding to this experiment. Time series of joint angles:
[0006] S1.2 Parametric Fitting of Excavation Trajectory: To transform variable-length time-series data into fixed-length neural network output labels, a cubic polynomial is used to regress and fit the joint angle trajectory of each acquisition; arbitrary joints are defined. trajectory function for:
[0007] For the The first experiment For each joint, construct the least squares objective function. To solve for the fitting coefficients : ; Combine all the fitted joint coefficients to form the first... The label vector of this experiment ; S1.3, Neural Network Training: Constructing a system containing... Training dataset of group samples Define mean squared error (MSE) as the loss function. Used to measure the network's predicted output With real labels Differences between them: ; in, The regularization term is used to prevent overfitting; the backpropagation algorithm is used to calculate the gradient of the loss function with respect to the weights and biases, and the gradient descent optimization algorithm is used to iteratively update the network parameters until the loss function converges, thus completing the model building.
[0008] Further, S2 includes the following steps: S2.1, The material surface point cloud set acquired by radar and preprocessed is as follows: ;in Indicates the first The spatial coordinates of a point in the excavator base coordinate system; S2.2 Using a second-order bivariate polynomial as the response surface function Surface fitting is performed on the scattered point cloud to extract the geometric features of the material surface; the response surface function Defined as: ; in, Represents a vector of basis functions of a second-order polynomial. Let represent the response surface coefficient vector to be solved, corresponding to the coefficients of the constant term, the first term, and the second term, respectively; the neural network is a multilayer feedforward network, and its forward propagation calculation model is defined as:
[0009]
[0010] in, This represents the weight matrix from the input layer to the hidden layer. This represents the hidden layer bias vector. Here is the activation function for the hidden layer; This represents the weight matrix from the hidden layer to the output layer. This represents the output layer bias vector. This represents the activation function of the output layer. This represents the hidden layer output vector; S2.3, Output parameter vector The cubic polynomial fitting coefficients for the trajectories of the boom, stick, and bucket joints are as follows: .
[0011] Further, S4 includes the following steps: S4.1 Multibody Dynamics Model Parameter Definition: The excavator dynamics equations are established based on the Lagrange method to calculate the driving torque of each joint. :
[0012] in, These represent the joint angle, velocity, and acceleration vectors, respectively. express Symmetric positive definite inertial matrix, its elements Includes the mass of each link Moment of inertia and the position parameters of the centroid; for Coriolis and centrifugal force matrices, including Christofel's symbolic terms; express The gravity term vector is obtained by taking the partial derivative of the gravitational potential energy of each link with respect to the angle. The Jacobian matrix for the bucket tooth tip velocity is used to represent the digging resistance in Cartesian space. Mapped to the load torque in the joint space; S4.2 Definition of the comprehensive optimization objective function: Define the optimization variable vector. Construct the objective function : ; Among them, energy consumption indicators The integral representing the absolute value of the hydraulic drive power over the entire cycle; ; Experience similarity index Calculation based on dynamic time warping algorithm: Let the planned trajectory sequence be... The reference trajectory sequence is , build The distance matrix, elements DTW cumulative distance Satisfies the definition of recursion:
[0013] final ; S4.3 Solving Algorithm Technical Parameters: The sequential quadratic programming algorithm is used to solve the problem, which transforms the nonlinear programming problem into a series of quadratic programming subproblems.
[0014] Furthermore, in S42, at the iteration point Construct sub-problems: ; ; in The Hessian matrix approximation for the Lagrange function. The search direction is determined by using line search; the step size is determined by line search. ,renew .
[0015] Furthermore, in step S32, in order to obtain the optimal coefficient vector... Construct the least squares optimization objective function This minimizes the sum of squared residuals in the height direction between the fitted surface and the actual measured point cloud: ; Using matrix differentiation, let... To obtain the coefficient vector Analytical solution:
[0016] in, Represents the height vector of all point clouds. express The design matrix, its first row element is The neural network prediction module uses this coefficient vector As input data, a set of reference trajectory parameters for skilled drivers is output through nonlinear mapping. .
[0017] Compared with the prior art, the present invention has the following advantages: This invention uses a sequential quadratic programming algorithm for iterative solution. The SQP algorithm solves the original nonlinear problem by approximating it as a quadratic programming (QP) subproblem at each iteration point. It has the advantages of fast convergence speed, high computational efficiency, and strong constraint handling ability, making it very suitable for this type of engineering optimization problem. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is a schematic diagram illustrating the composition and coordinate system definition of the unmanned excavator system in an embodiment of the present invention.
