Method for rapid regulation of earthwork balance in steep and narrow area of new energy mine

Abstract

The application discloses a steep and narrow area of new energy mine card earthwork excavation and filling balance rapid regulation method, belongs to the field of mine exploitation and earthwork engineering machinery automatic control technology. The method is aimed at the technical problems of low efficiency and poor real-time of excavation and filling balance regulation caused by the dependence on artificial judgment and periodic measurement in the prior art. The technical scheme points are: obtaining real-time point cloud data through a three-dimensional laser scanner; the vehicle-mounted processing unit compares the real-time point cloud data with the pre-stored designed terrain model within 0.3 s to 1.5 s to generate a real-time excavation and filling distribution map; based on this, the excavation and filling balance ratio is calculated, when the ratio exceeds the range of 0.95-1.05, the adjustment path pointing to the area with the largest difference between excavation and filling is planned, and the vehicle is controlled to execute at a speed of 5 km / h to 15 km / h. The method is mainly used for new energy mine card construction operation in steep and narrow mining areas, and realizes the rapid automatic regulation and control of earthwork excavation and filling balance.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology for mining and earthmoving machinery. More specifically, this invention relates to a rapid control method for earthwork excavation and filling balance in new energy mining operations in steep and narrow terrain. Background Technology

[0002] In open-pit mining and large-scale earthwork projects, the balance between excavation and filling is a core operational aspect. Achieving a dynamic balance between excavation and filling volumes is crucial for ensuring operational continuity, improving transportation efficiency, and reducing operating costs. Traditional excavation-filling balance operation methods rely primarily on manual experience and periodic measurements. Specifically, dispatchers direct mining trucks to load, transport, and unload earth and rock based on periodic survey results and experience. While this method can maintain basic operation in open, flat conventional operating areas, it exposes numerous technical bottlenecks when facing steep, narrow terrain.

[0003] First, in steep and narrow areas, the terrain changes rapidly and visibility is limited, making it difficult to achieve real-time dynamic balance of earthwork using manual judgment and periodic measurements. The fundamental reason is that manual judgment has a significant delay, failing to quickly perceive subtle changes in the terrain, while traditional surveying cycles can take hours or even days, resulting in severely delayed information on the cut and fill status. When an imbalance occurs, it cannot be detected immediately; adjustments are only made after the problem has accumulated to a certain extent, often requiring more time and resources for correction, severely impacting operational efficiency. Attempts have been made to improve this by increasing the frequency of manual inspections or increasing the density of measurement points, but these methods face difficulties such as high labor costs, insufficient data update speed to meet real-time control requirements, and the inability to guarantee personnel safety in dangerous and steep areas.

[0004] Secondly, to achieve automated control, a high-precision terrain model of the work area needs to be established as a benchmark. Current technologies often employ a single platform (such as only UAVs or only ground equipment) for terrain mapping. However, in steep and narrow areas, single-platform mapping has inherent limitations: UAV aerial surveys are easily obstructed by cliffs, creating blind spots and resulting in incomplete models; while relying solely on ground scanning is inefficient and makes it difficult to obtain a macroscopic overall terrain. Fusion of multi-source point cloud data from different platforms introduces new technical challenges. Point cloud data from different sources differ in resolution, coordinate systems, and perspectives, and direct fusion leads to registration errors. These errors are particularly significant in complex terrains and will propagate and accumulate in subsequent data processing chains, ultimately reducing the accuracy of the benchmark terrain model itself. This results in all automatic judgments and decisions based on this model being built on an unreliable foundation. Solving the problem of seamless fusion and accuracy control of multi-source data is a key challenge in improving system reliability.

[0005] Furthermore, even with an initial high-precision model, the terrain is constantly changing in a dynamic earthmoving environment. A static model quickly becomes "ineffective" as operations progress, failing to accurately reflect the current cut and fill status. Therefore, dynamically updating the designed terrain model is essential for maintaining system effectiveness. However, the dynamic updating of the model itself also presents stability challenges. Environmental noise at the work site, sensor measurement errors, and temporary disturbances generated by vehicle operations can all be misinterpreted by the model as valid terrain changes and updated accordingly. Such erroneous updates lead to model distortion, which in turn causes oscillations in system decisions—for example, vehicles may be repeatedly assigned to "false" work areas caused by noise. How to design an update mechanism that can both track real terrain evolution and effectively filter out interference while maintaining the long-term stability of the model itself is a serious challenge faced by automated control systems in practical applications. Summary of the Invention

[0006] One objective of this invention is to provide a rapid control method for earthwork excavation and filling balance in new energy mines in steep and narrow areas, which enables second-level perception and response to the terrain conditions in steep and narrow areas, significantly improving the timeliness and automation level of excavation and filling balance operations.

[0007] To achieve these objectives and other advantages of the present invention, according to one aspect of the present invention, a method for preparing a rapid control method for earthwork excavation and filling balance in new energy mines in steep and narrow areas is provided, comprising the following steps:

[0008] Step 1: Using the 3D laser scanner installed on the new energy mining truck, continuously scan the terrain of the ring-shaped working area with a radius of 50 m to 100 m centered on the new energy mining truck at a scanning frequency of 20 Hz to 40 Hz to obtain real-time point cloud data.

[0009] Step 2: Transmit the real-time point cloud data to the on-board processing unit of the new energy mining truck through the vehicle communication system. The on-board processing unit completes the processing of the real-time point cloud data within 0.3 s to 1.5 s, compares the real-time point cloud data with the pre-stored design terrain model, and generates a real-time excavation and filling distribution map.

[0010] Step 3: Based on the real-time excavation and filling distribution map, the on-board processing unit calculates the current excavation volume and filling volume, and calculates the ratio of excavation volume to filling volume as the excavation-filling balance ratio. When the excavation-filling balance ratio is lower than 0.95 or higher than 1.05, the on-board processing unit determines the adjustment path and adjustment speed of the new energy mining truck based on the real-time point cloud data. The adjustment path points to the area with the greatest difference between the excavation volume and the filling volume, and the adjustment speed is controlled between 5 km / h and 15 km / h.

[0011] Step 4: The on-board processing unit sends the adjustment path and speed to the control system of the new energy mining truck. The control system drives the electric drive mechanism and steering mechanism of the new energy mining truck according to the adjustment path and speed to execute the adjustment path and speed.

[0012] Preferably, the pre-stored design terrain model in step two is established through the following steps;

[0013] The UAV-borne LiDAR was used to scan the entire work area, while a ground-based 3D laser scanner was used to supplement the scan of steep cliffs and narrow ravines to obtain high-precision initial point cloud data.

[0014] The point cloud data acquired by the UAV-borne lidar and the ground-based 3D laser scanner are fused and registered to generate an initial digital elevation model of the work area with a resolution of 2 cm to 5 cm.

[0015] The digital design terrain model provided by the engineering design party is overlaid and compared with the initial digital elevation model. Areas with elevation deviations exceeding 10 cm to 15 cm are manually verified and corrected to generate a pre-stored design terrain model as a benchmark.

[0016] Preferably, the point cloud data fusion and registration includes a precision control step;

[0017] During the fusion registration process, the registration error between the UAV-borne LiDAR and the ground-based 3D laser scanner point cloud data is calculated. When the registration error is greater than 3 cm to 5 cm, the iterative nearest point algorithm is used for re-registration. The re-registered point cloud data is subjected to an overlap area consistency check, and point cloud data with an elevation difference of more than 2 cm to 4 cm in the overlap area are removed. In the final fused point cloud data, the registration error of 90% to 95% of the points is no greater than 1.5 times the resolution of the initial digital elevation model.

[0018] Preferably, the vehicle-mounted processing unit is equipped with a model update module, which dynamically updates the pre-stored design terrain model based on real-time point cloud data; wherein the dynamic update follows the following steps;

[0019] After completing one path adjustment, the on-board processing unit performs a second differential calculation on the real-time point cloud data collected and registered during this round of operation and the design terrain model before dynamic update to generate an incremental terrain dataset.

[0020] Statistical analysis was performed on the terrain increment dataset to calculate its average height change and the continuous area of ​​the changed region; only when the absolute value of the average height change is greater than 5 cm to 10 cm and the continuous area of ​​the changed region is greater than 2 m... 2 up to 5 m 2Only when this condition is met is the incremental data deemed valid.

