Multi-terrain photovoltaic cleaning crawler stability adaptive control method and system

Abstract

The application provides a kind of based on multi-terrain photovoltaic cleaning tracked vehicle stability adaptive control method and system, it is related to photovoltaic cleaning robot technical field, including by obtaining the multi-source sensing data of tracked vehicle, including attitude data, tracked pressure data, cleaning arm stress data and terrain scanning data, construct terrain three-dimensional point cloud model and carry out adaptive grid division;Based on multi-feature fusion identification terrain type and characteristic parameter, combine multi-source data to calculate tracked vehicle stability index;Adaptive entropy enhancement collaborative tracking algorithm is used to solve tracked torque distribution parameter and cleaning arm pose parameter;By constructing dynamic space-time constraint cost function, optimal path sequence is obtained by using progressive rolling horizon planning;Combined with predictive control strategy, generate tracked vehicle marching instruction and cleaning arm attitude instruction, realize the stable operation control of photovoltaic cleaning tracked vehicle.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic cleaning robots, and particularly relates to a stability adaptive control method and system based on a multi-terrain photovoltaic cleaning tracked vehicle. BACKGROUND

[0002] As an important part of clean energy, the power generation efficiency of photovoltaic power generation is directly affected by the cleanliness of the surface of photovoltaic modules. At present, large photovoltaic power stations generally use tracked cleaning vehicles for cleaning and maintenance of photovoltaic modules. Since photovoltaic power stations are mostly built in complex terrain such as mountains and deserts, cleaning operations face the challenge of stability brought by complex terrain. In order to ensure the quality of cleaning operations and the safety of equipment, the stability of tracked cleaning vehicles in multi-terrain environments needs to be accurately controlled.

[0003] The existing control method of photovoltaic cleaning tracked vehicles mainly has the following deficiencies: the traditional control method is mostly based on single sensor data, lacks the ability to fuse multi-source information, and cannot accurately perceive complex terrain characteristics; the existing stability control strategy often uses fixed parameter configuration, and cannot adaptively adjust to different terrain characteristics; the path planning method generally ignores the influence of terrain changes on the dynamics characteristics of tracked vehicles, resulting in instability risks in actual operation process; the cleaning arm pose control and tracked vehicle motion control are relatively independent, and lack a collaborative optimization mechanism.

[0004] In summary, in order to solve the stability control problem of photovoltaic cleaning tracked vehicles in complex terrain environments, a multi-source sensor information fusion terrain perception model is established, the adaptive entropy enhanced collaborative tracking algorithm is used to optimize the tracked torque distribution and cleaning arm pose parameters, and a dynamic space-time constrained progressive path planning method is used to realize the stable operation control of tracked vehicles in complex terrain environments. The present application effectively improves the environmental adaptability and operation safety of cleaning equipment, and has important engineering application value. SUMMARY

[0005] The embodiment of the present application provides a stability adaptive control method and system based on a multi-terrain photovoltaic cleaning tracked vehicle, which can solve the problems in the prior art.

[0006] The first aspect of the embodiment of the present application,

[0007] A stability adaptive control method based on a multi-terrain photovoltaic cleaning tracked vehicle is provided, comprising:

[0008] Obtaining attitude data, tracked pressure data, cleaning arm force data and terrain scanning data of the photovoltaic cleaning tracked vehicle, and constructing a terrain three-dimensional point cloud model;

[0009] The terrain three-dimensional point cloud model is adaptively meshed to obtain a plurality of mesh units, height correlation features between the mesh units are established, and terrain types and terrain feature parameters are determined based on multi-feature fusion;

[0010] A tracked vehicle stability index is calculated according to the attitude data, the track pressure data and the cleaning arm force data;

[0011] According to the terrain type, the terrain feature parameter and the tracked vehicle stability index, an adaptive entropy enhancement cooperative tracking algorithm is executed to solve tracked vehicle torque distribution parameters and cleaning arm pose parameters;

[0012] A dynamic space-time constraint cost function is constructed based on the tracked vehicle torque distribution parameters and the terrain feature parameters, and an optimal path sequence is obtained through progressive rolling horizon planning;

[0013] According to the optimal path sequence and the cleaning arm pose parameters, tracked vehicle marching instructions and cleaning arm attitude instructions are generated in combination with a predictive control strategy, so as to realize stable operation of the photovoltaic cleaning tracked vehicle.

[0014] In an alternative embodiment,

[0015] The terrain three-dimensional point cloud model is adaptively meshed to obtain a plurality of mesh units, height correlation features between the mesh units are established, and terrain types and terrain feature parameters are determined based on multi-feature fusion, which includes:

[0016] The terrain three-dimensional point cloud model is projected onto a horizontal plane, an initial edge length is calculated according to the total area and the total number of point clouds, and the point cloud model is divided into a plurality of initial mesh units;

[0017] The point cloud density distribution of the initial mesh units is calculated, a density flow field is constructed based on the density gradient, and a deformation vector is calculated to generate a deformed mesh, when the strain energy density of the deformed mesh exceeds a preset density threshold, the deformed mesh is split along the main direction of the density flow field to generate a target mesh unit;

[0018] The height difference value is calculated according to the height range in the target mesh unit, and the height gradient is calculated according to the height variation rate in the orthogonal direction; the height correlation features between the target mesh units are established, the height difference coefficient of adjacent units is calculated, and a space correlation matrix is constructed;

[0019] The height difference value, the height gradient and the space correlation matrix are combined into a feature vector, which is input into a multi-feature fusion terrain classification discriminant function, and the terrain type of the target mesh unit is determined based on the maximum likelihood criterion;

[0020] The terrain slope of the target mesh unit is calculated based on the height gradient, and the terrain roughness is calculated according to the integral deviation of the actual height from the least squares fitting plane;

[0021] The terrain type, terrain slope and terrain roughness are combined to form a terrain feature parameter vector, and terrain feature parameters of the target grid unit are determined.

[0022] In an alternative embodiment,

[0023] The point cloud density distribution of the initial grid unit is calculated, the density flow field is constructed based on the density gradient, and the deformation vector is calculated to generate the deformed grid. When the strain energy density of the deformed grid exceeds the preset density threshold, the deformed grid is split along the main direction of the density flow field to generate the target grid unit, comprising:

[0024] The number of point clouds of the initial grid unit is counted, the number of point clouds per unit area is calculated to obtain the point cloud density distribution, the X-direction density change rate and the Y-direction density change rate are calculated according to the point cloud density of the adjacent initial grid unit, and the density gradient vector is formed;

[0025] The density gradient vector is orthogonally rotated to obtain an orthogonal rotation vector, the density gradient vector and the orthogonal rotation vector are linearly combined to construct the density flow field, and the grid deformation vector is calculated according to the direction of the density flow field and the modulus of the density gradient vector;

[0026] The grid deformation vector is applied to the vertex of the initial grid unit to generate the deformed grid unit, and the strain energy density is calculated by calculating the deformation work per unit area in the deformed grid unit;

[0027] When the strain energy density exceeds the preset density threshold, the deformed grid unit is split along the main direction of the density flow field, the difference between the strain energy density and the preset density threshold is calculated, and the difference is substituted into an exponential function to determine the size of the split sub-grid;

[0028] The grid length ratio of the split sub-grid, the grid area change rate and the deformation vector difference of adjacent grid vertices are calculated. When the grid length ratio does not exceed the preset distortion threshold, the grid area change rate does not exceed the preset area change threshold, and the deformation vector difference does not exceed the product of the preset smoothing coefficient and the distance of the grid vertex, the target grid unit is determined.

[0029] In an alternative embodiment,

[0030] The track vehicle stability index is calculated according to the attitude data, the track pressure data and the cleaning arm force data, comprising:

[0031] The attitude data is received, including the pitch angle, the roll angle and the heading angle; the track pressure data is received, including the left track pressure distribution data and the right track pressure distribution data; the cleaning arm force data is received, including the torque data and the joint axial force data of each joint of the cleaning arm;

[0032] determine a boundary point coordinate of a track vehicle support polygon based on the attitude data, calculate distances from a track vehicle gravity center projection point to each side of the track vehicle support polygon, and determine a static stability margin as a ratio of a minimum value of the distances to a characteristic dimension of the track vehicle support polygon;

[0033] calculate a zero moment point coordinate of the track vehicle based on the track pressure data and the cleaning arm force data, determine a dynamic stability margin as a complement of a ratio of a distance between the zero moment point coordinate and a center coordinate of the track vehicle support polygon to a preset safety radius, and

[0034] determine a track vehicle stability index as a weighted sum of the static stability margin and the dynamic stability margin.

[0035] In an optional embodiment,

[0036] perform an adaptive entropy-enhanced cooperative tracking algorithm to solve the track torque distribution parameter and the cleaning arm pose parameter according to the terrain type, the terrain characteristic parameter, and the track vehicle stability index, including:

[0037] determine a parameter search space based on a preset track torque distribution parameter range and a preset cleaning arm pose parameter range, divide the parameter search space into a plurality of equal-probability subspaces, generate a cooperative tracking agent in each equal-probability subspace based on uniform distribution random sampling, and all the cooperative tracking agents generated in the equal-probability subspaces constitute a cooperative tracking agent group;

[0038] calculate a local entropy value for each cooperative tracking agent in the cooperative tracking agent group based on a neighborhood probability density, and determine a group overall entropy value as a mean value of the local entropy values;

[0039] construct a fitness function by weighted combination of a terrain evaluation value and a stability evaluation value according to the terrain type, the terrain characteristic parameter, and the track vehicle stability index;

[0040] calculate an adaptive step length adjustment coefficient based on the group overall entropy value, select a first agent according to the fitness function, and update parameters of agents in the cooperative tracking agent group in combination with the adaptive step length adjustment coefficient;

[0041] calculate a parameter dispersion degree of the cooperative tracking agent group, determine a group diversity index, and perform Gaussian random mutation when the group diversity index is less than a preset diversity threshold;

[0042] when a variation amount of an optimal parameter of continuous iterations is less than a preset convergence threshold, output a current optimal track torque distribution parameter and a current optimal cleaning arm pose parameter;

[0043] dynamically adjust a weight coefficient of the fitness function based on real-time updated terrain type and terrain characteristic parameter.

[0044] In an alternative embodiment,

[0045] A dynamic space-time constraint cost function is constructed based on the track torque distribution parameters and the terrain feature parameters, and an optimal path sequence is obtained through progressive hierarchical prediction in the time domain planning, including:

[0046] Based on the track torque distribution parameters, a torque distribution cost term is calculated through the left track torque parameters and the right track torque parameters, and based on the terrain feature parameters, a terrain feature cost term is calculated through the terrain slope parameters and the terrain roughness parameters, and a dynamic space-time constraint cost function is constructed by combining the time cost term, the path smoothing cost term, and the terrain feature cost term;

[0047] In the total planning time domain, a hierarchical prediction time domain architecture is constructed, the prediction time domain length is adaptively adjusted, a local optimal control sequence that minimizes the dynamic space-time constraint cost function is solved, the corresponding control instructions are executed and the prediction time domain is updated, and the cycle is iterated until the total planning time domain ends, and the planning result is obtained;

[0048] Based on the planning result, an optimal path sequence composed of position parameters, heading angle parameters, linear velocity parameters, and angular velocity parameters is generated.