[0020] Figure 2 This is a schematic diagram illustrating the principle of predicting the excavation trajectory of a skilled driver based on a neural network, as described in this invention.
[0021] Figure 3 This is a schematic diagram illustrating the force analysis of soil cutting resistance during the excavation process of the bucket in this invention.
[0022] Figure 4 This is a flowchart of the energy consumption optimization control method for the unmanned walking excavator of the present invention. Detailed Implementation
[0023] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0025] like Figure 1-4 As shown, this invention provides a method for planning the energy-optimal digging trajectory of an unmanned excavator based on the experience of a skilled driver, comprising the following steps: S1. The lidar scans the surface of the material to be excavated in front of the excavator, and uses a surface fitting algorithm to fit the point cloud data of the material surface to extract the material morphology characteristic parameters; excavation operation data of skilled drivers under different material surfaces are collected; S1 includes the following steps: S1.1, Data Collection by Skilled Drivers: Skilled drivers are invited to perform excavation operations on material surfaces with different topographical features. Angle sensors installed at the joints of the excavator's boom, stick, and bucket are used to collect data at a sampling frequency... Record Time series data of the first complete mining cycle; The dataset collected in this experiment is denoted as . ,in These are the material surface response surface coefficients corresponding to this experiment. Time series of joint angles:
[0026] S1.2 Parametric Fitting of Excavation Trajectory: To transform variable-length time-series data into fixed-length neural network output labels, a cubic polynomial is used to regress and fit the joint angle trajectory of each acquisition; arbitrary joints are defined. trajectory function for:
[0027] For the The first experiment For each joint, construct the least squares objective function. To solve for the fitting coefficients : ; Combine all the fitted joint coefficients to form the first... The label vector of this experiment ; S1.3, Neural Network Training: Constructing a system containing... Training dataset of group samples Define mean squared error (MSE) as the loss function. Used to measure the network's predicted output With real labels Differences between them: ; in, The regularization term is used to prevent overfitting; the backpropagation algorithm is used to calculate the gradient of the loss function with respect to the weights and biases, and the gradient descent optimization algorithm is used to iteratively update the network parameters until the loss function converges, thus completing the model building.
[0028] S2. Establish a neural network model to map the material morphology feature parameters to the reference excavation trajectory parameters of a skilled driver; S2 includes the following steps: S2.1, The material surface point cloud set acquired by radar and preprocessed is as follows: ;in Indicates the first The spatial coordinates of a point in the excavator base coordinate system; S2.2 Using a second-order bivariate polynomial as the response surface function Surface fitting is performed on the scattered point cloud to extract the geometric features of the material surface; the response surface function Defined as: ; in, Represents a vector of basis functions of a second-order polynomial. Let represent the response surface coefficient vector to be solved, corresponding to the coefficients of the constant term, the first term, and the second term, respectively; the neural network is a multilayer feedforward network, and its forward propagation calculation model is defined as:
[0029]
[0030] in, This represents the weight matrix from the input layer to the hidden layer. This represents the hidden layer bias vector. Here is the activation function for the hidden layer; This represents the weight matrix from the hidden layer to the output layer. This represents the output layer bias vector. This represents the activation function of the output layer. This represents the hidden layer output vector; S2.3, Output parameter vector The cubic polynomial fitting coefficients for the trajectories of the boom, stick, and bucket joints are as follows: .
[0031] S3. Based on soil failure theory, establish a digging resistance model and use the Lagrange equation to establish a multibody dynamics model of the excavator working device that includes the digging resistance term. In establishing an accurate dynamics model of the excavator digging process, by introducing a detailed soil-bucket interaction mechanical model, the digging resistance is incorporated as an external load into the multibody dynamics equation, thereby improving the physical realism of trajectory planning and the accuracy of energy consumption calculation.
[0032] S3.1, Excavation Resistance Modeling: To accurately calculate the external loads on the bucket during excavation, a soil failure model is used to describe the force situation when the bucket cuts the soil. The excavation resistance is decomposed into tangential resistance along the tangential direction of the bucket's movement trajectory. and normal resistance perpendicular to the trajectory of motion .
