Engineering plan intelligent arrangement method and system based on multi-constraint optimization

Abstract

This invention discloses an intelligent project scheduling method and system based on multi-constraint optimization, relating to the field of smart project management. It constructs a multi-dimensional constraint model for the project, including task sequence relationships, resource skill matrices, construction space grids and capacity, and personnel risk coefficients. An initial feasible plan is generated based on the critical path method and heuristic resource balancing. The initial plan is then iteratively optimized using a multi-objective optimization algorithm to obtain the Pareto optimal plan. This invention improves the rationality and feasibility of project scheduling by constructing a multi-dimensional constraint model that comprehensively integrates various constraints such as tasks, resources, space, and personnel risks. By combining the critical path method and multi-objective optimization algorithms, a globally optimal plan is generated, achieving synergistic optimization of schedule, cost, resource balance, and personnel risk, thereby improving project construction efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent engineering management technology, and in particular to an intelligent scheduling method and system for engineering plans based on multi-constraint optimization. Background Technology

[0002] As engineering construction scales expand and project complexity increases, the scientific and efficient nature of project planning plays a decisive role in project quality, schedule, and cost control. Currently, project planning largely relies on manual processes combined with traditional project management tools, primarily using the critical path method for schedule calculation, supplemented by simple resource allocation strategies to generate the plan. This approach struggles to comprehensively consider multi-dimensional constraints such as task timing, resource and skill matching, and construction space occupancy, easily leading to problems like unreasonable resource allocation and spatial conflicts.

[0003] Existing optimization schemes employ single intelligent algorithms for project planning, but these suffer from slow convergence and a tendency to get trapped in local optima, failing to achieve multi-objective collaborative optimization of schedule, cost, resource balance, and personnel risk. Furthermore, various disturbances occur during construction, such as personnel changes and design alterations. Traditional planning methods lack effective real-time response mechanisms, making rapid local adjustments difficult and prone to causing a disconnect between the plan and actual construction.

[0004] Furthermore, most existing solutions do not consider the impact of abnormal personnel behavior on plan execution, lack a dynamic update and feedback mechanism for personnel risk coefficients, and cannot optimize subsequent plan arrangements based on actual on-site execution conditions, further reducing the adaptability and reliability of engineering plans and making it difficult to meet the needs of modern intelligent engineering management. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method and system for intelligent scheduling of engineering plans based on multi-constraint optimization. The technical solution adopted is as follows: A smart scheduling method for engineering plans based on multi-constraint optimization includes the following steps: Step 1: Construct a multi-dimensional constraint model for the project, including task temporal relationships, resource skill matrix, construction space grid and capacity, and personnel risk coefficient; Step 2: Generate an initial feasible plan based on the critical path method and heuristic resource balancing, and iteratively optimize the initial plan using a multi-objective optimization algorithm to obtain the Pareto optimal plan. Step 3: Adopt a rolling time-domain strategy to divide the entire project planning period into multiple rolling windows. At the beginning of each window, based on the actual progress, resource availability status and abnormal events collected in real time by the Internet of Things, deviations are detected and local rescheduling is triggered. Step 4: During the rescheduling process, a substitute resource matching algorithm based on skill similarity is adopted, and combined with spatial grid capacity constraints, a fast local repair is performed to generate an adjusted plan. Step 5: Decompose the adjusted plan into work orders and distribute them for execution. At the same time, collect data on abnormal behavior on site, update the personnel risk coefficient, and feed the risk coefficient back into the multi-objective optimization model in Step 2 for subsequent plan optimization.

[0006] Optionally, the construction of the multi-dimensional constraint model for the project in step 1 specifically includes: defining a task set T, where each task contains a planned duration. Required resource types and quantities Construction location grid Prerequisite Task List Define a resource set R, and personnel with skill vectors. and risk coefficient The construction area is divided into three-dimensional grid units. Each grid has a capacity limit. And construct a resource spatiotemporal occupancy matrix. .

[0007] Optionally, generating the initial feasible plan in step 2 includes: calculating the earliest start time of the task using the critical path method. and latest start time The process involves identifying the critical path, then using a serial scheduling scheme generator combined with priority rules to generate an initial feasible plan while satisfying resource and space constraints. Finally, a multi-objective particle swarm optimization algorithm is used to optimize the initial plan, with particle position encoded as task start time offset. The fitness function includes four objectives: duration, cost, resource balance, and risk. The optimal solution is selected through Pareto non-dominated sorting.

[0008] Optionally, the rolling time-domain strategy in step 3 specifically includes: setting a rolling window length H, acquiring the current actual progress data at the beginning of each window, and calculating the deviation between the actual start time and the planned start time of the critical path task. If the judgment appears Exceeding the preset threshold If a disturbance event is detected, a rescheduling will be triggered. Disturbance events include available personnel falling below a set threshold and design changes; otherwise, the original plan will continue to be executed.