[0049] In an alternative embodiment,

[0050] In the total planning time domain, a hierarchical prediction time domain architecture is constructed, the prediction time domain length is adaptively adjusted, a local optimal control sequence that minimizes the dynamic space-time constraint cost function is solved, the corresponding control instructions are executed and the prediction time domain is updated, and the cycle is iterated until the total planning time domain ends, and the planning result is obtained, including:

[0051] The total planning time domain is divided into a macro planning layer time domain and a micro execution layer time domain, and the macro planning layer time domain includes multiple consecutive prediction periods, each prediction period corresponding to a micro execution layer time domain;

[0052] The environmental complexity of the current position is acquired by collecting environmental information, the deviation value between the actual trajectory and the planned trajectory is detected, the length of the prediction time domain is determined based on the environmental complexity, and the adjustment interval of the prediction time domain is determined based on the deviation value;

[0053] The current prediction time domain is divided into two adjacent time intervals according to a preset proportion, a first time interval and a second time interval are determined, a dynamic prediction model including track-ground interaction force is established in the first time interval to generate a motion trajectory, a kinematic prediction model based on steering constraints is established in the second time interval to generate a path direction, and the motion trajectory and the path direction are combined to generate a prediction trajectory;

[0054] Solve a control sequence that minimizes a dynamic space-time constraint cost function based on the predicted trajectory, execute the control instruction corresponding to the first time interval, update the prediction time domain to the starting point of the next prediction, and continue iteration, until the total planning time domain ends, to obtain a planning result.

[0055] A second aspect of the embodiment of the application,

[0056] A multi-terrain photovoltaic cleaning tracked vehicle stability adaptive control system is provided, comprising:

[0057] The first unit is configured to acquire attitude data, track pressure data, cleaning arm force data, and terrain scanning data of the photovoltaic cleaning tracked vehicle, and construct a terrain three-dimensional point cloud model.

[0058] The second unit is configured to perform adaptive grid division on the terrain three-dimensional point cloud model to obtain a plurality of grid units, establish height correlation features between the grid units, and determine terrain types and terrain feature parameters based on multi-feature fusion.

[0059] The third unit is configured to calculate a tracked vehicle stability index according to the attitude data, the track pressure data, and the cleaning arm force data.

[0060] The fourth unit is configured to execute an adaptive entropy enhancement cooperative tracking algorithm to solve tracked vehicle torque distribution parameters and cleaning arm pose parameters according to the terrain types, the terrain feature parameters, and the tracked vehicle stability index.

[0061] The fifth unit is configured to construct a dynamic space-time constraint cost function based on the tracked vehicle torque distribution parameters and the terrain feature parameters, and obtain an optimal path sequence through progressive rolling horizon planning.

[0062] The sixth unit is configured to generate tracked vehicle travel instructions and cleaning arm attitude instructions according to the optimal path sequence and the cleaning arm pose parameters, in combination with a predictive control strategy, to realize stable operation of the photovoltaic cleaning tracked vehicle.

[0063] A third aspect of the embodiment of the application,

[0064] An electronic device is provided, comprising:

[0065] A processor;

[0066] A memory for storing processor-executable instructions;

[0067] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0068] A fourth aspect of the embodiment of the application,

[0069] A computer readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method.

[0070] In the embodiment of the present application, by constructing a terrain three-dimensional point cloud model and an adaptive grid division technology, different terrain types and characteristic parameters can be accurately identified, accurate perception and analysis of complex and variable terrains are realized, the adaptability of the photovoltaic cleaning tracked vehicle in different photovoltaic scenes is improved, and the tracked vehicle can cope with various complex terrain environments such as slopes and uneven roads; a stability index is calculated in combination with attitude data, tracked pressure data and cleaning arm stress data, and the adaptive entropy-enhanced cooperative tracking algorithm is used to dynamically adjust tracked torque distribution and cleaning arm pose, real-time monitoring and active adjustment of the stability of the tracked vehicle are realized, the risk of tipping is effectively prevented, and the safety and reliability of the equipment in extreme working conditions are improved; a dynamic space-time constraint cost function and a progressive rolling horizon planning method are used to generate an optimal path sequence, and a predictive control strategy is used to adjust operating parameters in real time, which not only optimizes the operation efficiency, but also realizes energy consumption optimization and wear reduction during the cleaning process, prolongs the service life of the equipment, reduces the maintenance cost, and improves the cleaning quality of the photovoltaic panel and the overall efficiency of the cleaning operation. BRIEF DESCRIPTION OF DRAWINGS

[0071] Figure 1 A flowchart of the adaptive control method for the multi-terrain photovoltaic cleaning tracked vehicle is shown in the embodiment of the present application.

[0072] Figure 2 A grid strain energy density and deformation performance relationship diagram is shown in the embodiment of the present application.

[0073] Figure 3 A tracked vehicle terrain adaptability result comparison diagram is shown in the embodiment of the present application.

[0074] Figure 4 An adaptive time domain adjustment effect diagram in the complex terrain navigation process is shown in the embodiment of the present application. DETAILED DESCRIPTION

[0075] To make the purpose, technical scheme and advantages of the embodiment of the present application clearer, the technical scheme in the embodiment of the present application will be described clearly and completely in combination with the drawings in the embodiment of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0076] The technical scheme of the present application will be described in detail in specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in some embodiments.

[0077] Figure 1 A flowchart of a method for adaptive control of stability of a multi-terrain photovoltaic cleaning tracked vehicle according to an embodiment of the present application is shown in FIG. 1. As shown in FIG. 1, the method comprises the following steps: Figure 1

[0078] Obtaining attitude data, tracked vehicle pressure data, cleaning arm force data, and terrain scanning data of the photovoltaic cleaning tracked vehicle, and constructing a terrain three-dimensional point cloud model;

[0079] Adaptive grid division is performed on the terrain three-dimensional point cloud model to obtain a plurality of grid cells, height correlation features between the grid cells are established, and terrain types and terrain feature parameters are determined based on multi-feature fusion;

[0080] Calculating a tracked vehicle stability index according to the attitude data, tracked vehicle pressure data, and cleaning arm force data;

[0081] According to the terrain types, terrain feature parameters, and tracked vehicle stability index, an adaptive entropy-enhanced cooperative tracking algorithm is executed to solve tracked vehicle torque distribution parameters and cleaning arm pose parameters;

[0082] A dynamic space-time constraint cost function is constructed based on the tracked vehicle torque distribution parameters and the terrain feature parameters, and an optimal path sequence is obtained through progressive rolling horizon planning;

[0083] According to the optimal path sequence and the cleaning arm pose parameters, tracked vehicle travel instructions and cleaning arm attitude instructions are generated in combination with a predictive control strategy to realize stable operation of the photovoltaic cleaning tracked vehicle.

[0084] In an alternative embodiment, adaptive grid division is performed on the terrain three-dimensional point cloud model to obtain a plurality of grid cells, height correlation features between the grid cells are established, and terrain types and terrain feature parameters are determined based on multi-feature fusion, which comprises:

[0085] Projecting the terrain three-dimensional point cloud model onto a horizontal plane, calculating an initial edge length according to the total area and total number of the point cloud, and dividing into a plurality of initial grid cells;

[0086] Calculating the point cloud density distribution of the initial grid cells, constructing a density flow field based on the density gradient and calculating a deformation vector to generate a deformed grid, and when the strain energy density of the deformed grid exceeds a preset density threshold, splitting along the main direction of the density flow field to generate target grid cells;

[0087] Calculating a height difference value based on the height range within the target grid cells, and calculating a height gradient based on the height variation rate in the orthogonal direction; establishing height correlation features between the target grid cells, calculating a height difference coefficient of adjacent cells, and constructing a spatial correlation matrix;

[0088] ​The height difference value, height gradient and spatial correlation matrix are combined into a feature vector, which is input into a multi-feature fusion terrain classification discriminant function, and the terrain type of the target grid cell is determined based on the maximum likelihood criterion;

[0089] The terrain slope of the target grid cell is calculated based on the height gradient, and the terrain roughness is calculated by calculating the integral deviation of the actual height from the least squares fitting plane.

[0090] The terrain type, terrain slope and terrain roughness are combined into a terrain feature parameter vector to determine the terrain feature parameters of the target grid cell.

[0091] In a specific embodiment, the acquired terrain three-dimensional point cloud model is projected onto a horizontal plane, and the initial grid edge length is calculated based on the total area and total number of points of the point cloud. Assuming that a certain terrain point cloud data contains 100,000 points and covers an area of 1 square kilometer, the initial grid edge length can be set to 10 meters, so that the entire area is divided into 10,000 initial grid cells. The initial grid edge length calculation formula is equal to the square root of the point cloud coverage area divided by the grid division coefficient, which can be adjusted according to actual application requirements, and is usually valued at 10 to 20.

[0092] For each initial grid cell, the point cloud density distribution is calculated. For example, for the grid cell numbered (25, 30), the number of points inside is 124, and the area is 100 square meters, so the point cloud density is 1.24 points per square meter. The adjacent grid cell numbered (25, 31) has 78 points, and the density is 0.78 points per square meter. Based on the density value of the grid cell, a density distribution map of the entire area is constructed, and the density gradient is calculated. For adjacent grid cells, the density gradient is the density difference divided by the grid center distance, such as the density gradient of the above two grid cells is (1.24-0.78) / 10 = 0.046 / m2 / m.

[0093] A density flow field is constructed using the density gradient information, and the deformation vector of each grid cell is calculated accordingly. The deformation vector is consistent with the direction of the density gradient, and the size is proportional to the density gradient, and the proportion coefficient can be set to 0.5. The initial grid is deformed according to the deformation vector to obtain a deformed grid. The strain energy density of the deformed grid is calculated, and when the strain energy density exceeds a predetermined threshold (such as 0.15), the grid is split along the main direction of the density flow field. For example, the strain energy density of the grid cell near the ridge line is 0.23, which exceeds the threshold, so it is split into two sub-grids. After splitting, the final grid obtained is called the target grid cell, and the total number is usually 1.2 to 1.5 times the number of initial grids.

[0094] For each target grid cell, the height feature of the internal point cloud is calculated, and the height range, i.e. the height difference between the highest point and the lowest point, is calculated. For example, if the height of the highest point in a certain target grid cell is 325.8 meters and the height of the lowest point is 318.2 meters, the height range is 7.6 meters. This value is saved as the height difference value.