[0033] (1) Tangential excavation resistance
[0034] Based on the basic earthmoving equation, tangential excavation resistance is mainly generated by soil gravity, cohesion, adhesion, and inertial forces. Considering the relatively low conventional excavation speed, the inertial term is neglected, and the tangential resistance... The detailed calculation formula is as follows:
[0035] In the formula, Indicates the width of the bucket (m); Indicates the current cutting depth (m); This indicates soil density (kg / m³). This represents the acceleration due to gravity (m / s²). Indicates soil cohesion (Pa); This indicates the adhesion force (Pa) between the soil and the bucket. This represents the soil surface overload pressure (Pa). These represent dimensionless soil cutting resistance coefficients, which are related to the bucket cutting angle. soil internal friction angle Soil-metal friction angle and the angle of the fracture surface The function.
[0036] According to McKyes's two-dimensional wedge failure theory, the drag coefficient and The parsing expression is as follows: ; ; Among them, the angle of the fracture surface It can be obtained through the principle of least work, that is, satisfying... The angle.
[0037] (2) Normal excavation resistance Normal resistance Perpendicular to the bottom surface of the bucket, it mainly consists of the compaction reaction force and friction force of the bucket bottom plate against the soil. This is related to the tangential resistance. There is a coupling relationship between them, which can usually be represented as: ; in, This is the normal drag coefficient, typically taken as a value of [value missing]. It depends on the bucket cutting angle. And the degree of wear on the bucket floor plate. Alternatively, a more detailed decomposed model can be used to project the forces acting on the bucket teeth, sidewalls, and floor plate separately: ; In trajectory planning, the total resistance vector ultimately acting on the tip of the bucket teeth is... In Cartesian coordinates, this is represented as: ; in, Let be the rotation transformation matrix from the bucket local coordinate system to the base coordinate system.
[0038] S3.2, Multibody Dynamics Modeling of Excavators: The excavator's working device is simplified into a 3-DOF planar linkage mechanism consisting of the boom, stick, and bucket. The joint angle vector is defined as... These correspond to the joint angles of the boom, stick, and bucket, respectively. In step S3.2, to obtain the optimal coefficient vector... Construct the least squares optimization objective function This minimizes the sum of squared residuals in the height direction between the fitted surface and the actual measured point cloud: ; Using matrix differentiation, let... To obtain the coefficient vector Analytical solution:
[0039] in, Represents the height vector of all point clouds. express The design matrix, the first row element is The neural network prediction module uses this coefficient vector As input data, a set of reference trajectory parameters for skilled drivers is output through nonlinear mapping. .
[0040] The system's dynamic equations are established using the Lagrange method. The system's Lagrange function is also described. Defined as the total kinetic energy of the system With total potential energy Difference:
[0041] According to the Lagrange equation: ; Considering that the digging resistance acts as a generalized external force at the end of the system, the matrix form of the excavator's multibody dynamics equations is obtained: ; In the formula, The inertia matrix represents a symmetric positive definite matrix, which includes the mass moments of inertia of each link and coupling terms; Represents the matrix of Coriolis force and centrifugal force; This represents the gravity term vector, which is related to the mass and center of mass of each component. This represents the driving torque vector of each joint; The Jacobian matrix represents the velocity at the excavator's tip (bucket tooth tip), used to represent the digging resistance in Cartesian space at the tip. Mapped to joint space torque; This represents the digging resistance vector calculated in step S21. Since the excavator is driven by hydraulic cylinders, the joint torque... With hydraulic cylinder thrust The relationship between them is represented as
[0042] in This is the partial derivative matrix of the hydraulic cylinder length with respect to the joint angle (transmission ratio matrix). This refers to the output force of each hydraulic cylinder.
[0043] Ultimately, this dynamic model describes the dynamics of a given excavation trajectory. and soil environment The hydraulic cylinder driving force required by the system provides a dynamic constraint basis for subsequent energy consumption optimization planning.
[0044] S4. Define the optimal excavation trajectory planning model. Use parametric curves to parametrically describe the speed trajectory of each joint of the excavator. Use the speed curve control points and excavation time as optimization variables to construct an objective function and constraints that comprehensively consider system energy consumption and the similarity of the trajectory of a skilled driver.
[0045] S5. Using the reference excavation trajectory parameters predicted in step S2, calculate the initial values of NURBS control points and the estimated excavation time in reverse, and initialize the optimization variables. S6. Based on the current optimization variables, reconstruct the velocity, position and acceleration trajectories of the boom, stick and bucket in the entire time domain; S7. Based on the reconstructed trajectory and the excavation resistance and dynamics model in step S3, calculate the driving torque of each joint in real time. S8. Calculate the numerical value of the objective function under the current trajectory and the degree of violation of geometric, kinematic and dynamic constraints; S9. Determine whether the optimization process meets the convergence condition; if it does, output the optimal mining control command; if it does not, proceed to step S10. S10. Update the optimization variables using a nonlinear optimization algorithm and proceed to step S6.