[0009] Optionally, the rapid local repair in step 4 specifically includes: identifying completed tasks and unaffected tasks, and identifying the set of affected tasks. ,Will Sort by minimum slack rule; allocate resources to each task in turn; if the original resource is unavailable, calculate the skill similarity between the substitute resource and the original resource, and select the substitute resource with the highest skill similarity; check the spatial grid capacity; if there is a conflict, try to fine-tune the start time or change the position, and update the occupancy matrix; if no feasible solution can be found, expand the affected area and reschedule.

[0010] Optionally, the personnel risk coefficient in step 5 can be updated using the following incremental update formula: ; in This is the updated personnel risk coefficient. This is the learning rate for abnormal events, with a value range of 0.05 ≤ ≤0.2, A good performance learning rate The value range is 0.03≤ ≤0.15, These are abnormal events, including absenteeism, unauthorized absences, and overlapping activity patterns. The value range for a single abnormal event is 0.2. <0.5, the value range is 0.5 when multiple similar abnormal events are superimposed. 1; "Good performance" means the task is completed on time with no violations. The value for a single task's "good performance" rating is 0.2. <0.4, when multiple consecutive good performances are combined, the value is taken as 0.4. 1, The risk coefficient for personnel, with a value range of 0 ≤ ≤1.

[0011] Optionally, passive apathy is detected by personnel wearing devices. When the duration of a person remaining stationary without a task pause instruction exceeds a set duration threshold, it is determined as an abnormal event. Abnormal absence from duty is detected by electronic fence technology. When the duration of a person's trajectory outside the work area without a valid approval record exceeds a set time threshold, it is determined. Trajectory overlap is determined by calculating the similarity of the trajectories of two pairs of personnel using a dynamic time warping algorithm. When the similarity of the trajectories of personnel without collaborative tasks exceeds a set similarity threshold, it is determined.

[0012] Optionally, step 5, work order distribution, includes: generating a standard work order containing task ID, construction location grid, planned working hours, material list, and resource allocation information, pushing it to the corresponding personnel via mobile device, and tracking the work order status in real time: delivered, read, confirmed, started execution, completed; if the work order is not confirmed after a preset time, the set multi-channel reminders are automatically triggered.

[0013] Optionally, a risk objective is introduced into the fitness function of the multi-objective optimization algorithm, expressed as: ;in For personnel In the mission The participation weight in the critical path task ranges from 0.6 to 0.6. <1, for non-critical path tasks, the value range is 0≤ <0.6, optimizations are made to minimize the involvement of high-risk personnel in critical path tasks.

[0014] This intelligent engineering planning system based on multi-constraint optimization implements an intelligent engineering planning method based on multi-constraint optimization. The system includes a data integration module, an initial plan generation module, a rolling window manager, a local rescheduling engine, a work order distribution module, a status tracking module, and an abnormal behavior identification and feedback module. The data integration module collects basic engineering data and real-time IoT data to construct and maintain a multi-dimensional constraint model. The initial plan generation module generates an initial plan based on the critical path method and heuristic resource balancing, and optimizes it using a multi-objective particle swarm optimization algorithm. The rolling window manager divides the rolling window, monitors actual progress, and determines whether rescheduling is triggered. The local rescheduling engine performs local repairs based on skill similarity and spatial grid constraints when disturbances occur, generating an adjusted plan. The work order distribution module generates and pushes work orders. The status tracking module tracks the execution status. The abnormal behavior identification and feedback module monitors abnormal behaviors on-site in real time, updates personnel risk coefficients, and feeds these risk coefficients back to the initial plan generation module to optimize subsequent plans.

[0015] In summary, the present invention has at least one of the following beneficial technical effects: This invention provides an intelligent scheduling method and system for engineering plans based on multi-constraint optimization. By constructing a multi-dimensional constraint model, it comprehensively integrates various constraints such as tasks, resources, space, and personnel risks, thereby improving the rationality and feasibility of plan scheduling. Combining the critical path method and multi-objective optimization algorithms, it generates a globally optimal plan, achieving synergistic optimization of schedule, cost, resource balance, and personnel risks, thus improving the efficiency of engineering construction.

[0016] By employing a rolling time-domain strategy and a local rescheduling mechanism, the system can respond to disturbances during construction in real time, quickly adjust the plan, reduce deviations between the plan and actual construction, and ensure the orderly progress of the project. Through abnormal behavior identification and dynamic updates of personnel risk coefficients, a closed-loop optimization mechanism is formed, continuously improving the scientific rigor and adaptability of subsequent planning and reducing project execution risks.

[0017] Standardized work order distribution and status tracking enhance the standardization and transparency of task execution, reduce communication costs, improve project management efficiency, and help achieve intelligent project management. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the intelligent scheduling method for engineering plans based on multi-constraint optimization according to the present invention.