[0021] The incremental terrain dataset that has passed the validity verification is superimposed onto the pre-stored design terrain model in a weighted fusion manner to complete the iterative update of the model;

[0022] The model update module also includes an update stability control mechanism. This mechanism records the average change height of the current terrain increment dataset during each iteration update and compares it with the change height recorded in the previous two updates. When the change height sequence signs alternate between positive and negative for three consecutive updates, and the absolute values ​​of the two adjacent changes are both greater than 8 cm to 12 cm, it is determined to be an oscillation and update amplitude limitation is initiated. The update amplitude limitation is implemented by introducing a weighting factor that starts from 0.1 and increases in increments of 0.1 up to 0.5. This weighting factor is used to attenuate the fusion weight of the current terrain increment dataset.

[0023] Preferably, the generation of the real-time cut-and-fill distribution map in step two includes the following steps;

[0024] The onboard processing unit converts real-time point cloud data into a grid digital elevation model with a resolution of 10 cm to 20 cm;

[0025] The elevation difference matrix is ​​generated by performing grid-by-grid cell-by-grid cell-by-cell elevation difference calculations between the grid digital elevation model and the pre-stored design terrain model in the same coordinate system.

[0026] The elevation difference matrix is ​​processed by a region growing algorithm based on neighborhood connectivity, and adjacent grid cells with elevation changes greater than 15 cm to 30 cm are clustered into independent change regions.

[0027] Calculate the average elevation change value for each independent change area, and mark the area with a negative average elevation change value as the cut area and the area with a positive average elevation change value as the fill area.

[0028] By combining the spatial location information of all excavation and filling areas, a real-time excavation and filling distribution map is generated.

[0029] Preferably, the calculation of the current excavation volume and fill volume in step three is achieved through the following steps;

[0030] The vehicle-mounted processing unit extracts the boundary contours of all excavation and filling areas based on the real-time excavation and filling distribution map.

[0031] Each region is processed using triangulation to generate an irregular triangular mesh model;

[0032] Based on the spatial relationship between the triangular mesh model and the pre-stored design terrain model, the prism volume integration method is used to calculate the cut volume of each cut area and the fill volume of each fill area respectively.

[0033] During the calculation process, a real-time soil density calibration mechanism based on the vehicle-mounted weighing system is introduced. This mechanism is implemented in the following way: During the loading of the new energy mining truck, the vehicle-mounted processing unit synchronously acquires the actual load mass data measured by the vehicle-mounted weighing system, and dynamically updates the soil density value in combination with the calculated volume of the corresponding excavation area.

[0034] The updated earthwork density value is applied to the subsequent volume calculation process to achieve dynamic calibration of cut and fill volume calculation.

[0035] Preferably, the real-time soil density calibration mechanism further includes an error compensation step; the vehicle-mounted processing unit establishes an error propagation model for soil density, which considers a measurement error of 1% to 2% from the vehicle-mounted weighing system and a volume calculation error of 2% to 3% from the point cloud data; when dynamically updating the soil density value, a Kalman filter algorithm is used to filter the density measurement value; the filtered density value is combined with the error propagation model to compensate and correct the calculation results of excavation volume and fill volume; after compensation and correction, the estimated range of volume calculation error is between 3% and 5%.

[0036] Preferably, the determination of the adjustment path in step three is achieved through the following steps;

[0037] The vehicle-mounted processing unit identifies the target area with the greatest difference between the excavation volume and the filling volume based on the real-time excavation and filling distribution map, and uses the center coordinates of the target area as the path endpoint.

[0038] A path planning algorithm is used to search for paths in an obstacle map generated from real-time point cloud data. The cost function of this path planning algorithm includes two calculation terms: path length cost and terrain slope cost.

[0039] The initial path output by the path planning algorithm undergoes a safety check based on vehicle dynamics constraints. This check excludes path segments that exceed the maximum climbing angle of the new energy mining truck by 30% to 35%, as well as path segments that are less than the minimum turning radius of the new energy mining truck by 8 to 10 meters.

[0040] The path that has passed the security check is processed by a curve fitting algorithm to generate an executable path with continuous curvature.

[0041] During the execution of the executable path by the new energy mining truck, the 3D laser scanner performs real-time obstacle detection within a range of 20 m to 30 m in the direction of travel at a frequency of 5 Hz to 10 Hz.

[0042] When an obstacle is detected, the onboard processing unit initiates local path replanning, generates an obstacle avoidance path, and continues execution.

[0043] Preferably, the path planning algorithm employs a hierarchical planning strategy; the hierarchical planning strategy includes a global planning layer and a local planning layer; the global planning layer uses a path planning algorithm to perform coarse-grained path planning with a resolution of 1 m to 2 m, and a planning cycle of 10 s to 15 s; the local planning layer uses the same path planning algorithm to perform fine-grained path planning with a resolution of 0.2 m to 0.5 m, and a planning cycle of 1 s to 2 s; the global planning layer and the local planning layer are coupled through a path point sequence, and the local planning layer performs real-time fine planning among global path points; the on-board processing unit allocates dynamically adjustable computing resources to the global planning layer and the local planning layer; when the local planning layer triggers replanning more than 3 times within 1 second, the system automatically allocates more computing resources to the local planning layer.

[0044] Preferably, it also includes a system stability monitoring and control mode switching step;

[0045] The onboard processing unit monitors the following key system indicators in real time: the fluctuation frequency of the cut-fill balance ratio, the data integrity rate of real-time point cloud data, and the trigger frequency of local path replanning.

[0046] The system is considered to be in a state of instability risk when the following conditions are met simultaneously: the frequency of fluctuation of the cut-fill balance ratio exceeds 10 times per minute within 30 consecutive seconds, the data integrity rate of real-time point cloud data is less than 85%, and the frequency of local path replanning triggering exceeds 5 times per minute.

[0047] When the system is determined to be in a state of instability risk, the on-board processing unit automatically switches from fully automatic control mode to auxiliary control mode;

[0048] In the auxiliary control mode, the on-board processing unit simplifies the adjustment path to a straight line from the current location to the nearest cut-and-fill area, and limits the adjustment speed to 5 km / h to 8 km / h;

[0049] The system continuously monitors key indicators, and automatically switches back to fully automatic control mode after the indicators return to normal and remain stable for 10 seconds.

[0050] This invention offers at least the following advantages: The rapid control method for earthwork and rockfill balance in steep and narrow areas of new energy mining, as described in this invention, achieves second-level perception and response to terrain conditions in such areas by constructing a closed-loop control circuit of scanning-processing-decision-execution, significantly improving the timeliness and automation level of earthwork and rockfill balance operations. By integrating UAV and ground scanning data and implementing strict registration accuracy control, a high-precision and complete initial terrain benchmark is established, ensuring the reliability of subsequent judgments from the source. A model dynamic update mechanism based on saliency criteria and stability control is introduced, enabling the benchmark model to adapt to the continuous evolution of the operation process while effectively suppressing model distortion and decision oscillations caused by data noise. Mesh processing and region growing algorithms are used to achieve accurate identification and automatic classification of earthwork and rockfill areas. The volume calculation method, combined with real-time calibration and error compensation of vehicle-mounted weighing, significantly improves the accuracy of earthwork volume calculation. Layered path planning with integrated vehicle dynamics constraints and real-time obstacle detection ensures the executability and safety of the planned path. Finally, the system-level multi-indicator monitoring and degradation strategy provides the ultimate guarantee for stable operation under complex working conditions, forming a complete technical closed loop from bottom-level data to top-level decision-making, and jointly realizing rapid, accurate and robust control of earthwork excavation and filling balance under harsh terrain conditions.

[0051] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Detailed Implementation

[0052] The present invention will now be described in further detail with reference to specific embodiments, so that those skilled in the art can implement it based on the description.

[0053] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0054] It should be noted that, unless otherwise specified, the experimental methods described in the following implementation plan are all conventional methods, and the reagents and materials described are all commercially available unless otherwise specified.

[0055] This invention provides a method for rapid adjustment of earthwork cut-and-fill balance in steep and narrow new energy mining areas, comprising the following steps:

[0056] Step 1: Using the 3D laser scanner installed on the new energy mining truck, continuously scan the terrain of the ring-shaped working area with a radius of 50 m to 100 m centered on the new energy mining truck at a scanning frequency of 20 Hz to 40 Hz to obtain real-time point cloud data.