[0046] Example: The XCMG ET120 walking excavator is used as the specific control platform. For example... Figure 1 As shown, the unmanned excavator system is equipped with a LiDAR (Light Detection and Ranging) system mounted on the top of the cab for real-time scanning of the working environment in front of the excavator; high-precision angle encoders are installed at the joints of the boom, stick, and bucket for real-time feedback of joint angle information. This invention proposes an energy-optimal excavation trajectory planning method for unmanned excavators based on the experience of skilled drivers, the core logic of which is as follows: Figure 4 As shown, the specific implementation steps are as follows: S1: Material surface point cloud data acquisition and topographic feature parameterization Before excavation begins, the lidar is first activated to scan the material pile to be excavated in front of the excavator. Because the raw point cloud data contains noise such as airborne dust and is located in the radar coordinate system (…),… Therefore, the first step is to remove outliers using a filtering algorithm, and then use a calibration matrix to transform the point cloud data to the excavator base coordinate system. The processed material surface point cloud set is obtained below. .
[0047] To achieve a mathematical description of the material surface morphology, this embodiment employs the polynomial response surface method to fit the surface of the scattered point cloud. A second-order bivariate polynomial is set as the response surface function, as shown in formulas (3) and (4) in the specification. By constructing a least-squares optimization objective function, the response surface coefficient vector is solved. This vector It contains geometric feature information of the material surface (such as slope, curvature, etc.), which will serve as the input interface for subsequent neural network models.
[0048] S2: Skilled Driver Experience Trajectory Prediction Based on Neural Networks like Figure 2 As shown, this step aims to infer, based on the current material morphology, what excavation strategy a skilled driver would employ under similar working conditions. A backpropagation (BP) neural network or radial basis function (RBF) neural network model is pre-built and trained: Input layer: The material feature parameter vector extracted in step S1. .
[0049] Output layer: Outputs the polynomial fitting parameter set of the excavation trajectory of a skilled driver. This parameter set was obtained by performing a cubic polynomial fit on a large amount of practical data from skilled drivers (sampling frequency 50Hz).
[0050] Prediction process: The coefficient vector of the current operating condition... The trained network is input, and after forward propagation, it outputs the corresponding trajectory parameters. This provides a high-quality initial value containing "expert experience" for subsequent automatic planning.
[0051] S3: Establishing a model of excavator digging resistance and multibody dynamics To accurately assess excavation energy consumption, a dynamic model incorporating soil-machine interaction must be established. Excavation resistance modeling: such as... Figure 3 As shown, based on McKyes' two-dimensional wedge failure theory, the digging resistance experienced by the bucket is decomposed into tangential resistance. and normal resistance Tangential resistance primarily overcomes soil gravity, cohesion, and adhesion; its magnitude is related to the cutting depth. Cutting angle and soil physical parameters (density) Cohesion (etc.); the normal resistance is mainly composed of the compaction reaction force of the bucket bottom plate.
[0052] Dynamic mapping: The excavator's working device is simplified into a three-degree-of-freedom planar linkage mechanism, and the system's dynamic equations are established using the Lagrange equations: ; in, This is the joint driving torque vector. To extract the resistance vector, the model establishes the resistance vector from the joint motion state ( The precise mapping relationship between the required hydraulic drive torque and the required hydraulic torque is the basis for calculating energy consumption.
[0053] S4: Construct a trajectory planning model that integrates energy consumption optimization and experience; this step defines the mathematical optimization problem of trajectory planning.
[0054] Variable parameterization: To ensure trajectory smoothness and suitability for hydraulic system control, quintic non-uniform rational B-spline (NURBS) curves are used to parameterize the angular velocity trajectories of the boom, stick, and bucket, respectively. Selection The control points, excluding the first and last control points (where the speed is 0), and the total excavation time, will be calculated. As a vector of variables to be optimized .
