Artificial intelligence-based engineering resource scheduling optimization system and method

Abstract

This invention discloses an artificial intelligence-based engineering resource scheduling optimization system and method, belonging to the field of engineering management technology. It establishes a resource status perception network by deploying IoT sensors and constructs a progress benchmark system based on the project plan. By comparing the actual progress with the progress benchmark system, it identifies tasks with excessive deviations that need to be scheduled. It constructs a task urgency classification model, extracts the progress dynamics, task dependencies, and resource supply characteristics of the tasks to be scheduled, generates a resource demand urgency level for each task, and generates a scheduling demand description. It matches the scheduling demand with the resource status, constructs a multi-objective optimization function considering spatial distance, workload, and type suitability, obtains a resource allocation scheme by solving the function, sends scheduling instructions according to the allocation scheme, monitors the execution status, tracks resource location and task progress in real time, and re-triggers the scheduling process when a new deviation is detected, forming a closed-loop dynamic optimization mechanism.

Description

Technical Field

[0001] This invention relates to the field of engineering management technology, specifically to an artificial intelligence-based engineering resource scheduling and optimization system and method. Background Technology

[0002] In modern large-scale engineering construction, multiple sub-projects or work packages are often carried out in parallel. Efficient management of these projects heavily relies on precise coordination of schedules and resources. Traditional engineering resource scheduling methods primarily depend on the project manager's experience and static planning-based scheduling approaches, such as using Gantt charts and the critical path method to pre-determine resource allocation plans. However, the construction environment is highly dynamic and uncertain. Unexpected situations such as weather changes, material supply delays, and equipment failures frequently occur, easily leading to deviations between actual and projected progress. This static management model lacks real-time feedback and dynamic adjustment capabilities. When schedule deviations occur, managers struggle to grasp the overall resource status promptly and comprehensively, hindering the ability to quickly make optimal scheduling decisions. This results in delayed resource scheduling response, becoming a bottleneck restricting the improvement of engineering efficiency.

[0003] While some engineering projects have adopted project management software to digitally track progress and manage resources, these systems often have significant limitations. First, they typically treat progress management and resource management as two relatively independent modules, lacking an automated intelligent link between schedule deviation identification and resource reallocation, still requiring substantial manual intervention for analysis and decision-making. Second, when multiple projects experience simultaneous delays or advances, the system struggles to quickly calculate the optimal resource reallocation plan to balance overall progress. For example, a delay in one project might necessitate the transfer of critical equipment from other projects that are ahead of schedule. However, existing methods often fail to comprehensively consider the current location of resources, transfer costs, and the cascading impact on the subsequent progress of each project, easily leading to resource conflicts, localized resource idleness or shortages, and hindering cross-project resource collaboration and the maximization of overall benefits. Summary of the Invention

[0004] The purpose of this invention is to provide an artificial intelligence-based engineering resource scheduling and optimization system and method to solve the problems mentioned in the background art.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: an artificial intelligence-based engineering resource scheduling optimization method, comprising:

[0006] S100: By using IoT sensors deployed at construction sites, real-time quantity data and location coordinates of manpower, equipment and pipeline resources for each project are obtained, and a multi-dimensional feature vector containing resource type, working status and performance parameters is established to form a resource status perception network.

[0007] S200. Construct a schedule baseline system that includes task dependencies and time constraints based on project plan data;

[0008] S300: By comparing the actual progress data of each project with the progress benchmark system, identify the tasks to be scheduled that have a progress deviation exceeding a preset threshold.

[0009] S400: Collect historical project data to build a task urgency classification model, extract the progress dynamics, task dependencies and resource supply characteristics of the tasks to be scheduled, input the trained model, generate the resource demand urgency level of each task to be scheduled, and construct a scheduling demand description that includes the required resource type, quantity specifications and time window.

