A hydraulic engineering construction machinery group cooperative operation intelligent scheduling system

Abstract

The application relates to the field of construction scheduling control, in particular to a water conservancy engineering construction mechanical group collaborative operation intelligent scheduling system, which comprises the following steps: a geological mechanical coupling twin platform constructs a bidirectional coupling closed loop of a geological field and construction mechanical behavior according to real-time working condition data of a region to be constructed, obtains a dynamic geological attribute field quantity, and fits a geological disturbance sensitive coefficient of the region to be constructed; a cloud edge end collaborative network collects real-time working condition data through an intelligent agent node; a cloud computing center solves a global construction task scheduling scheme by adopting a multi-objective optimization algorithm; an edge computing node adopts a multi-agent deep reinforcement learning algorithm to perform real-time rolling optimization on the collaborative operation of construction machines in a local jurisdiction region according to a screened optimal local construction task distribution scheme, generates a local collaborative scheduling instruction, and schedules each construction machine. The scheduling system can realize real-time sensing of geological changes, reduce the probability of engineering accidents, and dynamically optimize the collaborative operation of the mechanical group.

Description

Technical Field

[0001] This invention relates to the field of construction scheduling and control, and in particular to an intelligent scheduling system for collaborative operation of a group of construction machinery in water conservancy projects. Background Technology

[0002] Water conservancy project construction typically involves complex operations such as large-scale earthwork excavation, filling, compaction, and material transportation. The efficiency of the coordinated operation of construction machinery directly affects the project's progress, quality, and cost. Traditional methods of scheduling construction machinery mainly rely on manual experience or fixed-path planning, which are difficult to adapt to complex working conditions such as dynamic changes in geological conditions, real-time fluctuations in machinery status, and interference from multiple coupled machines. This results in low equipment utilization, high energy consumption, and difficulty in ensuring construction quality.

[0003] In existing technologies, some construction management systems have incorporated positioning technology and communication networks to achieve real-time monitoring of machinery positions and the issuance of basic scheduling commands. However, these systems typically treat geological conditions as static prior information and cannot adaptively adjust scheduling strategies based on changes in geological properties and quantities during actual construction. Furthermore, existing scheduling methods often employ a centralized architecture, with all decisions calculated uniformly in the cloud. As the number of machines increases and the work area expands, problems arise such as high communication latency, poor real-time performance, and a high risk of single-point failures. Moreover, the optimization objectives for multi-machine collaborative operations are often limited to shortest paths or lowest energy consumption, lacking a comprehensive consideration of geological disturbance risks and dynamic interactions between machines, making it difficult to achieve optimal overall benefits.

[0004] Therefore, how to build an intelligent scheduling system that can perceive geological changes in real time and dynamically optimize the collaborative operation of machinery groups has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an intelligent scheduling system for collaborative operations of water conservancy engineering construction machinery groups.

[0006] In a first aspect, the present invention provides an intelligent scheduling system for collaborative operation of a group of construction machinery in water conservancy projects, the system comprising: The geomechanical coupling twin platform is used to construct a two-way coupling closed loop between the geological field and the behavior of construction machinery based on real-time working condition data of the area to be constructed, to obtain dynamic geological attribute field quantities, and to fit the geological disturbance sensitivity coefficient of the area to be constructed based on the dynamic geological attribute field quantities. The cloud-edge-device collaborative network includes intelligent agent nodes deployed on each construction machine, edge computing nodes that are communicatively connected to the intelligent agent nodes, and cloud computing centers that are communicatively connected to the edge computing nodes. The intelligent agent node is used to collect the real-time operating condition data. The cloud computing center is used to solve the global construction task scheduling scheme by using a multi-objective optimization algorithm based on the real-time working condition data, the dynamic geological attribute field quantity and the geological disturbance sensitivity coefficient, so as to obtain a global construction task scheduling scheme set. The edge computing node is used to select the optimal local construction task allocation scheme from the global construction task scheduling scheme set, and based on the optimal local construction task allocation scheme, the dynamic geological attribute field quantity, and the geological disturbance sensitivity coefficient, a multi-agent deep reinforcement learning algorithm is used to perform real-time rolling optimization of the collaborative operation of construction machinery within the local jurisdiction of the edge computing node, generating local collaborative scheduling instructions to schedule each construction machine; the multi-agent deep reinforcement learning algorithm is configured to introduce a geological disturbance cost reward into the reward function, and the geological disturbance cost reward is fitted based on the geological disturbance sensitivity coefficient; The geomechanical coupling twin platform includes: a static geological layer, a dynamic operation influence layer, a two-way closed-loop mapping module, and a dynamic fitting module for geological disturbance sensitivity coefficients; The static geological layer is used to store the a priori geological attribute field quantity constructed based on historical geological exploration data. The a priori geological attribute field quantity includes at least: geometric and stratigraphic distribution data, basic physical attribute data, mechanical parameter data, and initial state data. The dynamic operation influence layer is used to receive and store the real-time operating condition data in real time. The real-time operating condition data includes at least: vibration response data, cutting resistance data, and compaction feedback data. The bidirectional closed-loop mapping module is used to correct the prior geological attribute field quantity based on the real-time working condition data and through a pre-constructed physical mapping model to obtain the dynamic geological attribute field quantity. The dynamic fitting module for the geological disturbance sensitivity coefficient is used to dynamically fit the geological disturbance sensitivity coefficient of the area to be constructed based on the dynamic geological attribute field quantity and through a pre-constructed nonlinear mapping model of the disturbance sensitivity coefficient. The geological disturbance sensitivity coefficient is set to quantify the potential risk of geological disturbance in the area to be constructed when the construction machinery is operating.

[0007] Preferably, the bidirectional closed-loop mapping module is further configured to: calculate the correction gradient of different grid units of the static geological layer based on the dynamic geological attribute field quantity and the prior geological attribute field quantity, and perform adaptive refinement processing on the grid units based on the correction gradient quantity to improve the resolution of the grid units.

[0008] Preferably, the geomechanical coupled twin platform further includes: The mechanical operation parameter dynamic fitting module is used to dynamically adjust the rated operation efficiency of the construction machinery according to the geological disturbance sensitivity coefficient and the pre-constructed operation efficiency mapping function to obtain the adaptive operation efficiency of the construction machinery; and to dynamically adjust the rated movement speed of the construction machinery according to the geological disturbance sensitivity coefficient and the pre-constructed movement speed mapping function to obtain the adaptive movement speed of the construction machinery.

[0009] Preferably, the objective function of the multi-objective optimization algorithm includes: total completion time, total energy consumption, and geological disturbance risk; the geological disturbance risk is set to fit the geological disturbance sensitivity coefficient of the construction area where the construction task is located and the regional operation intensity of the construction area where the construction task is located.

[0010] Preferably, the constraints of the multi-objective optimization algorithm include at least: construction task allocation change constraints, work intensity constraints, and machinery task matching constraints. The construction task allocation change constraints are set such that the amount of construction task allocation change between any two adjacent construction task allocation cycles is less than or equal to a preset maximum change number. The work intensity constraints are set such that the work intensity of the construction area is less than or equal to a preset maximum area work intensity. The machinery task matching constraints are set such that the construction machinery should be matched with the construction task.

[0011] Preferably, the multi-objective optimization algorithm is the non-dominated sorting genetic algorithm II. In the solution of the non-dominated sorting genetic algorithm II, task chain encoding and mechanical task sequence encoding are used. The construction tasks that can be executed simultaneously within the task chain encoding are represented by sequence encoding to indicate the execution order. The mechanical task sequence encoding reflects the set of construction tasks for each construction machine. By combining the first entity corresponding to any of the task chain codes and the second entity corresponding to any of the mechanical task sequence codes, a feasible construction task scheduling scheme is obtained.

[0012] Preferably, the reward function of the multi-agent deep reinforcement learning algorithm further includes: construction task progress reward, operation efficiency reward and operation collaboration reward; the construction task progress reward is set to represent the proportion of completed construction tasks, the operation collaboration reward is set to penalize operation conflicts and operation waiting, and the geological disturbance cost reward is set to be fitted according to the operation intensity of the construction machinery and the geological disturbance sensitivity coefficient of the current location of the construction machinery.

[0013] Preferably, the edge computing node includes: A graph neural network model is used to obtain a collaborative feature vector based on the real-time working condition data of each agent node, the spatial topological relationship between the construction machinery, and the construction task dependency relationship between the construction machinery; the collaborative feature vector and the dynamic geological attribute field are spliced ​​and fused to obtain a spatiotemporal coupled dynamic vector, which is set as the state space for constructing the multi-agent deep reinforcement learning algorithm.