[0019] Figure 2 This is a schematic diagram of the intelligent scheduling system architecture for engineering plans according to a specific embodiment of the present invention. Detailed Implementation

[0020] The present invention will be further described in detail below with reference to the accompanying drawings.

[0021] This invention discloses an intelligent scheduling method and system for engineering plans based on multi-constraint optimization.

[0022] Reference Figure 1 and Figure 2 Example 1, an intelligent scheduling method for engineering plans based on multi-constraint optimization, includes the following steps: Step 1: Construct a multi-dimensional constraint model for the project, including task temporal relationships, resource skill matrix, construction space grid and capacity, and personnel risk coefficient; Step 2: Generate an initial feasible plan based on the critical path method and heuristic resource balancing, and iteratively optimize the initial plan using a multi-objective optimization algorithm to obtain the Pareto optimal plan. Step 3: Adopt a rolling time-domain strategy to divide the entire project planning period into multiple rolling windows. At the beginning of each window, based on the actual progress, resource availability status and abnormal events collected in real time by the Internet of Things, deviations are detected and local rescheduling is triggered. Step 4: During the rescheduling process, a substitute resource matching algorithm based on skill similarity is adopted, and combined with spatial grid capacity constraints, a fast local repair is performed to generate an adjusted plan. Step 5: Decompose the adjusted plan into work orders and distribute them for execution. At the same time, collect data on abnormal behavior on site, update the personnel risk coefficient, and feed the risk coefficient back into the multi-objective optimization model in Step 2 for subsequent plan optimization.

[0023] By adopting the above technical solutions and implementing a phased approach, intelligent scheduling and dynamic optimization of engineering plans are achieved. First, a multi-dimensional constraint model covering tasks, resources, space, personnel risks, and other factors is constructed to provide comprehensive constraints for plan scheduling, ensuring that the plan meets actual engineering needs. Based on the critical path method, task priorities and schedule baselines are defined. An initial feasible plan is generated using a heuristic resource balancing strategy, and then iteratively optimized through a multi-objective optimization algorithm to select the optimal plan that balances multiple objectives. A rolling time-domain strategy is adopted to break down long-term plans into multiple short-term windows, enabling dynamic monitoring and real-time adjustments to the plan. For disruptive events, local rescheduling is used to quickly correct plan deviations. Finally, through work order distribution and abnormal data feedback, a closed-loop optimization is formed, feeding back on-site execution data to subsequent plan optimizations, continuously improving the plan's adaptability and reliability.

[0024] Example 2, step 1 of constructing the multi-dimensional constraint model for the project specifically includes: defining a task set T, where each task includes a planned duration. Required resource types and quantities Construction location grid Prerequisite Task List Define a resource set R, and personnel with skill vectors. and risk coefficient The construction area is divided into three-dimensional grid units. Each grid has a capacity limit. And construct a resource spatiotemporal occupancy matrix. .

[0025] By adopting the above technical solutions, the core parameters of each task are clearly defined through the task set, and the time requirements, resource needs, construction locations, and dependencies of each task are clearly defined, ensuring the rationality of task scheduling. The resource set is defined to clarify the skill attributes and risk levels of personnel, providing a basis for resource matching and risk control. The construction area is divided into a three-dimensional spatiotemporal grid, and the capacity limitations of each grid are clearly defined. By constructing a resource spatiotemporal occupancy matrix, the occupancy status of resources in different spatiotemporal grids is recorded in real time, effectively avoiding spatial conflicts and resource overload problems, laying the foundation for subsequent plan optimization and conflict detection.

[0026] Example 3, step 2 of generating an initial feasible plan includes: calculating the earliest start time of the task using the critical path method. and latest start time The process involves identifying the critical path, then using a serial scheduling scheme generator combined with priority rules to generate an initial feasible plan while satisfying resource and space constraints. Finally, a multi-objective particle swarm optimization algorithm is used to optimize the initial plan, with particle position encoded as task start time offset. The fitness function includes four objectives: duration, cost, resource balance, and risk. The optimal solution is selected through Pareto non-dominated sorting.

[0027] By adopting the above technical solution, the critical path method is first used to calculate the earliest and latest start times of each task, accurately identifying the critical path and clarifying the core tasks affecting the overall project duration, providing a basis for subsequent resource allocation and priority ranking. A serial scheduling scheme generator, combined with priority rules, is used to allocate resources and arrange task sequences in an orderly manner, generating an initial feasible plan, while satisfying resource and space constraints. A multi-objective particle swarm optimization algorithm is then used to iteratively optimize the initial plan. By encoding the task start time offsets and combining them with a multi-objective fitness function covering project duration, cost, resource balance, and risk, the globally optimal plan is selected through Pareto non-dominated sorting, achieving multi-objective collaborative optimization.