[0010] S500: Match the scheduling requirement description with the resource status awareness network. Based on the spatial distance between the current location of the resource and the location of the task to be scheduled, the current workload status of the resource, and the type compatibility between the resource and the task, construct a multi-objective optimization function. By solving the multi-objective optimization function, generate a resource allocation scheme that maximizes the overall scheduling efficiency, in which priority is given to matching the resource with the closest spatial distance and the highest type compatibility for high urgency tasks.

[0011] S600: Based on the generated resource allocation scheme, send scheduling instructions to the corresponding resources and monitor the execution status of the instructions; track the movement trajectory of resource locations and the recovery status of task progress in real time; when a new progress deviation or resource status change is detected, re-trigger the scheduling process to form a closed-loop dynamic optimization mechanism.

[0012] Furthermore, S100 includes:

[0013] S101: Deploy heterogeneous IoT sensing devices within the construction site, including installing satellite positioning modules on mobile devices, adding operational status sensors to fixed equipment, and configuring wireless positioning beacons for human resources; each sensing device collects the spatial coordinates, working status, and performance indicators of the resources according to a preset sampling period, and transmits the collected raw data to the edge computing node via a wireless network;

[0014] S102: Standardize the raw resource data received by the edge computing node and construct a multi-dimensional feature vector V in a unified format; the structure of the feature vector V is defined as [resource unique identifier ID, resource type code, real-time location coordinates, current working status parameters, core performance parameters, data acquisition timestamp]; among which, the working status parameters are converted into numerical variables through a preset encoding system;

[0015] S103: The edge computing node preprocesses the standardized feature vector V, uses a threshold detection method to identify and remove abnormal data points, and uses an interpolation algorithm to complete the missing data sequence. The processed feature vector V is then uploaded to the cloud platform. The cloud platform stores the resource status in a partitioned database according to resource type, and visualizes the resource distribution status based on spatial positioning data, thus completing the construction of the resource status awareness network.

[0016] Furthermore, the S200 includes:

[0017] S201. Import project plan data from the project management system, parse the work task decomposition structure, and extract the attribute data of each task, including task number, duration, earliest start time, and latest finish time; establish a logical relationship matrix between tasks according to construction process requirements, and clarify the four types of dependency relationships between tasks: finish-start, start-start, finish-finish, and start-finish.

[0018] S202. Construct a project network diagram based on task dependencies, and use the critical path method for forward and backward computation: forward computation starts from the project starting point and calculates the earliest start time and earliest finish time of each task in topological order; backward computation starts from the project ending point and calculates the latest start time and latest finish time of each task; the critical path is determined by calculating the total float of the tasks, where the tasks with zero total float constitute the critical path of the project.

[0019] S203. Integrate resource requirement data with the schedule network to establish a schedule baseline system that includes a task attribute table, a dependency table, a time parameter table, and a resource allocation table. The task attribute table records the basic attributes of tasks, the dependency table stores the logical relationships between tasks in the form of an adjacency matrix, the time parameter table stores the time calculation results of each task, and the resource allocation table records the resource requirement plan for each task in different time periods. The schedule baseline system adopts a hierarchical storage structure, with the top layer being project-level nodes, the middle layer being sub-project-level nodes, and the bottom layer being task-level nodes. Each level of node is mapped hierarchically through parent-child association fields and stored in the project schedule database, generating a unique model version number.

[0020] Furthermore, the S300 also includes:

[0021] S301. Establish a dynamic comparison mechanism between actual progress data and the progress benchmark system. Obtain the actual start time and actual progress percentage of each task in real time through the project data interface. Align the actual progress data with the planned time parameters of the corresponding tasks in the progress benchmark system and calculate the time deviation ΔT and the progress completion rate deviation ΔP.