[0014] Preferably, the graph neural network model uses each of the construction machines within the local jurisdiction as a set of nodes, and the communication link, spatial distance, and coupling degree between any two construction machines as a set of edges, where the coupling degree represents the degree of coupling between the construction tasks corresponding to the construction machines.

[0015] This invention provides an intelligent scheduling system for collaborative operation of a group of construction machinery in water conservancy projects. Compared with the prior art, the embodiments of this invention have the following beneficial effects: The intelligent scheduling system for collaborative operation of water conservancy engineering construction machinery provided in this application constructs a geological-mechanical coupled twin platform. Utilizing real-time operating data from the construction machinery's operation, it employs Gaussian process regression to online correct prior geological attribute fields. This allows the geological-mechanical coupled twin platform to continuously approximate real geological conditions as construction progresses, overcoming the technical shortcomings of traditional geological exploration, such as sparse samples and distorted models. The dynamic fitting module for geological disturbance sensitivity coefficients generates spatially continuous geological disturbance sensitivity coefficients in real time, quantifying the potential risk levels of geological disturbances such as settlement, collapse, and excessive vibration at different locations during construction machinery operation. This provides a clear risk map for subsequent construction decisions, enabling construction paths and operational actions to proactively avoid high-risk areas and significantly reduce the probability of engineering accidents. As construction progresses, the accumulated disturbance event library and correction records are continuously enriched, the prediction accuracy of the Gaussian process regression model continuously improves, and the non-uniform grid is automatically refined in key areas, enabling the digital twin model to exhibit self-evolution capabilities over time. By decoupling global planning in the cloud and local optimization at the edge, decision latency is controlled to the millisecond level while ensuring global optimality, meeting the real-time control requirements of construction machinery. A geological disturbance cost reward based on digital twins is introduced into the reward function of multi-agent reinforcement learning, enabling work paths and actions to proactively avoid geologically risky areas and reduce the risk of engineering accidents such as landslides and subsidence. The Pareto optimal solution set output by the cloud-based multi-objective optimization algorithm provides a flexible scheduling selection space for the edge side, allowing dynamic adjustment of efficiency, energy consumption, and safety weights based on actual working conditions. By encoding the collaborative relationship between any two construction machines using graph neural networks, the operational conflicts and resource competition problems of large-scale construction machinery groups are effectively solved, improving overall construction efficiency. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of an intelligent scheduling system for collaborative operation of a group of construction machinery in water conservancy projects, according to an embodiment of the present invention. Figure Labels 1-Geological-mechanical coupled twin platform, 2-Cloud-edge-device collaborative network, 21-Intelligent agent node, 22-Edge computing node, 23-Cloud computing center. Detailed Implementation

[0017] The embodiments of the present invention are described in detail below with reference to the accompanying drawings. The embodiments are provided for illustrative purposes only and should not be construed as limiting the scope of the invention. The accompanying drawings are for reference and illustration only and do not constitute a limitation on the scope of protection of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of this invention.

[0018] In the description of this invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0019] This invention discloses an intelligent scheduling system for collaborative operation of a group of construction machinery in water conservancy projects, referring to... Figure 1 The system includes: The geological-mechanical coupling twin platform 1 is used to construct a two-way coupling closed loop between the geological field and the behavior of construction machinery based on the real-time working condition data of the area to be constructed, to obtain dynamic geological attribute field quantities, and to fit the dynamic geological attribute field quantities and the geological disturbance sensitivity coefficient of the area to be constructed.

[0020] In this embodiment, the geological machinery coupled twin platform 1 is built on a full-space intelligent GIS platform. This platform has the ability to fuse multi-source geological data, fine modeling of three-dimensional geological bodies, access real-time IoT sensing data, and dynamically update models, providing underlying support for the collaborative analysis of geological field and construction machinery behavior.

[0021] In a preferred embodiment of this application, the geomechanical coupled twin platform 1 includes a static geological layer and a dynamic operation influence layer. The static geological layer integrates borehole exploration data, remote sensing imagery, and a digital elevation model, and uses an implicit modeling algorithm to generate a static prior geological field for the area to be constructed. Furthermore, the static prior geological field is discretized into three-dimensional mesh elements using the finite element method. To balance computational accuracy and efficiency, unstructured hexahedral elements are used for mesh generation, with local refinement at geological interfaces, around structures, and in areas with expected high stress gradients. Each mesh element defines a prior geological attribute field quantity, serving as the initial state for model evolution. The geological attribute field quantity includes: geometric and stratigraphic distribution data, basic physical attribute data, mechanical parameter data, and initial state data.

[0022] Geometric and stratigraphic distribution data are used to define the spatial structure and topological relationships between geological bodies, identify the stratigraphic identifiers of each geological body, and establish contact relationships between strata, such as conformities, unconformities, and faults. The three-dimensional spatial coordinates of each stratigraphic interface are accurately recorded, and geostatistical interpolation methods are used to extend discrete borehole point data into continuous three-dimensional surfaces to accurately characterize complex geological structures such as stratigraphic undulations, pinch-outs, or lenses.

[0023] Basic physical property data include at least: density parameters, pore characteristics, and hydraulic parameters. Density parameters are generally taken as natural density. Pore characteristics are characterized by void ratio or porosity, which are key indicators for assessing soil compressibility and consolidation state. Hydraulic parameters include: natural water content, saturation, etc. For groundwater-bearing areas, the anisotropic tensor of permeability coefficient and the spatial distribution of groundwater level are further defined.

[0024] Mechanical parameters include at least: strength parameters, deformation parameters, constitutive model parameters, plasticity parameters, and contact characteristics. Strength parameters include total cohesion and total internal friction angle, suitable for working condition analysis during the construction phase. Deformation parameters include elastic modulus and Poisson's ratio; for soft soils or nonlinear materials, compression modulus or secant modulus are added as a reference. Constitutive model parameters include dilatation angle and failure ratio, used to describe the volume change and nonlinear strength evolution characteristics of soil during shearing. Plasticity parameters include liquid limit, plastic limit, and plasticity index. Liquid limit and plastic limit are key indicators for defining the critical state of soil moisture content, while the plasticity index is a key indicator for determining soil type and plasticity, providing a basis for evaluating construction mechanical properties. A larger plasticity index indicates stronger soil plasticity, higher compressibility, thixotropy, and sensitivity to construction disturbance, significantly impacting excavation, support, and pile foundation construction control. Contact characteristics include: soil-structure interface friction angle, soil-structure interface cohesion, and soil-structure strength reduction factor.

[0025] Initial state data should include at least the initial geostress field and the initial pore pressure field, which are used to establish the initial mechanical and seepage state of the formation and provide a benchmark for subsequent construction disturbance simulation.

[0026] In the preferred embodiment of this application, the objects included in the prior geological attribute field quantity are set according to the specific conditions of the area to be constructed, so as to reflect the geological characteristics of the strata of the area to be constructed.

[0027] The static geological layer constitutes the underlying prior knowledge base of the digital twin. The geological-mechanical coupling twin platform 1 also includes a dynamic operation influence layer, which receives and stores real-time operating condition data of construction machinery during operation through IoT data access services. The real-time operating condition data includes at least vibration response data and cutting resistance data, which are stored in the dynamic operation influence layer.

[0028] In a preferred embodiment of this application, a multi-source sensing system is deployed on the construction machinery to collect real-time operating data. For vibration response data, an accelerometer is used to acquire time-domain and frequency-domain vibration signals when the working device interacts with the geological body. The sampling frequency is set to 1000Hz–2000Hz to cover the main frequency characteristics of typical operating processes such as soil cutting, crushing, and vibratory compaction. The accelerometer is combined with a GNSS positioning module and a CAN bus data acquisition unit to achieve high-precision spatiotemporal alignment of the vibration response signal with the spatial unit and the operating status of the construction machinery. The vibration response data is used to reflect the comprehensive mechanical response of the strata under dynamic loads.