[0028] Example 4, the rolling time-domain strategy in step 3 specifically includes: setting a rolling window length H, acquiring the current actual progress data at the beginning of each window, and calculating the deviation between the actual start time and the planned start time of the critical path task. If the judgment appears Exceeding the preset threshold If a disturbance event is detected, a rescheduling will be triggered. Disturbance events include available personnel falling below a set threshold and design changes; otherwise, the original plan will continue to be executed.

[0029] By adopting the above technical solution and setting a reasonable rolling window length, the entire project planning period is divided into multiple consecutive short windows. At the beginning of each window, real-time IoT data is collected synchronously to obtain information on actual progress, resource availability, and abnormal events. By calculating the deviation between the actual and planned start times of critical path tasks, it is determined whether a preset threshold is exceeded. Simultaneously, it detects whether there are disturbances such as personnel shortages or design changes. If the rescheduling trigger conditions are met, a local rescheduling process is initiated; otherwise, the original plan continues to execute, ensuring that the plan can adapt to the actual site conditions in real time and reducing the disconnect between planning and construction.

[0030] Example 5, step 4 of the rapid local repair specifically includes: identifying completed tasks and unaffected tasks, and identifying the set of affected tasks. ,Will Sort by minimum slack rule; allocate resources to each task in turn; if the original resource is unavailable, calculate the skill similarity between the substitute resource and the original resource, and select the substitute resource with the highest skill similarity; check the spatial grid capacity; if there is a conflict, try to fine-tune the start time or change the position, and update the occupancy matrix; if no feasible solution can be found, expand the affected area and reschedule.

[0031] By adopting the above technical solution, during rescheduling, completed tasks and tasks unaffected by disturbances are first identified to avoid ineffective adjustments and accurately identify the set of affected tasks. Affected tasks are then sorted according to the minimum slack rule, prioritizing adjustments to critical tasks to ensure the overall project duration is not significantly impacted. When allocating resources to affected tasks, if the original resource is unavailable, the skill similarity between the substitute resource and the original resource is calculated, and the substitute resource with the highest matching degree is selected to ensure task execution quality. Simultaneously, the spatial grid capacity is checked; if spatial conflicts occur, they are resolved by fine-tuning task start times or changing construction locations, and the resource spatiotemporal occupancy matrix is ​​updated in real time to ensure resource and spatial constraints are met. If no feasible solution can be found, the affected area is expanded for rescheduling to ensure the feasibility of the plan.

[0032] Example 6: In step 5, the personnel risk coefficient is updated using the following incremental update formula: ; in This is the updated personnel risk coefficient. This is the learning rate for abnormal events, with a value range of 0.05 ≤ ≤0.2, A good performance learning rate The value range is 0.03≤ ≤0.15, These are abnormal events, including absenteeism, unauthorized absences, and overlapping activity patterns. The value range for a single abnormal event is 0.2. <0.5, the value range is 0.5 when multiple similar abnormal events are superimposed. 1; "Good performance" means the task is completed on time with no violations. The value for a single task's "good performance" rating is 0.2. <0.4, when multiple consecutive good performances are combined, the value is taken as 0.4. 1, The risk coefficient for personnel, with a value range of 0 ≤ ≤1.

[0033] By adopting the above technical solution, based on personnel's on-site performance, anomalies and positive performance are distinguished, and the impact of each on the risk coefficient is adjusted through a learning rate. Anomalies increase the personnel risk coefficient, while positive performance decreases it. This formula allows for real-time updates to the personnel risk coefficient, ensuring consistency between the risk coefficient and the actual performance of personnel. This provides accurate risk information for resource allocation in subsequent planning and enables reasonable management of high-risk personnel.

[0034] Example 7: Passive shirking is detected by personnel wearing devices. When the duration of a person remaining stationary without a task pause instruction exceeds a set duration threshold, it is determined as an abnormal event. Abnormal absence from post is detected by electronic fence technology. When the duration of a person's trajectory outside the work area without a valid approval record exceeds a set time threshold, it is determined. Trajectory overlap is calculated by dynamic time warping algorithm to determine the similarity of the trajectories of two people. When the similarity of the trajectories of people without collaborative tasks exceeds a set similarity threshold, it is determined.

[0035] By employing the above technical solutions, for passive resistance, wearable devices are used to collect personnel activity data. By comparing the duration of inactivity with task pause instructions, passive resistance is identified. For abnormal absences, electronic fence technology is used to delineate work areas. By monitoring whether personnel trajectories exceed these areas and whether valid approval records are available, abnormal absences are identified. For overlapping trajectories, a dynamic time warping algorithm is used to calculate the trajectory similarity of personnel without collaborative tasks. A similarity threshold is used to determine if there are any abnormal trajectory overlaps, ensuring that all types of abnormal behaviors can be accurately and efficiently identified.

[0036] Example 8, step 5 of the work order distribution includes: generating a standard work order containing task ID, construction location grid, planned working hours, material list, and resource allocation information, pushing it to the corresponding personnel via mobile terminal, and tracking the work order status in real time: delivered, read, confirmed, started execution, completed; if the work order is not confirmed after a preset time, the set multi-channel reminder is automatically triggered.

