A hierarchical scheduling optimization method and platform for multi-resource dynamic balancing of engineering projects

Abstract

The application discloses a layered scheduling optimization method and platform for multi-resource dynamic balancing of engineering projects, and belongs to the technical field of engineering project scheduling optimization. The method comprises the following steps: an engineering project data model containing an activity directed acyclic graph, multi-type resource requirements and milestone constraints is established; a resource coupling coefficient is calculated, and a feasible time window in the form of a multi-section closed interval union is generated for each activity; a resource pool quota is allocated among projects according to a resource urgency index at a strategic layer; adaptive rolling rescheduling is started according to four types of trigger conditions at an operation layer; and a cross-layer bidirectional closed loop is formed through step-by-step feedback. The platform comprises the following modules: a multi-source data acquisition module, a biological meteorological window management module, a resource demand prediction module, a three-layer scheduling engine module and the like. The application can simultaneously realize single-resource balancing and cross-resource collaboration, so that the scheduling result has biological meteorological sensitivity, and can quickly respond when a disturbance arrives, thereby relieving on-site work, improving resource pool utilization and reducing the deviation of the construction period.

Description

Technical Field

[0001] This invention belongs to the field of engineering project scheduling and optimization technology, specifically involving a hierarchical scheduling optimization method and platform for dynamic balancing of multiple resources in engineering projects. Background Technology

[0002] Engineering projects typically involve diverse and heterogeneous resources such as manpower, machinery, raw materials, seedlings and flowers, and hydropower. Their scheduling requires multi-objective collaborative optimization while satisfying constraints on activity sequence, resource capacity, and project timelines. This falls under the resource-constrained project scheduling problem and has been proven to be NP-hard. Existing engineering project scheduling technologies mainly fall into several categories. The first category is heuristic resource balancing methods, represented by the minimum moment method, which achieves balance for a single resource by minimizing the variance or moments of resource usage over time. This approach is widely used in various commercial project management software. The second category is metaheuristic optimization algorithms, represented by genetic algorithms, simulated annealing, ant colony optimization, and particle swarm optimization, used to find near-optimal solutions to large-scale resource-constrained project scheduling problems. The third category is rolling scheduling strategies, which transform dynamic time-domain scheduling problems into a sequence of finite-time-domain subproblems and reschedule according to a fixed window length to cope with dynamic disturbances. The fourth category is hierarchical, multi-layered scheduling architectures, often used in cloud computing resource scheduling, manufacturing plant production scheduling, and railway scheduling.

[0003] However, when the aforementioned existing technologies are applied to engineering projects, especially in scenarios such as ecological restoration, municipal landscaping, building construction, and water conservancy and hydropower, several technical problems remain that have not been effectively resolved. First, existing resource balancing methods only focus on minimizing the variance or moment of each resource on the time axis. They lack mathematical characterization of the asynchronous fluctuations between a large number of resource pairs that must be coordinated in engineering projects, such as concrete and transport vehicles, tower cranes and crane operators, seedlings and irrigation water, and personnel and machinery. This results in scheduling results that appear balanced in a single resource dimension but are poor in cross-resource coordination, which in turn manifests as idle work, equipment running idle, and resource waiting at the field level. Secondly, activities in landscaping and ecological restoration projects exhibit significant seasonality and weather dependence. For example, tree transplanting has two transplanting windows in spring and autumn; concrete curing requires specific temperatures; sowing and germination require specific temperature and humidity conditions; and rainfall is sensitive to earthwork and foundation construction. Existing scheduling methods often assume feasible start times for activities as a single continuous interval, failing to handle feasible time windows that are determined by both phenological segments of seedlings and meteorological rules. This results in insufficient executability of the scheduling results during the post-execution phase, requiring extensive manual adjustments. Thirdly, existing hierarchical scheduling architectures often involve unidirectional dispatch from upper to lower layers, with lower layers simply reporting their status after execution. The lack of a cross-layer resource reassignment mechanism makes it difficult to resolve resource conflicts between projects when multiple projects are running concurrently, leading to uneven overall resource pool utilization. Finally, existing rolling rescheduling methods mostly use a fixed window length. When encountering large disturbances, the window is too small to restore the stability of the plan, while the window is too large during the stable period, which causes unnecessary recomputation overhead. There is no mechanism to adaptively adjust the window length based on the disturbance magnitude and the remaining critical path slack, making it difficult to balance disturbance response speed and rescheduling cost.

[0004] Based on the above problems, there is an urgent need to design an engineering project scheduling optimization method and its supporting implementation platform that can simultaneously address multi-resource collaborative equilibrium modeling, biometeorological time window constraint modeling, cross-layer resource elastic reallocation, and adaptive disturbance response. Summary of the Invention

[0005] To address the problems existing in the background technology, this invention provides a hierarchical scheduling optimization method for dynamic balancing of multiple resources in engineering projects, comprising: S1: Establish an engineering project data model that includes an activity-directed acyclic graph, multiple types of resource requirements and capacity, and milestone constraints; S2: Calculate the resource pair coupling coefficients and generate feasible time windows in the form of multiple closed interval unions for each activity; S3: The strategic layer allocates resource pool quotas among projects based on the resource urgency index; S4: Under the constraints of the quota and feasible time window, the tactical layer schedules each activity with the goal of coupling multiple resource balance indices, and outputs the baseline schedule plan and resource time distribution. S5: During the execution of the operation layer, a rolling rescheduling is triggered according to preset conditions. The length of the rescheduling window is adaptively determined based on the next milestone and the preset upper and lower limits of the window, and the activities within the window are rearranged. S6: Execution data is fed back level by level between the operational layer, tactical layer, and strategic layer. The strategic layer updates the quota based on the feedback data, forming a cross-layer bidirectional closed loop.

[0006] Furthermore, step S1 includes: S11: Extract an activity list from construction organization design documents, building information models, or historical project archives, determine the number of activities, the duration of each activity, and the order of activities, and construct a directed acyclic graph of the activities; S12: Determine the quantity of various resources required per unit time during the execution period for each activity according to its process requirements, and form an activity and resource demand matrix; S13: Enter the total capacity limit of various resources in the resource pool, the project duration limit of each project, and the list of project milestone nodes; S14: After normalizing the above data, store it in the project database to obtain the project data model for use in steps S2 to S6.

[0007] Furthermore, step S2 includes: S21: For each of the two types of resources on the directed acyclic graph of the activities and Scan each activity and calculate the resources required simultaneously. With resources Number of activities And the resources needed or resources Total number of activities The resource-pair coupling coefficient is calculated using the following formula. : ; in: For resources With resources The coupling coefficient is dimensionless. To simultaneously require resources With resources The number of activities; For the need of resources or resources The total number of activities; / indicates division; and Indexing for different resource types; S22: Each Write a symmetric coupling matrix The coupling matrix Set the diagonal element to 1, and we get The dimensional coupling matrix is ​​used in step S4, where For the number of resource types; S23: Generate the feasible time windows for each activity according to the following sub-steps. : S231: For each activity Based on its process category, at least one construction environment constraint rule is retrieved from the biological and meteorological rule bases. The construction environment constraint rule specifies the feasible value range of at least one variable among temperature, humidity, wind speed, precipitation, or seedling phenological segment. S232: Using meteorological forecast data and seedling phenology calendar as input, the feasibility of each day in the scheduling time domain is determined by substituting the construction environment constraint rules into the data, and a binary label of feasible / infeasible for each day is obtained. S233: Merge consecutively marked as feasible dates into a closed interval segment. Sequentially number each consecutive feasible segment within the scheduling time domain. And use the union of the segments as the activity according to the following formula. The feasible time window : ; in: For the event The feasible time windows, a set in days; For the event The number of feasible window segments; , activities respectively No. The start and end dates of the segment window; Represents the union of sets; Indicates a closed interval; For activity indexing; For window segment index; S234: The coupling matrix and the feasible time windows for each activity. Store the data in the project database to obtain the coupling coefficient data and time window constraint data for use in step S4.

[0008] Furthermore, step S3 includes: S31: At each quota allocation trigger moment For each project and various resources The predicted demand for the next scheduling window is obtained from resource demand forecasting. Read the project from the execution log Resources As of Cumulative consumption at any time and resources in the resource pool As of Total cumulative consumption at any moment And read the current quota from the quota table. ; S32: Calculate the resource urgency index using the following formula. : ; in: For the project exist Resources at all times The urgency index is dimensionless; To predict demand, units and resources same; For the resources in the resource pool Total capacity limit, units and resources same; For resources in the resource pool As of Total cumulative consumption at any given time, in units and resources same; To prevent the denominator from being zero, a pre-defined positive small constant is used; This indicates taking the larger value; Index the project; Index for resource types; For time indexing; S33: For various resources The resources in the resource pool are calculated using the following formula. Total capacity The quotas were redistributed among the projects to obtain new allocations. : ; in: To reassign to the project resources Quotas, Units and Resources same; As defined in S32; Calculated according to S32; This represents the summation of all projects participating in the same resource pool; S34: Transfer the new quota The tactical level is issued and written into the quota table, replacing the original... This serves as a hard constraint for the activity scheduling in step S4.

[0009] Furthermore, step S4 employs a hybrid genetic algorithm (H-GA) to solve the problem, including: S41: Initialize the population. Each chromosome consists of two segments, the first segment being the activity priority sequence. The second section is for selecting gene sequences via a window. ,in: For the first The priority value of each activity is dimensionless. For the event The selected feasible time window index, , Obtained according to S233; The number of activities in the project data model; S42: Perform Serial Progress Generation Scheme (SSGS) decoding on each chromosome, according to the stated activity priority sequence. The start time of each activity is determined sequentially using a directed acyclic graph. And make it satisfy the time window constraint shown in the following formula: ; in: For the event The beginning moment; For the event The duration; For the event The window index used in this decoding was initially taken from the chromosome gene. It only applies to this decoding process and does not modify the chromosome gene sequence; , activities respectively No. The start and end dates of the segment window; S43: When the time window constraint described in S42 cannot be satisfied, the following decoding and repair rules shall be used to determine the cause. and And without modifying the chromosome gene sequence : (a) If the selected window segment Inside, Existence makes The earliest established Then Set to this earliest value, Set as ; (b) Otherwise in Search by segment index In the subsequent window segment, find the first one that satisfies... window segment ,Will Set as , Set as ; (c) If neither (a) nor (b) is feasible, then mark the activity as an infeasible solution and apply a preset penalty value to the fitness of the chromosome. S44: Perform a serial scan repair on solutions that violate resource capacity constraints, ensuring that activities executed simultaneously at any given time affect various resource types. The total amount occupied does not exceed the corresponding quota. ;in: The new quota calculated for S33; S45: Using the aforementioned coupled multi-resource equilibrium index As a fitness function, each chromosome is evaluated using the following formula: ; in: The multi-resource equilibrium index is dimensionless. for Time Resources Actual input, units and resources same; For resources The time average within the scheduling time domain, according to Calculation, Units and Resources same; This represents the total length of the scheduling time domain; To prevent the denominator from being zero, a pre-defined positive small constant is used; These are the weighting coefficients of the coupling terms, preset positive constants, and dimensionless. Calculated according to S21; This indicates taking the larger value; Indicates the time domain of scheduling Summation at each discrete time point; This represents the summation of all types of resources in the set of resource types; Indicates that for each satisfied Resource type Summation; For time indexing; and Index for resource types; S46: Perform selection, crossover, and mutation operations to generate the next generation population. After the number of iterations reaches a preset upper limit, the population will be... The sequence of activity start times and resource inputs corresponding to the individual with the smallest value are used as the baseline schedule and resource time distribution output.