[0095] The height variation rate in the orthogonal direction of the grid cell is calculated to determine the height gradient. The east-west direction and the north-south direction inside the grid cell are selected as the orthogonal directions, and the height variation rates in the two directions are calculated respectively. For example, the heights of the two end points in the east-west direction are 321.5 meters and 324.3 meters respectively, and the grid length is 12 meters, so the height variation rate in the east-west direction is (324.3-321.5) / 12=0.23 meters / meter; similarly, the height variation rate in the north-south direction is 0.31 meters / meter. Taking the average of the two, 0.27 meters / meter, as the height gradient of the grid cell.

[0096] The height correlation feature between the target grid cells is established, and the height difference coefficient of adjacent cells is calculated. For two adjacent grid cells, the height difference coefficient is the absolute value of the average height difference divided by the average of the height range. For example, the average heights of grid A and grid B are 322.4 meters and 320.1 meters respectively, and the height ranges are 7.6 meters and 6.8 meters respectively, so the height difference coefficient is |322.4-320.1| / ((7.6+6.8) / 2)=2.3 / 7.2=0.32.

[0097] According to the spatial position relationship between the target grid cells, a spatial correlation matrix is constructed. For example, for a 5x5 grid area, the spatial correlation of the central grid cell with the surrounding 24 grid cells is divided into three categories according to the distance: the correlation of the 8 adjacent grid cells is 0.8, the correlation of the grid cells one grid apart is 0.5, and the correlation of the remaining grid cells is 0.2.

[0098] The height difference value, height gradient and spatial correlation matrix of each target grid cell are combined into a feature vector, which is input into a multi-feature fusion terrain classification discriminant function. This function is based on the Bayesian conditional probability model, and calculates the conditional probability of the feature vector belonging to each terrain type, and selects the type with the maximum probability as the terrain type of the grid cell. For example, the conditional probabilities of the feature vector of a certain grid cell corresponding to the four terrain types of flat land, mountain, hilly land and gully are 0.15, 0.62, 0.18 and 0.05 respectively, so the terrain type of the grid cell is determined as mountain.

[0099] The terrain slope of the target grid cell is calculated based on the height gradient. The terrain slope is equal to the inverse tangent value of the height gradient, expressed in degrees. For example, if the height gradient is 0.27 meters / meter, the slope is 15.1 degrees.

[0100] The terrain roughness of the target grid cell is calculated. First, the plane equation of the point cloud in the grid cell is fitted by the least square method, and then the deviation of the actual height of each point from the height of the fitted plane is calculated, and the root mean square value of these deviations is taken as the roughness index. For example, if the sum of the squared deviations of the heights of 10 points in a certain grid cell is 1.82 square meters, then the roughness is (1.82 / 10) 1 / 2 = 0.43 meters.

[0101] The terrain type, terrain slope and terrain roughness are combined into a terrain feature parameter vector, which is used as the terrain feature parameter of the target grid cell. For example, the terrain feature parameter vector of a certain grid cell is {terrain type: mountain, slope: 15.1 degrees, roughness: 0.43 meters}.

[0102] Through adaptive grid division driven by density flow field, the characteristics of the terrain can be better preserved, and combined with the classification and discrimination mode of multi-feature fusion, the accurate identification and parameter extraction of complex terrain can be realized, which provides key technical support for applications such as unmanned aerial vehicle route planning and automatic driving terrain adaptability evaluation.

[0103] In an optional implementation, the point cloud density distribution of the initial grid cell is calculated, the density flow field is constructed based on the density gradient, and the deformation vector is calculated to generate a deformed grid. When the strain energy density of the deformed grid exceeds a preset density threshold, the deformed grid is split along the main direction of the density flow field to generate a target grid cell, comprising:

[0104] The number of point clouds in the initial grid cell is counted, the point cloud density distribution is calculated by counting the number of point clouds per unit area, the X-direction density change rate and the Y-direction density change rate are calculated according to the point cloud density of adjacent initial grid cells, and the density gradient vector is formed;

[0105] The density gradient vector is orthogonally rotated to obtain an orthogonal rotation vector, the density gradient vector and the orthogonal rotation vector are linearly combined to construct a density flow field, and the grid deformation vector is calculated according to the direction of the density flow field and the modulus of the density gradient vector;

[0106] The grid deformation vector is applied to the vertices of the initial grid cell to generate a deformed grid cell, and the strain energy density is calculated by calculating the deformation work per unit area in the deformed grid cell;

[0107] When the strain energy density exceeds the preset density threshold, the deformed grid cell is split along the main direction of the density flow field, the difference between the strain energy density and the preset density threshold is calculated, and the difference is substituted into an exponential function to determine the size of the split sub-grid;

[0108] The grid length ratio of the split sub-grid, the grid area change rate, and the deformation vector difference of adjacent grid vertices are calculated. When the grid length ratio is not more than a preset distortion threshold, the grid area change rate is not more than a preset area change threshold, and the deformation vector difference is not more than a product of a preset smoothing coefficient and the distance of the grid vertex, the target grid unit is determined.

[0109] In a specific embodiment, point cloud density analysis is performed on the initial grid unit. The entire point cloud data is projected onto the horizontal XY plane, and the entire region is divided into a plurality of initial grid units according to a preset initial grid length. For example, for point cloud data covering an area of 5000 square meters, an initial grid length of 10 meters is selected, and the entire region is divided into 50 initial grid units. The number of point clouds contained in each initial grid unit is counted, and the point cloud density distribution is obtained by dividing the grid unit area. For example, the grid unit numbered (3, 4) has an area of 100 square meters and contains 225 point clouds inside, so the point cloud density of the grid unit is 2.25 points per square meter.

[0110] For each initial grid unit, the density change rates in the X and Y directions are calculated respectively. Taking the grid unit numbered (3, 4) as an example, the point cloud density of the adjacent grid unit (4, 4) in the X direction is 1.85 points per square meter, so the X-direction density change rate is (1.85-2.25) / 10 = -0.04 points per square meter per meter; similarly, the point cloud density of the adjacent grid unit (3, 5) in the Y direction is 2.75 points per square meter, so the Y-direction density change rate is (2.75-2.25) / 10 = 0.05 points per square meter per meter. The X-direction density change rate and the Y-direction density change rate together constitute the density gradient vector of the grid unit, which is (-0.04, 0.05) points per square meter per meter.

[0111] The density gradient vector is orthogonally rotated to obtain an orthogonal rotation vector. Orthogonal rotation is to rotate the density gradient vector counterclockwise by 90 degrees, i.e., to rotate the vector (dx, dy) to (-dy, dx). For the density gradient vector (-0.04, 0.05) of the above grid unit, the orthogonal rotation vector is (-0.05, -0.04) points per square meter per meter.

[0112] The density gradient vector and the orthogonal rotation vector are linearly combined to construct a density flow field. The linear combination coefficients are α and β, respectively. Usually, α is 0.7 and β is 0.3. For the above grid unit, the density flow field vector is 0.7×(-0.04, 0.05)+0.3×(-0.05,-0.04) = (-0.043, 0.023) points per square meter per meter.

[0113] The grid deformation vector is calculated according to the direction of the density flow field and the modulus of the density gradient vector. The modulus of the density gradient vector is ((-0.04) 2 +(0.05) 2 ) 1 / 2 = 0.064 per square meter per meter. The size of the deformation vector is proportional to the modulus of the density gradient, and the direction is consistent with the vector of the density flow field. The proportional coefficient γ is usually 8. For the above grid cell, the grid deformation vector is 8 x 0.064 x (-0.043, 0.023) / ((-0.043) 2 +(0.023) 2 ) 1 / 2 = (-0.333, 0.179) meters.

[0114] The grid deformation vector is applied to each vertex of the initial grid cell to generate a deformed grid cell. For example, the coordinates of the four vertices of the initial grid cell are (30, 40), (40, 40), (40, 50), and (30, 50), and the coordinates after deformation are (29.667, 40.179), (39.667, 40.179), (39.667, 50.179), and (29.667, 50.179).

[0115] The strain energy density of the deformed grid cell is calculated, which is equal to the deformation work per unit area, and can be calculated by the displacement of the vertices before and after the grid deformation and the corresponding density gradient force. For the above grid cell, assume that the calculated strain energy density is 0.085.

[0116] It is determined whether the strain energy density exceeds the preset density threshold, which is usually set to 0.08; if the strain energy density exceeds the preset density threshold, the deformed grid cell is split along the main direction of the density flow field. For the above grid cell, 0.085 is greater than 0.08, and splitting is required. The main direction of the density flow field is the direction of the density flow field vector, i.e. the direction of (-0.043, 0.023).

[0117] The difference between the strain energy density and the preset density threshold is calculated, i.e. 0.085-0.08=0.005, and the difference is substituted into the exponential function f(x)=1-exp(-λx) to calculate the splitting coefficient, where λ is usually 200. For the difference 0.005, the splitting coefficient is 1-exp(-200x0.005)=0.632. The size of the sub-grid after splitting is determined according to the splitting coefficient and the original grid size, i.e. the side length of the sub-grid is the side length of the original grid multiplied by the splitting coefficient. For the original grid with a side length of 10 meters, the side length of the sub-grid after splitting is 10 x 0.632 = 6.32 meters.

[0118] The grid is split along the main direction of the density flow field, and the split line passes through the center point of the original grid and is perpendicular to the main direction of the density flow field. For the above grid unit, the center point coordinates are (34.667, 45.179), and the main direction of the density flow field is (-0.043, 0.023). The split line is perpendicular to the direction, and the direction vector is (0.023, 0.043).

[0119] The grid length ratio, grid area change rate, and deformation vector difference of adjacent grid vertices of the split sub-grid are calculated. The grid length ratio is the ratio of the longest side to the shortest side in the sub-grid, which is required to be not more than the preset distortion threshold, usually set to 1.5. The grid area change rate is the absolute value of the ratio of the sub-grid area to the original grid area minus 1, which is required to be not more than the preset area change threshold, usually set to 0.2. The deformation vector difference of adjacent grid vertices is the modulus value of the difference between the deformation vectors of adjacent grid vertices, which is required to be not more than the product of the preset smoothing coefficient and the distance of the grid vertices, and the preset smoothing coefficient is usually set to 0.1.

[0120] For example, the long sides of the two sub-grids after splitting are 8.5 meters and 8.3 meters, and the short sides are 6.2 meters and 6.4 meters, respectively. The grid length ratio is 8.5 / 6.2 = 1.37 and 8.3 / 6.4 = 1.30, respectively, both of which are less than the preset distortion threshold 1.5. The sub-grid areas are 52.7 square meters and 53.1 square meters, respectively, and the original grid area is 100 square meters. The grid area change rate is |52.7 / 100-1| = 0.473 and |53.1 / 100-1| = 0.469, respectively, both of which are greater than the preset area change threshold 0.2, and do not meet the conditions.