[0037] By adopting the above technical solution, standardized work orders containing core task information are generated, ensuring that executors can clearly obtain task requirements, resource information, and construction locations, avoiding information discrepancies. Work orders are pushed via mobile devices, enabling rapid task distribution and real-time tracking of the entire work order lifecycle, allowing for timely monitoring of task execution progress. For situations where work orders are not promptly confirmed after issuance, multi-channel reminders are automatically triggered to ensure that executors receive and confirm tasks in a timely manner, reducing communication costs and guaranteeing timely task initiation and orderly execution.

[0038] Example 9: The fitness function of the multi-objective optimization algorithm introduces a risk objective, expressed as: ;in For personnel In the mission The participation weight in the critical path task ranges from 0.6 to 0.6. <1, for non-critical path tasks, the value range is 0≤ <0.6, optimizations are made to minimize the involvement of high-risk personnel in critical path tasks.

[0039] By adopting the above technical solution, the risk target is calculated by summing the product of the personnel risk coefficient and the personnel's participation weight in the task, which can accurately reflect the overall risk level of the plan. Through the optimization of this target, the algorithm is guided to minimize the participation of high-risk personnel in critical path tasks during the iteration process, reduce the execution risk of critical tasks, and achieve multi-objective collaborative optimization of schedule, cost, resource balance and risk, thereby improving the reliability of the plan.

[0040] Example 10: An intelligent engineering plan orchestration system based on multi-constraint optimization. This system implements an intelligent engineering plan orchestration method based on multi-constraint optimization. The system includes a data integration module, an initial plan generation module, a rolling window manager, a local rescheduling engine, a work order distribution module, a status tracking module, and an abnormal behavior identification and feedback module. The data integration module collects basic engineering data and real-time IoT data to construct and maintain a multi-dimensional constraint model. The initial plan generation module generates an initial plan based on the critical path method and heuristic resource balancing, and optimizes it using a multi-objective particle swarm optimization algorithm. The rolling window manager divides the rolling window, monitors actual progress, and determines whether rescheduling is triggered. When disturbances occur, the local rescheduling engine performs local repairs based on skill similarity and spatial grid constraints, generating an adjusted plan. The work order distribution module generates and pushes work orders. The status tracking module tracks the execution status. The abnormal behavior identification and feedback module monitors abnormal behaviors on-site in real time, updates personnel risk coefficients, and feeds these risk coefficients back to the initial plan generation module to optimize subsequent plans.

[0041] By adopting the above technical solutions, the data integration module is responsible for the collection and processing of multi-source data, constructing and maintaining a multi-dimensional constraint model to provide data support for the entire system. The initial plan generation module generates and optimizes the initial plan, outputting the optimal plan. The scrolling window manager enables dynamic monitoring of the plan and determines the rescheduling trigger conditions. The local rescheduling engine quickly repairs the plan when disturbances occur, ensuring plan continuity. The work order distribution module and the status tracking module work together to distribute tasks and track progress, while the abnormal behavior identification and feedback module detects abnormal behavior, updates risk factors, and provides feedback, forming a closed-loop optimization to ensure the system can continuously adapt to the actual situation on site and improve the level of intelligence in project management.

[0042] The following specific embodiments illustrate the implementation principle of the present invention: This invention uses a large-scale industrial plant construction project as its implementation scenario. The project includes three major construction phases: civil engineering, electromechanical installation, and equipment commissioning, encompassing 50 construction tasks, 30 construction personnel, 8 types of construction equipment, 12 types of construction materials, and covering an area of ​​8000 square meters. The planned total construction period is 180 days. This embodiment fully implements the core process of the technical solution, incorporating all calculation formulas and actual calculation processes, and using quantitative data to demonstrate the advantages and efficiency of the solution.

[0043] Constructing a multi-dimensional constraint model for the project: Define the task set T as 50 construction tasks, with each task having a clearly defined planned duration. Required resource types and quantities Construction location grid Prerequisite Task List Such as foundation excavation tasks =10 days, requires 1 excavator and 3 civil workers, construction location is G8-5-0, no prerequisites.

[0044] Define a resource set R as 30 construction workers, and assign a skill vector to each worker. and initial risk coefficient The skill vector is quantified into a value of 0-1 according to the skill type. The initial risk coefficient is set to 0.1-0.5, with 0.2 for personnel with no abnormal records and 0.4 for personnel with minor violations.

[0045] The construction area was divided into three-dimensional grid units of 10m×10m×1 layer. Define the upper limit of capacity for each grid. The main construction area grid can accommodate a maximum of 2 large pieces of equipment and 5 construction workers, while the material storage area grid can store a maximum of 3 types of materials. At the same time, a resource spatiotemporal occupancy matrix is ​​constructed. It records the resource occupancy status of the grid at different times in real time, thus avoiding resource and space conflicts from the source.