[0029] For cutting resistance data, pressure sensors are installed in the large and small chamber pipelines of the bucket cylinder and stick cylinder, while the bucket posture and cutting depth information are obtained through cylinder displacement sensors. The cylinder thrust or pull force is calculated based on the pressure difference and effective area of ​​the large and small chamber pipelines. Combined with the geometric relationship of the bucket mechanism, the cutting resistance and resistance torque of the bucket or cutter during the cutting and tunneling process are obtained. To achieve accurate verification and dynamic updating of prior geological property fields, the geomechanical coupling twin platform 1 also includes a two-way closed-loop mapping module, used to verify and update prior geological property fields based on real-time operating data, generating dynamic geological property fields. There is a direct physical connection between deformation parameters and vibration response. In elastic half-space theory or layered medium theory, the surface vibration response characteristics have a definite functional relationship with the elastic modulus and shear modulus of the medium. For example, given the excitation force, the elastic modulus can be directly inverted from the measured surface displacement amplitude. In this application, a forward physical mapping model of elastic modulus and vibration response is established based on the transfer function or frequency response function, and then the modulus parameters are inverted through an optimization algorithm. The correlation between density parameters and vibration response is relatively indirect, mainly reflected in wave impedance and inertial effects. The propagation speed of vibration waves is related to both the density and modulus of the medium; therefore, density and modulus cannot be simultaneously decoupled from the single information of vibration wave velocity. Therefore, after the modulus parameter is dynamically updated, this application uses the surface wave method to extract the Rayleigh wave dispersion curve from the vibration response data, inverts the shear wave velocity profile, and obtains the wave velocity information. Then, a vibration physical mapping model describing the relationship between density parameter, modulus, and wave velocity information is established. The mapping relationship between shear wave velocity, longitudinal wave velocity, and natural density and elastic parameters is as follows: , in, Indicates shear wave velocity. Indicates shear modulus, , Indicates the longitudinal wave velocity. Indicates bulk modulus. , Represents Poisson's ratio. Indicates the elastic modulus. Representing the natural density, after the shear modulus inversion is completed, the natural density can be obtained through shear wave velocity inversion, thus achieving refined dynamic correction of the density parameter.

[0030] Furthermore, feature extraction is performed on the cutting resistance data. The extracted resistance features include peak cutting force, specific cutting energy, and resistance fluctuation pattern. Peak cutting force reflects the ultimate strength of the formation and is suitable for identifying hard interlayers and bedrock interfaces. The resistance fluctuation pattern can characterize the heterogeneity, stratification, and abnormal working conditions such as gravel and underground obstacles of the formation. Specific cutting energy is a comprehensive indicator for evaluating the machinability of the formation. Based on a pre-constructed mechanics-physical mapping model of cutting resistance and soil strength parameters, the measured cutting resistance is inverted online into strength parameters. For the construction of the mechanics-physical mapping model, this application is based on the cutting and rock breaking mechanics theory. Under ideal cutting conditions that ignore tool wear and have a constant inclination angle, the cutting resistance experienced by the tool is jointly determined by the soil cohesion, internal friction angle, cutting thickness, cutting width, and frictional characteristics of the tool-soil interface. The process of the tool cutting the formation is simplified into a plane strain shear failure model, specifically expressed as follows: in, This represents the measured cutting resistance. The mapping function represents the mechanical-physical mapping model. Indicates total cohesion. Indicates the total internal friction angle. Indicates the cutting thickness. Indicates the cutting width. Indicates the friction angle at the knife-soil interface. This represents the tool rake angle. In field operations, tool geometry and cutting thickness are obtained in real time through equipment operation status, while the tool-soil interface friction angle and tool rake angle are predetermined through indoor triaxial tests and cutting calibration tests. After substituting the above known parameters into the mechanical-physical mapping model, the measured cutting resistance becomes a function that is only related to the total cohesion and the total internal friction angle.

[0031] Using multiple sets of measured peak cutting forces as constraints, a mechanical response residual objective function is constructed: in, This represents the objective function of the mechanical response residual. This represents the measured peak cutting force. This represents the predicted peak cutting force output by the mechanical-physical mapping model.

[0032] The objective function is solved using the least squares method to fit the parameters of the mechanical-physical mapping model. After obtaining the measured cutting resistance, the cohesion and internal friction angle can be synchronously inverted online through the mechanical-physical mapping model.

[0033] Furthermore, the multi-dimensional parameters such as stratigraphic type, natural density, elastic modulus, total cohesion, and total internal friction angle obtained from vibration response and cutting resistance inversion are synergistically integrated into the prior geological attribute field to complete the verification and dynamic update of the prior geological field quantities. Finally, a refined and real-time dynamic geological attribute field is constructed, providing reliable geological parameter support for engineering construction optimization.

[0034] In a preferred embodiment of this application, the real-time operating data also includes compaction feedback data, which is obtained by collecting the vertical acceleration time history signal of the vibratory roller through a triaxial accelerometer installed on the axle or frame of the vibratory roller, and then performing frequency domain analysis. Since the compaction feedback data is directly related to the elastic parameters of the ground, the elastic parameters of the area to be constructed can also be inverted from the compaction feedback data. This data can then be weighted and summed with the elastic parameters obtained from vibration response data at other locations to improve the accuracy of the dynamic elastic parameters.

[0035] In the preferred embodiment of this application, the operating speed, engine speed, hydraulic flow, walking posture and other real-time working condition data of the construction machinery can also be collected. Through multi-source data fusion analysis, the dynamic updating, correction and verification of other parameters of the prior geological attribute field can be realized, thereby improving the accuracy and reliability of the geological and topographic coupled digital twin model.

[0036] To extend discrete observation point data to the entire continuous field, this invention employs Gaussian process regression as a surrogate model. First, a kernel function is defined to describe the correlation between any two points in space, and its hyperparameters are adaptively estimated by maximizing the marginal log-likelihood. Then, the measured attribute values ​​at discrete spatial locations obtained through inversion are used as observation data, combined with the prior geological attribute field quantities of the entire area to be constructed, to build a Gaussian process regression model. By solving for the posterior distribution, the posterior mean at any spatial location in the entire area to be constructed is obtained, which is the corrected dynamic geological attribute field quantity. When new discrete observation data arrives, a sequential update strategy is used to recalculate the posterior distribution, achieving online dynamic correction of the geological attribute field quantity. The correction of the geomechanical coupled twin platform adopts a triggering mechanism combining fixed periods and event-driven mechanisms; for example, correction of the geomechanical coupled twin platform is performed after a new construction task begins.

[0037] The bidirectional closed-loop mapping module is also used for adaptive refinement of non-uniform grids in static geological layers. Specifically, it calculates the correction gradient for each grid cell, where the correction is defined as the difference between the dynamic geological attribute field and the prior geological attribute field. When the magnitude of the correction gradient within a grid cell exceeds a preset correction gradient threshold, it is determined that the grid cell exhibits significant spatial variations in geological attributes, belonging to a geologically complex transition zone or a critical construction area. In this case, the grid is automatically refined for that area, splitting the parent grid cell into multiple sub-grid cells to improve the model resolution of that grid cell. Conversely, in areas with uniform geological attributes and no effective data feedback, the correction gradient remains at a low level, thus maintaining a coarse grid. Through this mechanism, a dynamic balance between computational resources and model accuracy is achieved.

[0038] In a preferred embodiment of this application, the geological machinery coupled twin platform 1 maintains a real-time dynamic mapping with each construction machine, and the real-time operating data of the construction machine is synchronized to the geological machinery coupled twin platform 1 in real time, driving the continuous updating of the geological machinery coupled twin platform 1.

[0039] In a preferred embodiment of this application, the geomechanical coupled twin platform 1 further includes a dynamic fitting module for geological disturbance sensitivity coefficients, used to dynamically fit the geological disturbance sensitivity coefficients of the area to be constructed based on the dynamically updated dynamic geological attribute field. The geological disturbance sensitivity coefficients are used to quantify the potential risk level of geological disturbances occurring in different spatial units during construction machinery operation, providing a quantitative basis for the geological disturbance cost reward in the reward function of multi-agent deep reinforcement learning.

[0040] The geological disturbance sensitivity coefficient is related not only to the corrected dynamic geological property field, but also to the correction gradient output by the two-way closed-loop mapping module. Therefore, this application defines the geological disturbance sensitivity coefficient as a nonlinear mapping function of multidimensional geological parameters, as follows: in, Indicates the sensitivity coefficient to geological disturbance. The nonlinear mapping function representing the geological disturbance sensitivity coefficient. Indicates natural water content. Indicates the plasticity index. Indicates total cohesion. Indicates the total internal friction angle. Indicates the shear expansion angle. Indicates the destruction ratio, Indicates the friction angle at the soil structure interface. Indicates the interfacial cohesion of soil structures. This represents the soil structure strength reduction factor. This represents the gradient of the correction amount.

[0041] In a preferred embodiment of this application, a nonlinear mapping model for the geological disturbance sensitivity coefficient is constructed using Gaussian process regression based on the nonlinear mapping function of the geological disturbance sensitivity coefficient. Gaussian process regression can simultaneously output the predicted mean and the predicted uncertainty, and the predicted uncertainty can be further used to quantify the confidence level of the disturbance risk estimate. The training sample set is provided by a historical disturbance event database. The geological disturbance sensitivity coefficient labels are represented by the first ratio of normalized settlement to settlement threshold, the vibration exceedance multiple, the second ratio of collapse area to collapse area threshold, and the weighted average of the soil structure damage degree. The parameters of the nonlinear mapping model for the disturbance sensitivity coefficient are fitted based on the training sample set to obtain the trained nonlinear mapping model. The geological disturbance sensitivity coefficient is fitted to the entire area to be constructed using the nonlinear mapping model to generate a geological disturbance sensitivity coefficient field. After each correction of the geomechanical coupling twin platform, the geological disturbance sensitivity coefficient is fitted in a timely manner to dynamically respond to the dynamic changes in the geological layers of the area to be constructed.