[0121] At this time, the position or direction of the split line needs to be adjusted, and the splitting is re-performed until all conditions are met. After multiple adjustments, the areas of the two sub-grids are 88.7 square meters and 87.2 square meters, respectively, and the area change rates are 0.113 and 0.128, respectively, both of which are less than the preset area change threshold 0.2. The maximum deformation vector difference of adjacent grid vertices is 0.85 meters, the grid vertex distance is 10 meters, the product of the preset smoothing coefficient and the grid vertex distance is 0.1 x 10 = 1 meter, and 0.85 is less than 1, which meets the conditions.

[0122] Therefore, the two sub-grids are determined as the target grid unit, and the corresponding original point cloud data is recorded, providing a basis for subsequent terrain feature analysis.

[0123] Existing point cloud processing techniques usually adopt uniform grid division or fixed structures such as octree, which are difficult to adapt to the complex changes of the terrain. For example, the traditional uniform grid division method has poor processing effect on the area with large change in point cloud density, and may cause redundant calculation in the sparse point cloud area and lose detailed information in the dense point cloud area. Although the octree method has certain adaptive ability, the fixed segmentation method cannot be optimized according to the terrain features.

[0124] The method of the embodiment is based on a grid generation method driven by a density flow field. Starting from the density distribution of the point cloud, the core concept of the density flow field is constructed, so that the grid division can be optimized along the main direction of the terrain features. The improvement is to improve the adaptability and expressiveness of the grid division to the terrain features. The density flow field is constructed by linear combination of the density gradient vector and the orthogonal rotation vector, the strain energy density is introduced to judge the splitting time, and the effectiveness of the splitting is ensured by multi-condition constraint.

[0125] Compared with the prior art, the method of the embodiment mainly embodies the advantages that: the concept of density flow field is introduced, so that the grid deformation and splitting can be performed along the main direction of the terrain change; the strain energy density is used as the splitting criterion, so that the adaptive splitting based on physical meaning is realized; and the quality and smooth transition of the grid after splitting are ensured by multiple constraints of the grid length ratio, the area change rate and the deformation vector difference.

[0126] The method of the embodiment can more accurately capture the terrain features, especially in the areas with large change in point cloud density, such as ridge, gully and other terrain transition areas. The grid division can be adaptively refined, so as to improve the accuracy of subsequent terrain analysis. The experimental results show that compared with the traditional uniform grid, the method improves the accuracy by about 35% in retaining the terrain features, and saves about 25% of the time in computing efficiency, which provides more reliable data support for the path planning and operation decision of the unmanned system in the complex terrain environment.

[0127] As Figure 2The simulation results of the relationship between the strain energy density and the mesh deformation and the splitting performance are shown. The horizontal axis represents the strain energy density value, and the vertical axis represents the mesh quality score (%) after deformation. The performance of the four methods under different strain energy densities is shown in the figure. When the strain energy density is 0.085, the mesh quality score of the technical solution reaches 92.7%, which is significantly higher than 78.4% of the splitting threshold method, 68.9% of the elastic deformation method and 61.2% of the fixed splitting method. When the strain energy density exceeds the preset density threshold (0.08), the mesh quality is maintained at a high level by introducing the mesh splitting strategy guided by the density flow field. Even in the extreme case of a strain energy density of 0.125, the quality score can still be maintained at 84.8%, while the quality scores of other methods rapidly decrease to 65.2%, 54.3% and 48.7%. This fully proves the superiority of the technical solution in processing high-density gradient regions. In particular, in the interval of strain energy density of 0.095 to 0.115, the average gap between the technical solution and the suboptimal method reaches 18.6 percentage points. When the strain energy density is 0.105, the mesh quality score of the technical solution is 89.3%, while the splitting threshold method is 72.6%, with a gap of 16.7%. The preset density threshold of 0.08 marked in the figure is a key parameter of the technical solution. Above this threshold, the density flow field main direction splitting algorithm is activated, so that the mesh quality score can be maintained at a high level without significant decrease with the increase of strain energy density.

[0128] In the embodiment, local refinement is realized based on point cloud density to enhance the geometric expression capability of the target region; a density flow field is constructed based on density gradient to guide the mesh to preferentially deform and split in high strain regions, thereby improving the adaptability of the model to non-uniform data; a grid length ratio, an area change rate and a smoothness constraint are introduced to effectively suppress mesh distortion and ensure the stability and accuracy of subsequent calculations; and a splitting scale is controlled by strain energy density and an exponential function to balance local feature capture and global structure rationality.

[0129] In an optional implementation, calculating the tracked vehicle stability index according to the attitude data, the track pressure data and the cleaning arm force data comprises:

[0130] The attitude data includes a pitch angle, a roll angle and a heading angle; the track pressure data includes left track pressure distribution data and right track pressure distribution data; and the cleaning arm force data includes torque data and joint axial force data of each joint of the cleaning arm;

[0131] Based on the attitude data, a boundary point coordinate of a tracked vehicle support polygon is determined, a distance from a tracked vehicle gravity center projection point to each side of the tracked vehicle support polygon is calculated, and a minimum value of the distance and a ratio of the minimum value to a characteristic size of the tracked vehicle support polygon are determined as a static stability margin;

[0132] calculating zero moment point coordinates of the tracked vehicle according to the tracked vehicle pressure data and the cleaning arm force data, determining a supplement of a ratio between a distance between the zero moment point coordinates and tracked vehicle support polygon center coordinates and a preset safety radius as a dynamic stability margin;

[0133] determining a weighted sum of the static stability margin and the dynamic stability margin as a tracked vehicle stability index.

[0134] In a specific embodiment, the tracked vehicle is generally used for cleaning operations in complex terrain environments, and its stability is crucial to operation safety. The method of the present embodiment obtains attitude data, tracked vehicle pressure data, and cleaning arm force data during tracked vehicle operation, then calculates a static stability margin and a dynamic stability margin based on these data, and finally fuses the two stability margins by weighting to obtain a final stability index.

[0135] The tracked vehicle is equipped with data acquisition equipment such as attitude sensors, tracked vehicle pressure sensors, and cleaning arm torque sensors. The attitude sensor acquires pitch angle, roll angle, and heading angle data of the tracked vehicle; the tracked vehicle pressure sensor acquires pressure distribution data of the left and right tracks; and the cleaning arm torque sensor acquires torque and axial force data of each joint of the cleaning arm.

[0136] Attitude data acquisition can be achieved by an inertial measurement unit (IMU) installed on the tracked vehicle body. This unit typically integrates an accelerometer, a gyroscope, and a magnetometer, and can acquire real-time attitude information of the tracked vehicle relative to the ground. For example, in an actual test, the acquired attitude data are: a pitch angle of 5.2 degrees, a roll angle of 3.7 degrees, and a heading angle of 178.5 degrees.

[0137] Tracked vehicle pressure data acquisition is achieved by a pressure sensor array installed in the contact area between the tracked vehicle and the ground, which can acquire the pressure distribution of the left and right tracks along the length direction. In actual application, 8 pressure measurement points are arranged on the left and right tracks, respectively, in the key areas of the front, middle, and rear parts of the tracks. For example, the left track pressure distribution is [120, 135, 142, 150, 148, 140, 128, 115] kPa, and the right track pressure distribution is [118, 132, 145, 153, 150, 142, 130, 112] kPa.

[0138] Cleaning arm force data are acquired by force / torque sensors installed at each joint of the cleaning arm. For a tracked vehicle equipped with a three-degree-of-freedom cleaning arm, the joint torque data are [250, 180, 120] N·m, and the joint axial force data are [800, 650, 400] N.

[0139] The boundary point coordinates of the support polygon of the tracked vehicle are determined based on the attitude data. The support polygon refers to a polygon formed by the boundary of the support area of the tracked vehicle in contact with the ground. In the case of ideal flat ground, the support polygon is approximately rectangular; in complex terrain conditions, the influence of the attitude of the tracked vehicle on the support polygon needs to be considered.

[0140] Specifically, a tracked vehicle body coordinate system is established, with the origin at the geometric center of the vehicle body, the x-axis pointing in the direction of the vehicle head, the y-axis pointing to the left side of the vehicle body, and the z-axis vertically upward. Then, according to the tracked vehicle size parameters (length L is 1.8 meters, width B is 0.6 meters) and the tracked vehicle attitude data, the coordinates of the four corner points of the tracked vehicle in the world coordinate system are determined through coordinate transformation.

[0141] Taking the above attitude data as an example, after coordinate transformation, the four boundary point coordinates of the support polygon are obtained: left front corner point (-0.9, 0.3, -0.082) meters, right front corner point (-0.9, -0.3, -0.065) meters, left rear corner point (0.9, 0.3, 0.094) meters, and right rear corner point (0.9, -0.3, 0.077) meters. Note that the z-coordinate here reflects the height difference caused by the pitch angle and roll angle.

[0142] The distances from the projection point of the tracked vehicle's center of gravity to the edges of the tracked vehicle's support polygon are calculated. The position of the center of gravity of the tracked vehicle can be determined by design parameters or obtained through dynamic analysis. In this example, the center of gravity is at (0.05, 0, 0.35) meters, and its projection point on the horizontal plane is at (0.05, 0, 0) meters.

[0143] The perpendicular distances from the projection point to the edges of the support polygon are calculated, resulting in four distance values: front edge distance 0.85 meters, right edge distance 0.3 meters, rear edge distance 0.85 meters, and left edge distance 0.3 meters. The minimum value 0.3 meters is taken as the key parameter for judging stability. The ratio of this minimum distance value to the characteristic size of the support polygon (which can be taken as the radius of the inscribed circle of the support polygon, 0.3 meters) is determined as the static stability margin, which is 1.0 in this example.

[0144] The zero moment point (ZMP) coordinates of the tracked vehicle are calculated based on the tracked pressure data and the force data of the cleaning arm. The zero moment point is the point where the sum of the moments of all forces acting on the system is zero, and is an important indicator for evaluating dynamic stability.

[0145] The calculation process first needs to integrate the tracked pressure distribution data to determine the positions of the left and right tracked pressure centers. According to the previous pressure distribution data, the left tracked pressure center is calculated to be 0.98 meters from the front end of the tracked vehicle, and the right tracked pressure center is calculated to be 1.02 meters from the front end of the tracked vehicle.

[0146] The zero moment point coordinates are calculated by the principle of moment balance combined with the force data of the cleaning arm. Considering the contribution of the cleaning arm posture and force to the overall moment of the system, the calculation results show that the zero moment point coordinates are (0.12, -0.05, 0) meters.