[0046] This step, through standardized quantification and matrix modeling, takes only 2 hours for the initial modeling and conflict pre-detection, which is significantly more efficient than the 3 days of traditional manual management. It also achieves full-dimensional constraint coverage and reduces the constraint omission rate to 0.

[0047] Generate and optimize an initial feasible plan: Critical path method calculation: Calculate the earliest start time for all tasks. and latest start time The calculation formula is: If there is no prerequisite task, then , For the subsequent tasks, the final task Total planned construction period ； Calculations showed that 12 tasks, including foundation excavation, rebar tying, concrete pouring, main power supply line laying, and core equipment commissioning, had a slack of 0 and were identified as critical path tasks, forming the core link of the overall project duration, with an identification accuracy of 100%.

[0048] Initial plan generation and optimization: A serial scheduling scheme generator combined with critical task priority rules is used to allocate resources to each task while meeting resource and space constraints, generating an initial feasible plan within 4 hours. Based on this, a multi-objective particle swarm optimization algorithm is employed, encoding particle positions as task start time offsets. Four optimization objectives are time, cost, resource balance, and risk, with the risk objective function being the most prominent. The calculation formula is: The total number of tasks is 50. The set of personnel assigned to task i. The weight of personnel participation in the task is set, with a value ranging from 0.6 to 0.6 for critical path tasks. <1, for non-critical path tasks, the value range is 0≤ <0.6. Taking the main power supply line laying task as an example, two electromechanical workers are assigned, each with a risk coefficient of 0.2. The core operation participation weight is 0.9, and the auxiliary operation participation weight is 0.7, resulting in a calculated risk value of 0.32. The initial overall project risk target value is 8.6, which decreases to 3.2 after 50 algorithm iterations, representing a 62.8% reduction in overall risk. The optimal plan is selected through Pareto non-dominated sorting, achieving a plan executability of 95%.

[0049] Dynamic monitoring of rolling time-domain policies: Set the rolling window length H = 7 days, and divide the total project duration of 180 days into 26 rolling windows. At the beginning of each window, calculate the deviation between the actual start time and the planned start time of the critical path task. The calculation formula is: Actual Start Time / Planned Start Time A deviation threshold δ = 1 day is set. When the number of available personnel falls below the threshold or a disturbance event such as a design change is detected, a local rescheduling is triggered. In the 8th window, it was detected that two electromechanical workers responsible for laying the main power supply line suddenly took leave, and the number of available electromechanical workers fell below the set threshold. The system completed the judgment and triggered rescheduling within 5 minutes, which is 24 times more efficient than the traditional manual response time of 2 hours. This step, through 7-day short-window high-frequency monitoring combined with real-time data collection from the Internet of Things, achieved 100% data accuracy and triggered only 6 necessary reschedulings throughout the entire process, with no invalid adjustments.

[0050] Local rescheduling for rapid repair: In response to disturbance events, 25 unaffected tasks, such as civil engineering foundation construction and equipment commissioning preparation, were first identified. Then, 3 affected tasks, namely main power supply line laying, distribution box installation, and lighting circuit commissioning, were precisely identified. After sorting them according to the minimum slack rule, the critical path tasks were adjusted first.

[0051] For tasks where the original resource is unavailable, calculate the skill similarity between the substitute resource and the original resource. The calculation formula is: Skill Similarity = Substitute Original Substitute Original

[0052] The original power supply installation worker's skill vector was [0.95, 0.8, 0.7], and the substitute worker's skill vector was [0.9, 0.85, 0.75]. The calculated skill similarity was 0.998, allowing for the rapid selection of the substitute worker with the highest skill match, with a matching time of only 1 minute. Simultaneously, the system checked the construction space grid capacity and found equipment conflicts in the main power supply line laying work grid. The system automatically adjusted the task start time by one day and updated the resource spatiotemporal occupancy matrix, with conflict resolution taking ≤5 minutes.

[0053] The entire local rescheduling process took only 10 minutes to generate the adjusted complete plan, maintaining a total project duration of 180 days without delay. Compared to the traditional manual adjustment time of 1 day, the efficiency was improved by more than 140 times. Moreover, the adjusted plan was seamlessly integrated with the original plan, with no resource or space conflicts.

[0054] Work order distribution and closed-loop feedback: The adjusted plan is broken down into standardized work orders, each containing a task ID, construction location grid, planned working hours, material list, and resource allocation information. These work orders are pushed to relevant personnel via mobile devices, and their entire lifecycle status is tracked in real time: delivered, read, confirmed, started, and completed. A preset confirmation time of 2 hours is set for work orders; if no confirmation is received within this timeframe, a dual-channel reminder is automatically triggered via mobile message and SMS. The on-time confirmation rate for all work orders is 100%, and the information transmission accuracy rate is 100%, with no delays caused by information omissions.