[0057] Step 2: Transmit the real-time point cloud data to the on-board processing unit of the new energy mining truck through the vehicle communication system. The on-board processing unit completes the processing of the real-time point cloud data within 0.3 s to 1.5 s, compares the real-time point cloud data with the pre-stored design terrain model, and generates a real-time excavation and filling distribution map.

[0058] Step 3: Based on the real-time excavation and filling distribution map, the on-board processing unit calculates the current excavation volume and filling volume, and calculates the ratio of excavation volume to filling volume as the excavation-filling balance ratio. When the excavation-filling balance ratio is lower than 0.95 or higher than 1.05, the on-board processing unit determines the adjustment path and adjustment speed of the new energy mining truck based on the real-time point cloud data. The adjustment path points to the area with the greatest difference between the excavation volume and the filling volume, and the adjustment speed is controlled between 5 km / h and 15 km / h.

[0059] Step 4: The on-board processing unit sends the adjustment path and speed to the control system of the new energy mining truck. The control system drives the electric drive mechanism and steering mechanism of the new energy mining truck according to the adjustment path and speed to execute the adjustment path and speed.

[0060] In the above technical solution, the 3D laser scanner can be a 32-line or 64-line mechanical rotating lidar. This scanner is rigidly mounted on the protective frame on top of the mining truck cab, with its scanning center axis tilted downwards at an angle of 5 to 10 degrees to optimize scanning coverage of the nearby ground. The scanner operates at a fixed frequency of 20 Hz, emitting laser pulses and receiving return signals to generate point cloud data containing 3D coordinates and reflection intensity. The point cloud data is transmitted to the onboard processing unit via shielded twisted-pair cables, generating approximately 300,000 valid points per second, sufficient to describe terrain features within an 80-meter radius.

[0061] The onboard processing unit can be an industrial-grade onboard computer equipped with a multi-core CPU and a dedicated GPU computing module. This processing unit is installed in a shockproof cabinet inside the mining truck's cab and connects to the scanner and control system via an onboard Ethernet connection. The real-time point cloud data processing procedure includes: first, outlier filtering and voxel grid downsampling are applied to the input point cloud, reducing the data volume to 60% of the original data; then, an iterative nearest-point algorithm is used to register the processed real-time point cloud with a pre-stored design terrain model, controlling the registration error to within 3 centimeters; finally, a real-time cut-and-fill distribution map is generated through elevation difference calculation. The entire processing takes approximately 0.8 seconds.

[0062] The cut-fill balance calculation and route planning module is integrated into the onboard processing unit and implemented through specialized algorithm software. This module first extracts all elevation change areas from the real-time cut-fill distribution map and clusters adjacent change areas using a region growing algorithm. Then, it calculates the volume of each area based on a triangular mesh model, summing the total cut and fill volumes. When the cut-fill balance ratio reaches 0.97, the route planning program is initiated. Route planning uses the A* algorithm, comprehensively considering path length and terrain slope factors to generate a safe path from the current location to the target area, while setting the driving speed to 8 kilometers per hour. The planned path and speed parameters are sent to the vehicle control system via the CAN bus.

[0063] Through the above implementation methods, a complete automated control chain from terrain perception to vehicle execution is established, enabling continuous monitoring and rapid response to the terrain conditions of the work area. The use of industrial-grade hardware ensures the system's reliability under harsh working conditions, while the optimized algorithm process guarantees the timeliness of data processing and decision-making. This method can effectively replace the traditional work mode that relies on manual judgment, achieving automated control of earthwork excavation and filling balance in steep and narrow terrain conditions, thus improving operational safety and efficiency.

[0064] In other technical solutions, the pre-stored design terrain model in step two is established through the following steps;

[0065] The UAV-borne LiDAR was used to scan the entire work area, while a ground-based 3D laser scanner was used to supplement the scan of steep cliffs and narrow ravines to obtain high-precision initial point cloud data.

[0066] The point cloud data acquired by the UAV-borne lidar and the ground-based 3D laser scanner are fused and registered to generate an initial digital elevation model of the work area with a resolution of 2 cm to 5 cm.

[0067] The digital design terrain model provided by the engineering design party is overlaid and compared with the initial digital elevation model. Areas with elevation deviations exceeding 10 cm to 15 cm are manually verified and corrected to generate a pre-stored design terrain model as a benchmark.

[0068] In the above technical solution, during the data acquisition phase, the UAV-borne LiDAR can be a lightweight laser scanning system mounted on a multi-rotor UAV platform. The ground-based 3D laser scanner can be a pulsed scanning measurement system. The UAV flies along a predetermined route over the work area, maintaining a flight altitude of 80 to 100 meters, with a scanning field of view of 60 degrees and a point cloud density of 200 points per square meter. The ground scanner is mounted on a stable tripod and placed in areas where the UAV cannot scan, such as the bottom of ravines and the edge of cliffs, with a scanning distance set to 50 to 150 meters, performing multi-station supplementary scanning of key terrain features. The point cloud data acquired by both devices is stored in standard LAS format.

[0069] In the data processing stage, point cloud fusion and registration can be performed using professional point cloud processing software running on a workstation equipped with a high-performance graphics processor. The software first performs coordinate system unification on the point cloud data acquired by the UAV and ground scanning, and then performs initial registration through feature point matching. Subsequently, an iterative nearest-point algorithm is used for fine registration, with 100 iterations and a registration error threshold of 5 cm. The registered point cloud is then gridded to generate a digital elevation model with a resolution of 3 cm. During processing, consistency checks are performed on the point clouds in overlapping areas, and outliers with elevation differences greater than 4 cm are removed.

[0070] During the verification and correction phase, the digital terrain model provided by the engineering design team can be a 3D surface model in CAD format. The design model and the measured digital elevation model are imported into the same coordinate system and overlaid for comparison. Spatial analysis functions are used to detect elevation differences between the two. When an elevation deviation exceeding 12 centimeters is detected, a surveying engineer conducts a manual assessment based on the actual site conditions. During the correction process, the measured terrain data is prioritized, and areas in the design model that clearly do not conform to the actual terrain features are appropriately adjusted. Finally, a design terrain model that closely matches the actual terrain is generated as the benchmark.

[0071] By combining UAVs and ground scanning equipment, the problem of blind spots in complex terrain caused by single scanning methods was effectively overcome, ensuring the integrity and accuracy of the initial terrain data. A professional point cloud processing workflow was employed to achieve precise fusion of multi-source data, establishing a high-resolution digital terrain model. Through manual verification and correction, significant deviations between the design and the actual situation were further eliminated, providing a reliable benchmark for subsequent real-time cut-fill balance control, thus ensuring the operational accuracy of the entire system from the data source.

[0072] In other technical solutions, the fusion and registration of the point cloud data includes a precision control step;

[0073] During the fusion registration process, the registration error between the UAV-borne LiDAR and the ground-based 3D laser scanner point cloud data is calculated. When the registration error is greater than 3 cm to 5 cm, the iterative nearest point algorithm is used for re-registration. The re-registered point cloud data is subjected to an overlap area consistency check, and point cloud data with an elevation difference of more than 2 cm to 4 cm in the overlap area are removed. In the final fused point cloud data, the registration error of 90% to 95% of the points is no greater than 1.5 times the resolution of the initial digital elevation model.

[0074] In the above technical solution, during the registration error calculation and re-registration stages, professional 3D data processing software with point cloud registration capabilities can be selected. This software runs on a workstation equipped with a sixteen-core processor and sixty-four GB of memory. The software first calculates the initial registration error between the UAV and the ground-scanned point cloud. When the registration error reaches 4 cm, it automatically triggers an iterative nearest-point algorithm for re-registration. During the re-registration process, the maximum number of iterations is set to two hundred, and the convergence threshold is 0.001 meters. The matching error between the two point clouds is minimized by continuously optimizing the rigid body transformation matrix.

[0075] In the consistency check of overlapping areas, a point cloud difference analysis algorithm can be used to process the re-registered data. This algorithm first establishes an index of the overlapping areas of the two point clouds, and then calculates the three-dimensional Euclidean distance between corresponding point pairs. For point pairs with an elevation difference exceeding 3 cm, the system automatically marks them as inconsistencies. The inconsistency removal process employs a two-way verification mechanism, simultaneously comparing differences from both the UAV point cloud and the ground point cloud to ensure the uniformity and fairness of the removal criteria.