[0147] The distance between the zero moment point coordinates and the support polygon center coordinates (0, 0, 0) meters is calculated as 0.13 meters. The preset safety radius is taken as 0.25 meters, and the ratio of the distance between the zero moment point and the polygon center to the safety radius is 0.52, and the complement value 0.48 is determined as the dynamic stability margin.

[0148] The static stability margin and the dynamic stability margin are weighted and fused to obtain the tracked vehicle stability index. The weight coefficient can be adjusted according to actual operation requirements, and under normal circumstances, the weight of the static stability margin can be taken as 0.6, and the weight of the dynamic stability margin can be taken as 0.4. In this example, the tracked vehicle stability index is calculated as 1.0 × 0.6 + 0.48 × 0.4 = 0.792.

[0149] According to the calculated stability index, different thresholds can be set to correspond to different stability states: when the stability index is greater than 0.8, the system is highly stable and can perform high-difficulty operations; when the stability index is between 0.5 and 0.8, the system is in a normal stable state and is suitable for regular operations; when the stability index is between 0.3 and 0.5, the system stability is insufficient and the operation intensity needs to be reduced; and when the stability index is less than 0.3, the system is in a critical stable state and should immediately stop operation and adjust the posture.

[0150] The embodiment can be further extended, including considering the influence of terrain factors on the support polygon, introducing a filtering algorithm to improve data reliability, and performing time series analysis on the stability index to predict potential instability, etc. These extensions can further improve the accuracy and reliability of the tracked vehicle stability evaluation, and adapt to more complex operating environments.

[0151] In this embodiment, by determining the support polygon boundary point coordinates, calculating the ratio of the minimum value of the projection of the center of gravity to each edge to the characteristic size of the polygon, the stability margin of the vehicle in the static state is quantified, providing a basis for evaluating the safety of the vehicle under the influence of external dynamics; the zero moment point is calculated combined with the tracked pressure and cleaning arm force data, and compared with the support polygon center coordinates, and the complement value of the distance ratio is used to determine the dynamic stability margin, so that the real-time stability performance can still be evaluated in time when the vehicle is disturbed or the load changes; by weighting the static stability margin and the dynamic stability margin, a comprehensive tracked vehicle stability index is formed, providing an intuitive and quantitative evaluation index for the overall stability of the vehicle, facilitating monitoring and early warning.

[0152] In an optional embodiment, the adaptive entropy-enhanced cooperative tracking algorithm is executed to solve the track torque distribution parameters and the cleaning arm pose parameters according to the terrain type, the terrain characteristic parameters and the tracked vehicle stability index, and includes:

[0153] Based on the preset track torque distribution parameter range and the preset cleaning arm pose parameter range, a parameter search space is determined, the parameter search space is divided into a plurality of equal-probability subspaces, a cooperative tracking agent is generated in each equal-probability subspace based on a uniform distribution random sampling, and all the cooperative tracking agents generated in the equal-probability subspaces constitute a cooperative tracking agent group;

[0154] A local entropy value is calculated for each cooperative tracking agent in the cooperative tracking agent group based on a neighborhood probability density, and a mean value of the local entropy values is determined as a group overall entropy value;

[0155] According to the terrain type, the terrain characteristic parameters and the tracked vehicle stability index, a fitness function is constructed by weighted combination of the terrain evaluation value and the stability evaluation value;

[0156] An adaptive step length adjustment coefficient is calculated based on the group overall entropy value, a first agent is selected according to the fitness function, and parameters of the agents in the cooperative tracking agent group are updated in combination with the adaptive step length adjustment coefficient;

[0157] A parameter dispersion degree of the cooperative tracking agent group is calculated, a group diversity index is determined, and a Gaussian random mutation is performed when the group diversity index is less than a preset diversity threshold;

[0158] When a variation of the optimal parameters in continuous iterations is less than a preset convergence threshold, the current optimal track torque distribution parameters and the cleaning arm pose parameters are outputted;

[0159] The weight coefficients of the fitness function are dynamically adjusted based on the real-time updated terrain type and terrain characteristic parameters.

[0160] When the tracked vehicle performs a complex terrain cleaning operation, the track torque needs to be reasonably distributed and the cleaning arm pose needs to be adjusted according to the terrain type, the terrain characteristic parameters and the stability index, so as to ensure the efficiency and safety of the cleaning operation.

[0161] The parameter search space is determined according to the preset track torque distribution parameter range and the preset cleaning arm pose parameter range. The track torque distribution parameters include a torque ratio TL / TR of the left and right tracks, and the value range is set to [0.5, 2.0]. The cleaning arm pose parameters include an extension length L of the cleaning arm (range [0.8m, 1.5m]), a pitch angle α (range [-30°, 45°]) and a yaw angle β (range [-60°, 60°]).

[0162] The parameter search space is divided into multiple equal-probability subspaces. In a specific implementation, the interval [0.5, 2.0] of the torque ratio TL / TR is uniformly divided into 15 subintervals, the interval [0.8 m, 1.5 m] of the stretch length L is uniformly divided into 10 subintervals, the interval [-30°, 45°] of the pitch angle a is uniformly divided into 15 subintervals, and the interval [-60°, 60°] of the yaw angle b is uniformly divided into 20 subintervals, thereby forming a grid division of the multi-dimensional parameter search space.

[0163] A cooperative tracking agent is randomly generated based on a uniform distribution within each equal-probability subspace. For example, in the subspace [1.1, 1.2] x [1.0 m, 1.1 m] x [0°, 5°] x [-10°, -5°], a cooperative tracking agent with a torque ratio of 1.15, a stretch length of 1.05 m, a pitch angle of 2.5°, and a yaw angle of -7.8° is randomly generated. A total of 200 cooperative tracking agents are generated in all subspaces to form a cooperative tracking agent group.

[0164] The local entropy value of each agent in the cooperative tracking agent group is calculated to reflect the uncertainty of the parameter distribution. The local entropy value is calculated as follows: first, a neighborhood radius r of each agent is determined (for example, 10% of the diagonal length of the parameter space), the number of agents n within the radius r is counted, then the probability density p of the region is calculated, and the local entropy value h is calculated. For example, there are 8 agents in the neighborhood of a certain agent, and the total number of agents is 200, so the probability density p of the region is 8 / 200 = 0.04, and the local entropy value h is -p x ln(p) = 0.128. The average value of the local entropy values of all agents is taken as the overall entropy value H of the group, and the overall entropy value H of the group in the initial state is about 0.85.

[0165] An adaptability function is constructed according to the terrain type, terrain characteristic parameters, and tracked vehicle stability index. The terrain evaluation value ET is determined according to the terrain type: 0.9 for flat terrain, 0.7 for gentle slope terrain, 0.5 for steep slope terrain, and 0.3 for step terrain. The terrain characteristic parameters include the slope angle θ (ranging from 0° to 30°) and the terrain roughness R (ranging from 0 to 1). The stability evaluation value ES is calculated based on the tracked vehicle stability index SI, and the higher the SI value, the larger the ES value. The adaptability function F is constructed by weighted combination of the terrain evaluation value and the stability evaluation value: F = w1 x ET + w2 x ES, where w1 and w2 are weight coefficients, and in the initial state, w1 = 0.4 and w2 = 0.6.

[0166] An adaptive step size adjustment coefficient λ is calculated based on the group overall entropy value H. When the H value is large (for example, H>0.7), λ takes a smaller value (such as 0.3); when the H value is small (for example, H<0.3), λ takes a larger value (such as 0.8), to balance global search and local convergence. The agent with the highest fitness value is selected as the first agent, and the remaining agents update the parameters based on the first agent: new parameter value = original parameter value + λ × (first agent parameter value - original parameter value) + random disturbance. For example, the first agent torque ratio is 1.3, the torque ratio of an agent is 1.1, λ = 0.5, and the random disturbance is 0.02, then the updated torque ratio of the agent is 1.1 + 0.5 × (1.3-1.1) + 0.02 = 1.22.

[0167] The parameter dispersion degree of the cooperative tracking agent group is calculated to determine the group diversity index D. The parameter dispersion degree is obtained by calculating the standard deviation of each parameter in the agent group. The group diversity index D is obtained by normalizing and weightedly averaging the standard deviations of the four parameters. When D is less than a preset diversity threshold DT (for example, DT = 0.15), Gaussian random variation is performed: 30% of the agents are randomly selected, and Gaussian random disturbance with a mean of 0 and a standard deviation of 0.1 is added to their parameters, to increase the diversity of the group and prevent falling into local optimum.

[0168] The above steps are iteratively executed. When the optimal parameter variation of 10 consecutive iterations is less than a preset convergence threshold (for example, the torque ratio variation is less than 0.01, the stretch length variation is less than 0.01m, and the angle variation is less than 0.5°), the current optimal track torque distribution parameter and cleaning arm pose parameter are output.

[0169] The weight coefficients of the fitness function are dynamically adjusted based on the real-time updated terrain type and terrain feature parameters. When it is detected that the terrain is flat or gently sloping, the weight w1 of the terrain evaluation value is increased (for example, adjusted to 0.6), and the weight w2 of the stability evaluation value is reduced (for example, adjusted to 0.4); when it is detected that the terrain is steep or stepped, the weight w1 of the terrain evaluation value is reduced (for example, adjusted to 0.3), and the weight w2 of the stability evaluation value is increased (for example, adjusted to 0.7). At the same time, the weight is adjusted according to the terrain roughness R: the larger the R value, the larger the w2.

[0170] Exemplarily, a certain tracked vehicle is working on a steep slope terrain with a slope angle of 20° and a terrain roughness of 0.6, and the initial stability index is 0.72. After 50 iterations of the adaptive entropy-enhanced cooperative tracking algorithm, the optimal tracked torque distribution parameters are obtained: torque ratio TL / TR = 1.35 (the left tracked torque is greater than the right tracked torque, which is beneficial to lateral stability); the optimal cleaning arm pose parameters are: extension length L = 0.95 m (shorter extension length reduces the center of gravity), pitch angle a = 15° (lift up to adapt to the slope), yaw angle β = -10° (slightly deflect inward to balance the weight). After using this set of parameters, the tracked vehicle stability index is improved to 0.86, and the cleaning efficiency is improved by about 25%.

[0171] In this embodiment, the entire parameter search space is divided into multiple equal-probability subspaces using the preset tracked torque distribution parameter range and cleaning arm pose parameter range, and then the cooperative tracking agents are uniformly sampled in each subspace to effectively cover the global search space and improve the search efficiency and comprehensiveness; the local entropy value under the neighborhood probability density is calculated for each cooperative tracking agent, and the mean value thereof is taken as the group overall entropy value, which can comprehensively reflect the dispersion degree of the internal parameter distribution of the agent group and the search diversity, thereby providing a basis for subsequent adaptive step length adjustment and parameter updating; the terrain type, terrain characteristic parameters and tracked vehicle stability index are combined to construct the fitness function, so that the search process not only considers the stability of the target device, but also takes into account the environmental adaptability, further improving the scientificity of parameter selection and the reliability of practical application; the adaptive step length adjustment coefficient is calculated according to the group overall entropy value, the agent parameters are dynamically updated, and at the same time the diversity index is used to monitor the parameter dispersion degree of the group, when the diversity is insufficient, the search space is expanded through Gaussian random mutation, ensuring that the algorithm has sufficient exploration ability in the global range and preventing falling into local optimum.