[0010] Furthermore, the hybrid genetic algorithm H-GA employs a twin-population co-evolutionary structure for solving the problem. This twin-population co-evolutionary structure operates internally within the tactical layer solution process, distinguishing it from the macroscopic hierarchical architecture comprised of the strategic, tactical, and operational layers. S47: Within the tactical layer solution process, maintain the quota recommendation subpopulation POP_Q and the scheduling subpopulation POP_S; the quota recommendation vector for each chromosome coding item and resource pair in POP_Q. In POP_S, each chromosome encodes an activity priority sequence according to the two-segment coding structure defined in S41. Selecting gene sequences with a window ;in: Items encoded for the POP_Q chromosome Resources Recommended quota values, units and resources same; S48: After each preset migration cycle In each generation, the elite chromosome with the best fitness is selected from POP_Q, and its encoded quota proposal vector is used. As a hard constraint for each chromosome in POP_S to execute steps S42 to S46, each chromosome in POP_S re-enters the decoding and fitness evaluation process under this hard constraint; where: For migration cycle; S49: Select the coupling multi-resource balance index from POP_S Extract the peak resource requirements of each item and resource pair from the smallest elite chromosome. Calculate the fitness correction term for each chromosome in POP_Q using the following formula. : ; in: This is the fitness correction term for the POP_Q chromosome, and it is dimensionless. Items extracted from POP_S elite chromosomes Resources Peak demand in the current phase, in units and resources same; As defined in S47; The first position on the right side of the equals sign indicates negation, and the area inside the parentheses indicates subtraction; This indicates taking the larger value, with a lower bound of 0; This indicates the summation of all projects and resources; The original fitness of the POP_Q chromosome is superimposed, causing POP_Q to evolve towards a direction of increased quota recommendations for peak demand items and resources, thereby enabling the twin populations to co-evolve through bidirectional migration between S48 and S49.

[0011] Furthermore, step S5 includes: S51: The operation layer collects actual on-site progress, resource consumption and meteorological data on a daily basis, and calculates the deviation of each activity from the baseline schedule in real time; S52: Determine whether any of the following four types of triggering conditions are true: (i) Periodic triggering: The time interval since the last rescheduling reaches the preset rescheduling period. ; (ii) Resource gap trigger: There is a project With resources This makes the resource gap more than Exceeding the preset resource gap ratio threshold ; (iii) Critical path slack triggering: Remaining critical path slack Not greater than zero; (iv) External disturbance trigger: Preset external disturbance event is detected; in: The preset rescheduling period; , , As defined in S31; The preset resource gap ratio threshold is dimensionless. for Remaining critical path slack at each time step; To constrain the contract period; The project completion date is calculated based on the current progress using the critical path method. This indicates taking the larger value; if any condition is met, proceed to step S53. S53: Calculate the rescheduling window length using the following formula. : ; in: for The rescheduling window length is specified at any given time. The planned completion date for the next project milestone; This is the preset lower limit for window length; This is the preset maximum window length. ,in and These represent taking the smaller value and taking the larger value, respectively. Indexed by the current time; S54: Time interval The start time of activities other than those already committed remains unchanged as a constraint, and the start time falls within the interval. The activities within are shown in the mixed objective function as follows: Solve again: ; in: The objective function value for the rescheduling phase is dimensionless. Calculated according to S45; The deviation from the sensitivity coefficient is a preset positive constant, dimensionless; For the set of time intervals corresponding to the rescheduling window, in units; Indicates the time at which the beginning falls into Summing up all activities; For rescheduled activities The beginning moment; Activities in the baseline plan The start time; replace the interval in the original plan with the solution result. At the start of the internal activity, the updated progress plan is obtained.

[0012] Furthermore, step S6 includes: S61: The operation layer uploads the actual start time, actual end time, actual consumption of various resources, and on-site disturbance events of each activity to the tactical layer on a daily cycle. S62: The tactical layer updates the predicted demand based on the uploaded data. And extract the items in the next scheduling window. Resources Peak actual input The strategic layer mentioned above is reported; among which: For the project Resources Peak demand within this scheduling window, in units and resources same; S63: The strategic layer receives the report from the tactical layer. With the updated The revised forecast demand is constructed using the following formula. and with Alternative Recalculate the urgency index according to S32 to S34. With the new quota : ; in: The revised forecast demand, in units and resources same; As defined in S31; As defined in S62; This indicates taking the larger value; this step ensures that peak demand items receive a corresponding increase in quota in the next allocation cycle, while the proportional allocation formula in S33 guarantees this. That is, the total capacity constraint of the resource pool is conserved and not violated; S64: The strategic layer will update the new quota. The tactical layer is then issued, triggering it to perform step S4, a local rescheduling, on the affected items, thus creating a two-way closed loop between the upper and lower layers.

[0013] This invention provides a hierarchical scheduling optimization platform for engineering projects, including a multi-source data acquisition module interconnected by a data bus, a biological and meteorological window management module, a resource demand prediction module, a three-layer scheduling engine module, a disturbance response module, and a digital twin visualization module; The outputs of the biological and meteorological window management module and the resource demand prediction module are all connected to the input of the three-layer scheduling engine module. The three-layer scheduling engine module includes a cascaded strategic layer solver, tactical layer solver, and operational layer scheduler. There is a feedback channel between the three layers from the lower layer to the upper layer. The output of the three-layer scheduling engine module is connected to the digital twin visualization module and the field operation terminal. The output of the disturbance response module is connected to the input of the three-layer scheduling engine module.

[0014] In a preferred embodiment, the multi-source data acquisition module includes an IoT gateway and several edge nodes. The edge nodes are electrically connected to a weather station, a soil moisture sensor, a tower crane load sensor, a construction vehicle positioning terminal, a personnel positioning card, and a material identification reader. The IoT gateway is connected to the data bus and performs format normalization processing on the raw data before sending it to other modules. The biological and meteorological window management module includes a seedling phenology database, a construction process temperature and humidity rule database, and a meteorological forecast interface. The seedling phenology database and the construction process temperature and humidity rule database are both connected to the meteorological forecast interface. The output end of the biological and meteorological window management module outputs feasible time window data. The resource demand forecasting module is connected to a historical project database and an activity template library. The output of the resource demand forecasting module outputs the resource demand forecast data to the input of the strategic layer solver. In the three-layer scheduling engine module, the strategic layer solver, the tactical layer solver, and the operational layer scheduler communicate bidirectionally via a message bus. The data categories transmitted by the message bus include: quota data issued by the strategic layer solver to the tactical layer solver; activity start and end time instruction data issued by the tactical layer solver to the operational layer scheduler; actual activity start and end time data and actual resource consumption data reported by the operational layer scheduler to the tactical layer solver; and peak resource demand data reported by the tactical layer solver to the strategic layer solver. The strategic layer solver is used for resource pool quota allocation, the tactical layer solver is used for activity scheduling, and the operational layer scheduler is used for daily job instruction generation and execution monitoring. The disturbance response module includes a trigger condition discrimination unit and a rescheduling and invocation unit. The input of the trigger condition discrimination unit is connected to the output of the multi-source data acquisition module, and the output of the rescheduling and invocation unit is connected to the input of the operation layer scheduler. The digital twin visualization module uses a fusion of a 3D geographic information system view and a building information model view as the base map, and overlays and displays Gantt layers, resource histogram layers, coupled multi-resource balance index curve layers, urgency index heatmap layers, and time window conflict alarm layers. It also has an interactive interface for project managers to access and drill down.

[0015] The beneficial effects achieved by this invention are as follows: First, this invention introduces a coupled multi-resource balance index at the tactical level as the optimization objective for activity scheduling. This index is composed of the normalized variance of each resource relative to its own time mean and the normalized sum of squared deviations of resource pairs weighted by their coupling coefficients. This allows the scheduling process to simultaneously consider the stability of individual resources on the time axis and the synchronicity of strongly coupled resource pairs in the time distribution. The resource pair coupling coefficient uses the co-occurrence tightness of two types of resources in the project activity set as an objective measure, incorporating it as a weight source into the objective function. This allows the hybrid genetic algorithm to preferentially select for the synchronicity of strongly coupled resource pairs in the search direction. As demonstrated in the embodiments, the resource time distribution obtained by the method of this invention is generally stable across various resource dimensions, and the peaks and troughs of strongly coupled resource pairs exhibit near-synchronous fluctuations on the time axis. This fundamentally overcomes the asynchronous phenomenon of strongly coupled resources caused by the traditional minimum moment scheduling method, which only considers the time-series fluctuations of individual resources and ignores the synergy of resource pairs. Consequently, it alleviates the tendency for on-site idle work, equipment idling, and resource waiting.