[0076] In the final accuracy control stage, a point cloud quality assessment module can be used to perform statistical analysis on the fused overall data. This module randomly selects one thousand uniformly distributed sample points from the fused point cloud and calculates their positional deviation from the reference point cloud. When 92% of the sample points are found to have a registration error within 2 cm, the fusion result is considered to meet the requirements. For areas that do not meet the standards, the system generates an accuracy distribution map to guide operators in performing local fine-tuning.

[0077] By establishing a rigorous registration accuracy control process, the problem of error accumulation during the fusion of multi-source point cloud data was effectively solved. An automated re-registration mechanism ensured the accuracy of point cloud matching, while consistency checks of overlapping areas eliminated systematic deviations between different data sources. Finally, statistical quality control methods guaranteed the overall reliability of the fused data, providing an accurate data foundation for subsequent digital elevation model generation, thereby improving the benchmark data quality of the entire earthwork cut-and-fill balance control system.

[0078] In other technical solutions, the vehicle-mounted processing unit is equipped with a model update module, which dynamically updates the pre-stored design terrain model based on real-time point cloud data; wherein, the dynamic update follows the following steps;

[0079] After completing one path adjustment, the on-board processing unit performs a second differential calculation on the real-time point cloud data collected and registered during this round of operation and the design terrain model before dynamic update to generate an incremental terrain dataset.

[0080] Statistical analysis was performed on the terrain increment dataset to calculate its average height change and the continuous area of ​​the changed region; only when the absolute value of the average height change is greater than 5 cm to 10 cm and the continuous area of ​​the changed region is greater than 2 m... 2 up to 5 m 2 Only when this condition is met is the incremental data deemed valid.

[0081] The incremental terrain dataset that has passed the validity verification is superimposed onto the pre-stored design terrain model in a weighted fusion manner to complete the iterative update of the model;

[0082] The model update module also includes an update stability control mechanism. This mechanism records the average change height of the current terrain increment dataset during each iteration update and compares it with the change height recorded in the previous two updates. When the change height sequence signs alternate between positive and negative for three consecutive updates, and the absolute values ​​of the two adjacent changes are both greater than 8 cm to 12 cm, it is determined to be an oscillation and update amplitude limitation is initiated. The update amplitude limitation is implemented by introducing a weighting factor that starts from 0.1 and increases in increments of 0.1 up to 0.5. This weighting factor is used to attenuate the fusion weight of the current terrain increment dataset.

[0083] In the above technical solution, during the data acquisition and differential calculation stages, an industrial-grade vehicle-mounted computer with point cloud processing capabilities can be selected as the hardware platform. This computer is installed in a shockproof cabinet within the mining truck's cab and connected to a 3D laser scanner via a gigabit Ethernet interface. After the mining truck completes a path adjustment, the system automatically initiates the data acquisition process, registering the newly acquired real-time point cloud data with the design terrain model stored on a solid-state drive. After registration, the elevation difference between the two models is calculated using a spatial differential algorithm, generating an incremental terrain dataset. This dataset is stored in system memory as a two-dimensional matrix, with each grid cell corresponding to a 10cm x 10cm area on the actual ground.

[0084] In the validity determination stage, an embedded analysis module can be used to perform statistical processing on the terrain incremental dataset. This module first calculates the average elevation change of all grid cells in the dataset. When the average elevation change reaches 8 cm, it further performs connected region analysis. The system uses an eight-neighborhood region growing algorithm to identify continuous change areas and calculates the projected area of ​​each connected region. The system sets the validity determination thresholds as follows: the average elevation change is not less than 8 cm and the continuous change area is not less than 4 square meters. Only data that simultaneously meets both conditions will be marked as valid incremental data.

[0085] During the model update execution phase, a weighted fusion algorithm can be used to integrate effective incremental data into the existing model. This algorithm assigns a weight of 0.3 to the new incremental data and a weight of 0.7 to the existing model data, calculating the new elevation value for each grid cell through a weighted average. During the update process, the system establishes a data version management mechanism, generating backup files with each update to ensure a rollback to a previous stable version in case of anomalies. The entire update process is executed in a background thread of the onboard computer, without affecting the normal operation of other real-time control functions.

[0086] During the oscillation detection phase, an embedded processing module with real-time analysis capabilities can be used. This module is installed in the expansion slot of the onboard processing unit and communicates with the main processor via a PCIe interface. Internally, the module runs a specially written oscillation detection algorithm that continuously records the average change in height for each terrain increment dataset and stores the three most recent update data in a circular buffer. When three consecutive updates of change in height are detected to be +12 cm, -11 cm, and +13 cm respectively, the system automatically determines that an oscillation pattern has occurred. At this point, both the alternation of the direction of change and the amplitude of the change exceed the set 10 cm threshold.

[0087] In the update amplitude limit execution stage, a programmable logic controller (PLC) or a microprocessor with similar functionality can be selected. This processor is mounted on the circuit board of the vehicle control unit and connected to the main system via a digital signal interface. Upon receiving an oscillation determination signal, the processor initiates the weighting factor control program, which starts with an initial value of 0.1 and gradually increases the weighting factor in increments of 0.1. During each model update, the system multiplies the current weighting factor by the terrain increment dataset, achieving gradual control of the update amplitude until the weighting factor reaches its maximum value of 0.5.

[0088] In the history management update phase, non-volatile memory can be selected as the data storage medium. This memory is installed in the shockproof enclosure of the vehicle system and connected to the main system via a SATA interface. The system records the timestamp, change data, and weighting factor values ​​for each update in a fixed format, forming a complete historical sequence. This data is stored in a circular buffer, retaining the most recent one hundred update records, while important data is periodically backed up to the vehicle's solid-state drive for subsequent analysis and diagnostics.

[0089] By establishing a model update mechanism based on strict validity criteria, dynamic synchronization between the designed terrain model and the actual terrain was achieved, ensuring the timeliness of the baseline data. Statistical analysis methods were used to screen for changed areas, effectively distinguishing between actual earthwork changes and temporary disturbances, avoiding model distortion caused by data noise. The weighted fusion method maintained model stability while incorporating new data, and the version management mechanism provided data security guarantees, together forming a reliable dynamic maintenance system for the terrain model.

[0090] By establishing an oscillation detection mechanism based on continuous trend analysis, abnormal fluctuations during model updates can be identified in a timely manner. A progressively adjusted weighting factor control strategy effectively suppresses update amplitude while maintaining model update capability, avoiding model distortion caused by over-updates. A complete update history provides data support for system status monitoring and fault diagnosis, together forming a model update system capable of long-term stable operation.

[0091] In some other technical solutions, the generation of the real-time excavation and filling distribution map in step two includes the following steps;

[0092] The onboard processing unit converts real-time point cloud data into a grid digital elevation model with a resolution of 10 cm to 20 cm;

[0093] The elevation difference matrix is ​​generated by performing grid-by-grid cell-by-grid cell-by-cell elevation difference calculations between the grid digital elevation model and the pre-stored design terrain model in the same coordinate system.

[0094] The elevation difference matrix is ​​processed by a region growing algorithm based on neighborhood connectivity, and adjacent grid cells with elevation changes greater than 15 cm to 30 cm are clustered into independent change regions.

[0095] Calculate the average elevation change value for each independent change area, and mark the area with a negative average elevation change value as the cut area and the area with a positive average elevation change value as the fill area.

[0096] By combining the spatial location information of all excavation and filling areas, a real-time excavation and filling distribution map is generated.

[0097] In the above technical solution, an industrial control computer with point cloud processing capabilities can be selected during the data preprocessing stage. This computer is installed on a standardized guide rail in the vehicle-mounted control cabinet and connected to the scanner via a copper-core shielded cable. The system downsamples the real-time acquired point cloud data using voxel grid filtering to generate a grid digital elevation model with a resolution of 15 cm. The elevation value of each grid cell is obtained by averaging all point cloud data falling within that grid, while simultaneously recording the point cloud density information of each grid cell. Grids with a point cloud density lower than 5 points per square meter are specially marked.

[0098] In the elevation difference calculation stage, a professional geographic information system (GIS) core processing library can be selected. This software module runs on a dedicated computing card of the vehicle-mounted computer, accessing the data files of the gridded digital elevation model and the pre-stored design terrain model via memory mapping. The system first unifies the two models to the same plane coordinate system and elevation datum, then performs elevation subtraction operations on each grid cell to generate an elevation difference matrix. For grid cells with elevation changes exceeding 20 centimeters, the system marks them as areas of significant change and indexes them for subsequent processing.