[0022] S302. A tiered threshold method is used to identify tasks with significant deviations. Basic thresholds [ΔT_base, ΔP_base] are set, where ΔT_base is the basic threshold for time deviation, determined by analyzing the normal time fluctuation range of non-critical path tasks in historical projects and taking its statistical upper limit at confidence level ρ; ΔP_base is the basic threshold for schedule deviation, set based on project management schedule tolerance standards; for critical path tasks, a path criticality coefficient α is defined, where α = min(λ, total task float / total project duration); where λ is a preset minimum... Small sensitivity coefficient, 0 < λ < 1; for critical path tasks, the basic threshold is scaled by α to obtain the critical path decision threshold [α × ΔT_base, α × ΔP_base]; establish a mapping relationship between task path type and decision threshold, non-critical path tasks use the basic decision threshold [ΔT_base, ΔP_base]; critical path tasks use the scaled decision threshold [α × ΔT_base, α × ΔP_base]; when the |ΔT| and |ΔP| of a task both exceed the decision threshold of its corresponding path type, it is marked as a preliminary candidate task;

[0023] S303. Construct a deviation impact assessment model. The specific implementation process includes:

[0024] Establish an evaluation system encompassing three dimensions: path criticality, deviation propagation risk, and resource dependency complexity.

[0025] Path criticality K: Used to quantify the criticality of a task in the schedule network. It is calculated as the reciprocal of the total float of the task, i.e., K = 1 / total float of the task. When the total float is 0, the total project duration is used as the denominator.

[0026] Deviation propagation risk R: Used to assess the potential impact of the task's schedule deviation on subsequent tasks. It is calculated by analyzing the topology of the task in the dependency network. The formula is: R = number of direct successor tasks + γ × number of successor tasks on the critical path, where γ is the critical path impact weighting factor, γ>1, and is determined based on the statistical regularity of deviation propagation in historical project data.

[0027] Resource dependency complexity C: This measures the degree to which a task depends on scarce resources. It is calculated by multiplying the resource scarcity index by the resource demand. The resource scarcity index Sq represents the current scarcity of resource type q, and it is calculated as Sq = 1 - (current available quantity of resource type q / current total demand of resource type q). Therefore, C = Σ(Sq × the demand of the task for resource q).

[0028] Establish the comprehensive influence weight calculation formula: W=β1×K+β2×R+β3×C, where the weight coefficients β1, β2, and β3 are determined by the analytic hierarchy process and satisfy β1+β2+β3=1; based on the calculated weight value W, all preliminary candidate tasks are sorted in descending order, and the top M tasks are selected to generate the final list of tasks to be scheduled.

[0029] Furthermore, the S400 includes:

[0030] Historical project data was collected to construct a training sample set. The sample features included three dimensions: dynamic progress features (time deviation, progress completion deviation rate, and deviation duration in days); task dependency features (total task float, number of subsequent critical tasks, and total number of subsequent dependent tasks); and resource supply features (scarcity of required resource types, time required for resource allocation, and quantity of functionally equivalent resources). A gradient boosting decision tree algorithm was employed, based on the dynamic progress features and task dependency features. The system uses a combination of feature groups and resource supply feature groups as input features. Based on the post-scheduling deviation correction effect, resource utilization efficiency, and the delay time of subsequent tasks on the critical path, it comprehensively evaluates the urgency of resource demand for tasks in historical scheduling scenarios through weighted evaluation, and sets a comprehensive score threshold to classify them into four levels: urgent, high, medium, and low, which serve as the model's label. The model is trained by dividing the samples through k-fold cross-validation. After training, the contribution of each feature is determined through feature importance analysis. A screening threshold is set to retain high-contribution features and delete low-importance features. The model is retrained based on the optimized feature combination, and finally, a classification model that can output the probability distribution of tasks at the four resource demand urgency levels of urgent, high, medium, and low is constructed.

[0031] By extracting relevant features of the tasks to be scheduled and inputting them into a trained classification model, the probability distribution of tasks at four levels (urgent, high, medium, and low) is obtained. The level corresponding to the highest probability is selected as the final urgency level of the task, and differentiated resource allocation rules are formulated based on the urgency level to construct a complete demand description framework: First, a resource demand list is generated based on task requirements and workload, including the type, technical specifications, and quantity requirements of the required resources; then, time constraint parameters are determined based on the task time parameters in the schedule baseline system, including the expected start time, the latest start time, and the expected duration; then, the urgency level is directly used as a quantified scheduling priority indicator; finally, based on the resource status awareness network, functionally equivalent backup resource types are found for critical resources to form resource alternatives; finally, all the above elements are integrated into a unified structured data format to generate a complete scheduling demand description.