[0016] Secondly, this invention explicitly incorporates activity-level biometeorological constraints into the scheduling model as feasible time windows in the form of multi-segment closed interval unions. It also utilizes a dual-segment chromosome structure jointly encoded by window selection genes and activity priority sequences, along with complete decoding and repair rules, to participate in the genetic algorithm solution. This ensures that the scheduling results naturally follow the executable time range determined by seedling phenological segments, construction process temperature and humidity rules, and meteorological forecast data, avoiding rework and quality losses caused by neglecting external environmental factors in traditional scheduling. Furthermore, this invention implements a hybrid genetic algorithm within the tactical layer solution process using a dual-population co-evolutionary structure. The quota suggestion subpopulation and the scheduling subpopulation exchange elite solution information through bidirectional migration, expanding the original one-dimensional chromosome search space into a two-dimensional search space jointly for quotas and scheduling. Experimental examples show that the method of this invention can stably converge to a smaller level of multi-resource imbalance under the same iteration upper limit, and the dispersion of results from multiple independent repeated runs is relatively small, overcoming the inherent defects of single-population genetic algorithms that are prone to getting trapped in local optima and have insufficient stability in solution results.

[0017] Third, this invention constructs a hierarchical architecture with three levels of solvers—strategic, tactical, and operational—working collaboratively, and achieves bidirectional closed-loop feedback between layers through a resource urgency index. The resource urgency index uses the ratio of a project's predicted resource demand to the remaining available capacity of the resource pool as a metric, enabling the strategic layer to dynamically adjust quotas based on the actual competitive situation of each project in the resource pool at each quota allocation trigger. The execution performance of the operational and tactical layers, after correcting the predicted demand, is fed back into the strategic layer's urgency calculation process. The strategic layer updates its quotas based on the feedback data and issues a local rescheduling trigger to the tactical layer. The proportional allocation formula ensures that the total capacity constraint of the resource pool remains constant throughout each redistribution process. Experimental examples demonstrate that this invention's bidirectional closed-loop mechanism overcomes the cyclical imbalance phenomenon of intensified resource competition near milestones and insufficient resource utilization far from milestones in the traditional static quota model. This improves resource pool utilization in multi-project concurrent scenarios and significantly alleviates resource conflicts between projects.

[0018] Fourth, this invention employs a rolling rescheduling mechanism at the operational layer that parallelly assesses four trigger conditions: cycle triggering, resource gap triggering, critical path slack triggering, and external disturbance triggering. This allows the rescheduling process to be activated immediately upon disturbance arrival without waiting for the cycle to expire. The invention adaptively determines the rescheduling window length based on the remaining time of the next engineering milestone and the upper and lower limits of the preset window length. This ensures sufficient search space when the window is far from the milestone and shrinks as the window approaches the milestone to avoid ineffective long-term rescheduling. Furthermore, it uses the original plan deviation as an additional objective to suppress the cascading oscillations of rescheduling on field operations. Experimental examples demonstrate that the adaptive window and multi-trigger condition mechanism of this invention significantly reduce the cumulative plan deviation under disturbance scenarios compared to the fixed window and fixed quota schemes, simultaneously improving both disturbance response speed and plan stability. The hierarchical scheduling optimization platform of this invention is composed of a multi-source data acquisition module, a biometeorological window management module, a resource demand prediction module, a three-layer scheduling engine module, a disturbance response module, and a digital twin visualization module, which are coordinated by a data bus. It integrates the perception, prediction, decision-making, execution, and feedback links into a complete engineering closed loop. It has good compatibility with existing construction management information systems and can be deployed as a module plug-in on the basis of existing project management software. It is convenient for engineering practice and promotion in various engineering scenarios such as ecological restoration, municipal landscaping, and water conservancy and hydropower. Attached Figure Description

[0019] Figure 1 This is a bar chart comparing the coupling multi-resource balance index of Examples 1, 2, and 3 with Comparative Examples 1, 2, 3, and 4 under 30 Monte Carlo experiments.

[0020] Figure 2 This is a comparison chart of resource time distribution between Example 1 and Comparative Examples 1, 2, and 3, where (a) is a stacked area chart of resource time distribution in Example 1, (b) is a stacked area chart of resource time distribution in Comparative Example 1, (c) is a stacked area chart of resource time distribution in Comparative Example 2, and (d) is a stacked area chart of resource time distribution in Comparative Example 3.

[0021] Figure 3 This is a comparison chart of the convergence curves of the hybrid genetic algorithm in Examples 1, 2, and 3 and Comparative Example 3.

[0022] Figure 4 The charts show the adaptive rescheduling comparisons of Examples 1 and 3 with Comparative Examples 2 and 4. (a) is a comparison chart of the rescheduling window length of each method changing over time, and (b) is a comparison chart of the cumulative plan deviation of each method changing over time.

[0023] Figure 5 This is a flowchart of a hierarchical scheduling optimization method for dynamic balancing of multiple resources in engineering projects, based on the present invention. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] The method and platform of this invention will be described in detail below using an engineering project scheduling scenario. This embodiment takes a multi-project parallel scenario consisting of several ecological and environmental protection, municipal construction, water conservancy projects, and landscaping construction tasks as an example. These construction tasks share limited resources of construction personnel, machinery, seedlings, and special equipment. The overall process of the method in this embodiment includes six main steps, S1 to S6, which are executed collaboratively by three levels of solvers: strategic, tactical, and operational, and a cross-layer closed loop is achieved through a bidirectional feedback channel. Each step will be explained in detail below.

[0026] Reference Figure 5 Step S1 is the construction phase of the project data model, which is further subdivided into sub-steps S11 to S14. In sub-step S11, the data acquisition unit extracts an activity list from the construction organization design documents, the Building Information Model (BIM), and historical project archives to determine the number of activities. Duration of each activity The sequence of activities is determined, and a directed acyclic graph (DAG) of activities is constructed. BIM is the industry standard for digital modeling of engineering information. The nodes of the DAG correspond to each activity, and the directed edges correspond to the sequence constraints between activities. In sub-step S12, the quantity of various resources required per unit time during the execution period is determined for each activity according to its process requirements, forming an activity-resource demand matrix. The rows of the matrix correspond to activities, the columns correspond to resource types, and the matrix element values ​​represent the demand intensity of the corresponding resource. In sub-step S13, the total capacity limit of various resources in the resource pool is entered. The document lists the project duration limits and milestone nodes for each project. These milestone nodes are key time points stipulated in the contract and must be adhered to as hard time constraints during the scheduling process. In sub-step S14, the above data is normalized and stored in the project database to obtain the project data model for subsequent steps S2 to S6. Normalization refers to performing unit conversion and data format standardization on numerical data to facilitate subsequent algorithm reading and processing.

[0027] Step S2 is the stage for generating coupling coefficients and feasible time windows, which is further subdivided into sub-steps S21 to S234. In sub-step S21, for each pair of resources... and Scan each activity on the DAG and calculate the resources required simultaneously. With resources Number of activities And the resources needed or resources Total number of activities And calculate the resource-to-coupling coefficient according to the following formula. : ; in For resources With resources The coupling coefficient is dimensionless. To simultaneously require resources With resources The number of activities, expressed in units; For the need of resources or resources The total number of activities, expressed in units of; and Indexes are provided for different resource types. Its form is similar to the Jaccard similarity coefficient in statistics, reflecting the degree of co-occurrence of two types of resources in an activity set; because Therefore The value ranges from 0 to 1, when the two types of resources are completely independent. When the value is 0, it is fully shared. Set to 1. In sub-step S22, set each Write a symmetric coupling matrix The coupling matrix Set the diagonal element to 1, and we get The dimensional coupling matrix is ​​used in step S4, where The number of resource types; this matrix will serve as the weight source for subsequent calculations of the coupled multi-resource equilibrium index.

[0028] Sub-step S23 generates feasible time windows for each activity according to biological and meteorological rules, and is further subdivided into S231 to S234. In sub-step S231, for each activity... Based on its process category, at least one construction environment constraint rule is retrieved from the biological and meteorological rule bases. This construction environment constraint rule specifies the feasible value range of at least one variable among temperature, humidity, wind speed, precipitation, or seedling phenological segments. Taking sprinkler irrigation seedling planting as an example, the constraint rule may include conditions such as the average daily temperature being within a certain range, the cumulative daily precipitation being below a certain upper limit, the wind speed being below a certain upper limit, and planting being completed before sunset. In sub-step S232, using meteorological forecast data and the seedling phenological calendar as input, each day within the scheduling time domain is substituted into the construction environment constraint rule for feasibility assessment, resulting in a binary label of feasibility or infeasibility for each day. The logic of feasibility assessment is to perform Boolean operations on each meteorological and phenological variable of the day according to the logical expressions stored in the rule base. If all variables fall within the feasible range, the day is marked as feasible; otherwise, it is marked as infeasible. In sub-step S233, consecutively marked feasible dates are merged into a closed interval segment. Sequentially number each consecutive feasible segment within the scheduling time domain. And use the union of the segments as the activity according to the following formula. Feasible time window : ; in For the event The feasible time window is a set of days as the unit of time; For the event The number of feasible window segments, in segments; , activities respectively No. The start and end dates of the segment window, in days; Represents the union of sets; Indicates a closed interval; For activity indexing; For window segment indexing. In sub-step S234, the coupling matrix is... and feasible time windows for each activity The coupling coefficient data and time window constraint data are stored in the project database to obtain the data for use in step S4. Step S2 transforms external biometeorological laws into activity-level time constraints, giving various process-sensitive activities a priori executable time range during subsequent scheduling, thus avoiding rework and quality losses caused by neglecting environmental factors in traditional scheduling.

[0029] Step S3 allocates resource pool capacity as quotas among projects at the strategic level based on the resource urgency index, further subdivided into sub-steps S31 to S34. In sub-step S31, at each quota allocation trigger moment... For each project and various resources The resource demand forecasting module obtains the predicted demand for the next scheduling window based on a historical project database and an activity template library, using either a time-series regression method or an empirical proportional method. Read the project from the execution log. Resources As of Cumulative consumption at any time and resources in the resource pool As of Total cumulative consumption at any moment ; and read the current quota from the quota table In sub-step S32, the resource urgency index is calculated using the following formula. : ; in For the project exist Resources at all times The urgency index is dimensionless; To predict demand, units and resources same; For resources in the resource pool Total capacity limit, units and resources same; For resources in the resource pool As of Total cumulative consumption at any given time, in units and resources same; To prevent the denominator from being zero, a pre-defined positive small constant is used; This indicates taking the larger value; Index the project; Index for resource types; This is a time index, in days (d). The preferred value is a small positive constant that is much smaller than the order of magnitude of the resource capacity, and can be taken as . The value is on the order of one ten-thousandth to one thousandth to ensure the stability of the denominator and not affect the allocation result. The urgency index... The physical meaning of the project The ratio of future demand to the remaining available capacity of the entire resource pool; a higher ratio indicates a greater demand for resources from the project. The competition is more intense; by using the remaining available capacity instead of the current quota of a single project as the denominator, the timing misalignment problem of calculating urgency based on the quota of the previous cycle is avoided, so that the urgency reflects the competitive situation of each project for the total remaining resources in the resource pool.