[0042] As construction progresses, newly occurring disturbance events are continuously added to the training sample set. The dynamic fitting module for the geological disturbance sensitivity coefficient adopts a sparse Gaussian process or an online variational inference strategy to optimize the nonlinear mapping model of the disturbance sensitivity coefficient online, thereby achieving the adaptive evolution of the disturbance sensitivity coefficient field over time.

[0043] In a preferred embodiment of this application, the constructed geomechanical coupled twin platform 1 utilizes real-time operating data fed back during construction machinery operation and employs Gaussian process regression to correct prior geological attribute field quantities online. This allows the geomechanical coupled twin platform 1 to continuously approximate real geological conditions as construction progresses, overcoming the technical shortcomings of traditional geological exploration, such as sparse samples and distorted models. The dynamic fitting module for geological disturbance sensitivity coefficients generates spatially continuous geological disturbance sensitivity coefficients in real time, quantifying the potential risk levels of geological disturbances such as settlement, collapse, and excessive vibration at different locations during construction machinery operation. This provides a clear risk map for subsequent construction decisions, enabling construction paths and operational actions to proactively avoid high-risk areas and significantly reduce the probability of engineering accidents. As construction progresses, the accumulated disturbance event library and correction records are continuously enriched, the prediction accuracy of the Gaussian process regression model continuously improves, and the non-uniform grid is automatically refined in key areas, enabling the digital twin model to exhibit self-evolution capabilities over time.

[0044] In a preferred embodiment of this application, a cloud-edge-device collaborative network 2 is used to achieve clustered and intelligent collaborative operation of multiple construction machines. The cloud-edge-device collaborative network 2 adopts a three-layer distributed architecture, consisting of intelligent agent nodes 21, edge computing nodes 22, and a cloud computing center. The intelligent agent nodes 21, edge computing nodes 22, and cloud computing center interact with each other via communication protocols such as 5G, industrial Ethernet, or Wi-Fi 6. Specifically, the device side consists of the intelligent agent nodes 21 deployed on each construction machine, the edge side consists of the edge computing nodes 22 communicating with the intelligent agent nodes 21, and the cloud side consists of the cloud computing center communicating with the edge computing nodes 22. The intelligent agent nodes 21, edge computing nodes 22, and cloud computing center 23 form a three-level distributed collaborative architecture, enabling layered collaboration in data acquisition, local computing, edge decision-making, global scheduling, and model iteration.

[0045] The intelligent agent node 21 is integrated into the onboard controller, industrial computer, or edge gateway of each construction machine, and includes at least a data acquisition unit, an inference computing unit, an actuator control unit, and a first communication unit. The data acquisition unit includes, but is not limited to, accelerometers, pressure sensors, RTK-GNSS high-precision positioning modules, and onboard CAN bus interfaces, used to collect real-time working condition data, operating parameters, machine pose data, environmental data, and position data. The inference computing unit, using an embedded processor, is used to run lightweight models and perform filtering, feature extraction, and outlier removal on the raw sensor data. The actuator control unit receives and executes control commands from the edge computing node 22 to drive the machinery to perform actions. The first communication module supports 5G, 4G, Wi-Fi 6, Bluetooth, industrial Ethernet, and low-power wide-area network, ensuring low latency and high reliability of data uploads, and uploading processed data to the edge computing node 22 periodically or via event triggering.

[0046] Based on the construction characteristics of the area to be constructed, this application divides the area into several local jurisdictional zones. Each local jurisdictional zone deploys an edge computing node 22, responsible for the real-time optimization and collaborative control of construction machinery within that zone. The edge computing node receives data reported by all agent nodes 21 within its local jurisdictional zone, stores the data locally, and uploads it to the cloud computing center 23 as needed. Based on the global construction task scheduling scheme set issued by the cloud computing center 23, and combining local real-time data with the edge model, low-latency, high-reliability local decision-making is performed, and control commands are issued to the agent nodes 21.

[0047] Cloud computing center 23, deployed in a remote cloud data center, aggregates all data uploaded by all edge computing nodes 22, performs persistent storage, big data analysis and visualization, and, based on global construction plans, resource status and environmental information, performs global modeling and construction task planning, generating a global construction task scheduling scheme set. Cloud computing center 23 uses full historical data to train or update the geomechanical coupling twin platform 1 and distributes it to edge computing nodes 22.

[0048] The cloud-edge-device collaborative network 2 adopted in this application achieves millisecond-level real-time response compared to traditional centralized control or stand-alone operation modes, ensuring construction safety and accuracy. Edge computing nodes 22 cache a subset of the local geological-mechanical coupled twin platform, enabling independent decision-making and control even during network interruptions, ensuring operational continuity. The combination of the cloud computing center 23 and edge computing nodes 22 balances global optimization with local real-time performance, achieving efficient allocation of construction resources and dynamic adjustment of work paths and actions.

[0049] The cloud computing center 23 is used to solve the global construction task scheduling scheme based on the real-time working condition data collected by the intelligent agent node 21, combined with the operating parameters, mechanical pose data, environmental data and location data, and the geological machinery coupled twin platform 1, using a multi-objective optimization algorithm. The solved global construction task scheduling scheme set is then deployed to the corresponding edge computing node 22.

[0050] In a preferred embodiment of this application, the cloud computing center models the global construction task allocation problem as a multi-objective optimization problem. The optimization objective function includes: total completion time, total energy consumption, and geological disturbance risk, specifically represented as follows: in, This represents the global optimization objective. This indicates the total completion time. This represents the total energy consumption item. Indicates the risk of geological disturbance. The weighting coefficient for the total completion time item. This represents the weighting coefficient of the total energy consumption item. This represents the weighting coefficient of the geological disturbance risk term. .

[0051] The total completion time is represented as follows: in, Indicates the first The completion time of the construction machinery.

[0052] The formula for calculating the completion time is: in, Indicates allocation to construction machinery The set of construction tasks Construction machinery Complete the construction task Homework time, Construction machinery Travel time between different regions.

[0053] The calculation method for working time varies depending on the type of construction machinery. Taking a road roller as an example, the calculation is performed using feedback from the machine's travel speed, vibration frequency, and compaction degree collected by the intelligent agent node 21. The calculation formula is as follows: in, Indicates the first The construction area for each construction task is extracted from the geological-mechanical coupling twin platform. This indicates the number of compaction passes, determined by construction process parameters and dynamically adjusted based on real-time compaction feedback. Construction machinery The adaptive operation efficiency is dynamically adjusted based on real-time data collected by the intelligent agent nodes and combined with the geological disturbance sensitivity coefficient. This indicates the width of the machinery in operation, such as the width of a road roller's wheel. Compaction feedback dynamically adjusts the number of passes to adapt to the needs of actual working conditions.

[0054] In a preferred embodiment of this application, the efficiency of mechanical operation is dynamically adjusted based on the geological disturbance sensitivity coefficient of the construction area. Therefore, the geological-mechanical coupling twin platform 1 of this application further includes: a dynamic fitting module for mechanical operation parameters, used to dynamically adjust the rated operating efficiency of the construction machinery based on the geological disturbance sensitivity coefficient and a pre-constructed operating efficiency mapping function to obtain the adaptive operating efficiency of the construction machinery. Specifically, the edge computing unit receives high-precision location data reported by the intelligent nodes of each construction machine, and obtains the geological disturbance sensitivity coefficient of the current machine position from the geological disturbance sensitivity coefficient field cached in the geological-mechanical coupling twin platform 1 using spatial interpolation. Finally, the geological disturbance sensitivity coefficient is converted into an operating efficiency adjustment coefficient using a pre-fitted operating efficiency mapping function of the geological disturbance sensitivity coefficient and the operating efficiency adjustment coefficient of different construction machines. The rated operating efficiency of the construction machinery is dynamically adjusted using the operating efficiency adjustment coefficient to obtain the adaptive operating efficiency of the construction machinery, which is then expressed as: in, Construction machinery The work efficiency adjustment coefficient Construction machinery The rated operating efficiency.