[0055] Objective function construction: Constructing a comprehensive objective function It consists of two parts: energy consumption items Based on the dynamic model in step S3, the absolute values of the hydraulic drive power of each joint throughout the entire excavation cycle are integrated to characterize the total energy consumption of the excavation operation. (Empirical deviation term) The Dynamic Time Warping (DTW) algorithm is used to calculate the similarity distance between the planned trajectory and the reference trajectory of the skilled driver predicted in step S2. Since the operation time of the automatically mined data may differ from that of the skilled driver, DTW can effectively solve the nonlinear alignment problem on the time axis compared to Euclidean distance, ensuring that the planned trajectory reproduces the driving skills of the skilled driver to the greatest extent in terms of shape and trend.
[0056] Constraint settings: Set strict physical constraints, including: geometric constraints (starting / ending pose), kinematic constraints (joint angle, velocity, acceleration limits), dynamic constraints (maximum thrust limit of hydraulic cylinder), and geometric envelope constraints to avoid collisions.
[0057] S5: Initialization of optimization variables; To avoid nonlinear optimization getting trapped in local optima and to improve solution speed, this invention does not use random initialization, but instead utilizes the skilled driver trajectory parameters predicted in step S2. By reverse calculation, the polynomial trajectory of a skilled driver is converted into the form of control points on a NURBS curve, and this set of control points is used as the optimization variable vector. initial value .
[0058] S6 to S10: Iterative solution based on Sequential Quadratic Programming (SQP); The above nonlinear programming problem is solved using the Sequential Quadratic Programming algorithm, specifically through a closed-loop iterative process: Trajectory Reconstruction (S6): In the first stage In the next iteration, based on the current optimization variables (i.e., NURBS control points and time), and reconstruct the velocity, position and acceleration curves of the boom, stick and bucket in the entire time domain using basis functions.
[0059] Dynamics calculation (S7): Substitute the reconstructed motion state into the resistance model and dynamic equations of step S3, and calculate the driving torque and hydraulic cylinder thrust required for each joint at each moment in real time.
[0060] Evaluation and Constraint Check (S8): Calculate the objective function value corresponding to the current trajectory. And check whether there are any violations of constraints such as maximum digging force and speed limits.
[0061] Convergence criterion (S9): Calculate the first-order optimality residuals under the Karush-Kuhn-Tucker (KKT) conditions. If the residuals are less than the preset tolerance... (For example If the algorithm converges, the current optimal control variable is output.
[0062] Update Iteration (S10): If convergence is not achieved, construct a quadratic programming (QP) subproblem at the current point, calculate the search direction and step size, and update the optimization variables to obtain the desired result. Then return to step S6 to enter the next cycle.
[0063] Finally, the globally optimal control variables obtained through the above steps will be discretized into a sequence of instructions that can be executed by the excavator's underlying controller, driving the hydraulic system to complete an automatic, efficient, and energy-saving excavation operation.
[0064] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In the above embodiments of the present invention, the descriptions of each embodiment have their own emphasis; parts not described in detail in a certain embodiment can be referred to in the relevant descriptions of other embodiments. It should be understood that the disclosed technical content in the several embodiments provided in this application can be implemented in other ways.
[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for optimal energy consumption trajectory planning of an unmanned excavator based on experienced skilled operator, characterized in that, The method comprises the following steps: S1, the laser radar scans the material surface to be excavated in front of the excavator, uses a curved surface fitting algorithm to fit the point cloud data of the material surface, and extracts material topographic feature parameters; excavating operation data of a skilled driver under different material surfaces is collected; S2, a neural network model is established, and the material topographic feature parameters are mapped to reference excavation trajectory parameters of the skilled driver; S3, a digging resistance model is established based on a soil failure theory, and a multi-body dynamics model of the working device of the excavator containing a digging resistance term is established by using a Lagrange equation; S4, an optimal excavation trajectory planning model is defined, a parameterized curve is used to parameterize the joint speed trajectory of the excavator, the control points of the speed curve and the excavation time are used as optimization variables, a target function and constraint conditions are constructed by comprehensively considering the system energy consumption and the trajectory similarity of the skilled driver; S5, the initial value of the NURBS control point and the estimated excavation time are obtained by reverse calculation by using the reference excavation trajectory parameters predicted in step S2, and the optimization variables are initialized; S6, the speed, position and acceleration trajectories of the boom, dipper and bucket in the full time domain are reconstructed according to the current optimization variables; S7, based on the reconstructed trajectory and the digging resistance and dynamics model in step S3, the driving torque of each joint is calculated in real time; S8, the numerical value of the target function under the current trajectory and the violation degree of the geometric, kinematic and dynamic constraints are calculated; S9, it is judged whether the optimization process meets the convergence condition; If yes, the optimal excavation control instruction is output; If no, go to step S10; S10, the optimization variables are updated by using a nonlinear optimization algorithm, and go to step S6. 