[0030] In sub-step S33, the resources in the resource pool are allocated according to the following formula. Total capacity The quotas were redistributed among the projects to obtain new allocations. : ; in To reassign to the project resources Quotas, Units and Resources same; As defined in S32; Calculated according to S32; Indicates multiplication; This represents the summation over all projects participating in the same resource pool. The formula is the standard form of proportional allocation, and the summation over all projects can be derived as follows: Therefore, the total allocation of quotas always equals the upper limit of the total capacity of the resource pool, and the total capacity constraint of the resource pool is conserved. In sub-step S34, the new quotas are... Issue tactical level and write into quota table to replace original This serves as a hard constraint for the activity scheduling in step S4.

[0031] Step S4 employs a hybrid genetic algorithm (H-GA) at the tactical level, under quota and time window constraints, to solve the scheduling problem involving multiple resource coupling and equilibrium, outputting the baseline schedule and resource time distribution. This is further subdivided into sub-steps S41 to S46. H-GA stands for Hybrid Algorithm, which incorporates specific problem-solving strategies into the standard genetic algorithm. In sub-step S41, the population is initialized. Each chromosome consists of two segments: the first segment is the activity priority sequence. The second section is for selecting gene sequences via a window. ,in For the first The priority value of each activity is dimensionless. For the event The selected feasible time window index, The value ranges from 1 to between, Obtained according to S233; The number of activities in the project data model is expressed in units. The priority values ​​are generated by assigning priority to a predetermined proportion of chromosomes in the population in ascending order of total float time, and assigning priority to the remaining proportion of chromosomes in the population by random numbers. The total float time is the difference between the latest completion time and the earliest start time of the activity. The total float time is obtained by the critical path method (CPM) cited in S52 below. The smaller the total float time, the more critical the activity, and prioritizing its scheduling helps to compress the project duration.

[0032] In sub-step S42, the Serial Progress Generation Scheme (SSGS) is decoded for each chromosome according to the activity priority sequence. The start time of each activity is determined sequentially using the activity DAG. And make it satisfy the time window constraint shown in the following formula: ; in For the event The start time, in days; For the event The duration, in days; For the event The window index used in this decoding was initially taken from the chromosome gene. It only applies to this decoding process and does not modify the chromosome gene sequence; , activities respectively No. The start and end dates of the segment window, in days; This indicates that the subset contains; This represents a closed interval. The logic of the SSGS is to find the earliest start time that satisfies the preconditions and resource constraints for each activity in priority order.

[0033] In sub-step S43, the case where the time window constraint cannot be satisfied is handled, and the following decoding and repair rules are used to determine the issue. and Without modifying chromosome gene sequences Rule a is that if the selected window segment... Memory is in use The earliest established Then Set to this earliest value, Set as Rule b is: if rule a has no solution, then... Search by segment index In the subsequent window segment, find the first one that satisfies... window segment ,Will Set as , Set as Rule c states that if neither rule a nor rule b has a feasible solution, the activity is marked as infeasible for decoding, and a preset penalty value is applied to the fitness of that chromosome. This rule provides a complete repair strategy for the decoding process, ensuring that any chromosome can be decoded to a feasible or penalized state, preventing the genetic algorithm from losing its search direction due to discarding infeasible chromosomes. Simultaneously, the chromosome genes are not modified during the decoding process, ensuring a clear distinction between the genetic algorithm's genotype and phenotype, consistent with the standard operational semantics of genetic algorithms.

[0034] In sub-step S44, a serial scan repair is performed on solutions that violate resource capacity constraints, ensuring that activities executed simultaneously at each time point affect various types of resources. The total amount occupied does not exceed the corresponding quota. The serial scan repair method scans from left to right along the time axis. When resource overflow is encountered, the activity that started later is shifted to the right to a non-overflow position, thereby restoring resource capacity constraints. In sub-step S45, the Coupling Multi-Resource Balance Index (CMRBI) is used as the fitness function, and each chromosome is evaluated according to the following formula; CMRBI is the target index constructed by this scheme for the multi-resource coupling balance requirement: ; Where CMRBI is the Coupling Multi-Resource Balancing Index, which is dimensionless; for Time Resources Actual input, units and resources same; For resources The time average within the scheduling time domain, according to Calculation, Units and Resources same; The total length of the scheduling time domain, in seconds (d). A pre-defined positive small constant, with the same meaning as described in S32, is used to prevent the denominator from being zero. ; These are the weighting coefficients of the coupling terms, preset positive constants, and dimensionless. Calculated according to S21; This indicates taking the larger value; This indicates the time domain from the 1st to the 1st. Sum of the discrete moments of the day; This represents the summation of all types of resources in the set of resource types; Indicates that for each satisfied Sum of resource types; This is a time index, in days (d). and This serves as an index for resource types. The CMRBI consists of two items. The first item is the normalized variance of each resource type relative to its own average value, reflecting the degree of uniformity in the temporal distribution of each resource type in a single dimension; the smaller the value, the more stable the resource input. The normalized variance of the first item is the square of the coefficient of variation, a standard indicator in statistics for measuring the dispersion of a distribution. The second item is the normalized sum of squares of deviations of resource pairs weighted by the coupling coefficient, reflecting the synchronicity of strongly coupled resource pairs in the temporal distribution; the smaller the value, the more consistent the input rhythm of strongly coupled resource pairs. To balance the two weights, the preferred value is a positive constant that makes the two values ​​comparable in magnitude, which can be pre-calibrated using the two historical values ​​of the items. of Protection Resources that are zero in the scheduling time domain will not cause division by zero errors, thus ensuring that the objective function is robust to computation for all resource types.

[0035] In sub-step S46, selection, crossover, and mutation operations are performed to generate the next generation population. After the number of iterations reaches a preset upper limit, the activity start time sequence and resource input sequence corresponding to the individual with the smallest CMRBI value in the population are output as the baseline schedule and resource time distribution. This step follows the basic framework of genetic algorithms. The selection operation can use roulette wheel or tournament methods, the crossover operation can use two-point crossover for priority sequences and uniform crossover for window selection gene sequences, and the mutation operation randomly perturbs individual gene loci. The iteration upper limit can be preset according to the problem size, preferably between 200 and 1000 generations for small and medium-sized engineering projects. The execution results of sub-steps S41 to S46 are the baseline schedule and resource time distribution.

[0036] To further improve the solution quality of H-GA, the H-GA can be solved using a twin-population co-evolutionary structure. This twin-population co-evolutionary structure operates within the tactical layer solution process, distinct from the macroscopic hierarchical architecture consisting of the strategic, tactical, and operational layers. In engineering implementation, it manifests as two parallel algorithm instances within the same tactical layer solver; specifically, it includes sub-steps S47 to S49. In sub-step S47, quota suggestion sub-population POP_Q and scheduling sub-population POP_S are maintained within the tactical layer solution process. The quota suggestion vectors for each chromosome coding item and resource pair in POP_Q are... In POP_S, each chromosome encodes an activity priority sequence according to the two-segment coding structure defined in S41. Selecting gene sequences with a window ;in Items encoded for the POP_Q chromosome Resources Recommended quota values, units and resources The same applies. The evolutionary goal of POP_Q is to find a quota combination that, after local fine-tuning based on the quotas issued at the strategic level, can achieve better scheduling performance. The evolutionary goal of POP_S is to find the scheduling scheme with the minimum CMRBI given the quotas. In sub-step S48, after each preset migration cycle... In each generation, the elite chromosome with the best fitness is selected from POP_Q, and its encoded quota proposal vector is... As a hard constraint for each chromosome in POP_S when executing S42 to S46, each chromosome in POP_S re-enters the decoding and fitness evaluation process under this hard constraint; among which The migration cycle is measured in generations. In substep S49, the elite chromosome with the lowest CMRBI value is selected from POP_S, and the peak resource requirements for each item and resource pair are extracted. Calculate the fitness correction term for each chromosome in POP_Q using the following formula. : ; in This is the fitness correction term for the POP_Q chromosome, and it is dimensionless. Items extracted from POP_S elite chromosomes Resources Peak demand in the current phase, in units and resources same; As defined in S47; The first position on the right side of the equals sign indicates negation, and the area inside the parentheses indicates subtraction; This indicates taking the larger value, with a lower bound of 0; This represents the summation of all items and resources. The original fitness of the POP_Q chromosome is superimposed, causing POP_Q to evolve towards suggestions for increasing quotas for peak demand items and resource pairs. This, in turn, leads to the co-evolution of the twin populations through bidirectional migration between S48 and S49. This co-evolutionary structure of the twin populations, through information exchange between POP_Q and POP_S, enables parallel iteration of quota optimization and scheduling optimization, making the solution obtained by H-GA more closely match the actual peak resource demand and preventing a single population from getting trapped in local optima. The preferred value is a positive integer that is relatively small compared to the total number of iterations. Usually, the total iteration process can include several to a dozen transitions.

[0037] Step S5 triggers rolling rescheduling according to preset conditions during the operation layer execution. The rescheduling window length is adaptively determined based on the next milestone and preset upper and lower window limits, and activities within the window are rearranged. This is further subdivided into sub-steps S51 to S54. In sub-step S51, the operation layer collects actual on-site progress, resource consumption, and weather data on a daily basis, and calculates the deviation of each activity from the baseline progress plan in real time. In sub-step S52, it is determined whether any of the following four triggering conditions are met. Condition i is periodic triggering, where the time interval since the last rescheduling reaches a preset rescheduling cycle. Condition ii is triggered by a resource shortage, and a project exists. With resources This makes the resource gap more than Exceeding the preset resource gap ratio threshold Condition iii triggers the critical path relaxation, and the remaining critical path relaxation... Not greater than zero. Condition iv is triggered by an external disturbance, detecting a preset external disturbance event. The preset rescheduling period is expressed in days (d). , , As defined in S31; The preset resource gap ratio threshold is dimensionless. for The remaining critical path relaxation at each time step, in days (d). Contractual time constraints, unit: days; The project completion date is calculated based on the current progress using the Critical Path Method (CPM), and is expressed in days (d). This indicates taking the larger value. The earliest start time EST(i) and earliest finish time of each activity are obtained by forward traversal of the activity DAG, and the latest start time and latest finish time LFT(i) are obtained by reverse traversal. The criticality of an activity is measured by the total float LFT(i) - EST(i), and activities with zero total float constitute the critical path. The preferred value is an integer multiple of a week or half a week to facilitate on-site docking. The preferred value is a decimal reflecting acceptable deviation, typically between 0.1 and 0.3. The physical meaning of the resource gap ratio is the project... Resources The proportion of the forecasted demand exceeding the remaining quota relative to the original quota, and this proportion is non-zero only when the forecasted demand is greater than the remaining quota.