[0099] In the region identification and classification stage, a computer vision algorithm library with image processing capabilities can be selected. This library is integrated into the main control program through an application programming interface (API) and uses the eight-neighbor connectivity criterion for region growing. The system clusters adjacent grids with significant changes into independent regions, with a minimum area threshold of 3 square meters for each region. For each independent region, the average elevation change value of all grids within it is calculated. Regions with a negative average change value less than -5 cm are labeled as cut areas, and regions with a positive average change value greater than 5 cm are labeled as fill areas. The final real-time cut-fill distribution map is output as a vector layer, containing the geometric boundaries and attribute information of each region.

[0100] By establishing a systematic processing flow from point cloud to grid, rapid identification and accurate quantification of terrain change areas were achieved. The gridded data processing method improved computational efficiency and ensured real-time system response. A region growing algorithm based on neighborhood connectivity accurately identified continuous change areas, avoiding fragmented erroneous judgments. Combined with an automatic classification mechanism based on elevation change direction, intelligent identification of cut-and-fill areas was achieved, providing a reliable data foundation for subsequent earthwork balance calculations.

[0101] In other technical solutions, the calculation of the current excavation volume and fill volume in step three is achieved through the following steps;

[0102] The vehicle-mounted processing unit extracts the boundary contours of all excavation and filling areas based on the real-time excavation and filling distribution map.

[0103] Each region is processed using triangulation to generate an irregular triangular mesh model;

[0104] Based on the spatial relationship between the triangular mesh model and the pre-stored design terrain model, the prism volume integration method is used to calculate the cut volume of each cut area and the fill volume of each fill area respectively.

[0105] During the calculation process, a real-time soil density calibration mechanism based on the vehicle-mounted weighing system is introduced. This mechanism is implemented in the following way: During the loading of the new energy mining truck, the vehicle-mounted processing unit synchronously acquires the actual load mass data measured by the vehicle-mounted weighing system, and dynamically updates the soil density value in combination with the calculated volume of the corresponding excavation area.

[0106] The updated earthwork density value is applied to the subsequent volume calculation process to achieve dynamic calibration of cut and fill volume calculation.

[0107] In the above technical solution, an industrial control computer with computer vision processing capabilities can be selected for the boundary extraction and meshing stages. This computer is installed in a 19-inch standard rack within the vehicle-mounted control cabinet and connected to the data acquisition system via shielded twisted-pair cables. The system first reads the vector boundary data of each region from the real-time cut-and-fill distribution map, and simplifies the boundaries using the Douglas-Puk algorithm, with a simplification distance tolerance set to 0.5 meters. Subsequently, constrained Delaunay triangulation is performed on each region, generating an irregular triangular mesh model composed of triangular facets. The maximum side length of the triangles is limited to 1.5 meters to ensure that the model accurately describes the terrain features without becoming overly complex.

[0108] During the volume calculation phase, a professional 3D spatial analysis software library can be selected. This software library runs on the 64-bit operating system environment of the onboard computer and is integrated into the main control program through an application programming interface. The system uses the prism volume integration method for calculation based on the generated triangular mesh model and the pre-stored design terrain model. Specifically, each triangular mesh is projected onto the design terrain surface to form a triangular prism. The volume of each triangular prism is calculated and summed to obtain the total volume of the region. During the calculation, the orientation of each triangular prism is determined; the volume of the cut area is taken as negative, and the volume of the fill area is taken as positive. Finally, the total cut volume and the total fill volume are summed to obtain the total volume.

[0109] During the density calibration phase, resistance strain gauge load cells installed on the mine truck's suspension system can be used. These sensors are connected to the on-board weighing terminal via a waterproof junction box, and the weighing terminal communicates with the main control computer via a CAN bus. During the loading process, the system collects weighing data in real time. When the detected load mass stabilizes for more than 3 seconds, the system records the mass data at that moment. The system compares this mass data with the calculated volume of the corresponding excavation area, divides the mass by the volume to obtain the current earthwork density value, and uses this density value to calibrate subsequent volume calculations.

[0110] By establishing a complete processing chain from boundary extraction to volume calculation, precise quantification of cut and fill volumes was achieved. The use of a triangular mesh model better adapts to complex terrain features, improving the accuracy of volume calculations. Combined with the real-time density calibration mechanism of the vehicle-mounted weighing system, calculation errors caused by variations in earthwork density were effectively eliminated, ensuring the reliability of cut-fill balance judgments. The entire calculation process was automated, providing accurate data support for earthwork balance control.

[0111] In other technical solutions, the real-time soil density calibration mechanism also includes an error compensation step; the vehicle-mounted processing unit establishes an error propagation model for soil density, which considers a measurement error of 1% to 2% from the vehicle-mounted weighing system and a volume calculation error of 2% to 3% from the point cloud data; when dynamically updating the soil density value, a Kalman filter algorithm is used to filter the density measurement value; the filtered density value is combined with the error propagation model to compensate and correct the calculation results of excavation volume and fill volume; after compensation and correction, the estimated range of volume calculation error is between 3% and 5%.

[0112] In the above technical solution, during the error propagation model establishment stage, an industrial control computer with matrix operation capabilities can be selected. This computer is installed in a shockproof cabinet within the vehicle-mounted system and connected to each sensor interface module via a backplane bus. The system constructs a mathematical model based on error propagation theory, setting the measurement error of the vehicle-mounted weighing system to 1.5% and the point cloud data volume calculation error to 2.5%. A covariance matrix is ​​established to describe the influence of these error sources on the final density calculation; the matrix parameters are initialized based on sensor calibration data and historical measurement statistics.

[0113] In the Kalman filtering stage, an embedded processor with digital signal processing capabilities can be selected. This processor is installed in a dedicated slot on the main control board and communicates with the main processor via a high-speed serial interface. The filtering algorithm sets the process noise variance to 0.01 and the measurement noise variance to 0.04, recursively processing the density measurements through two steps: state prediction and state update. Each time new density measurement data is input, the filter calculates the optimal density estimate based on the system model and the observation model, and simultaneously updates the error covariance matrix.

[0114] During the compensation and correction execution phase, an arithmetic logic unit (ALU) with floating-point operation capabilities can be selected. This unit is integrated inside the onboard processor and connected to the cache memory via an internal bus. The system substitutes the density estimate output from the Kalman filter into the error propagation model to calculate the correction coefficients for the current volume calculation result. The correction coefficients are applied to the real-time volume calculation result through a multiplier, and a range check is performed on the corrected result to ensure that the output value is within a reasonable range. The compensated and corrected volume data is stored in a dual-port RAM for subsequent processing.

[0115] By establishing a complete error propagation model, the system can accurately describe the cumulative impact of errors in each measurement stage on the final result. The Kalman filter algorithm is used to optimize the density measurements, effectively reducing the interference of random errors. The compensation and correction mechanism based on the error model significantly improves the accuracy of volume calculation results, providing a more reliable data foundation for cut-fill balance control. The entire error compensation process is automated, ensuring that the system maintains stable measurement accuracy under complex working conditions.

[0116] In other technical solutions, the determination of the adjustment path in step three is achieved through the following steps;

[0117] The vehicle-mounted processing unit identifies the target area with the greatest difference between the excavation volume and the filling volume based on the real-time excavation and filling distribution map, and uses the center coordinates of the target area as the path endpoint.

[0118] A path planning algorithm is used to search for paths in an obstacle map generated from real-time point cloud data. The cost function of this path planning algorithm includes two calculation terms: path length cost and terrain slope cost.

[0119] The initial path output by the path planning algorithm undergoes a safety check based on vehicle dynamics constraints. This check excludes path segments that exceed the maximum climbing angle of the new energy mining truck by 30% to 35%, as well as path segments that are less than the minimum turning radius of the new energy mining truck by 8 to 10 meters.

[0120] The path that has passed the security check is processed by a curve fitting algorithm to generate an executable path with continuous curvature.

[0121] During the execution of the executable path by the new energy mining truck, the 3D laser scanner performs real-time obstacle detection within a range of 20 m to 30 m in the direction of travel at a frequency of 5 Hz to 10 Hz.

[0122] When an obstacle is detected, the onboard processing unit initiates local path replanning, generates an obstacle avoidance path, and continues execution.