[0055] For the travel time, simulation is performed using a geomechanical coupling twin platform 1. Considering constraints such as terrain accessibility and machine turning radius, the shortest feasible path is simulated, and the rated travel speed is determined based on the machine model and terrain. For example, the travel speed is higher on flat ground and needs to be reduced on slopes or soft ground. In a preferred embodiment of this application, the dynamic fitting module for machine operation parameters is also used to dynamically adjust the rated travel speed of the construction machinery based on the geological disturbance sensitivity coefficient and a pre-constructed travel speed mapping function, obtaining an adaptive travel speed for the construction machinery. Then, based on the adaptive travel speed and the travel distance fitted by the geomechanical coupling twin platform 1, the travel time is obtained.

[0056] In a preferred embodiment of this application, both the work efficiency mapping function and the movement speed mapping function are fitted using experimental data. For different construction machinery, the changes in the first geological disturbance sensitivity coefficient corresponding to different work efficiencies are collected experimentally. Experimental data with changes in the first geological disturbance sensitivity coefficient less than a threshold value are selected to form a first experimental dataset. Similarly, the changes in the second geological disturbance sensitivity coefficient corresponding to different movement speeds are collected experimentally. Experimental data with changes in the second geological disturbance sensitivity coefficient less than a threshold value are selected to form a second experimental dataset. Finally, the first and second experimental datasets are fitted separately to obtain the corresponding work efficiency mapping function and movement speed mapping function.

[0057] Total energy consumption is expressed as follows: in, This indicates the number of construction machines. Construction machinery Total energy consumption.

[0058] in, Construction machinery The energy consumption of the operation is related to the operation time and adaptive operation efficiency. Construction machinery The energy consumption of mobile movement is related to the distance traveled, the adaptive movement speed, and the terrain slope.

[0059] The formula for calculating the energy consumption of the operation is: in, Construction machinery Adaptive work efficiency The energy consumption function per unit time is given below, based on the construction machinery. Fitting historical construction data.

[0060] Mobile energy consumption was simulated using a geomechanical coupled twin platform 1. The formula for calculating mobile energy consumption is as follows: in, Construction machinery Adaptive movement speed The energy consumption function per unit time for construction machinery is given below. Historical movement data fitting, This indicates the additional energy consumption for climbing, calculated by the terrain slope and construction machinery of the geomechanical coupling twin platform. Historical movement data calculation.

[0061] The geological disturbance risk item is represented as follows: in, Indicates the number of construction tasks. Indicates the first Geological disturbance sensitivity coefficient of the construction area where the construction task is located Indicates the first The regional work intensity of a construction task reflects the total construction energy of all construction machinery within that area. The construction intensity of each piece of construction machinery is calculated based on its adaptive work efficiency.

[0062] The constraints of a multi-objective optimization model include at least: constraints on changes in construction task allocation, constraints on work intensity, and constraints on matching machinery tasks. The constraint on changes in construction task allocation is expressed as follows: in, This indicates the current construction task allocation cycle number. The construction task was assigned to the first... The result of the construction machinery This indicates the number of the previous construction task allocation cycle. The construction task was assigned to the first... The result of the construction machinery This indicates the maximum number of changes that can be changed in each redistribution, as preset.

[0063] The work intensity constraint is expressed as: in, Indicates the first The maximum area operation intensity for each construction task is set based on dynamic geological properties, surrounding environmental information, and historical construction data.

[0064] In a preferred embodiment of this application, the maximum area operation intensity is determined using the following method: A first maximum area operation intensity is obtained based on the sensitive stratum identification results; for example, the maximum area operation intensity for ordinary roadbeds is 5, and for soft soil areas it is 4. A second maximum area operation intensity is obtained based on surrounding environmental information; for example, the maximum area operation intensity for adjacent buildings is 3, and the maximum area operation intensity directly above historical relics / pipelines is 1 (high-intensity operations are prohibited). After determining the central area of ​​a special building, the maximum area operation intensity outside the central area is calculated by importing GIS data and calculating the distance. Historical construction data of the construction area is collected, and a third geological disturbance sensitivity coefficient is obtained based on this data; for example, areas where settlement, collapse, or pipeline damage has occurred during historical construction are marked as high-risk areas, and the corresponding maximum area operation intensity is set to 1. The minimum value among the first, second, and third maximum area operation intensities is taken as the final maximum area operation intensity for the construction area, and this value is correspondingly coupled within the geomechanical coupling twin platform 1.

[0065] Construction machinery for water conservancy projects is diverse, including specialized and general-purpose machinery. Specialized machinery can only be used for a single, specific construction task, while general-purpose machinery can be applied to various tasks. Therefore, this application defines a machinery construction task compatibility parameter, which is expressed as follows: ,in, Construction machinery Can perform construction tasks , Construction machinery Unworkable construction task Therefore, the mechanical task matching constraint in this application is expressed as: In a preferred embodiment of this application, the Non-dominated Sorting Genetic Algorithm II is used as the solution framework for the multi-objective optimization model. The Non-dominated Sorting Genetic Algorithm II searches for a set of non-dominated solutions that balance the various optimization objectives on the Pareto optimal front by simulating selection, crossover, and mutation in biological evolution.

[0066] In a preferred embodiment of this application, based on the construction technology of water conservancy projects, the various construction tasks of the entire construction process are divided into different construction task chains according to their construction correlation. Specifically, construction tasks with sequential constraints are grouped into the same construction task chain, and the priority of construction tasks is set according to their execution order. For example, if the cutting construction task can only be carried out after the cutting construction task is completed, then the cutting construction task has a higher priority than the pressing construction task. Construction tasks that can be executed simultaneously have the same priority. The execution order of construction tasks with the same priority only affects the travel distance of the machinery, energy consumption, and total completion time; any order is a feasible solution. For construction tasks that are not interrelated, they are divided into separate construction task chains. Different construction task chains can be executed simultaneously, but their execution order is affected by the allocation of construction tasks to each piece of machinery.

[0067] Based on the task chain partitioning results, the non-dominated sorting genetic algorithm II of this application employs two encoding methods: task chain encoding and mechanical task sequence encoding. For each construction task chain, task chain encoding is used. Within each construction task chain, there are partial order constraints; construction tasks with dependencies are arranged in a fixed, unchangeable order and do not require encoding. Construction tasks of the same priority can be executed in parallel; therefore, a sequence encoding is used to represent the execution order of a set of construction tasks of the same priority, including synchronous execution. An example of the encoding structure for a task chain encoded individual is as follows: A (priority 1) → {B, C} (priority 2) → D (priority 3), where A, B, C, and D all represent specific task items.

[0068] The task chain code for the entire area to be constructed is represented as follows: in, This represents the task chain code for the entire area to be constructed. Represents the construction task chain Task chain encoding individuals, This indicates the number of task chains in the entire area to be constructed. Indicates the task chain index.

[0069] In a preferred embodiment of this application, the mechanical task sequence for the entire area to be constructed is encoded as follows: in, This represents the sequence code of mechanical tasks for the entire area to be constructed. Construction machinery The set of construction tasks is defined, and the length of the construction task sets for different machines can vary. The union of all construction task sets covers all construction tasks. For construction tasks that allow multi-machine collaboration, the same construction task can appear in the construction task sets of multiple machines, indicating that it is completed collaboratively by multiple machines.

[0070] In a specific embodiment of this application, the construction tasks include: A (primary cutting), B (primary compaction), C (secondary cutting), and D (secondary compaction). (Inspection); Construction machinery includes: Excavator I (cutting), Excavator II (cutting), Road Roller I (compacting), Road Roller II (compacting), and Inspection Vehicle (inspection). A possible sequence of machinery tasks is coded as follows: The above mechanical task sequence coding indicates that: Construction task A is performed by excavator I alone; construction task C is performed by excavator I and excavator II together; construction task B is performed by roller I; and construction task D is performed by roller I and roller II together. The inspection is carried out by the testing vehicle.

[0071] Furthermore, for the machine task sequence encoding, population initialization is performed, randomly determining which construction tasks each machine will execute, generating a construction task set for each machine. For construction tasks allowing multi-machine collaboration, a collaboration coefficient is randomly determined, and then a specified number of machines are randomly selected from the compatible machine set. To avoid the construction task set of a single machine becoming too long, an upper limit is set on the number of construction tasks for each machine, controlling that the number of construction tasks for each machine does not exceed its preset capacity limit. After initialization, it is checked whether each construction task appears in the sequence of at least one compatible machine; if not, it is randomly supplemented and assigned. For the task chain encoding, population initialization is performed, randomly determining the execution order of construction tasks of the same priority level within the same construction task chain, generating a construction task execution sequence for each construction task chain.