[0038] If any condition is met, proceed to sub-step S53. In sub-step S53, calculate the rescheduling window length using the following formula. : ; in for The length of the time-based rescheduling window, in d; The planned completion date for the next project milestone, expressed in days (d). This is the preset lower limit for window length, in units of d; This is the preset maximum window length, in d (d). , and These represent taking the smaller value and taking the larger value, respectively. To represent subtraction; This is the current time index, in seconds (d). The `clip` function restricts the first parameter within the parentheses to the interval formed by the second and third parameters; it is a commonly used truncation function in numerical computation. and The optimal value depends on the project cycle and rescheduling cost, typically Set at approximately one week, Set the window to approximately one month; too short a window will cause frequent reordering and disrupt the site, while too long a window will result in a loss of timely response to approaching milestones.

[0039] In sub-step S54, the time interval is... The start time of activities other than those already committed remains unchanged as a constraint, and the start time falls within the interval. The activities within are shown in the mixed objective function as follows: Solve again: ; in The objective function value for the rescheduling phase is dimensionless; CMRBI is calculated according to S45. The deviation from the sensitivity coefficient is a preset positive constant, dimensionless; This is the set of time intervals corresponding to the rescheduling window, in days (d). Indicates the time at which the beginning falls into Summing up all activities; For rescheduled activities The start time, in days; Activities in the baseline plan The starting time is in days; the solution result replaces the interval in the original plan. At the start time of the internal activities, the updated progress plan is obtained. The first item, CMRBI, maintains the resource balance level after rescheduling; the second item... The product of the deviation from the original plan and the sum of the absolute values ​​of the deviations at the start time is used to penalize the degree of deviation of the rescheduling from the original plan, so as to avoid large fluctuations in on-site operations caused by each rescheduling. The optimal value for is a positive constant that makes the two terms comparable; if it is too small, the plan loses stability, and if it is too large, the equilibrium objective fails. During the rescheduling phase, the solver can reuse the H-GA framework from step S4, replacing the fitness function from CMRBI with... The rest of the process remains unchanged.

[0040] Step S6 implements hierarchical feedback between the operational, tactical, and strategic layers. The strategic layer updates quotas based on the feedback data, forming a cross-layer bidirectional closed loop. This is further subdivided into sub-steps S61 to S64. In sub-step S61, the operational layer uploads the actual start and end times of each activity, the actual consumption of various resources, and on-site disturbance events to the tactical layer on a daily cycle. This upload is completed through an Internet of Things (IoT) communication link, and the data format follows a unified data model predefined by the platform. In sub-step S62, the tactical layer updates the predicted demand based on the uploaded data. And extract the items in the next scheduling window. Resources Peak actual input Report to the strategic level; among which For the project Resources Peak demand within this scheduling window, in units and resources Same; the extraction method is for Take the maximum value according to the time series, the For the project exist Resources at all times The actual amount of input.

[0041] In sub-step S63, the strategic layer receives reports from the tactical layer. With the updated The revised forecast demand is constructed using the following formula. and with Alternative Recalculate the urgency index according to S32 to S34. With new quota : ; in The revised forecast demand, in units and resources same; As defined in S31; As defined in S62; This indicates taking the larger value. This sub-step ensures that peak demand items receive a corresponding increase in quota in the next allocation cycle, while the proportional allocation formula in S33 guarantees this. equal This ensures that the total capacity constraint of the resource pool is maintained. The mechanism reverse-engineers the actual peak values ​​observed at the tactical layer into the quota allocation process at the strategic layer, avoiding insufficient response to peak scenarios based on historical data predictions. Compared to the simple approach of directly adding quotas to a specific project or resource pair, this mechanism automatically reduces quotas from other projects and resource pairs through global proportional allocation in S33, maintaining the physical realizability of the system's total resource pool capacity. In sub-step S64, the strategic layer updates the new quotas... The tactical layer is issued, triggering the tactical layer to execute step S4 local rescheduling for the affected projects, thus obtaining a two-way closed loop between the upper and lower layers. The local rescheduling only solves for the activities involved in the affected projects and resources, while the start time of unaffected activities remains unchanged, in order to reduce the amount of computation and on-site disturbance.

[0042] The following section describes the hierarchical scheduling optimization platform for engineering projects implementing the described method. The platform comprises six functional modules: a multi-source data acquisition module, a biological and meteorological window management module, a resource demand prediction module, a three-layer scheduling engine module, a disturbance response module, and a digital twin visualization module. These six modules are interconnected via a data bus. The data bus is implemented using a high-performance message middleware, supporting both publish-subscribe and point-to-point communication modes, ensuring decoupling of the six modules in physical or virtual deployment.

[0043] The multi-source data acquisition module includes an IoT gateway and several edge nodes. These edge nodes are electrically connected to a weather station, soil moisture sensor, tower crane load sensor, construction vehicle positioning terminal, personnel positioning cards, and material identification readers. The IoT gateway connects to a data bus and uses the Message Queuing Telemetry Transmission Protocol (MQTT) to receive raw data uploaded by the edge nodes. It then normalizes the raw data according to a platform-predefined unified data model before sending it to other modules. MQTT is a lightweight messaging protocol widely used in the Internet of Things (IoT) field. The function of the multi-source data acquisition module is to provide the entire platform with real-time, uniformly formatted field data streams, including meteorological data, soil moisture, mechanical load, personnel and vehicle locations, and material flow status. It serves as a unified entry point for subsequent modules to perceive the field status.

[0044] The biological and meteorological window management module includes a seedling phenology database, a construction process temperature and humidity rule database, and a weather forecast interface. The seedling phenology database stores phenological calendars and suitable planting periods for common garden seedlings, while the construction process temperature and humidity rule database stores environmental constraints imposed by common construction processes. Both the seedling phenology database and the construction process temperature and humidity rule database are connected to the weather forecast interface. The weather forecast interface receives measured data from weather stations from the multi-source data acquisition module and integrates forecast data from external meteorological services to obtain a daily weather forecast sequence within the scheduling time domain. The output of the biological and meteorological window management module outputs feasible time window data. The function of the biological and meteorological window management module is to transform external environmental constraints into activity-level time window inputs, providing a time constraint basis for the tactical layer solver. This allows scheduling to no longer rely solely on purely numerical time windows as input, but instead possesses the phenological and meteorological sensitivity required for business scenarios such as ecological environmental protection and landscaping.

[0045] The resource demand forecasting module is connected to a historical project database and an activity template library. The historical project database stores activity and resource consumption records for completed projects, while the activity template library stores standard consumption coefficients for various resources based on typical construction activities. The resource demand forecasting module forecasts resource demand within a future scheduling window using either a time-series regression method or an empirical proportional method. The time-series regression method is suitable for scenarios with abundant historical data, while the empirical proportional method is suitable for estimating demand based on project volume and standard consumption coefficients when historical data is limited. The resource demand forecasting module outputs the predicted resource demand data to the input of the strategic layer solver.

[0046] The three-layer scheduling engine module consists of a cascaded strategic layer solver, a tactical layer solver, and an operational layer scheduler, and is equipped with a feedback channel from the lower layer to the upper layer. The output of the three-layer scheduling engine module is connected to the digital twin visualization module and the field operation terminal. The strategic layer solver, tactical layer solver, and operational layer scheduler communicate bidirectionally via a message bus. The data categories transmitted by the message bus include quota data issued by the strategic layer solver to the tactical layer solver, activity start and end time instruction data issued by the tactical layer solver to the operational layer scheduler, actual activity start and end time data and actual resource consumption data reported by the operational layer scheduler to the tactical layer solver, and peak resource demand data reported by the tactical layer solver to the strategic layer solver. The strategic layer solver is used for resource pool quota allocation, deploying algorithm instances corresponding to S31 to S34; the tactical layer solver is used for activity scheduling, deploying H-GA instances and twin-population co-evolutionary structures corresponding to S41 to S46; the operational layer scheduler is used for daily-level job instruction generation and execution monitoring, deploying rolling rescheduling logic corresponding to S51 to S54. This three-layer architecture decreases the decision granularity from long-term and macro-project level to medium-term, project-internal activity level, then to daily and on-site operation level. Each layer solver optimizes independently at its appropriate time scale, while achieving overall collaboration through bidirectional feedback; the quotas issued by the upper layer are hard constraints on the lower layer, and the performance reported by the lower layer serves as decision input for the upper layer.

[0047] The disturbance response module includes a trigger condition discrimination unit and a rescheduling invocation unit. The input of the trigger condition discrimination unit is connected to the output of the multi-source data acquisition module, and it continuously discriminates according to the four types of trigger conditions described in S52. The output of the rescheduling invocation unit is connected to the input of the operation layer scheduler, and it invokes the operation layer scheduler to execute S53 and S54 when any trigger condition is met. The disturbance response module is independent of the main solution process, decoupling the triggering and invocation of rescheduling from the scheduling algorithm itself, which facilitates the platform's future expansion to include more trigger types.

[0048] The digital twin visualization module uses a fusion of a 3D Geographic Information System (GIS) view and a Building Information Modeling (BIM) view as its base map. It overlays Gantt layers, resource histogram layers, coupled multi-resource equilibrium index curve layers, urgency index heatmap layers, and time window conflict alarm layers. An interactive interface is provided for project managers to access and drill down. The GIS view is integrated with the BIM view through a unified spatial coordinate system. The base map displays the spatial relationships between the project's geographical area and various building components in 3D. The layers are overlaid on the base map in 2D or symbolic form. The interactive interface supports layer switching, timeline dragging, and click-to-search operations. The purpose of the digital twin visualization module is to present the solution results and execution status to project managers in an intuitive and visual way, facilitating decision-making and communication.