[0032] Furthermore, the S500 includes:

[0033] S501: The cloud platform receives a list of tasks to be scheduled and their corresponding scheduling requirements from S400. For each task to be scheduled, it retrieves candidate resources that meet the conditions within the current time and task time window from the resource status database to form an initial candidate resource set. The selection conditions are: the resource type matches the primary or backup type required by the task; the resource is currently idle or can be made available within the required time window based on its task progress; and the expected idle period of the resource can fully cover the time window required for task execution.

[0034] S502. Using resource allocation scheme A as the decision variable, where Aij represents whether task i is allocated to resource j, Aij=1 indicates allocation, Aij=0 indicates no allocation, and each task can only be allocated to resources in its initial candidate resource set; define three core optimization objectives to construct a multi-objective optimization function:

[0035] The first optimization objective F1(A) is the total scheduling distance, which is the sum of the path distances for all allocated resources to move from their current location to their corresponding task location. The path distance is calculated using the spatial Euclidean distance between the real-time location coordinates of the resources and the coordinates of the task location.

[0036] The second optimization objective F2(A) is the total workload imbalance, which is the sum of squared deviations between the actual working time of each resource and the average working time of all resources. The actual working time is the sum of the cumulative time of the tasks already executed by the resource and the expected execution time of the tasks allocated this time. The average working time is the arithmetic mean of the actual working times of all resources involved in the allocation.

[0037] The third optimization objective, F3(A), is a negative total type fit. The type fit is set according to the degree of functional matching between resources and tasks: when the resource type is completely consistent with the primary type required by the task, the fit value is 1.0; when the resource type is a backup type required by the task and has similar functions, the fit value ranges from 0.6 to 0.9. The specific value is determined by a preset functional similarity evaluation matrix. This matrix compares the core functional parameters of the resource with the functional parameters required by the task, calculates the parameter matching rate, and sets the value rules. The total type fit is the sum of the type fit of all allocated resources.

[0038] The final multi-objective optimization function is set as: Minimize F(A)=[F1(A),F2(A),-F3(A)]; where Minimize F(A) means that by optimizing resource allocation scheme A, the total scheduling distance F1(A), the total workload imbalance F2(A), and the negative total type fit -F3(A) are minimized, thereby achieving the comprehensive optimization of resource scheduling in three dimensions: spatial efficiency, load balance, and functional matching.

[0039] S503. Using integer encoding, each chromosome corresponds to a complete resource allocation scheme. The chromosome length is equal to the number of tasks to be scheduled, m. The value of the i-th gene position is the resource number j, where j belongs to the initial candidate resource set of task i, indicating that task i is assigned to resource j. An initial population of size N is randomly generated. For each task i, a resource allocation is randomly selected from its initial candidate resource set. The fitness function is used to evaluate the merits of each allocation scheme in the population. The fitness function is obtained by weighted summation of three optimization objectives, and the formula is: Fitness = (w1 × Distance optimization value) + (w2 × Load balancing optimization value) + (w3 × Fit optimization value); where w1, w2, and w3 are weight coefficients, satisfying w1 + w2 +w3=1, and dynamically adjusted according to the distribution of task urgency levels; Distance optimization value = baseline distance / total scheduling distance (F1(A)); Load balancing optimization value = 1 / (1 + total workload imbalance (F2(A))); Fit optimization value = total type fit (F3(A)) / maximum possible fit; Baseline distance is the sum of the maximum possible distances between all tasks and candidate resources; Maximum possible fit is the sum of fit when all tasks are allocated primary resources;