[0123] In the above technical solution, a vehicle-mounted industrial computer with parallel computing capabilities can be selected for the target area identification and path planning stages. This computer is installed in a shockproof cabinet at the rear of the driver's cab and connected to the sensing system via a high-speed Ethernet connection. The system first calculates the difference in cut and fill volume for each area based on a real-time cut and fill distribution map, selecting the area with the largest difference as the target, and using the center of its circumscribed rectangle as the path endpoint. Path planning uses the A* algorithm, searching within an obstacle map generated from real-time point clouds. The algorithm's cost function has a path length weight of 0.6, a terrain slope weight of 0.4, and a search step size of 0.5 meters.

[0124] During the safety verification and path optimization phase, a path processing module with geometric analysis capabilities can be selected. This module runs in the onboard computer's real-time operating system and communicates with the planning module via shared memory. The system performs segmented checks on the planned initial path, excluding sections with a gradient exceeding 32% and turning radii less than 9 meters. The verified path is then smoothed using a B-spline curve fitting algorithm, with the curve order set to 3 and the control point spacing set to 2 meters, ensuring that the generated path curvature is continuous and conforms to the vehicle's kinematic characteristics.

[0125] During the path execution and obstacle handling phases, a combination of a 3D laser scanner and a real-time control system can be used. The scanner is mounted on the front protective frame of the vehicle and connected to the control unit via a dedicated cable. As the vehicle travels along the planned path, the scanner performs a fan-shaped scan of a 25-meter area ahead at a frequency of 8Hz. When an obstacle is detected to be less than 1.5 meters from the path centerline, the system immediately initiates local replanning. Local planning uses the same path search algorithm to generate obstacle avoidance paths between global path points, ensuring continuous vehicle operation.

[0126] An integrated path planning and safety verification mechanism ensures the feasibility of planned paths in complex terrain. A multi-factor weighted cost function balances path length and safety, while rigorous dynamic constraint verification prevents vehicles from getting stuck in impassable conditions. Real-time obstacle detection and rapid replanning capabilities guarantee the continuity of the driving process. The entire path determination method achieves high efficiency while ensuring safety.

[0127] In other technical solutions, the path planning algorithm adopts a hierarchical planning strategy; the hierarchical planning strategy includes a global planning layer and a local planning layer; the global planning layer uses a path planning algorithm to perform coarse-grained path planning with a resolution of 1 m to 2 m and a planning cycle of 10 s to 15 s; the local planning layer uses the same path planning algorithm to perform fine-grained path planning with a resolution of 0.2 m to 0.5 m and a planning cycle of 1 s to 2 s; the global planning layer and the local planning layer are coupled through a path point sequence, and the local planning layer performs real-time fine planning among global path points; the on-board processing unit allocates dynamically adjustable computing resources to the global planning layer and the local planning layer; when the local planning layer triggers replanning more than 3 times within 1 second, the system automatically tilts more computing resources toward the local planning layer.

[0128] In the above technical solution, during the construction phase of the hierarchical planning system, a vehicle-mounted industrial computer with multi-core processing capabilities can be selected as the hardware platform. This computer is installed in a shockproof control cabinet and connected to various functional modules via a backplane bus. The system divides the planning task into two independent layers: the global planning layer runs on two dedicated computing cores, processing terrain data for the entire work area at a resolution of 1.5 meters, with a planning cycle set to 12 seconds; the local planning layer runs on another set of computing cores, processing the area surrounding the vehicle at a resolution of 0.3 meters, with a planning cycle set to 1.5 seconds. The two layers exchange path point sequence data via shared memory.

[0129] During the planning task execution phase, the same path planning algorithm can be used but run at different scales. The global planning layer uses the A* algorithm to process the low-resolution obstacle map, generating a coarse path sequence containing 20 to 30 path points, with each path point spaced 8 to 10 meters apart. The local planning layer receives the global path point sequence and performs fine-grained planning between adjacent global path points, generating a detailed path containing 5 to 8 path points, with the path point spacing controlled at 1 to 2 meters. Both planner layers use the same cost function configuration to ensure consistency in the planning strategy.

[0130] During the dynamic allocation phase of computing resources, the system resource management module can be used for real-time monitoring and allocation. This module runs on an independent management core and continuously monitors the replanning trigger frequency of the local planning layer. When the number of replanning events reaches 3 within 1 second, the resource management module automatically adjusts the computing core allocation ratio, temporarily allocating two computing cores from the global planning layer to the local planning layer. The system also sets a duration threshold for resource allocation; if the high-frequency replanning state continues for 5 seconds, the system will automatically revert to the original resource allocation scheme.

[0131] By establishing a hierarchical planning architecture, the system achieves a reasonable allocation of path planning tasks at different granularities, ensuring both the rationality of the global path and the real-time performance of local obstacle avoidance. A dynamic resource allocation mechanism enables the system to adaptively adjust computing resources according to actual operating conditions, prioritizing the response speed of local planning in complex environments. This design effectively balances the contradiction between planning quality and computational efficiency, providing reliable technical support for the stable operation of vehicles in steep and confined areas.

[0132] Other technical solutions also include system stability monitoring and control mode switching steps;

[0133] The onboard processing unit monitors the following key system indicators in real time: the fluctuation frequency of the cut-fill balance ratio, the data integrity rate of real-time point cloud data, and the trigger frequency of local path replanning.

[0134] The system is considered to be in an unstable risk state when the following conditions are met simultaneously: the cut-fill balance ratio fluctuates more than 10 times per minute within 30 consecutive seconds, the data integrity rate of real-time point cloud data is less than 85%, and the local path replanning trigger frequency exceeds 5 times per minute.

[0135] When the system is determined to be in a state of instability risk, the on-board processing unit automatically switches from fully automatic control mode to auxiliary control mode;

[0136] In the auxiliary control mode, the on-board processing unit simplifies the adjustment path to a straight line from the current location to the nearest cut-and-fill area, and limits the adjustment speed to 5 km / h to 8 km / h;

[0137] The system continuously monitors key indicators, and automatically switches back to fully automatic control mode after the indicators return to normal and remain stable for 10 seconds.

[0138] In the above technical solution, during the system status monitoring phase, a multi-channel data acquisition card can be used in conjunction with an industrial-grade microprocessor. The monitoring system is installed in a dedicated slot in the vehicle-mounted control cabinet and connects to various sensors via analog and digital input modules. The system acquires cut-fill balance ratio data at a frequency of 1Hz, calculating its fluctuations per minute; simultaneously, it calculates the point cloud data integrity rate, i.e., the ratio of valid points to the theoretical total number of points; and records the number of local path replanning triggers. This data is transmitted to the main processing unit via an RS485 bus for real-time analysis and storage.

[0139] During the instability determination phase, a programmable logic controller (PLC) with logic processing capabilities can be used. This PLC is mounted on a rail in a shockproof enclosure and connected to the main system via an optocoupler isolation module. The system is configured with three thresholds for instability determination: a cut-fill balance ratio fluctuation frequency of 12 times per minute, a point cloud data integrity rate of 85%, and a local path replanning trigger frequency of 6 times per minute. When these three conditions are simultaneously met for 40 consecutive seconds, the controller outputs an instability signal, which is transmitted to the main controller via a relay module.

[0140] During the control mode switching and recovery phase, a dual-mode programmable controller can be used. This controller is installed adjacent to the main control board and controls mode switching via a digital output module. When switching to auxiliary control mode, the system uses a straight-line path planning algorithm to select the cut-and-fill area closest to the vehicle's current position; the straight-line distance is calculated using the Euclidean algorithm. The vehicle speed is limited to 7 kilometers per hour. The system continuously monitors key indicators, and when all indicators return to normal and remain stable for 12 seconds, the controller automatically switches back to fully automatic control mode.

[0141] By establishing a system stability monitoring mechanism with multi-parameter joint judgment, abnormal operating states of the system can be identified in a timely manner. An automatic mode switching strategy is adopted, which can quickly degrade to a more reliable control mode when the system is at risk of instability, ensuring operational safety. Reasonable recovery conditions are set to ensure that the system can automatically return to normal operation after the operating conditions improve, reducing the need for manual intervention. This comprehensive monitoring and protection mechanism enhances the system's adaptability and operational reliability under complex operating conditions.

[0142] Example 1

[0143] In steep and narrow mining areas, system deployment is required before implementing this method. A 32-line mechanical rotating lidar is installed on the protective frame on top of the cab of the new energy mining truck. This device is fixed with a rigid bracket, and the scanning center axis is tilted downwards at an 8-degree angle. An industrial-grade onboard computer, equipped with a multi-core CPU and a dedicated GPU computing module, is installed in a shockproof cabinet inside the cab. It connects to the lidar via gigabit Ethernet and communicates with the vehicle control system via a CAN bus. The control system directly drives the electric drive mechanism and steering mechanism of the mining truck, completing the hardware platform setup.