[0049] Through the implementation of the above methods and platform, this solution can address the complex scenarios of parallel projects involving multiple types of engineering projects, such as ecological and environmental protection, municipal engineering, water conservancy, and landscaping. It incorporates the tightly coupled demands of various resources into a unified objective function, integrates external meteorological and biological factors into activity-level time constraints, and achieves dynamic coordination of resource allocation, activity scheduling, and on-site execution through a two-way closed loop between the strategic, tactical, and operational layers. Compared to traditional single-resource minimum moment methods or fixed-window rolling rescheduling methods, this solution improves upon multi-resource coupling balance, response speed to external disturbances, and timely response to approaching milestones. The implementation of the solution is completed collaboratively by the platform's six functional modules via a data bus. It exhibits good compatibility with existing construction management information systems and can be deployed as a module plug-in on existing project management software, facilitating its promotion in engineering practice.

[0050] Example 1: An ecological and environmentally friendly constructed wetland construction project includes 30 activities, involving six types of resources: excavators, loaders, construction vehicles, green seedlings, ecological stone materials, and construction personnel. The scheduling time domain is 90 days, and project milestones are set on days 40, 80, and 120. This example is implemented according to the method of the present invention, wherein... Take 1.0, Take 1.0, Pick 1 / 1000 Take 7 days, Take 0.15, Take 7 days, Take 30 days, Take 10.

[0051] According to sub-steps S11 to S14 of step S1 of this invention, the activity list and the sequence of activities are extracted from the construction organization design document, a directed acyclic graph of activities is constructed, the activity and resource requirement matrix is ​​determined, the total capacity of the resource pool and three milestone nodes are entered, and the project data model is formed and stored in the project database. According to sub-steps S21 to S234 of step S2 of this invention, the coupling coefficients of 15 resource pairs are calculated. Fill in the symmetric coupling matrix Based on the matching rules of biological and meteorological rule bases, and combined with seasonal weather forecasts and seedling phenological calendars, feasible time windows for each activity are obtained. Some of these activities involve planting seedlings. It consists of 2 to 3 closed interval segments. According to sub-steps S31 to S34 of step S3 in this invention, the resource demand prediction module provides the predicted demand for the future scheduling window. Read the cumulative consumption Calculate the urgency index The new quota is obtained by proportional allocation formula. The tactical layer is then deployed. Following sub-steps S41 to S46 of step S4 in this invention, the dual-segment chromosome population is initialized, SSGS decoding is performed, time window constraints are handled according to the decoding repair rules, resource capacity constraints are repaired using serial scanning, CMRBI fitness is evaluated, and after 800 generations of crossover and mutation iterations, the baseline progress plan and resource time distribution are output. Simultaneously, the dual-segment population co-evolutionary structure from sub-steps S47 to S49 of step S4 in this invention is enabled to accelerate convergence. Following sub-steps S51 to S54 of step S5 in this invention, the operation layer collects data daily, and the disturbance response module continuously judges based on four types of triggering conditions. Upon triggering, the rescheduling window length is calculated using the clip formula. and with a mixed objective function Solve again. According to sub-steps S61 to S64 of step S6 of the present invention, to... Inject into the next cycle to form a cross-layer bidirectional closed loop.

[0052] Example 2 involves a comprehensive renovation project of urban roads and drainage networks, comprising 60 activities and involving 8 types of resources, including pipeline laying machinery, concrete pump trucks, tower cranes, earthmoving vehicles, steel structure components, asphalt pavers, construction personnel, and special equipment. The scheduling time domain is 180 days. The difference from Example 1 lies in the larger project scale and tighter resource coupling; among which... We set the weight to 1.5 to increase the weight of the coupling term. Take 0.8, Take 14 days, Take 0.20, Take 10 days, Take 45d, Take 15. Pick 1 / 2000; the remaining steps are performed in the same manner as in Example 1.

[0053] Example 3: A landscaping group concurrently undertakes three engineering projects: one park landscaping project, one municipal square project, and one water feature restoration project. The three projects share a common resource pool containing five types of shared resources: construction workers, excavators, water trucks, green seedlings, and decorative stone. Each project independently comprises 20 to 25 activities, with a total scheduling timeframe of 150 days. Project milestones are set on days 50, 100, and 150. This example... Take 0.8, To improve the sensitivity to plan stability, we set it to 1.2. Take 10 days, Take 0.18, Take 7 days, Take 35 days, Take 12, Pick 1 / 1000; This embodiment focuses on testing the bidirectional closed loop of step S6 of the present invention and the dynamic allocation of urgency quota in step S3 of the present invention. The execution methods of the remaining steps are the same as those in embodiment 1.

[0054] Comparative Example 1, without calculating the resource-to-coupling coefficient. The sub-steps S21 and S22 in step S2 of this invention are omitted, and in sub-step S45 of step S4 of this invention, only the first term of the CMRBI fitness function is retained, and the second term of the coupling weighting is removed; the parameters of other steps are the same as in Example 1. This comparative example corresponds to the traditional single-resource minimum torque scheduling method in engineering practice.

[0055] Comparative Example 2 omits all sub-steps S31 to S34 of step S3 in this invention. The quota is issued proportionally based on the initial estimate of each project's duration, without dynamic allocation based on urgency. Other steps are the same as in Example 1. This comparative example corresponds to the static quota allocation mode commonly seen in engineering practice.

[0056] Comparative Example 3 omits sub-steps S47 to S49 of step S4 in this invention, which describes the co-evolutionary structure of twin populations, and retains only the evolution of a single population; other steps are the same as in Example 1. This comparative example corresponds to a standard genetic algorithm implementation without introducing a co-evolutionary mechanism.

[0057] Comparative Example 4 simplifies sub-step S53 of step S5 of the present invention to a fixed window length. Taking 14d, the resource gap trigger, critical path slack trigger, and external disturbance trigger discrimination in sub-step S52 are omitted, and only the periodic trigger is retained; other steps are the same as in Example 3. This comparative example corresponds to the common equal-period fixed-window rolling rescheduling mode in engineering practice.

[0058] Experiment 1, to verify the effectiveness of the CMRBI fitness function in step S4 of this invention on multi-resource coupling equilibrium, conducted 30 independent Monte Carlo runs on seven methods (Examples 1, 2, 3, and Comparative Examples 1, 2, 3, and 4), respectively, and statistically analyzed the mean and standard deviation of the CMRBI values. The initial random number seed was different for each run, and the activity and resource demand matrices were randomly perturbed within a reasonable engineering range to simulate the diversity of actual engineering projects. The experimental results are as follows: Figure 1 As shown.

[0059] from Figure 1 As can be seen, the mean CMRBI values ​​of Examples 1, 2, and 3 are 0.42, 0.48, and 0.45, respectively, with corresponding standard deviations of 0.05, 0.06, and 0.05. The mean CMRBI value of Comparative Example 1 using the minimum moment method is 1.10, with a corresponding standard deviation of 0.15; the mean CMRBI value of Comparative Example 2 using the fixed quota method is 0.85, with a corresponding standard deviation of 0.12; the mean CMRBI value of Comparative Example 3 using the single-population GA is 0.55, with a corresponding standard deviation of 0.08; and the mean CMRBI value of Comparative Example 4 using the fixed window method is 0.60, with a corresponding standard deviation of 0.10. The CMRBI values ​​of each example are significantly lower than those of the comparative examples, with Comparative Example 1 being approximately 162% higher than Example 1, Comparative Example 2 approximately 102% higher, Comparative Example 3 approximately 31% higher, and Comparative Example 4 approximately 43% higher. Figure 1 It can also be seen that the standard deviation of each embodiment does not exceed 0.06, while the standard deviation of each comparative example reaches a maximum of 0.15, indicating that the method of the present invention is stable in multiple repeated experiments. The CMRBI objective function of step S4 of the present invention includes two terms: normalized variance and weighted sum of squared deviations. The minimum moment method only considers the temporal fluctuation of a single resource and ignores the synergy of resource pairs, so comparative example 1 loses control in the second term; the fixed quota method does not respond to urgency at the quota level, so comparative example 2 experiences increased competition near the milestone, causing both terms to be higher than the embodiment; the single-population GA and fixed window methods lose co-evolution and adaptive triggering, respectively, and their performance is between comparative example 1 and each embodiment. The present invention The coupling coefficient, in the form of Jaccard similarity, characterizes the co-occurrence tightness of two types of resources in the activity set. It is incorporated as a weighted weight into the second term of CMRBI, so that H-GA tends to select the synchronization of strongly coupled resource pairs when solving. Therefore, the resource timing of the embodiment is optimized in two dimensions at the same time, and the final CMRBI value is significantly reduced compared to the comparative example 1 which only considers the stationarity of a single resource.

[0060] Experimental Example 2, to verify the shaping effect of the CMRBI terms in step S4 of this invention on the timing of resource input, used data from the same project in Example 1. Step S4 was replaced with the methods of Comparative Examples 1, 2, and 3, while keeping other steps the same, resulting in four sets of resource time distribution data. Each set of data recorded the actual input of six types of resources daily over a 90-day scheduling time domain, and these data were then superimposed to form a stacked area map. The experimental results are as follows: Figure 2 As shown, Figure 2 (a) corresponds to Example 1. Figure 2 (b) Response ratio 1, Figure 2 (c) Corresponding ratio 2, Figure 2 The proportion of response to (d) is 3.

[0061] from Figure 2 As can be seen in (a), the time distribution curves of the six types of resources in Example 1 are generally stable. The peaks and troughs of strongly coupled resource pairs, such as excavators and loaders, fluctuate almost synchronously on the time axis, and the input of various resources does not fluctuate drastically throughout the entire scheduling time domain. Figure 2 As can be seen in Figure (b), the resource curves in Comparative Example 1 exhibit obvious sawtooth fluctuations, with a prominent peak-valley misalignment phenomenon in strongly coupled resource pairs. During certain periods, a resource is at its peak while its coupled resources are at a low level. Figure 2 As can be seen in Comparative Example 2, due to the fixed quota, resource competition intensifies and peaks near the milestone period, while resource utilization is insufficient further away from the milestone period, resulting in a periodic peak-and-trough curve. Figure 2 As can be seen in (d), the stationarity of the resource curves in Comparative Example 3 is between that of Example 1 and Comparative Example 1, but the coupled resource pairs are still not synchronized. This result directly proves the feasibility of the method of this invention in resource timing shaping, because the curves in Example 1 exhibit the stationarity and synchronization characteristics expected in engineering practice, and resource scheduling on-site can be directly executed according to the schedule corresponding to this curve. The causal relationship between this result and the innovation of this invention is as follows: the second term of CMRBI, the weighted sum of squared coupling deviations, will... As weights incorporated into the objective function, H-GA tends to select chromosomes with synchronous fluctuations of two types of strongly coupled resources during the solution process. Mechanistically, this is equivalent to injecting the synergy of resource pairs as prior knowledge into the search direction of the optimizer. Therefore, the peak-valley synchronization characteristic of strongly coupled resources in Example 1 is the result of the second optimization term of CMRBI. Comparative Example 1 loses this shaping effect due to the absence of the second term. Comparative Example 2 is driven by external hard constraints rather than coupling-sensitive optimization due to fixed quotas. Although Comparative Example 3 retains both terms of CMRBI, the single-population GA gets trapped in a local optimum.