[0040] The optimization process involves several steps: First, a small number of individuals are randomly selected from the current population. The individuals with the highest fitness are chosen to advance to the next generation. Gene exchange is performed on these selected individuals, with random exchange points chosen to swap gene segments between two parent individuals, generating new offspring. Individuals are then randomly adjusted, with resources allocated to certain tasks replaced by other available resource numbers for the same task according to a preset probability. The individuals with the highest fitness in each generation are retained to advance directly to the next generation, ensuring that the best solution is not lost during optimization. Iterative optimization stops when the number of generations reaches a preset maximum or when the improvement in the best fitness across multiple generations is less than a set threshold. From the final set of optimized solutions, allocation priorities are determined based on the urgency level of each task. Priority is given to resource allocation schemes with the shortest spatial distance and highest type fit for high-urgency tasks. The final output is the resource allocation scheme that maximizes overall scheduling efficiency.

[0041] An AI-based engineering resource scheduling and optimization system includes a resource status perception module, a schedule baseline modeling module, a deviation task identification module, an urgency and demand generation module, a resource allocation optimization module, and a scheduling closed-loop monitoring module.

[0042] The resource status perception module acquires real-time quantity data and location coordinates of manpower, equipment and pipeline resources for each project through IoT sensors deployed at the construction site, and establishes a multi-dimensional feature vector containing resource type, working status and performance parameters to form a resource status perception network.

[0043] The schedule baseline modeling module constructs a schedule baseline system that includes task dependencies and time constraints based on project plan data;

[0044] The deviation task identification module identifies scheduled tasks whose progress deviations exceed a preset threshold by comparing the actual progress data of each project with the progress benchmark system.

[0045] The urgency and demand generation module collects historical project data to build a task urgency classification model, extracts the progress dynamics, task dependencies and resource supply characteristics of the tasks to be scheduled, inputs them into the trained model, generates the resource demand urgency level of each task to be scheduled, and constructs a scheduling demand description that includes the required resource type, quantity specifications and time window.

[0046] The resource allocation optimization module matches the scheduling requirement description with the resource status awareness network. Based on the spatial distance between the current location of the resource and the location of the task to be scheduled, the current workload status of the resource, and the type compatibility between the resource and the task, a multi-objective optimization function is constructed. By solving the multi-objective optimization function, a resource allocation scheme that maximizes the overall scheduling efficiency is generated, in which the resource with the closest spatial distance and the highest type compatibility is matched first for high urgency tasks.

[0047] The scheduling closed-loop monitoring module sends scheduling instructions to the corresponding resources and monitors the execution status of the instructions according to the generated resource allocation scheme; it tracks the movement trajectory of resource locations and the recovery status of task progress in real time; when a new progress deviation or resource status change is detected, the scheduling process is re-triggered, forming a closed-loop dynamic optimization mechanism.

[0048] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0049] This invention constructs a resource status awareness network by deploying IoT sensors to collect multi-dimensional resource data in real time, solving the problem of information lag in traditional scheduling and providing accurate basis for scheduling decisions. Simultaneously, it builds a progress benchmark system based on project plans and combines it with a deviation detection mechanism to identify tasks to be scheduled, preventing high-impact tasks from being overlooked and ensuring the accuracy of task priority determination. Machine learning technology is used to generate task urgency levels, supporting differentiated resource allocation and ensuring that high-urgency tasks receive priority response. A dynamic weighted multi-objective optimization function combined with a genetic algorithm is used to align optimization objectives with task priorities, improving the rationality of resource allocation. Finally, a closed-loop monitoring and dynamic adjustment mechanism responds in real time to changes in resource status and progress deviations, improving resource utilization efficiency and reducing the risk of critical path delays. Attached Figure Description

[0050] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0051] Figure 1 This is a flowchart of an artificial intelligence-based engineering resource scheduling optimization method. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Please see Figure 1 This invention provides a technical solution: an artificial intelligence-based engineering resource scheduling optimization method, comprising the following steps:

[0054] S100: By using IoT sensors deployed at construction sites, real-time quantity data and location coordinates of manpower, equipment and pipeline resources for each project are obtained, and a multi-dimensional feature vector containing resource type, working status and performance parameters is established to form a resource status perception network.