[0144] Before the operation began, a multi-rotor UAV equipped with a lightweight laser scanning system was used to perform a full-area scan of the work area, maintaining a flight altitude of 90 meters and a point cloud density of 200 points per square meter. Simultaneously, a pulsed scanning measurement system was used on the ground to supplement the scan of steep cliffs and narrow ravines, with a scanning distance of 100 meters. The acquired multi-source point cloud data was imported into professional point cloud processing software, and fusion and registration were performed using an iterative nearest-point algorithm. The registration error threshold was set at 4 centimeters, and point cloud data with elevation differences exceeding 3 centimeters were discarded, ultimately generating an initial digital elevation model with a resolution of 3 centimeters. This model was then overlaid and compared with a CAD format 3D surface model provided by the engineering design team. Areas with elevation deviations exceeding 12 centimeters were manually verified and corrected, generating a pre-stored design terrain model and storing it on the vehicle's onboard computer's solid-state drive.

[0145] During real-time operation, a 3D laser scanner continuously scans an 80-meter radius annular area at a frequency of 30Hz to acquire real-time point cloud data. The vehicle-mounted processing unit completes data processing within 0.8 seconds: first, it converts the real-time point cloud into a 15-centimeter resolution grid digital elevation model, and performs elevation difference calculations with a pre-stored design terrain model to generate an elevation difference matrix; then, it uses a region growing algorithm to cluster adjacent grid cells with elevation changes greater than 20 centimeters into independent change regions; and it labels the cut and fill areas according to the positive and negative values ​​of the average elevation change values ​​to generate a real-time cut and fill distribution map. Based on this distribution map, it extracts the regional boundary contours, uses triangular meshing to generate an irregular triangular mesh model, and calculates the cut and fill volumes using the prism volume integration method. Simultaneously, it introduces a real-time soil density calibration mechanism based on the vehicle-mounted weighing system, combining a Kalman filter algorithm to filter the density measurements, achieving dynamic calibration and error compensation for volume calculations.

[0146] When the calculated cut-fill balance ratio reaches 0.97, the system initiates the path planning process. The onboard processing unit identifies the target area with the greatest difference in cut-fill volume based on the real-time cut-fill distribution map and employs a hierarchical planning strategy for path planning. The global planning layer performs coarse-grained path planning at a resolution of 1.5 meters, with a planning cycle of 12 seconds; the local planning layer performs fine-grained path planning at a resolution of 0.3 meters, with a planning cycle of 1.5 seconds. The path planning uses the A* algorithm, with a cost function including two calculation terms: path length and terrain slope, with weights set to 0.6 and 0.4 respectively. The planned path undergoes safety verification, excluding path segments with a slope exceeding 32% and path segments with a turning radius less than 9 meters. A curvature-continuous executable path is then generated through B-spline curve fitting. While the vehicle executes the path at a speed of 8 km / h, a 3D laser scanner performs real-time obstacle detection within a 25-meter radius ahead at a frequency of 8 Hz. Upon detecting an obstacle, local path replanning is immediately initiated.

[0147] During the model update phase, after each path adjustment, the system performs a secondary difference calculation between the latest collected real-time point cloud data and the current design terrain model to generate an incremental terrain dataset. When an average change in height reaches 8 centimeters and a continuous change in area reaches 4 square meters, it is considered valid incremental data and is integrated into the design terrain model with a weight of 0.3. Simultaneously, the system monitors the update trend through a stability control mechanism. When three consecutive updates show alternating directions and a change amplitude exceeding 10 centimeters, an update amplitude limit is activated, using a weighting factor that gradually increases from 0.1 to control the update amplitude.

[0148] The system stability monitoring module monitors the fluctuation frequency of the cut-fill balance ratio, the integrity rate of point cloud data, and the frequency of local path replanning triggers in real time. If, within 40 seconds, a fluctuation frequency of 12 times per minute, a data integrity rate below 85%, and a replanning trigger frequency of 6 times per minute are simultaneously detected, the system is deemed to be in an unstable risk state and automatically switches to auxiliary control mode. In this mode, straight-line path planning is used, with a speed limited to 7 kilometers per hour, until the system indicators return to normal and remain stable for 12 seconds, after which it automatically reverts to fully automatic mode.

[0149] Through the above complete implementation process, rapid and precise control of earthwork excavation and filling balance in new energy mining areas under complex terrain conditions of steep and narrow terrain was achieved.

[0150] Comparative Example 1: Traditional manual judgment versus periodic measurement method

[0151] This comparative model employs a traditional work mode that relies entirely on manual experience and periodic measurements. Two surveyors are on-site, conducting topographic surveys every 8 hours using a total station, with a 20-meter interval between measurement points. After manual recording, the measurement data is returned to the office for processing, generating a topographic map using CAD software and comparing it with the design drawings; this entire process takes approximately 10 hours. Mine truck drivers rely entirely on hand signals and whistle commands from on-site supervisors, who visually determine the cut-and-fill areas. Vehicle routes are chosen by the drivers based on experience, and driving speeds are adjusted arbitrarily between 5-25 km / h depending on road conditions. The design model is updated monthly, during which time changes in terrain cannot be reflected in a timely manner.

[0152] Comparative Example 2: Single-Platform Mapping vs. Static Model Method

[0153] This comparative study employed a single UAV platform for terrain mapping, maintaining a static model. Aerial surveying was conducted using only a fixed-wing UAV equipped with a lidar, flying at an altitude of 200 meters, achieving a point cloud density of 50 points per square meter. Due to the high flight altitude and UAV limitations, effective scanning of steep cliff faces was impossible, resulting in missing data in the model. The acquired point cloud data was directly used to generate a digital elevation model with a resolution of 15 centimeters, without verification and correction against ground measurement data. Throughout the subsequent three-month work period, this design model remained in its initial state, failing to update as the work progressed, leading to systematic biases during real-time comparisons.

[0154] Comparative Example 3: Simplified Calculation vs. Fixed Path Planning Methods

[0155] This comparative model employs a simplified calculation method and a fixed path planning strategy. Earthwork volume calculation uses the average density method, uniformly set at 2.0 tons / cubic meter, without real-time calibration. Volume calculation is based on a regular grid model, using a prism formula for simplified calculation, ignoring terrain surface features. Path planning is completed once before operation, considering only straight-line distance and ignoring terrain slope. Vehicles travel at a fixed speed of 12 km / h, stopping and waiting for manual replanning upon encountering obstacles. Due to insufficient calculation accuracy and the path's inability to adapt to real-time conditions, frequent manual intervention and adjustments are required.

[0156] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and embodiments shown and described herein.