[0062] Experiment 3, to verify the effectiveness of the twin-population co-evolutionary structure in sub-steps S47 to S49 of the present invention in terms of solution efficiency, conducted 30 independent runs of each of the four methods (Example 1, Example 2, Example 3, and Comparative Example 3), recording the mean and 95% confidence interval of the optimal fitness of each generation of CMRBI. The population size for each run was set to 100, the iteration limit was 1000 generations, and the initial random number seed was different. The experimental results are as follows: Figure 3 As shown.

[0063] from Figure 3 As can be seen, the optimal CMRBI of Example 1 decreased to 0.46 in the 150th generation and stabilized at 0.42 in the 300th generation; the optimal CMRBI of Example 2 decreased to 0.60 in the 200th generation and stabilized at 0.49 in the 500th generation; the optimal CMRBI of Example 3 decreased to 0.55 in the 200th generation and stabilized at 0.46 in the 500th generation; and the optimal CMRBI of Comparative Example 3 decreased to 0.62 in the 500th generation and decreased to 0.55 in the 1000th generation, and still did not stabilize. All three examples converged faster and had lower final values ​​than Comparative Example 3, and had narrower confidence interval bandwidths. From the shape of the confidence interval bands, it can be seen that the dispersion of the 30 independent runs of the examples is small, while the dispersion of Comparative Example 3 is large and it has not fully converged by the end of the iteration. The results directly demonstrate the feasibility of the twin-population co-evolution mechanism of this invention, because the embodiment stably converges to a significantly lower fitness under the same iteration limit, indicating that the algorithm can actually generate an effective scheduling scheme. This invention maintains the quota suggestion subpopulation POP_Q and the scheduling subpopulation POP_S in parallel within the tactical layer solution process. POP_Q transmits the elite solution as a hard constraint to POP_S through sub-step S48, and POP_S feeds back the peak resource demand to POP_Q as a fitness correction term through sub-step S49. The bidirectional migration guides the search directions of the two subpopulations, which is mechanistically equivalent to expanding the original one-dimensional chromosome search space into a two-dimensional joint search space of quota and scheduling, avoiding single-population GA from getting trapped in local optima. Simultaneously, the division of labor in the chromosome coding dimension by the twin-population co-evolution allows each subpopulation to focus on its assigned sub-problem, resulting in a search efficiency higher than the hybrid coding search of a single population.

[0064] Experiment Example 4 verifies the improvement in disturbance response capability achieved by the adaptive rescheduling window and four types of triggering conditions in step S5 of the present invention. This experiment performs simulations on the same disturbance sequence injected into four methods: Example 1, Example 3, and Comparative Examples 2 and 4. The disturbance sequence contains five different types of disturbance events, occurring on days 20, 45, 75, 110, and 130, respectively. The disturbance types cover mechanical failures, rainstorm delays, and design changes. Experimental Results Figure 4 As shown, Figure 4In the middle (a), the rescheduling window length is... The curves showing the window length over time are presented, including examples 1, 3, and 4 (comparative example 4). Comparative example 2 is not included because its window rescheduling mechanism is the same as that of example 1. Figure 4 (a) is drawn separately; Figure 4 (b) represents the cumulative planned deviation. The curves showing the change over time include the cumulative deviations of four methods: Example 1, Example 3, Comparative Example 2, and Comparative Example 4.

[0065] from Figure 4 As can be seen in Example 1 and Example 3 (a) The curve exhibits a step-like adaptive change, with the remaining time until the next milestone being greater than [missing information]. hour Pick The remaining time is and Between Take the remaining time; if the remaining time is less than... hour Shrink to After the milestone It jumps back to a larger value. (Comparative Example 4) The milestone distance is 14 days without response. Figure 4 In (a), the milestone nodes of Examples 1 and 3 are marked with gray vertical dotted lines, and the occurrence times of the five disturbance events are marked with red vertical dashed lines and red inverted triangles. These markings indicate that even when multiple disturbance events arrive successively, the window length of the embodiments can still respond according to the distance to the next milestone. Figure 4As can be seen in (b), after 5 disturbance events, the final cumulative deviation of Example 1 was 28 days, Example 3 was 35 days, Comparative Example 2 was 82 days, and Comparative Example 4 was 105 days. At the time of the disturbance, the slope of the curve of Example 1 increased slightly and then tended to stabilize, while the slope of the curve of Comparative Example 4 remained high for a long time. The cumulative plan deviation in the embodiment is only about 1 / 3 to 1 / 4 of that in the comparative example, indicating that the present invention can achieve rapid response and small disturbances during the execution of engineering projects. After the disturbance arrives, the resource gap trigger, critical path slack trigger, or external disturbance trigger in the embodiment immediately activates rescheduling without waiting for periodic triggers, reducing the response delay to the daily level. The clip formula of the present invention adaptively adjusts the window length according to the proximity of the next milestone, leaving sufficient search space for the optimizer when far from the milestone, and shrinking the window to avoid invalid long-term rescheduling when approaching the milestone. The comparative example 4 can only respond when it is triggered by the periodic trigger, and in the worst case, the disturbance delay is close to one cycle. Because the quota in the comparative example 2 is fixed, the resource gap after the disturbance cannot be resolved by quota redistribution, which leads to continuous accumulation of plan deviations. The present invention has three mechanisms: multiple trigger conditions, adaptive window, and two-way closed loop, which makes it superior to the comparative example in terms of disturbance response, plan stability, and resource redistribution.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A hierarchical scheduling optimization method for dynamic balancing of multiple resources in engineering projects, characterized in that, include: S1: Establish an engineering project data model that includes an activity-directed acyclic graph, multiple types of resource requirements and capacity, and milestone constraints; S2: Calculate the resource pair coupling coefficients and generate feasible time windows in the form of multiple closed interval unions for each activity; S3: The strategic layer allocates resource pool quotas among projects based on the resource urgency index; S4: Under the constraints of the quota and feasible time window, the tactical layer schedules each activity with the goal of coupling multiple resource balance indices, and outputs the baseline schedule plan and resource time distribution. S5: During the execution of the operation layer, a rolling rescheduling is triggered according to preset conditions. The length of the rescheduling window is adaptively determined based on the next milestone and the preset upper and lower limits of the window, and the activities within the window are rearranged. S6: Execution data is fed back level by level between the operational layer, tactical layer, and strategic layer. The strategic layer updates the quota based on the feedback data, forming a cross-layer bidirectional closed loop.

2. The method according to claim 1, characterized in that, Step S1 includes: S11: Extract an activity list from construction organization design documents, building information models, or historical project archives, determine the number of activities, the duration of each activity, and the order of activities, and construct a directed acyclic graph of the activities; S12: Determine the quantity of various resources required per unit time during the execution period for each activity according to its process requirements, and form an activity and resource demand matrix; S13: Enter the total capacity limit of various resources in the resource pool, the project duration limit of each project, and the list of project milestone nodes; S14: After normalizing the above data, store it in the project database to obtain the project data model for use in steps S2 to S6.

3. The method according to claim 1, characterized in that, Step S2 includes: S21: For each of the two types of resources on the directed acyclic graph of the activities and Scan each activity and calculate the resources required simultaneously. With resources Number of activities And the resources needed or resources Total number of activities The resource-pair coupling coefficient is calculated using the following formula. : ； in: For resources With resources The coupling coefficient is dimensionless. To simultaneously require resources With resources The number of activities; For the need of resources or resources The total number of activities; / indicates division; and Indexing for different resource types; S22: Each Write a symmetric coupling matrix The coupling matrix Set the diagonal element to 1, and we get The dimensional coupling matrix is ​​used in step S4, where For the number of resource types; S23: Generate the feasible time windows for each activity according to the following sub-steps. : S231: For each activity Based on its process category, at least one construction environment constraint rule is retrieved from the biological and meteorological rule bases. The construction environment constraint rule specifies the feasible value range of at least one variable among temperature, humidity, wind speed, precipitation, or seedling phenological segment. S232: Using meteorological forecast data and seedling phenology calendar as input, the feasibility of each day in the scheduling time domain is determined by substituting the construction environment constraint rules into the data, and a binary label of feasible / infeasible for each day is obtained. S233: Merge consecutively marked as feasible dates into a closed interval segment. Sequentially number each consecutive feasible segment within the scheduling time domain. And use the union of the segments as the activity according to the following formula. The feasible time window : ； in: For the event The feasible time windows, a set in days; For the event The number of feasible window segments; , activities respectively No. The start and end dates of the segment window; Represents the union of sets; Indicates a closed interval; For activity indexing; For window segment index; S234: The coupling matrix and the feasible time windows for each activity. Store the data in the project database to obtain the coupling coefficient data and time window constraint data for use in step S4.

4. The method according to claim 1, characterized in that, Step S3 includes: S31: At each quota allocation trigger moment For each project and various resources The predicted demand for the next scheduling window is obtained from resource demand forecasting. Read the project from the execution log Resources As of Cumulative consumption at any time and resources in the resource pool As of Total cumulative consumption at any moment And read the current quota from the quota table. ; S32: Calculate the resource urgency index using the following formula. : ； in: For the project exist Resources at all times The urgency index is dimensionless; To predict demand, units and resources same; For the resources in the resource pool Total capacity limit, units and resources same; For resources in the resource pool As of Total cumulative consumption at any given time, in units and resources same; To prevent the denominator from being zero, a pre-defined positive small constant is used; This indicates taking the larger value; Index the project; Index for resource types; For time indexing; S33: For various resources The resources in the resource pool are calculated using the following formula. Total capacity The quotas were redistributed among the projects to obtain new allocations. : ； in: To reassign to the project resources Quotas, Units and Resources same; As defined in S32; Calculated according to S32; This represents the summation of all projects participating in the same resource pool; S34: Transfer the new quota The tactical level is issued and written into the quota table, replacing the original... This serves as a hard constraint for the activity scheduling in step S4.