[0172] As Figure 3As shown, the scheme of the embodiment is compared with three existing technologies in terms of terrain adaptability under different terrain types. From the data in the figure, it can be seen that the scheme of the embodiment performs well in all test terrains, especially in complex terrain conditions. In steep slope terrain (slope 42°), the terrain adaptability score of the technical scheme reaches 87.6, which is 12.3 points higher than the parameter adjustment method based on PID control; in soft mud terrain, the adaptability score is 83.2, which is 15.7 points higher than the parameter optimization method based on genetic algorithm; in rocky uneven road surface, the adaptability score reaches 92.1, which is 18.9 points higher than the self-adaptive technology based on fuzzy control; in composite terrain, the adaptability score is 85.9, which is superior to all comparison technologies. The data shows that the scheme of the embodiment, through the adaptive step adjustment based on group entropy and the diversity maintenance mechanism, optimizes the cleaning arm pose parameters and the track torque distribution, greatly improves the stability and work efficiency of the system in various terrains. The figure also presents the iteration convergence speed of different algorithms. The scheme of the embodiment can reach a stable state after an average of 15.2 iterations, while the parameter adjustment method based on PID control needs 25.3 iterations, the parameter optimization method based on genetic algorithm needs 27.8 iterations, and the self-adaptive technology based on fuzzy control needs 28.5 iterations, proving the efficiency of the scheme. It is worth noting that in the terrain mutation area (marked as T4 area in the figure), the real-time adjustment capability of the scheme of the embodiment makes its adaptability decrease by only 8.3%, which is much lower than the decrease of the parameter adjustment method based on PID control (19.7%), the parameter optimization method based on genetic algorithm (23.4%) and the self-adaptive technology based on fuzzy control (15.7%).

[0173] In an alternative embodiment, a dynamic space-time constraint cost function is constructed based on the track torque distribution parameters and the terrain feature parameters, and the optimal path sequence is obtained through progressive hierarchical prediction time domain planning, including:

[0174] Based on the track torque distribution parameters, the torque distribution cost term is calculated through the left track torque parameter and the right track torque parameter, and based on the terrain feature parameters, the terrain feature cost term is calculated through the terrain slope parameter and the terrain roughness parameter, and the dynamic space-time constraint cost function is constructed by combining the time cost term and the path smoothing cost term;

[0175] In the total planning time domain, a hierarchical prediction time domain architecture is constructed, the prediction time domain length is adaptively adjusted, the local optimal control sequence that minimizes the dynamic space-time constraint cost function is solved, the corresponding control instructions are executed and the prediction time domain is slidingly updated, and the cycle iteration is repeated until the total planning time domain ends, to obtain the planning result;

[0176] Based on the planning result, an optimal path sequence composed of position parameters, heading angle parameters, linear velocity parameters and angular velocity parameters is generated.

[0177] In one specific embodiment, a dynamic spatiotemporal constraint cost function is constructed, which consists of four main parts: track torque distribution cost term, terrain feature cost term, time cost term, and path smoothness cost term.

[0178] The track torque distribution cost term is calculated by the left and right track torque parameters. Specifically, when the robot is moving, the real-time values of the left track torque TL and the right track torque TR are recorded. The torque difference AT = |TL-TR| is calculated, and a torque balance coefficient KT (a typical value is 0.7) is set. If the torque difference is greater than a preset threshold Tthreshold (e.g., 50 Nm), the torque distribution cost CT = KT x (AT-Tthreshold); otherwise, CT = 0. This design allows the robot to maintain a balanced force on the left and right tracks as much as possible, improving the stability of movement.

[0179] The terrain feature cost term takes into account the terrain slope and roughness. The slope angle a (unit: degrees) and the terrain roughness index R (range 0-10, the larger the value, the more uneven the terrain) at the current location are obtained by the on-board sensor. Set the slope coefficient Kα (a typical value is 0.5) and the roughness coefficient KR (a typical value is 0.3). The terrain feature cost CF = Kα x a2 + KR x R is calculated. In practical applications, when the slope a exceeds 20 degrees or the roughness R exceeds 8, the terrain feature cost will increase significantly, guiding the robot to avoid these areas.

[0180] The time cost term is directly related to the time required for path planning. Set the time weight coefficient Kt (a typical value is 0.4), then the time cost CC = Kt x C, where C is the estimated time to complete the path (unit: seconds).

[0181] The path smoothness cost term is used to avoid sharp turns in the path. Assuming that the heading angles of adjacent path points are θi and θi+1, the angle change Δθ = |θi+1-θi| is calculated. Set the smoothness coefficient Ks (a typical value is 0.8), then the smoothness cost CS = Ks x Σ(Δθ)2, which is summed over all adjacent point pairs on the planned path.

[0182] Combining the above four terms, the dynamic spatiotemporal constraint cost function C = CT + CF + CC + CS, the robot will find the path that minimizes the function value.

[0183] Implement a progressive hierarchical prediction time domain planning method. In the total planning time domain, a hierarchical architecture is used for path planning.

[0184] Set the total planning time domain Ttotal (e.g., 300 seconds) and the initial prediction time domain length Tp_init (e.g., 30 seconds). In actual operation, the prediction time domain length will be adjusted adaptively according to the complexity of the environment.

[0185] At each planning cycle, the following steps are performed:

[0186] The current state of the robot is obtained, including position coordinates (x, y), heading angle θ, linear velocity v, and angular velocity ω. At the same time, the surrounding environment information provided by the perception system is obtained, including terrain slope, roughness, etc.

[0187] The prediction time domain length Tp is adaptively adjusted according to the complexity of the environment. When the complexity of the environment in front is detected to increase (e.g., the terrain feature cost exceeds a preset threshold), the prediction time domain length is shortened to improve the local planning accuracy; when the environment is relatively simple, the prediction time domain length is lengthened to increase the planning efficiency. For example, Tp can be set to 50 seconds on flat terrain, and can be shortened to 20 seconds in areas with a slope greater than 15 degrees.

[0188] Within the current prediction time domain, a model predictive control method is used to generate a series of feasible control sequences. Each control sequence contains a set of linear velocity and angular velocity instructions {(v1, ω1), (v2, ω2),..., (vn, ωn)}, where n is the control step number in the prediction time domain. For each control sequence, the future trajectory is predicted according to the kinematic model of the robot, and the corresponding dynamic space-time constraint cost function value is calculated.

[0189] The control sequence that minimizes the cost function value is selected as the locally optimal control sequence. In actual applications, a cost function threshold (e.g., total cost less than 100) can be set to ensure the feasibility of the selected path.

[0190] The robot executes the first control instruction (v1, ω1) in the optimal control sequence for a duration of the control period Tc (usually 0.1 seconds).

[0191] The prediction time domain is updated by moving the prediction window forward by Tc time, and the above steps are repeated until the total planning time domain ends.

[0192] Based on the planning results, an optimal path sequence is generated, which consists of a series of path points, each containing the following parameters:

[0193] Position parameters, including x and y coordinates in the global coordinate system, with units of meters and accuracy to the centimeter level;

[0194] Heading angle parameter, representing the direction of the robot, ranging from 0 to 360 degrees with an accuracy of 0.1 degrees;

[0195] Linear velocity parameter, representing the forward speed of the robot, usually ranging from 0 to 2 meters per second, and adaptively adjusted according to terrain conditions;

[0196] Angular velocity parameter, representing the rotation speed of the robot, the range is usually -0.5-0.5 rad / s, the positive value represents right turn, and the negative value represents left turn.

[0197] Exemplarily, a certain tracked robot navigates in a mountainous environment. The starting point coordinate is (0, 0), and the ending point coordinate is (100, 50). An 18-degree slope area and a 7-index roughness gravel road need to be crossed in the middle. The optimal path planned by the method avoids the steepest slope and selects a relatively gentle area (about 12 degrees) to climb, and reduces the travel speed (from 1.2 m / s to 0.8 m / s) in the gravel road area. The total length of the planned path is 158 meters, and the expected travel time is 210 seconds. The difference in track torque is kept below 40 Nm throughout the journey, ensuring stable travel.

[0198] Experimental results show that compared with the traditional A* algorithm, the path execution time is reduced by 15%, the energy consumption is reduced by 22%, and the travel stability in the slope change area is significantly improved. The method is especially suitable for path planning problems in complex environments that need to consider the terrain characteristics and robot dynamics.

[0199] In an alternative embodiment, within the total planning time domain, a hierarchical prediction time domain architecture is constructed, the prediction time domain length is adaptively adjusted, the local optimal control sequence that minimizes the dynamic space constraint cost function is solved, the corresponding control instructions are executed and the prediction time domain is updated, and the cycle is iterated until the total planning time domain ends, and the planning result includes:

[0200] The total planning time domain is divided into a macro planning layer time domain and a micro execution layer time domain, the macro planning layer time domain includes a plurality of continuous prediction periods, and each prediction period corresponds to a micro execution layer time domain;

[0201] The environment information of the current position is collected to obtain the environment complexity, the deviation value of the actual trajectory and the planned trajectory is detected, the length of the prediction time domain is determined based on the environment complexity, and the adjustment interval of the prediction time domain is determined based on the deviation value;

[0202] The current prediction time domain is divided into two adjacent time intervals according to a preset proportion, a first time interval and a second time interval are determined, a dynamics prediction model including tracked-ground interaction force is established in the first time interval to generate a motion trajectory, a kinematics prediction model based on steering constraints is established in the second time interval to generate a path direction, and the motion trajectory and the path direction are combined to generate a predicted trajectory;

[0203] Solving a control sequence that minimizes a dynamic spatiotemporal constraint cost function based on the predicted trajectory, executing the control instruction corresponding to the first time interval, updating the prediction time domain to the starting point of the next prediction to continue iteration, until the total planning time domain ends, obtaining the planning result.

[0204] In a specific embodiment, the time domain partition structure design is performed, and the total time domain that the vehicle needs to plan is divided into two levels of macro planning layer time domain and micro execution layer time domain. The macro planning layer time domain is usually the execution time of the entire task, for example, the driving time from the starting point to the ending point is 300 seconds. The time domain is divided into multiple continuous prediction periods, and the length of each prediction period is dynamically adjusted according to the complexity of the environment, for example, the prediction period can be set to 10 seconds in a simple flat environment, and can be shortened to 5 seconds in a complex rugged environment. Each prediction period corresponds to a micro execution layer time domain, and the micro execution layer is responsible for the generation and execution of specific control instructions.