[0055] During work order execution, wearable devices, electronic fences, and dynamic time warping algorithms are used to accurately identify three types of abnormal events: passive resistance, abnormal absence from duty, and overlapping activity patterns. Simultaneously, good performance with timely task completion and no violations is recorded. The personnel risk coefficient is incrementally updated, calculated using the following formula: ; This is the learning rate for abnormal events, with a value range of 0.05 ≤ ≤0.2, in this embodiment, it is taken as 0.15; A good performance learning rate The value range is 0.03≤ ≤0.15, in this embodiment we take 0.1; These are abnormal events, including absenteeism, unauthorized absences, and overlapping activity patterns. The value range for a single abnormal event is 0.2. <0.5, the value range is 0.5 when multiple similar abnormal events are superimposed. 1; "Good performance" means the task is completed on time with no violations. The value for a single task's "good performance" rating is 0.2. <0.4, when multiple consecutive good performances are combined, the value is taken as 0.4. 1, The risk coefficient for personnel, with a value range of 0 ≤ ≤1.

[0056] A construction worker triggered a passive resistance incident. =0.3, the risk coefficient before the update was 0.2, and the calculated value after the update is 0.245. A substitute electromechanical worker achieved one good performance. =0.3, the risk coefficient before the update was 0.2, and the calculated value after the update is 0.17. A worker with an initial risk coefficient of 0.4, who triggered 2 abnormal absences and completed 3 good performances, =0.6, =0.5, and the calculated updated risk coefficient is 0.44.

[0057] The updated personnel risk coefficients are fed back to the multi-objective optimization model in real time, serving as the core parameter for the next rolling window plan optimization. In subsequent optimizations, the task participation weight of high-risk personnel is reduced, and they are not assigned to critical path tasks. Low-risk personnel are prioritized for assignment to core critical tasks. The overall risk target for the next window project decreases from 3.2 to 2.8, a further reduction of 12.5%. This step achieves dynamic closed-loop updates of the personnel risk coefficients. In the later stages of the project, the average personnel risk decreased from 0.25 to 0.18, and the proportion of high-risk personnel decreased from the initial 10% to 0. No safety accidents occurred during construction.

[0058] This solution was implemented throughout the entire project, with parameter values ​​strictly adhering to technical requirements: plan generation efficiency was improved by over 40 times compared to traditional manual methods, and rescheduling response efficiency was improved by over 140 times; resource and space conflict rate was reduced to 0, and the construction process was continuous and uninterrupted; personnel risks were accurately and dynamically managed, with a safety accident rate of 0; the project was completed 5 days ahead of schedule, optimizing the construction period by 2.8%, and construction costs were reduced by 8% compared to expectations. Compared to the 12% cost overrun of traditional manual methods, costs were optimized by 20%; data exchange between system modules and automatic command triggering improved overall management efficiency by 80% and reduced management costs by 30%. It can be widely applied to the planning and management of various large-scale projects and is adapted to the needs of modern smart engineering management.

[0059] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. An engineering plan intelligent arrangement method based on multi-constraint optimization, characterized in that, Includes the following steps: Step 1: Construct a multi-dimensional constraint model for the project, including task temporal relationships, resource skill matrix, construction space grid and capacity, and personnel risk coefficient; Step 2: Generate an initial feasible plan based on the critical path method and heuristic resource balancing, and iteratively optimize the initial plan using a multi-objective optimization algorithm to obtain the Pareto optimal plan. Step 3: Adopt a rolling time-domain strategy to divide the entire project planning period into multiple rolling windows. At the beginning of each window, based on the actual progress, resource availability status and abnormal events collected in real time by the Internet of Things, deviations are detected and local rescheduling is triggered. Step 4: During the rescheduling process, a substitute resource matching algorithm based on skill similarity is adopted, and combined with spatial grid capacity constraints, a fast local repair is performed to generate an adjusted plan. Step 5: Decompose the adjusted plan into work orders and distribute them for execution. At the same time, collect data on abnormal behavior on site, update the personnel risk coefficient, and feed the risk coefficient back into the multi-objective optimization model in Step 2 for subsequent plan optimization.

2. The intelligent scheduling method for engineering plans based on multi-constraint optimization according to claim 1, characterized in that, The construction of the engineering multi-dimensional constraint model in step 1 specifically includes: defining a task set T, each task containing a planned duration , required resource types and quantities , construction location grid , pre-task list ; defining a resource set R, personnel having a skill vector and risk coefficient ; dividing the construction area into three-dimensional grid units , each grid having an upper limit of capacity , and constructing a resource space-time occupation matrix .

3. The intelligent scheduling method for engineering plans based on multi-constraint optimization according to claim 2, characterized in that, Step 2, generating the initial feasible plan, includes calculating the earliest start time of the task using the critical path method. and latest start time The process involves identifying the critical path, then using a serial scheduling scheme generator combined with priority rules to generate an initial feasible plan while satisfying resource and space constraints. Finally, a multi-objective particle swarm optimization algorithm is used to optimize the initial plan, with particle position encoded as task start time offset. The fitness function includes four objectives: duration, cost, resource balance, and risk. The optimal solution is selected through Pareto non-dominated sorting.