[0072] Further, decoding and adaptive calculations are performed. The goal of decoding is to simulate the construction timeline of the construction machinery based on the set of construction tasks for each piece of construction machinery in the mechanical task sequence encoding and the execution sequence of construction tasks in the task chain encoding, to calculate the start and end times of each construction task, and then to calculate the global optimization objective value. During the decoding process, the task chain encoding information is used as external input to constrain the start time of construction tasks. The mechanical task sequence encoding does not need to explicitly express the order within the chain; it only needs to ensure that subsequent construction tasks can only begin after the preceding construction task within the chain is completed during decoding.

[0073] Furthermore, based on the geomechanical coupled twin platform 1, the construction task execution process is simulated. The initialized task chain coded individuals and mechanical task sequence coded individuals are arbitrarily combined to generate a composite individual, which constitutes the parent composite population. An event-driven simulation process is then performed on each composite individual. Specifically, a start-up time is maintained for each machine, representing the earliest time when the machine can begin executing the next construction task; a predecessor completion time is maintained for each construction task, initially set to 0. If a construction task has a predecessor construction task within the chain, the predecessor completion time records the latest completion time among all predecessor construction tasks. Based on the construction task set of each construction machine in the mechanical task sequence encoding and the construction task execution sequence of the task chain encoding, all construction machines participating in the execution of that construction task are determined.

[0074] Furthermore, for each group of individuals, based on its task chain encoding, the initial construction tasks of all construction task chains are added to the ready queue. The set of construction machinery to execute the initial construction task is determined based on the machine task sequence encoding. When the ready queue is not empty, a construction task is retrieved from the ready queue, and the earliest possible start time of the construction task is calculated. When there is no preceding construction task, assuming that a construction task can begin as long as there is at least one construction machine, the earliest possible start time is calculated as follows: in, Indicates construction task The earliest possible start time, Indicates construction task A collection of construction machinery, Construction machinery Available start time, including construction machinery The time required to move from the previous construction task location to the current construction task location.

[0075] When construction tasks When there are preceding construction tasks, the earliest possible start time can be calculated using the following formula: in, Indicates construction task The completion time of the precursor.

[0076] Furthermore, the entire process of the construction task execution is simulated in the geomechanical coupling twin platform 1, and the operation time is calculated. The end time of the construction task is then recorded, and the end time is expressed as follows: in, Indicates construction task End time, Indicates construction task Homework time.

[0077] Update the availability time of all construction machinery in the construction machinery collection, if the construction task... If there are subsequent construction tasks within the chain, update the completion time of the predecessor of that subsequent construction task and add it to the ready queue. After all construction tasks are completed, calculate the final available time of each machine, i.e., the total running time of that machine, and determine the working time, movement time, and movement distance of each construction machine. Then, based on the working time, movement time, and movement distance, calculate the objective function value of this composite individual.

[0078] Furthermore, a fast non-dominated sorting and Pareto stratification are performed. In each evolutionary iteration, based on the objective function value, a fast non-dominated sorting is performed on all combinations of individuals in the current population, dividing the population into multiple Pareto front layers. All combinations of individuals not dominated by any other individual form the first layer. After removing individuals from the first layer, the remaining combinations of individuals not dominated by any other individual form the second layer, and so on, until all combinations of individuals are stratified.

[0079] Furthermore, crowding degree calculation is performed. To maintain the diversity of the solution set and avoid premature convergence of the population to a local optimum, crowding degree distance is introduced as a density estimation index for composite individuals within the same non-dominated layer. For composite individuals in the same non-dominated layer, crowding degree represents the density of other composite individuals around that composite individual. The greater the crowding degree, the sparser the surrounding area of ​​that composite individual, and the higher the diversity contribution.

[0080] Furthermore, a tournament selection process is employed, using a binary tournament selection method to choose parental combinations from the current population for subsequent crossover and mutation operations. The selection is based on Pareto stratification results and density estimation indices. Two combinations are randomly selected from the population, and the combination with the lower non-dominated layer or greater crowding in the same layer is chosen to enter the mating pool. This process is repeated until the size of the mating pool matches the initial population size.

[0081] Furthermore, crossover and mutation operations are performed sequentially on the combined individuals in the mating pool to generate offspring populations. Since each combined individual in this application includes task chain-coded individuals and machine task sequence-coded individuals, crossover and mutation operators are designed for each type of coding, resulting in offspring task chain-coded individuals and offspring machine task sequence-coded individuals. Specifically, for each task chain-coded individual, sequential crossover is performed on construction task execution sequences of the same priority to generate offspring task chain-coded individuals. For each machine task sequence-coded individual, two machines are randomly selected, and the construction tasks in the construction task sets of the two parent individuals are swapped with a certain probability, ensuring no duplicate construction tasks within the sequence. Further mutation operations are performed: a machine is randomly selected, and a construction task is randomly removed from its construction task set and inserted into the construction task set of another compatible machine, performing insertion mutation. A machine is randomly selected, and a construction task with a compatibility parameter of 1 is randomly added to its construction task set, performing addition mutation. A machine is randomly selected, and a construction task is randomly removed from its construction task set, performing removal mutation. After crossover and mutation operations, check if each construction task appears in at least one construction task set. If not, assign it randomly. Check if the construction tasks in each construction machine's task set meet compatibility requirements. If not, remove incompatible construction tasks and randomly insert them into other compatible machines. Any combination of the offspring task chain encoded individuals and the offspring machine task sequence encoded individuals yields offspring combined individuals, which constitute the offspring combined population.

[0082] The elite preservation strategy, to prevent the optimal solution from being lost during evolution, merges the parent and offspring populations to form a temporary population. A fast non-dominated sort is applied to this temporary population, resulting in a stratified structure. Starting from the first stratum, populations are added to the next generation layer by layer until the population size exceeds the initial size. Within each stratum, populations are sorted in descending order of crowding and added sequentially until the initial population size is reached. Unselected populations are eliminated. This elite preservation strategy ensures that at least one of the best individuals from the previous generation is retained in each generation and maintains population diversity through crowding selection.

[0083] Repeat the above steps of non-dominated sorting, crowding calculation, tournament selection, crossover mutation, and elite retention until the termination condition is met. The final output is the Pareto optimal solution set, which consists of all individuals in the first layer of the non-dominated layer, serving as the global construction task scheduling scheme set. Each individual in the global construction task scheduling scheme set represents a feasible construction task scheduling scheme. A construction task scheduling scheme includes at least: the execution sequence of each construction task, the set of construction machinery participating in the execution of each construction task, and the start time, working time, and movement time of each construction machine.

[0084] In the preferred embodiment of this application, the sequential constraints within the construction task chain are decoupled from cross-chain parallel scheduling, ensuring both strict satisfaction of process dependencies and full exploitation of the optimization space for parallel operations. Mechanical task sequence encoding can naturally express multi-machine collaborative operations and mechanical serial constraints, adapting to the complex scenarios in water conservancy projects where specialized and general-purpose machinery coexist. Event-driven decoding simulation accurately reflects resource competition, movement time, and process waiting during construction, providing a reliable simulation foundation for multi-objective function calculation. Customized crossover mutation operators and repair mechanisms ensure that all individuals satisfy hard constraints such as construction task coverage, mechanical compatibility, and process dependencies during the evolution process, ensuring that the search always occurs within the feasible solution space.

[0085] Edge computing node 22, serving as the core computing unit for local collaborative scheduling, is equipped with a lightweight geomechanical coupled twin platform 1 and a pre-trained graph neural network model. Upon receiving the global construction task scheduling scheme set, it divides the global construction task scheduling scheme set into several local construction task allocation scheme sets according to the local jurisdiction area. Each edge node adapts the optimal local construction task allocation scheme for the current working condition from the local construction task allocation scheme set based on the real-time working condition data of the construction machinery within its local jurisdiction area.

[0086] In a preferred embodiment of this application, due to the limited computing resources of the edge computing node 22, it is necessary to perform multi-level filtering on the set of local construction task allocation schemes within its local jurisdiction. This application adopts a two-layer filtering method of matching rule filtering and ranking selection. In the matching rule filtering stage, terrain disturbance index matching and mechanical cluster resource status matching are set. The logical design of terrain disturbance index matching is as follows: the edge computing node 22 calculates the theoretical terrain disturbance index of all local construction task allocation schemes, compares the theoretical terrain disturbance index with the terrain disturbance index tolerance value of the local jurisdiction, and if the theoretical terrain disturbance index exceeds the terrain disturbance index tolerance value of the area, the construction task allocation scheme is eliminated.

[0087] The formula for calculating the theoretical terrain disturbance index is: in, Indicates a local jurisdiction Theoretical topographic disturbance index. Indicates a local jurisdiction Internal construction task index Indicates a local jurisdiction Internal construction tasks The geological disturbance sensitivity coefficient of the construction area Indicates a local jurisdiction Internal construction tasks The intensity of work in the construction area.