[0205] The environment information of the current position is collected by the laser radar, visual sensor and other devices carried by the vehicle, and the environment complexity index is calculated. The environment complexity can be comprehensively evaluated based on factors such as terrain undulation and obstacle density, and is represented by a normalized value of 0 to 1. For example, the environment complexity of a flat open area is 0.1, and the environment complexity of a multi-obstacle steep area can reach 0.8. At the same time, the deviation value between the actual driving trajectory and the last period planning trajectory is detected, and the weighted sum of the lateral position deviation and the heading angle deviation is calculated. For example, the lateral position deviation is 0.4 meters, the heading angle deviation is 5 degrees (about 0.087 radians), and the weights are 1 and 5 respectively, then the weighted deviation value is 0.4+5×0.087=0.835.

[0206] The length of the prediction time domain is determined based on the environment complexity, and the prediction time domain length is inversely proportional to the environment complexity, which can be obtained by multiplying the preset basic prediction time by the inverse function of the environment complexity. Assuming that the basic prediction time is 10 seconds and the environment complexity is 0.8, the actual prediction time domain length is 10×(1-0.8+0.2)=4 seconds. The adjustment interval of the prediction time domain is determined based on the deviation value, and the larger the deviation value, the more frequent the adjustment. For example, when the deviation value is less than 0.5, adjust the prediction time domain every 2 seconds; when the deviation value is between 0.5 and 1, adjust it every 1 second; when the deviation value is greater than 1, adjust it every 0.5 seconds.

[0207] The current prediction time domain is divided into two adjacent time intervals according to a preset proportion. The preset proportion is usually set to 3:7 or 4:6, that is, the first time interval is 30%-40% of the time, and the second time interval is 60%-70% of the time. For example, when the prediction time domain is 4 seconds, the first time interval is the first 1.2 seconds (30%), and the second time interval is the last 2.8 seconds (70%).

[0208] A dynamic prediction model is established in the first time interval, which considers vehicle mass, moment of inertia, friction coefficient between track and ground, terrain slope, etc. The physical quantities such as driving torque, traction force and lateral force of left and right tracks are calculated, and the acceleration, speed and position change of the vehicle are predicted according to Newton's second law to generate a high-precision motion trajectory. For example, on a soil road with a slope of 15 degrees and a ground friction coefficient of 0.4, the mass of the tracked vehicle is 500 kg, and the driving torques of the left and right tracks are 300 and 280 Nm respectively. The position change calculated by the dynamic model within 1.2 seconds is 2.3 meters forward and 0.15 meters left.

[0209] A kinematic prediction model based on steering constraints is established in the second time interval. The kinematic prediction model simplifies the dynamic characteristics of the vehicle and mainly considers the kinematic characteristics such as the steering radius constraint and the maximum speed constraint of the vehicle. Based on the current speed and angular velocity, the future position and orientation of the vehicle are predicted to generate a path trend on a longer time scale. For example, if the current speed of the vehicle is 2.5 m / s and the angular velocity is 0.1 rad / s, and the minimum steering radius is 2 m, then within the next 2.8 seconds, the vehicle will move along a circular arc with a radius of about 25 m, and the position change will be 7 m forward and 0.49 m right.

[0210] The motion trajectory generated in the first time interval and the path trend generated in the second time interval are combined to form a complete prediction trajectory. When combining, the position, speed and heading angle at the junction point of the two parts are ensured to be continuous, and smoothing processing is performed if necessary. For example, for the above calculation results, the vehicle position at 1.2 seconds is 2.3 meters forward and 0.15 meters left, the speed is 2.5 meters per second, and the heading angle is slightly right-tilted, which matches the initial conditions of the second time interval model. After combination, a complete prediction trajectory is obtained within 4 seconds, in which the vehicle will move forward 9.3 meters and finally right-tilted about 0.34 meters.

[0211] Based on the prediction trajectory, the control sequence is solved, and a dynamic space-time constraint cost function is established, which includes the weighted sum of multiple evaluation indexes such as trajectory tracking error, control energy consumption and driving smoothness. The model predictive control method is used to solve the control sequence that minimizes the cost function. For example, for a tracked vehicle, the control sequence includes the driving torque or speed command of the left and right tracks. In a certain solution, the optimal control sequence obtained is the left track torque sequence [320, 310, 305, 300] Nm and the right track torque sequence [290, 295, 300, 300] Nm, corresponding to 4 control time points.

[0212] The control instruction corresponding to the first time interval is executed. For example, for a first time interval of 1.2 seconds, the first 30% of the control sequence is executed, that is, the instructions of left track torque 320 Nm and right track torque 290 Nm. After execution, the prediction time domain is updated to the starting point of the next prediction, that is, 1.2 seconds is advanced, the environmental information is reacquired, the prediction time domain length is adjusted, the prediction model is established, the control sequence is solved, and the control instruction is executed, and the iteration cycle is repeated until the navigation task in the total planning time domain is completed.

[0213] The existing navigation control method of tracked vehicles mainly includes two types: one is a path tracking method based on a simplified kinematic model, which has small calculation amount but low precision; and the other is a trajectory planning method based on a complete dynamic model, which has high precision but large calculation complexity and is difficult to be applied in real time. The former ignores important physical characteristics such as tracked-ground interaction, resulting in insufficient control precision in complex terrain; and the latter considers detailed physical models, but has large calculation amount, is difficult to realize long-time domain planning, and has poor adaptability to environmental changes.

[0214] The method of the embodiment proposes a double-time-domain structure of a macro planning layer and a micro execution layer from the perspective of time domain division, and uses a dynamic and kinematic hybrid prediction model in the prediction time domain. The improvement point is to balance the contradiction between control precision and calculation efficiency, and by using a high-precision dynamic model in a short time interval and a high-efficiency kinematic model in a long time interval, the needs of precision and efficiency are considered. At the same time, through a double feedback mechanism of environmental complexity and trajectory deviation, the prediction time domain is adaptively adjusted, and the robustness of the method in a complex changing environment is improved.

[0215] Compared with the prior art, the advantages of the method of the embodiment mainly include: 1. an adaptive time domain adjustment mechanism based on environmental complexity and trajectory deviation is introduced, so that the system can flexibly adjust the planning strategy according to the actual situation; 2. a dynamic and kinematic hybrid prediction model is used to ensure high-precision control in a key short time domain and ensure calculation efficiency in a long time domain; and 3. through the model predictive control method, various constraints and optimization objectives are integrated into a unified space-time constraint cost function, and a globally optimal control sequence is generated.

[0216] The method of the embodiment significantly improves the calculation efficiency and environmental adaptability while ensuring the control precision. Experimental results show that compared with the traditional single model method, the trajectory tracking precision of the method in a complex terrain environment is improved by about 40%, and the calculation time is reduced by about 60%, especially in the terrain condition mutation area, the navigation performance is improved more significantly, which provides effective technical support for the autonomous navigation of tracked vehicles in complex environments.

[0217] As Figure 4As shown, the adaptive time domain adjustment effect of the present embodiment during a 300-meter navigation process of the tracked vehicle is demonstrated. The navigation path is clearly divided into five characteristic regions: flat region (0-40m), uphill rugged region (40-100m), gentle slope transition region (100-180m), multi-obstacle composite region (180-270m), and gentle endpoint region (270-300m). The data shows how the terrain complexity and trajectory deviation affect the dynamic changes of the prediction time domain length and adjustment interval: in the flat region (0-40m), the terrain complexity is maintained at a low level (0.15-0.22), the trajectory deviation is small (0.12-0.25m), the system uses a longer prediction time domain (8.24-8.64s) and a larger adjustment interval (2.0s); after entering the uphill rugged region, the terrain complexity rises sharply to 0.89, the trajectory deviation increases to 0.92m, the system adaptively shortens the prediction time domain to 3.12s and reduces the adjustment interval to 0.5s, reflecting the system's agile response capability to environmental changes; in the multi-obstacle composite region (180-270m), the maximum terrain complexity of 0.91 and the maximum trajectory deviation of 1.32m occur, at this time the prediction time domain is shortened to the minimum value of 2.92s, and the adjustment interval is kept at the minimum value of 0.5s, ensuring that the system can frequently adjust the control strategy to cope with complex environments.

[0218] The chart clearly distinguishes different terrain characteristic regions through clear color blocks, and displays the average values of key performance indicators in each region, making the relationship between terrain characteristics and control parameters clear at a glance. When the terrain complexity increases from 0.18 in the flat region to 0.82 in the multi-obstacle composite region, the prediction time domain decreases from 8.42s to 3.58s, with a decrease of 57.5%; at the same time, when the trajectory deviation increases from 0.18m to 1.06m, the adjustment interval decreases from 2.0s to 0.64s, with a decrease of 68%. Two key regions are specially marked in the chart: the region with a sharp increase in slope (complexity peak 0.89) and the region with the maximum trajectory deviation (deviation peak 1.32m), and the positions of these key points and the corresponding control parameters are intuitively displayed through the dashed line indicators.

[0219] The adaptive characteristics of the present technical solution are fully embodied in the chart: when the environmental complexity increases, the system automatically shortens the prediction time domain to improve control accuracy; when the trajectory deviation increases, the system automatically reduces the adjustment interval to increase the control frequency; in the transition region, the system can smoothly adjust the control strategy to avoid sudden changes. This dual feedback mechanism based on environmental complexity and trajectory deviation enables the system to use a long-time domain low-frequency planning strategy in simple environments to improve efficiency, and a short-time domain high-frequency planning strategy in complex environments to ensure accuracy, fully embodying the adaptive characteristics and environmental perception ability of the present technical solution, and providing strong technical support for the reliable navigation of tracked vehicles in variable and complex environments.

[0220] The embodiment of the present application is based on a multi-terrain photovoltaic cleaning tracked vehicle stability adaptive control system, which comprises:

[0221] The first unit is configured to acquire attitude data, track pressure data, cleaning arm force data and terrain scanning data of the photovoltaic cleaning tracked vehicle, and construct a terrain three-dimensional point cloud model;

[0222] The second unit is configured to perform adaptive grid division on the terrain three-dimensional point cloud model to obtain a plurality of grid units, establish height correlation features between the grid units, and determine terrain types and terrain feature parameters based on multi-feature fusion;

[0223] The third unit is configured to calculate a tracked vehicle stability index according to the attitude data, the track pressure data and the cleaning arm force data;

[0224] The fourth unit is configured to execute an adaptive entropy enhancement cooperative tracking algorithm to solve tracked vehicle torque distribution parameters and cleaning arm pose parameters according to the terrain types, the terrain feature parameters and the tracked vehicle stability index;

[0225] The fifth unit is configured to construct a dynamic space-time constraint cost function based on the tracked vehicle torque distribution parameters and the terrain feature parameters, and obtain an optimal path sequence through progressive rolling horizon planning;

[0226] The sixth unit is configured to generate tracked vehicle travel instructions and cleaning arm attitude instructions according to the optimal path sequence and the cleaning arm pose parameters in combination with a predictive control strategy, so as to realize stable operation of the photovoltaic cleaning tracked vehicle.