4. The intelligent scheduling method for engineering plans based on multi-constraint optimization according to claim 3, characterized in that, The rolling time domain strategy in step 3 specifically includes: setting a rolling window length H, at the beginning of each window, obtaining current actual progress data, calculating the deviation of the actual start time of the critical path task from the planned start time ; if it is judged that more than a preset threshold or a disturbance event is detected, triggering rescheduling, the disturbance event including available personnel below a set threshold and design changes, otherwise continuing to execute the original plan.

5. The intelligent scheduling method for engineering plans based on multi-constraint optimization according to claim 4, characterized in that, Step 4, the rapid local repair, specifically includes: identifying completed tasks and unaffected tasks, and identifying the set of affected tasks. ,Will Sort by minimum slack rule; allocate resources to each task in turn; if the original resource is unavailable, calculate the skill similarity between the substitute resource and the original resource, and select the substitute resource with the highest skill similarity; check the spatial grid capacity; if there is a conflict, try to fine-tune the start time or change the position, and update the occupancy matrix; if no feasible solution can be found, expand the affected area and reschedule.

6. The intelligent scheduling method for engineering plans based on multi-constraint optimization according to claim 5, characterized in that, In step 5, the personnel risk coefficient is updated using the following incremental update formula: ； in This is the updated personnel risk coefficient. This is the learning rate for abnormal events, with a value range of 0.05 ≤ ≤0.2, This is the good performance learning rate, with a value ranging from 0.03 to 1. ≤0.15, These are abnormal events, including absenteeism, unauthorized absences, and overlapping activity patterns. The value range for a single abnormal event is 0.

2. <0.5, the value range is 0.5 when multiple similar abnormal events are superimposed. 1; "Good performance" means the task is completed on time with no violations. The value for a single task's "good performance" rating is 0.

2. <0.4, when multiple consecutive good performances are combined, the value is taken as 0.

4. 1, The risk coefficient for personnel, with a value range of 0 ≤ ≤1.

7. The intelligent scheduling method for engineering plans based on multi-constraint optimization according to claim 6, characterized in that, Passive apathy is detected by personnel wearing devices. When the duration of a person remaining stationary without a task pause instruction exceeds a set duration threshold, it is considered an abnormal event. Abnormal absence from duty is detected by electronic fence technology. When the duration of a person's trajectory outside the work area without a valid approval record exceeds a set time threshold, it is considered an abnormal event. Trajectory overlap is detected by calculating the similarity of the trajectories of two pairs of personnel using a dynamic time warping algorithm. When the similarity of the trajectories of personnel without collaborative tasks exceeds a set similarity threshold, it is considered an abnormal event.

8. The intelligent scheduling method for engineering plans based on multi-constraint optimization according to claim 7, characterized in that, Step 5, work order distribution, includes: generating a standard work order containing task ID, construction location grid, planned working hours, material list, and resource allocation information, pushing it to the corresponding personnel via mobile device, and tracking the work order status in real time: delivered, read, confirmed, started execution, completed; if the work order is not confirmed within a preset time after being issued, the set multi-channel reminders are automatically triggered.

9. The intelligent scheduling method for engineering plans based on multi-constraint optimization according to claim 8, characterized in that, The fitness function of the multi-objective optimization algorithm incorporates a risk objective, expressed as: ;in For personnel In the mission The participation weight in the critical path task ranges from 0.6 to 0.

6. <1, for non-critical path tasks, the value range is 0≤ <0.6, optimizations are made to minimize the involvement of high-risk personnel in critical path tasks.

10. An intelligent scheduling system for engineering plans based on multi-constraint optimization, characterized in that, To implement the intelligent scheduling method for engineering plans based on multi-constraint optimization as described in claim 9, the system includes a data integration module, an initial plan generation module, a rolling window manager, a local rescheduling engine, a work order distribution module, a status tracking module, and an abnormal behavior identification and feedback module. The data integration module is used to collect basic engineering data and real-time IoT data, and to construct and maintain a multi-dimensional constraint model. The initial plan generation module generates an initial plan based on the critical path method and heuristic resource balancing, and optimizes it using a multi-objective particle swarm optimization algorithm. The scroll window manager is used to divide scroll windows, monitor actual progress, and determine whether to trigger rescheduling. When a disturbance occurs, the local rescheduling engine performs local repair based on skill similarity and spatial grid constraints, generating an adjustment plan; the work order distribution module generates and pushes work orders; the status tracking module tracks the execution status; and the abnormal behavior identification and feedback module monitors abnormal behavior on-site in real time, updates personnel risk coefficients, and feeds back the risk coefficients to the initial plan generation module to optimize subsequent plans.

Images