Claims

1. A rapid control method for earthwork excavation and filling balance in steep and narrow new energy mining areas, characterized in that, include: Step 1: Using the 3D laser scanner installed on the new energy mining truck, continuously scan the terrain of the ring-shaped working area with a radius of 50 m to 100 m centered on the new energy mining truck at a scanning frequency of 20 Hz to 40 Hz to obtain real-time point cloud data. Step 2: Transmit the real-time point cloud data to the on-board processing unit of the new energy mining truck through the vehicle communication system. The on-board processing unit completes the processing of the real-time point cloud data within 0.3 s to 1.5 s, compares the real-time point cloud data with the pre-stored design terrain model, and generates a real-time excavation and filling distribution map. The pre-stored design terrain model is established through the following steps; The UAV-borne LiDAR was used to scan the entire work area, while a ground-based 3D laser scanner was used to supplement the scan of steep cliffs and narrow ravines to obtain high-precision initial point cloud data. The point cloud data acquired by the UAV-borne lidar and the ground-based 3D laser scanner are fused and registered to generate an initial digital elevation model of the work area with a resolution of 2 cm to 5 cm. The digital design terrain model provided by the engineering design party is overlaid and compared with the initial digital elevation model. Areas with elevation deviations exceeding 10 cm to 15 cm are manually verified and corrected to generate a pre-stored design terrain model as a benchmark. Step 3: Based on the real-time excavation and filling distribution map, the on-board processing unit calculates the current excavation volume and filling volume, and calculates the ratio of excavation volume to filling volume as the excavation-filling balance ratio. When the excavation-filling balance ratio is lower than 0.95 or higher than 1.05, the on-board processing unit determines the adjustment path and adjustment speed of the new energy mining truck based on the real-time point cloud data. The adjustment path points to the area with the greatest difference between the excavation volume and the filling volume, and the adjustment speed is controlled between 5 km / h and 15 km / h. Step 4: The on-board processing unit sends the adjustment path and adjustment speed to the control system of the new energy mining truck. The control system drives the electric drive mechanism and steering mechanism of the new energy mining truck according to the adjustment path and adjustment speed to execute the adjustment path and adjustment speed. The vehicle-mounted processing unit is equipped with a model update module, which dynamically updates the pre-stored design terrain model based on real-time point cloud data; the dynamic update follows the following steps. After completing one path adjustment, the on-board processing unit performs a second differential calculation on the real-time point cloud data collected and registered during this round of operation and the design terrain model before dynamic update to generate an incremental terrain dataset. Statistical analysis was performed on the terrain increment dataset to calculate its average height change and the continuous area of ​​the changed region; only when the absolute value of the average height change is greater than 5 cm to 10 cm and the continuous area of ​​the changed region is greater than 2 m... 2 up to 5 m 2 Only when this condition is met is the incremental data deemed valid. The incremental terrain dataset that has passed the validity verification is superimposed onto the pre-stored design terrain model in a weighted fusion manner to complete the iterative update of the model; The model update module also includes an update stability control mechanism. This mechanism records the average change height of the current terrain increment dataset during each iteration update and compares it with the change height recorded in the previous two updates. When the change height sequence signs alternate between positive and negative for three consecutive updates, and the absolute values ​​of the two adjacent changes are both greater than 8 cm to 12 cm, it is determined to be an oscillation and update amplitude limitation is initiated. The update amplitude limitation is implemented by introducing a weighting factor that starts from 0.1 and increases in increments of 0.1 up to 0.

5. This weighting factor is used to attenuate the fusion weight of the current terrain increment dataset.

2. The rapid control method for earthwork excavation and filling balance in steep and narrow new energy mining areas as described in claim 1, characterized in that, The fusion and registration of the point cloud data includes a precision control step; During the fusion registration process, the registration error between the UAV-borne LiDAR and the ground-based 3D laser scanner point cloud data is calculated. When the registration error is greater than 3 cm to 5 cm, the iterative nearest point algorithm is used for re-registration. The re-registered point cloud data is subjected to an overlap area consistency check, and point cloud data with an elevation difference of more than 2 cm to 4 cm in the overlap area are removed. In the final fused point cloud data, the registration error of 90% to 95% of the points is no greater than 1.5 times the resolution of the initial digital elevation model.

3. The rapid control method for earthwork excavation and filling balance in steep and narrow new energy mining areas as described in claim 1, characterized in that, The generation of the real-time excavation and filling distribution map in step two includes the following steps; The onboard processing unit converts real-time point cloud data into a grid digital elevation model with a resolution of 10 cm to 20 cm; The elevation difference matrix is ​​generated by performing grid-by-grid cell-by-cell elevation difference calculations between the grid digital elevation model and the pre-stored design terrain model in the same coordinate system. The elevation difference matrix is ​​processed by a region growing algorithm based on neighborhood connectivity, and adjacent grid cells with elevation changes greater than 15 cm to 30 cm are clustered into independent change regions. Calculate the average elevation change value for each independent change area, and mark the area with a negative average elevation change value as the cut area and the area with a positive average elevation change value as the fill area. By combining the spatial location information of all excavation and filling areas, a real-time excavation and filling distribution map is generated.

4. The rapid control method for earthwork excavation and filling balance in steep and narrow new energy mining areas as described in claim 1, characterized in that, The calculation of the current excavation volume and fill volume in step three is achieved through the following steps; The vehicle-mounted processing unit extracts the boundary contours of all excavation and filling areas based on the real-time excavation and filling distribution map. Each region is processed using triangulation to generate an irregular triangular mesh model; Based on the spatial relationship between the triangular mesh model and the pre-stored design terrain model, the prism volume integration method is used to calculate the cut volume of each cut area and the fill volume of each fill area respectively. During the calculation process, a real-time soil density calibration mechanism based on the vehicle-mounted weighing system is introduced. This mechanism is implemented in the following way: During the loading of the new energy mining truck, the vehicle-mounted processing unit synchronously acquires the actual load mass data measured by the vehicle-mounted weighing system, and dynamically updates the soil density value in combination with the calculated volume of the corresponding excavation area. The updated earthwork density value is applied to the subsequent volume calculation process to achieve dynamic calibration of cut and fill volume calculation.

5. The rapid control method for earthwork excavation and filling balance in steep and narrow new energy mining areas as described in claim 4, characterized in that, The real-time soil density calibration mechanism also includes an error compensation step; the vehicle-mounted processing unit establishes an error propagation model for soil density, which considers a measurement error of 1% to 2% from the vehicle-mounted weighing system and a volume calculation error of 2% to 3% from the point cloud data; when dynamically updating the soil density value, a Kalman filter algorithm is used to filter the density measurement value; the filtered density value is combined with the error propagation model to compensate and correct the calculation results of excavation volume and fill volume; after compensation and correction, the estimated range of volume calculation error is between 3% and 5%.

6. The rapid control method for earthwork excavation and filling balance in steep and narrow new energy mine areas as described in claim 1, characterized in that, The determination of the adjusted path in step three is achieved through the following steps; The vehicle-mounted processing unit identifies the target area with the greatest difference between the excavation volume and the filling volume based on the real-time excavation and filling distribution map, and uses the center coordinates of the target area as the path endpoint. A path planning algorithm is used to search for paths in an obstacle map generated from real-time point cloud data. The cost function of this path planning algorithm includes two calculation terms: path length cost and terrain slope cost. The initial path output by the path planning algorithm undergoes a safety check based on vehicle dynamics constraints. This check excludes path segments that exceed the maximum climbing angle of the new energy mining truck by 30% to 35%, as well as path segments that are less than the minimum turning radius of the new energy mining truck by 8 to 10 meters. The path that has passed the security check is processed by a curve fitting algorithm to generate an executable path with continuous curvature. During the execution of the executable path by the new energy mining truck, the 3D laser scanner performs real-time obstacle detection within a range of 20 m to 30 m in the direction of travel at a frequency of 5 Hz to 10 Hz. When an obstacle is detected, the onboard processing unit initiates local path replanning, generates an obstacle avoidance path, and continues execution.

7. The rapid control method for earthwork excavation and filling balance in steep and narrow new energy mining areas as described in claim 6, characterized in that, The path planning algorithm adopts a hierarchical planning strategy, which includes a global planning layer and a local planning layer. The global planning layer uses a path planning algorithm to perform coarse-grained path planning with a resolution of 1 m to 2 m and a planning cycle of 10 s to 15 s. The local planning layer uses the same path planning algorithm to perform fine-grained path planning with a resolution of 0.2 m to 0.5 m, and the planning cycle is 1 s to 2 s; The global planning layer and the local planning layer are coupled through a sequence of path points. The local planning layer performs real-time fine planning between global path points. The on-board processing unit allocates dynamically adjustable computing resources to the global planning layer and the local planning layer. When the local planning layer triggers replanning more than 3 times within 1 second, the system automatically allocates more computing resources to the local planning layer.

8. The rapid control method for earthwork excavation and filling balance in steep and narrow new energy mining areas as described in claim 1, characterized in that, It also includes system stability monitoring and control mode switching steps; The onboard processing unit monitors the following key system indicators in real time: the fluctuation frequency of the cut-fill balance ratio, the data integrity rate of real-time point cloud data, and the trigger frequency of local path replanning. The system is considered to be in a state of instability risk when the following conditions are met simultaneously: the frequency of fluctuation of the cut-fill balance ratio exceeds 10 times per minute within 30 consecutive seconds, the data integrity rate of real-time point cloud data is less than 85%, and the frequency of local path replanning triggering exceeds 5 times per minute. When the system is determined to be in a state of instability risk, the on-board processing unit automatically switches from fully automatic control mode to auxiliary control mode; In the auxiliary control mode, the on-board processing unit simplifies the adjustment path to a straight line from the current location to the nearest cut-and-fill area, and limits the adjustment speed to 5 km / h to 8 km / h; The system continuously monitors key indicators, and automatically switches back to fully automatic control mode after the indicators return to normal and remain stable for 10 seconds.