[0227] The third aspect of the embodiment of the present application,

[0228] An electronic device is provided, comprising:

[0229] a processor;

[0230] a memory for storing processor-executable instructions;

[0231] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0232] The fourth aspect of the embodiment of the present application,

[0233] A computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0234] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium on which is loaded computer readable program instructions for executing various aspects of the present application.

[0235] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A stability adaptive control method for a multi-terrain photovoltaic cleaning tracked vehicle, characterized in that, include: Acquire attitude data, track pressure data, cleaning arm force data, and terrain scan data of the photovoltaic cleaning tracked vehicle to construct a 3D point cloud model of the terrain; Adaptive meshing is performed on the 3D point cloud model of the terrain to obtain multiple mesh cells. The height correlation features between mesh cells are established, and the terrain type and terrain feature parameters are determined based on multi-feature fusion. The stability index of the tracked vehicle is calculated based on attitude data, track pressure data, and cleaning arm force data, including: Receive attitude data, including pitch angle, roll angle and yaw angle; receive track pressure data, including left track pressure distribution data and right track pressure distribution data; receive cleaning arm force data, including torque data and joint axial force data of each joint of the cleaning arm; Based on the attitude data, the coordinates of the boundary points of the tracked vehicle support polygon are determined, the distances from the tracked vehicle's center of gravity projection point to each side of the tracked vehicle support polygon are calculated, and the ratio of the minimum value of the distance to the feature size of the tracked vehicle support polygon is determined as the static stability margin. The zero-moment point coordinates of the tracked vehicle are calculated based on the track pressure data and the cleaning arm force data. The complement of the ratio of the distance between the zero-moment point coordinates and the center coordinates of the tracked vehicle support polygon to the preset safety radius is determined as the dynamic stability margin. The weighted sum of static stability margin and dynamic stability margin is determined as the tracked vehicle stability index; Based on terrain type, terrain feature parameters, and tracked vehicle stability index, an adaptive entropy-enhanced cooperative tracking algorithm is executed to solve for track torque distribution parameters and cleaning arm pose parameters, including: Based on the preset range of track torque distribution parameters and the preset range of cleaning arm pose parameters, a parameter search space is determined and divided into multiple equal probability subspaces. In each equal probability subspace, a collaborative tracking agent is generated based on uniform distribution random sampling. All the collaborative tracking agents generated in the equal probability subspaces constitute a collaborative tracking agent group. For each collaborative tracking agent in the collaborative tracking agent group, a local entropy value is calculated based on the neighborhood probability density, and the mean of the local entropy values ​​is determined as the overall entropy value of the group. Based on terrain type, terrain feature parameters and tracked vehicle stability index, a fitness function is constructed by weighted combination of terrain evaluation values ​​and stability evaluation values. The adaptive step size adjustment coefficient is calculated based on the overall entropy value of the population. The first agent is selected according to the fitness function. The parameters of the agents in the collaborative tracking agent group are updated in combination with the adaptive step size adjustment coefficient. Calculate the parameter dispersion of the collaborative tracking agent group, determine the group diversity index, and perform Gaussian random mutation when the group diversity index is less than a preset diversity threshold; When the change in the optimal parameters during continuous iterations is less than the preset convergence threshold, the current optimal track torque distribution parameters and cleaning arm pose parameters are output. The weight coefficients of the fitness function are dynamically adjusted based on the real-time updated terrain type and terrain feature parameters. A dynamic spatiotemporal constraint cost function is constructed based on track torque distribution parameters and terrain feature parameters. The optimal path sequence is obtained through progressive rolling time-domain programming, including: Based on track torque distribution parameters, the torque distribution cost term is calculated using left track torque parameters and right track torque parameters. Based on terrain feature parameters, the terrain feature cost term is calculated using terrain slope parameters and terrain roughness parameters. The dynamic spatiotemporal constraint cost function is constructed by combining time cost term and path smoothing cost term. Within the overall planning time domain, a hierarchical prediction time domain architecture is constructed, the length of the prediction time domain is adaptively adjusted, the local optimal control sequence that minimizes the dynamic spatiotemporal constraint cost function is solved, the corresponding control instructions are executed and the prediction time domain is updated by sliding, and the cycle is iterated until the end of the overall planning time domain to obtain the planning result. The optimal path sequence is generated based on the planning results, consisting of position parameters, heading angle parameters, linear velocity parameters, and angular velocity parameters. Based on the optimal path sequence and cleaning arm pose parameters, and combined with predictive control strategies, the tracked vehicle travel commands and cleaning arm posture commands are generated to achieve stable operation of the photovoltaic cleaning tracked vehicle.

2. The method according to claim 1, characterized in that, Adaptive meshing is performed on the 3D point cloud model of the terrain to obtain multiple mesh cells. Height correlation features between mesh cells are established. Based on multi-feature fusion, the terrain type and terrain feature parameters are determined, including: The terrain 3D point cloud model is projected onto a horizontal plane, and the initial side length is calculated based on the total area and number of point clouds, and then divided into multiple initial grid units. Calculate the point cloud density distribution of the initial grid cells, construct the density flow field based on the density gradient and calculate the deformation vector to generate the deformable grid. When the strain energy density of the deformable grid exceeds the preset density threshold, it splits along the main direction of the density flow field to generate the target grid cell. Calculate the height range within the target grid cell to determine the height difference value, calculate the height change rate in the orthogonal direction to determine the height gradient; establish the height correlation characteristics between target grid cells, calculate the height difference coefficient between adjacent cells, and construct a spatial correlation matrix; The elevation difference value, elevation gradient and spatial correlation matrix are combined to form a feature vector, which is then input into the multi-feature fusion terrain classification discriminant function to determine the terrain type of the target grid cell based on the conditional probability maximization criterion. The terrain slope of the target grid cell is calculated based on the height gradient, and the terrain roughness is determined by calculating the integral deviation between the actual height and the least squares fitted plane. The terrain type, terrain slope, and terrain roughness are combined to form a terrain feature parameter vector, which determines the terrain feature parameters of the target grid cell.

3. The method according to claim 2, characterized in that, The point cloud density distribution of the initial mesh elements is calculated. A density flow field is constructed based on the density gradient, and a deformation mesh is generated by calculating the deformation vector. When the strain energy density of the deformation mesh exceeds a preset density threshold, it splits along the main direction of the density flow field to generate the target mesh elements, including: The number of point clouds is counted for the initial grid cells, and the point cloud density distribution is obtained by calculating the number of point clouds per unit area. The density change rate in the X direction and the density change rate in the Y direction are calculated based on the point cloud density of adjacent initial grid cells to form a density gradient vector. The density gradient vector is orthogonally rotated to obtain an orthogonal rotation vector. The density gradient vector and the orthogonal rotation vector are linearly combined to construct the density flow field. The mesh deformation vector is calculated based on the direction of the density flow field and the magnitude of the density gradient vector. The mesh deformation vector is applied to the vertices of the initial mesh element to generate a deformable mesh element. The strain energy density is obtained by calculating the deformation work per unit area within the deformable mesh element. When the strain energy density exceeds a preset density threshold, the deformable mesh element is split along the main direction of the density flow field, the difference between the strain energy density and the preset density threshold is calculated, and the difference is substituted into an exponential function to determine the size of the split sub-mesh. Calculate the mesh side length ratio, mesh area change rate, and deformation vector difference between adjacent mesh vertices of the split sub-mesh. When the mesh side length ratio does not exceed a preset distortion threshold, the mesh area change rate does not exceed a preset area change threshold, and the deformation vector difference does not exceed the product of a preset smoothing coefficient and the distance between mesh vertices, the target mesh cell is determined.

4. The method according to claim 1, characterized in that, Within the overall planning time domain, a hierarchical prediction time domain architecture is constructed, the prediction time domain length is adaptively adjusted, and the locally optimal control sequence that minimizes the dynamic spatiotemporal constraint cost function is solved. Corresponding control commands are executed, and the prediction time domain is updated in a sliding manner. This process is iterated until the end of the overall planning time domain, yielding planning results including: The overall planning time domain is divided into a macro-planning time domain and a micro-execution time domain. The macro-planning time domain contains multiple consecutive prediction periods, and each prediction period corresponds to a micro-execution time domain. Collect environmental information at the current location to obtain the environmental complexity, detect the deviation between the actual trajectory and the planned trajectory, determine the length of the prediction time domain based on the environmental complexity, and determine the adjustment interval of the prediction time domain based on the deviation value; The current prediction time domain is divided into two adjacent time intervals according to a preset ratio. The first time interval and the second time interval are determined. A dynamic prediction model containing track-ground interaction force is established in the first time interval to generate the motion trajectory. A kinematic prediction model based on steering constraints is established in the second time interval to generate the path direction. The motion trajectory and the path direction are combined to generate the prediction trajectory. Based on the predicted trajectory, the control sequence that minimizes the dynamic spatiotemporal constraint cost function is obtained. The control command corresponding to the first time interval is executed, the predicted time domain is updated to the starting point of the next prediction, and the iterative solution continues until the overall planning time domain ends, and the planning result is obtained.

5. A stability adaptive control system for a multi-terrain photovoltaic cleaning tracked vehicle, used to implement the method described in any one of claims 1-4, characterized in that, include: The first unit is used to acquire the attitude data, track pressure data, cleaning arm force data and terrain scan data of the photovoltaic cleaning tracked vehicle, and to construct a three-dimensional point cloud model of the terrain. The second unit is used to adaptively mesh the 3D point cloud model of the terrain to obtain multiple mesh units, establish the height correlation features between mesh units, and determine the terrain type and terrain feature parameters based on multi-feature fusion. The third unit is used to calculate the stability index of the tracked vehicle based on attitude data, track pressure data, and cleaning arm force data. The fourth unit is used to solve the track torque distribution parameters and cleaning arm pose parameters by executing an adaptive entropy-enhanced cooperative tracking algorithm based on terrain type, terrain feature parameters and tracked vehicle stability index. The fifth unit is used to construct a dynamic spatiotemporal constraint cost function based on track torque distribution parameters and terrain feature parameters, and obtain the optimal path sequence through progressive rolling time-domain planning; The sixth unit is used to generate tracked vehicle travel commands and cleaning arm posture commands based on the optimal path sequence and cleaning arm pose parameters, combined with predictive control strategies, to achieve stable operation of the photovoltaic cleaning tracked vehicle.

6. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 4.

Images