[0088] The logical design for matching the resource status of the machinery cluster is as follows: check whether the number of currently available machines, remaining power, remaining fuel, and hydraulic system status meet the minimum requirements of the scheme. If not, the construction task allocation scheme is removed.

[0089] After the filtering layer, in the ranking and selection stage, a comprehensive cost function is introduced to rank the remaining construction task allocation schemes. The comprehensive cost function is expressed as: in, Indicates a local jurisdiction The estimated total time for implementing the partial construction task allocation plan. Indicates a local jurisdiction Estimated energy consumption for implementing the partial construction task allocation plan. , and This represents the weighting coefficient, which is dynamically adjusted according to the current construction stage. For example, it is increased when expediting the construction. The value increases when environmental protection requirements are raised. The value of .

[0090] Furthermore, based on the comprehensive cost function values ​​of the remaining construction task allocation schemes, a lightweight sorting method is used to select the optimal local construction task allocation scheme.

[0091] Under the selected optimal local construction task allocation scheme, edge computing node 22, with a preset rolling time window as the period, uses a multi-agent deep reinforcement learning algorithm based on the geological machinery coupled twin platform 1 and the pre-built graph neural network model to perform real-time rolling optimization of the operation paths and actions of construction machinery within the local jurisdiction. The graph neural network model uses each construction machinery within the local jurisdiction as a set of nodes, and the communication link, spatial distance, and coupling degree between any two construction machinery as an edge set. The coupling degree of construction machinery represents the degree of coupling between the construction tasks corresponding to two construction machinery, which is determined according to the optimal local construction task allocation scheme. For example, under the optimal local construction task allocation scheme, after construction machinery I completes its construction task, only construction machinery II can execute the next construction task, so the coupling degree between construction machinery I and construction machinery II is 1; when construction machinery I completes its construction task, both construction machinery II and construction machinery III can complete the next construction task, so the coupling degree between construction machinery I and construction machinery II is 1 / 2, and the coupling degree between construction machinery I and construction machinery III is also 1 / 2. The graph neural network model is used to obtain a collaborative feature vector based on the real-time operating data of each agent node, the spatial topological relationships between construction machinery, and the construction task dependencies between construction machinery. The collaborative feature vector and the dynamic geological attribute field are then concatenated and fused to obtain a spatiotemporally coupled dynamic vector, which is used to construct the state space of the multi-agent deep reinforcement learning algorithm. Specifically, based on the geological machinery coupling twin platform 1, the dynamic geological attribute field of the current construction area is obtained. Further, the pre-built graph neural network model is used to extract the spatial topological relationships and construction task dependencies between construction machinery. Based on the real-time operating data of each agent node, the spatial topological relationships between construction machinery, and the construction task dependencies between construction machinery, high-dimensional state features of each construction machinery are extracted, and a collaborative feature vector is generated through a message passing mechanism. The collaborative feature vector is concatenated and fused with the dynamic geological attribute field to obtain the spatiotemporally coupled dynamic vector. This spatiotemporally coupled dynamic vector is input into the local scheduling model constructed using the multi-agent deep reinforcement learning algorithm to construct the state space of the multi-agent deep reinforcement learning algorithm. In the local scheduling model, each agent relies on the spatiotemporally coupled dynamic vector of local observations to make distributed decisions and generate local collaborative scheduling instructions to control the running trajectory and work intensity of construction machinery.

[0092] In a preferred embodiment of this application, during the agent training phase, a multi-agent proximal policy optimization algorithm is used for collaborative decision-making. The reward function of the multi-agent proximal policy optimization algorithm is designed as follows: in, This indicates a reward for the progress of the construction task. This represents a work efficiency bonus, calculated based on the amount of work completed per unit of time. This indicates rewards for collaborative work and penalties for conflicts and waiting times between machines. This indicates the cost and reward for geological disturbance. , , and This represents the weight coefficient of each reward item, satisfying... .

[0093] Construction task progress reward is defined as: in, Indicates time Completed construction volume This indicates the total volume of construction work that needs to be completed.

[0094] The work efficiency bonus is defined as: in, This indicates the actual construction volume per unit time. This represents the construction volume per unit time, which is fitted using historical data.

[0095] The definition of collaborative task reward is: in, Indicates time The cumulative duration of path conflicts and work area conflicts involving construction machinery within a local jurisdiction. Indicates time The cumulative time that construction machinery spends waiting for other construction machinery or when resources are idle within a local jurisdiction. This represents the conflict penalty coefficient, used to control the penalty weight for conflicts. This represents the waiting penalty coefficient, used to control the penalty weight for waiting. The longer the conflict and waiting time, the lower the reward value for job collaboration.

[0096] The geological disturbance cost reward is defined as: in, Construction machinery At any moment The intensity of the operation is characterized by vibration amplitude and cutting force. This indicates the length of the integration time window, i.e., a rolling time window. Construction machinery Current location Construction machinery Current location The geological disturbance sensitivity coefficient. The geological disturbance cost, reward, and penalty for overwork enable the optimization of work paths and actions to adapt to changes in geological conditions under the constraints of the global construction task scheduling scheme.

[0097] Edge computing node 22 employs a model predictive control framework for rolling optimization within fixed time windows. At the beginning of each fixed time window, based on the current dynamic geological attribute field of the geomechanical coupled twin platform 1 and the collaborative feature vector output by the graph neural network model, the operating path and sequence of actions for the construction machinery within the next time window are planned. The actions of the first time step are executed, and in the next fixed time window, the plans are re-planned to achieve adaptive rolling optimization.

[0098] In a preferred embodiment of this application, the cloud computing center 23 periodically and event-triggeredly resolves the global construction task scheduling scheme set and distributes the new Pareto front to the edge computing node 22, updating its decision constraint boundaries. The edge computing node 22 uploads real-time working condition data during execution to the cloud computing center 23 for offline optimization and parameter adjustment of the multi-objective optimization algorithm model.

[0099] In the preferred embodiment of this application, by decoupling global planning in the cloud and local optimization at the edge, the decision latency is controlled to the millisecond level while ensuring global optimality, thus meeting the real-time control requirements of construction machinery. A geological disturbance cost reward based on digital twins is introduced into the reward function of multi-agent reinforcement learning, enabling work paths and actions to proactively avoid geologically risky areas and reduce the risk of engineering accidents such as landslides and subsidence. The Pareto optimal solution set output by the cloud-based multi-objective optimization algorithm provides a flexible scheduling selection space for the edge side, allowing for dynamic adjustment of efficiency, energy consumption, and safety weights based on actual working conditions. By encoding the collaborative relationship between any two construction machines using a graph neural network, the operational conflicts and resource competition problems of large-scale construction machinery groups are effectively solved, improving overall construction efficiency.

[0100] This embodiment provides an intelligent scheduling system for collaborative operation of a group of construction machinery in water conservancy projects. It addresses the technical problem of constructing an intelligent scheduling system capable of real-time sensing of geological changes and dynamic optimization of collaborative machinery operation. The system includes: a geological-mechanical coupling twin platform, used to construct a two-way coupled closed loop between the geological field and the behavior of the construction machinery based on real-time operating data of the area to be constructed, obtaining dynamic geological attribute field quantities, and fitting the geological disturbance sensitivity coefficient of the area to be constructed based on these dynamic geological attribute field quantities; and a cloud-edge-device collaborative network, including intelligent agent nodes deployed on each construction machine, edge computing nodes communicating with the intelligent agent nodes, and a cloud computing center communicating with the edge computing nodes. The intelligent agent nodes are used to collect real-time operating data; the cloud computing center is used to perform multi-objective optimization based on the real-time operating data, dynamic geological attribute field quantities, and geological disturbance sensitivity coefficient. A global construction task scheduling scheme is obtained by solving a global construction task scheduling scheme set using a computational algorithm. Edge computing nodes are used to select the optimal local construction task allocation scheme from the global scheme set. Based on the optimal local allocation scheme, dynamic geological attribute fields, and geological disturbance sensitivity coefficients, a multi-agent deep reinforcement learning algorithm is used to perform real-time rolling optimization of the collaborative operation of construction machinery within the local jurisdiction of the edge computing nodes, generating local collaborative scheduling instructions for each piece of construction machinery. The multi-agent deep reinforcement learning algorithm is configured to introduce a geological disturbance cost reward into the reward function, which is fitted based on the geological disturbance sensitivity coefficient. The intelligent scheduling system for collaborative operation of hydraulic engineering construction machinery groups provided in this application constructs a geological machinery coupled twin platform. This platform utilizes real-time operating data fed back during construction machinery operation and employs Gaussian process regression to correct prior geological attribute fields online. This allows the geological machinery coupled twin platform to continuously approximate real geological conditions as construction progresses, overcoming the technical shortcomings of sparse samples and distorted models in traditional geological exploration. The dynamic fitting module for geological disturbance sensitivity coefficients generates spatially continuous geological disturbance sensitivity coefficients in real time, quantifying the potential risk levels of geological disturbances such as settlement, collapse, and excessive vibration at different locations during construction machinery operation. This provides a clear risk map for subsequent construction decisions, enabling construction paths and operational actions to proactively avoid high-risk areas and significantly reduce the probability of engineering accidents. As construction progresses, the accumulated disturbance event library and correction records are continuously enriched, the prediction accuracy of the Gaussian process regression model continues to improve, and the non-uniform grid is automatically refined in key areas, enabling the digital twin model to exhibit self-evolution capabilities over time. By decoupling global planning in the cloud and local optimization at the edge, decision latency is controlled to the millisecond level while ensuring global optimality, meeting the real-time control requirements of construction machinery. A geological disturbance cost reward based on digital twins is introduced into the reward function of multi-agent reinforcement learning, enabling operational paths and actions to proactively avoid geologically risky areas and reduce the risk of engineering accidents such as collapse and settlement.The Pareto optimal solution set output by the cloud-based multi-objective optimization algorithm provides a flexible scheduling selection space for the edge side, which can dynamically adjust the efficiency, energy consumption, and safety weights according to actual working conditions. By encoding the collaborative relationship between any two construction machines through graph neural networks, the problem of operational conflicts and resource competition in large-scale construction machine groups is effectively solved, thereby improving overall construction efficiency.