5. The method according to claim 3, characterized in that, Step S4 is solved using a hybrid genetic algorithm (H-GA), including: S41: Initialize the population. Each chromosome consists of two segments, the first segment being the activity priority sequence. The second section is for selecting gene sequences via a window. ,in: For the first The priority value of each activity is dimensionless. For the event The selected feasible time window index, , Obtained according to S233; The number of activities in the project data model; S42: Perform Serial Progress Generation Scheme (SSGS) decoding on each chromosome, according to the stated activity priority sequence. The start time of each activity is determined sequentially using a directed acyclic graph. And make it satisfy the time window constraint shown in the following formula: ； in: For the event The beginning moment; For the event The duration; For the event The window index used in this decoding was initially taken from the chromosome gene. It only applies to this decoding process and does not modify the chromosome gene sequence; , activities respectively No. The start and end dates of the segment window; S43: When the time window constraint described in S42 cannot be satisfied, the following decoding and repair rules shall be used to determine the cause. and And without modifying the chromosome gene sequence : (a) If the selected window segment Inside, Existence makes The earliest established Then Set to this earliest value, Set as ; (b) Otherwise in Search by segment index In the subsequent window segment, find the first one that satisfies... window segment ,Will Set as , Set as ; (c) If neither (a) nor (b) is feasible, then mark the activity as an infeasible solution and apply a preset penalty value to the fitness of the chromosome. S44: Perform a serial scan repair on solutions that violate resource capacity constraints, ensuring that activities executed simultaneously at any given time affect various resource types. The total amount occupied does not exceed the corresponding quota. ;in: The new quota calculated for S33; S45: Using the aforementioned coupled multi-resource equilibrium index As a fitness function, each chromosome is evaluated using the following formula: ； in: The multi-resource equilibrium index is dimensionless. for Time Resources Actual input, units and resources same; For resources The time average within the scheduling time domain, according to Calculation, Units and Resources same; This represents the total length of the scheduling time domain; To prevent the denominator from being zero, a pre-defined positive small constant is used; These are the weighting coefficients of the coupling terms, preset positive constants, and dimensionless. Calculated according to S21; This indicates taking the larger value; Indicates the time domain of scheduling Summation at each discrete time point; This represents the summation of all types of resources in the set of resource types; Indicates that for each satisfied Resource type Summation; For time indexing; and Index for resource types; S46: Perform selection, crossover, and mutation operations to generate the next generation population. After the number of iterations reaches a preset upper limit, the population will be... The sequence of activity start times and resource inputs corresponding to the individual with the smallest value are used as the baseline schedule and resource time distribution output.

6. The method according to claim 5, characterized in that, The hybrid genetic algorithm H-GA employs a twin population co-evolutionary structure for solving the problem. This twin population co-evolutionary structure operates internally within the tactical layer solution process, and is distinct from the macroscopic hierarchical architecture comprised of the strategic, tactical, and operational layers. S47: Within the tactical layer solution process, maintain the quota recommendation subpopulation POP_Q and the scheduling subpopulation POP_S; the quota recommendation vector for each chromosome coding item and resource pair in POP_Q. In POP_S, each chromosome encodes an activity priority sequence according to the two-segment coding structure defined in S41. Selecting gene sequences with a window ;in: Items encoded for the POP_Q chromosome Resources Recommended quota values, units and resources same; S48: After each preset migration cycle In each generation, the elite chromosome with the best fitness is selected from POP_Q, and its encoded quota proposal vector is used. As a hard constraint for each chromosome in POP_S to execute steps S42 to S46, each chromosome in POP_S re-enters the decoding and fitness evaluation process under this hard constraint; where: For migration cycle; S49: Select the coupling multi-resource balance index from POP_S Extract the peak resource requirements of each item and resource pair from the smallest elite chromosome. Calculate the fitness correction term for each chromosome in POP_Q using the following formula. : ； in: This is the fitness correction term for the POP_Q chromosome, and it is dimensionless. Items extracted from POP_S elite chromosomes Resources Peak demand in the current phase, in units and resources same; As defined in S47; The first position on the right side of the equals sign indicates negation, and the area inside the parentheses indicates subtraction; This indicates taking the larger value, with a lower bound of 0; This indicates the summation of all projects and resources; The original fitness of the POP_Q chromosome is superimposed, causing POP_Q to evolve towards a direction of increased quota recommendations for peak demand items and resources, thereby enabling the twin populations to co-evolve through bidirectional migration between S48 and S49.

7. The method according to claim 4, characterized in that, Step S5 includes: S51: The operation layer collects actual on-site progress, resource consumption and meteorological data on a daily basis, and calculates the deviation of each activity from the baseline schedule in real time; S52: Determine whether any of the following four types of triggering conditions are true: (i) Periodic triggering: The time interval since the last rescheduling reaches the preset rescheduling period. ; (ii) Resource gap trigger: There is a project With resources This makes the resource gap more than Exceeding the preset resource gap ratio threshold ; (iii) Critical path slack triggering: Remaining critical path slack Not greater than zero; (iv) External disturbance trigger: Preset external disturbance event is detected; in: The preset rescheduling period; , , As defined in S31; The preset resource gap ratio threshold is dimensionless. for Remaining critical path slack at each time step; To constrain the contract period; The project completion date is calculated based on the current progress using the critical path method. This indicates taking the larger value; if any condition is met, proceed to step S53. S53: Calculate the rescheduling window length using the following formula. : ； in: for The rescheduling window length is specified at any given time. The planned completion date for the next project milestone; This is the preset lower limit for window length; This is the preset maximum window length. ,in and These represent taking the smaller value and taking the larger value, respectively. Indexed by the current time; S54: Time interval The start time of activities other than those already committed remains unchanged as a constraint, and the start time falls within the interval. The activities within are shown in the mixed objective function as follows: Solve again: ； in: The objective function value for the rescheduling phase is dimensionless. Calculated according to S45; The default positive constant is a dimensionless value used to deviate from the sensitivity coefficient. For the set of time intervals corresponding to the rescheduling window, in units; Indicates the time at which the beginning falls into Summing the various activities; For rescheduled activities The beginning moment; Activities in the baseline plan The start time; replace the interval in the original plan with the solution result. At the start of the internal activity, the updated progress plan is obtained.

8. The method according to claim 4, characterized in that, Step S6 includes: S61: The operation layer uploads the actual start time, actual end time, actual consumption of various resources, and on-site disturbance events of each activity to the tactical layer on a daily cycle. S62: The tactical layer updates the predicted demand based on the uploaded data. And extract the items in the next scheduling window. Resources Peak actual input The strategic layer mentioned above was reported; among which: For the project Resources Peak demand within this scheduling window, in units and resources same; S63: The strategic layer receives the report from the tactical layer. With the updated The revised forecast demand is constructed using the following formula. and with Alternative Recalculate the urgency index according to S32 to S34. With the new quota : ； in: The revised forecast demand, in units and resources same; As defined in S31; As defined in S62; This indicates taking the larger value; this step ensures that peak demand items receive a corresponding increase in quota in the next allocation cycle, while the proportional allocation formula in S33 guarantees this. That is, the total capacity constraint of the resource pool is conserved and not violated; S64: The strategic layer will update the new quota. The tactical layer is then issued, triggering it to perform step S4, a local rescheduling, on the affected items, thus creating a two-way closed loop between the upper and lower layers.

9. A hierarchical scheduling optimization platform for engineering projects implementing the method of any one of claims 1 to 8, characterized in that: It includes a multi-source data acquisition module interconnected via a data bus, a biological and meteorological window management module, a resource demand forecasting module, a three-layer scheduling engine module, a disturbance response module, and a digital twin visualization module; The outputs of the biological and meteorological window management module and the resource demand prediction module are all connected to the input of the three-layer scheduling engine module. The three-layer scheduling engine module includes a cascaded strategic layer solver, tactical layer solver, and operational layer scheduler. There is a feedback channel between the three layers from the lower layer to the upper layer. The output of the three-layer scheduling engine module is connected to the digital twin visualization module and the field operation terminal. The output of the disturbance response module is connected to the input of the three-layer scheduling engine module.

10. The platform according to claim 9, characterized in that: The multi-source data acquisition module includes an IoT gateway and several edge nodes. The edge nodes are electrically connected to a weather station, a soil moisture sensor, a tower crane load sensor, a construction vehicle positioning terminal, a personnel positioning card, and a material identification reader. The IoT gateway is connected to the data bus and performs format normalization processing on the raw data before sending it to other modules. The biological and meteorological window management module includes a seedling phenology database, a construction process temperature and humidity rule database, and a meteorological forecast interface. The seedling phenology database and the construction process temperature and humidity rule database are both connected to the meteorological forecast interface. The output end of the biological and meteorological window management module outputs feasible time window data. The resource demand forecasting module is connected to a historical project database and an activity template library. The output of the resource demand forecasting module outputs the resource demand forecast data to the input of the strategic layer solver. In the three-layer scheduling engine module, the strategic layer solver, the tactical layer solver, and the operational layer scheduler communicate bidirectionally via a message bus. The data categories transmitted by the message bus include: quota data issued by the strategic layer solver to the tactical layer solver; activity start and end time instruction data issued by the tactical layer solver to the operational layer scheduler; actual activity start and end time data and actual resource consumption data reported by the operational layer scheduler to the tactical layer solver; and peak resource demand data reported by the tactical layer solver to the strategic layer solver. The strategic layer solver is used for resource pool quota allocation, the tactical layer solver is used for activity scheduling, and the operational layer scheduler is used for daily job instruction generation and execution monitoring. The disturbance response module includes a trigger condition discrimination unit and a rescheduling and invocation unit. The input of the trigger condition discrimination unit is connected to the output of the multi-source data acquisition module, and the output of the rescheduling and invocation unit is connected to the input of the operation layer scheduler. The digital twin visualization module uses a fusion of a 3D geographic information system view and a building information model view as the base map, and overlays and displays Gantt layers, resource histogram layers, coupled multi-resource balance index curve layers, urgency index heatmap layers, and time window conflict alarm layers. It also has an interactive interface for project managers to access and drill down.

Images