2.The method of claim 1, wherein, The S1 comprises the following steps: S1.1, skilled operator operation data collection: invite skilled operators to operate excavators to dig on different material surfaces, use angle sensors installed at the joints of the excavator's boom, stick and bucket to collect data at a sampling frequency of 100 Hz record the timing data of the incomplete digging cycle; the first The data set collected in the second experiment is denoted as , where is the response surface coefficient corresponding to the material surface of the experiment, is the joint angle time series: S1.2, Parameterized fitting of the trajectory: To convert the variable-length time-series data into fixed-length neural network output labels, a cubic polynomial is used to regress and fit each collected joint angle trajectory; the trajectory function of any joint is defined as: For the first experiment, the least squares objective function was constructed for the first joint to solve for the fitting coefficients : ； All the joint relationship sets obtained by fitting are combined to form the first Label vector of the second experiment ; S1.3, Neural network training: Constructing a training dataset containing group samples ; Defining Mean Squared Error (MSE) as the loss function for measuring the difference between the network's predicted output and the true label : ； wherein, denotes the regularization term to prevent overfitting; the error back propagation algorithm is used to calculate the gradient of the loss function with respect to the weights and biases, and the gradient descent optimization algorithm is used to iteratively update the network parameters until the loss function converges, completing the model building. 3.The method of claim 1, wherein, The S2 comprises the following steps: S2.1, the set of material surface point clouds collected by radar and after pretreatment is ; wherein represents the spatial position coordinates of the i th point in the base coordinate system of the excavator. S2.2, using a second order bivariate polynomial as response surface function surface fitting the scattered point cloud to extract the geometry of the material surface; the response surface function is defined as: ； wherein, represents a second order polynomial basis function vector, represents a response surface coefficient vector to be solved, corresponding to constant term, first order term and second order term coefficients respectively; the neural network is a multi-layer feedforward network, and a forward propagation calculation model thereof is defined as: wherein, represents an input layer to hidden layer weight matrix, represents a hidden layer bias vector, is a hidden layer activation function; represents a hidden layer to output layer weight matrix, represents an output layer bias vector, represents an output layer activation function, represents a hidden layer output vector; S2.3, the output parameter vector Cubic polynomial fit coefficients containing the three joint trajectories of the boom, stick, and bucket: 。 4.The method of claim 1, wherein, The S4 comprises the following steps: S4.1, Polybody dynamics model parameter definition: based on Lagrange method to establish the dynamics equation of excavator, used to calculate the driving torque of each joint : wherein, θ, v, a represent joint angle, velocity and acceleration vectors, respectively; θ, v, a represent joint angle, velocity and acceleration vectors, respectively; symmetric positive definite inertia matrix, whose elements are containing the mass of each link , moment of inertia , and center of mass position parameters; is Coriolis and centrifugal force matrix, containing Christoffel symbol terms; is gravity term vector, obtained by taking the partial derivative of the potential energy of each link with respect to the angle; is the bucket tooth tip velocity Jacobian matrix, used to map the digging resistance in Cartesian space to the joint space load torques; S4.2, comprehensive optimization objective function definition: define the optimization variable vector , construct the objective function : ； The energy consumption index is represents the integral of the absolute value of the hydraulic drive power in the whole cycle; ； Experience similarity index Based on dynamic time warping algorithm: let the planning trajectory sequence be , the reference trajectory sequence be , construct the distance matrix of , element ; the DTW cumulative distance satisfies the recursive definition: Finally ; S4.3, algorithm technical parameters are solved: a sequential quadratic programming algorithm is used to solve the problem, and a nonlinear programming problem is converted into a series of quadratic programming sub-problems.
5. The method of claim 4, wherein, In the S4.3, at the iteration point , the sub-problem is constructed: ； ； where is the Hessian matrix approximation of the Lagrangian function, is the search direction; the step size is determined by line search , update .
6. The method of claim 3, wherein, In the S3.2, in order to obtain the optimal coefficient vector , a least square optimization objective function is constructed , so that the residual square sum of the fitting surface and the actual measured point cloud in the height direction is minimum: ； Using matrix differentiation, let... To obtain the coefficient vector Analytical solution: wherein, represents the height vector of all point clouds, represents the design matrix of the first row element is ; the neural network prediction module takes the coefficient vector as the input layer data, and outputs the reference trajectory parameter set of the skilled driver through nonlinear mapping.