[0101] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the various embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as these combinations of technical features do not contradict each other, they should be considered within the scope of this specification.

[0102] The above-described embodiments are merely preferred embodiments of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A smart scheduling system for collaborative operation of a group of construction machinery in water conservancy projects, characterized in that, The system includes: The geomechanical coupling twin platform is used to construct a two-way coupling closed loop between the geological field and the behavior of construction machinery based on real-time working condition data of the area to be constructed, to obtain dynamic geological attribute field quantities, and to fit the geological disturbance sensitivity coefficient of the area to be constructed based on the dynamic geological attribute field quantities. The cloud-edge-device collaborative network includes intelligent agent nodes deployed on each construction machine, edge computing nodes that are communicatively connected to the intelligent agent nodes, and cloud computing centers that are communicatively connected to the edge computing nodes. The intelligent agent node is used to collect the real-time operating condition data. The cloud computing center is used to solve the global construction task scheduling scheme by using a multi-objective optimization algorithm based on the real-time working condition data, the dynamic geological attribute field quantity and the geological disturbance sensitivity coefficient, so as to obtain a global construction task scheduling scheme set. The edge computing node is used to select the optimal local construction task allocation scheme from the global construction task scheduling scheme set, and based on the optimal local construction task allocation scheme, the dynamic geological attribute field quantity, and the geological disturbance sensitivity coefficient, a multi-agent deep reinforcement learning algorithm is used to perform real-time rolling optimization of the collaborative operation of construction machinery within the local jurisdiction of the edge computing node, generating local collaborative scheduling instructions to schedule each construction machine; the multi-agent deep reinforcement learning algorithm is configured to introduce a geological disturbance cost reward into the reward function, and the geological disturbance cost reward is fitted based on the geological disturbance sensitivity coefficient; The geomechanical coupling twin platform includes: a static geological layer, a dynamic operation influence layer, a two-way closed-loop mapping module, and a dynamic fitting module for geological disturbance sensitivity coefficients; The static geological layer is used to store the a priori geological attribute field quantity constructed based on historical geological exploration data. The a priori geological attribute field quantity includes at least: geometric and stratigraphic distribution data, basic physical attribute data, mechanical parameter data, and initial state data. The dynamic operation influence layer is used to receive and store the real-time operating condition data in real time. The real-time operating condition data includes at least: vibration response data, cutting resistance data, and compaction feedback data. The bidirectional closed-loop mapping module is used to correct the prior geological attribute field quantity based on the real-time working condition data and through a pre-constructed physical mapping model to obtain the dynamic geological attribute field quantity. The dynamic fitting module for the geological disturbance sensitivity coefficient is used to dynamically fit the geological disturbance sensitivity coefficient of the area to be constructed based on the dynamic geological attribute field quantity and through a pre-constructed nonlinear mapping model of the disturbance sensitivity coefficient. The geological disturbance sensitivity coefficient is set to quantify the potential risk of geological disturbance in the area to be constructed when the construction machinery is operating.

2. The intelligent scheduling system for collaborative operation of water conservancy engineering construction machinery as described in claim 1, characterized in that, The bidirectional closed-loop mapping module is further configured to: calculate the correction gradient of different grid units of the static geological layer based on the dynamic geological attribute field quantity and the prior geological attribute field quantity, and perform adaptive refinement processing on the grid units based on the correction gradient quantity to improve the resolution of the grid units.

3. The intelligent scheduling system for collaborative operation of water conservancy engineering construction machinery as described in claim 1, characterized in that, The geomechanical coupling twin platform also includes: The dynamic fitting module for mechanical operation parameters is used to dynamically adjust the rated operation efficiency of the construction machinery according to the geological disturbance sensitivity coefficient and the pre-constructed operation efficiency mapping function to obtain the adaptive operation efficiency of the construction machinery; and to dynamically adjust the rated movement speed of the construction machinery according to the geological disturbance sensitivity coefficient and the pre-constructed movement speed mapping function to obtain the adaptive movement speed of the construction machinery.

4. The intelligent scheduling system for collaborative operation of water conservancy engineering construction machinery as described in claim 1, characterized in that, The objective function of the multi-objective optimization algorithm includes: total completion time, total energy consumption, and geological disturbance risk. The geological disturbance risk is set to fit the geological disturbance sensitivity coefficient of the construction area where the construction task is located and the regional operation intensity of the construction area where the construction task is located.

5. The intelligent scheduling system for collaborative operation of water conservancy engineering construction machinery as described in claim 1, characterized in that, The constraints of the multi-objective optimization algorithm include at least: construction task allocation change constraints, work intensity constraints, and machinery task matching constraints. The construction task allocation change constraints are set such that the amount of construction task allocation change between any two adjacent construction task allocation cycles is less than or equal to a preset maximum change number. The work intensity constraints are set such that the work intensity of the construction area is less than or equal to a preset maximum area work intensity. The machinery task matching constraints are set such that the construction machinery should be matched with the construction task.

6. The intelligent scheduling system for collaborative operation of water conservancy engineering construction machinery as described in claim 1, characterized in that, The multi-objective optimization algorithm uses the non-dominated sorting genetic algorithm II. In the solution of the non-dominated sorting genetic algorithm II, task chain encoding and mechanical task sequence encoding are used. The construction tasks that can be executed simultaneously within the task chain encoding are represented by sequence encoding to indicate the execution order. The mechanical task sequence encoding reflects the set of construction tasks for each construction machine. By combining the first entity corresponding to any of the task chain codes and the second entity corresponding to any of the mechanical task sequence codes, a feasible construction task scheduling scheme is obtained.

7. The intelligent scheduling system for collaborative operation of water conservancy engineering construction machinery as described in claim 1, characterized in that, The reward function of the multi-agent deep reinforcement learning algorithm also includes: construction task progress reward, operation efficiency reward and operation collaboration reward; the construction task progress reward is set to represent the proportion of completed construction tasks, the operation collaboration reward is set to penalize operation conflicts and operation waiting, and the geological disturbance cost reward is set to be fitted according to the operation intensity of the construction machinery and the geological disturbance sensitivity coefficient of the current location of the construction machinery.

8. The intelligent scheduling system for collaborative operation of water conservancy engineering construction machinery as described in claim 1, characterized in that, The edge computing nodes include: A graph neural network model is used to obtain a collaborative feature vector based on the real-time working condition data of each agent node, the spatial topological relationship between the construction machinery, and the construction task dependency relationship between the construction machinery; the collaborative feature vector and the dynamic geological attribute field are spliced ​​and fused to obtain a spatiotemporal coupled dynamic vector, which is set as the state space for constructing the multi-agent deep reinforcement learning algorithm.

9. The intelligent scheduling system for collaborative operation of water conservancy engineering construction machinery as described in claim 8, characterized in that, The graph neural network model uses each of the construction machines in the local jurisdiction as a set of nodes, and the communication link, spatial distance and coupling degree between any two construction machines as a set of edges. The coupling degree of the construction machines represents the degree of coupling between the construction tasks corresponding to the construction machines.

Images