[0055] S200. Construct a schedule baseline system that includes task dependencies and time constraints based on project plan data;

[0056] S300: By comparing the actual progress data of each project with the progress benchmark system, identify the tasks to be scheduled that have a progress deviation exceeding a preset threshold.

[0057] S400: Collect historical project data to build a task urgency classification model, extract the progress dynamics, task dependencies and resource supply characteristics of the tasks to be scheduled, input the trained model, generate the resource demand urgency level of each task to be scheduled, and construct a scheduling demand description that includes the required resource type, quantity specifications and time window.

[0058] S500: Match the scheduling requirement description with the resource status awareness network. Based on the spatial distance between the current location of the resource and the location of the task to be scheduled, the current workload status of the resource, and the type compatibility between the resource and the task, construct a multi-objective optimization function. By solving the multi-objective optimization function, generate a resource allocation scheme that maximizes the overall scheduling efficiency, in which priority is given to matching the resource with the closest spatial distance and the highest type compatibility for high urgency tasks.

[0059] S600: Based on the generated resource allocation scheme, send scheduling instructions to the corresponding resources and monitor the execution status of the instructions; track the movement trajectory of resource locations and the recovery status of task progress in real time; when a new progress deviation or resource status change is detected, re-trigger the scheduling process to form a closed-loop dynamic optimization mechanism.

[0060] S100 includes:

[0061] S101: Deploy heterogeneous IoT sensing devices within the construction site, including installing satellite positioning modules on mobile devices, adding operational status sensors to fixed equipment, and configuring wireless positioning beacons for human resources; each sensing device collects the spatial coordinates, working status, and performance indicators of the resources according to a preset sampling period, and transmits the collected raw data to the edge computing node via a wireless network;

[0062] S102: Standardize the raw resource data received by the edge computing node and construct a multi-dimensional feature vector V in a unified format; the structure of the feature vector V is defined as [resource unique identifier ID, resource type code, real-time location coordinates, current working status parameters, core performance parameters, data acquisition timestamp]; among which, the working status parameters are converted into numerical variables through a preset encoding system;

[0063] S103: The edge computing node preprocesses the standardized feature vector V, uses a threshold detection method to identify and remove abnormal data points, and uses an interpolation algorithm to complete the missing data sequence. The processed feature vector V is then uploaded to the cloud platform. The cloud platform stores the resource status in a partitioned database according to resource type, and visualizes the resource distribution status based on spatial positioning data, thus completing the construction of the resource status awareness network.

[0064] S200 includes:

[0065] S201. Import project plan data from the project management system, parse the work task decomposition structure, and extract the attribute data of each task, including task number, duration, earliest start time, and latest finish time; establish a logical relationship matrix between tasks according to construction process requirements, and clarify the four types of dependency relationships between tasks: finish-start, start-start, finish-finish, and start-finish.

[0066] S202. Construct a project network diagram based on task dependencies, and use the critical path method for forward and backward computation: forward computation starts from the project starting point and calculates the earliest start time and earliest finish time of each task in topological order; backward computation starts from the project ending point and calculates the latest start time and latest finish time of each task; the critical path is determined by calculating the total float of the tasks, where the tasks with zero total float constitute the critical path of the project.

[0067] S203. Integrate resource requirement data with the schedule network to establish a schedule baseline system that includes a task attribute table, a dependency table, a time parameter table, and a resource allocation table. The task attribute table records the basic attributes of tasks, the dependency table stores the logical relationships between tasks in the form of an adjacency matrix, the time parameter table stores the time calculation results of each task, and the resource allocation table records the resource requirement plan for each task in different time periods. The schedule baseline system adopts a hierarchical storage structure, with the top layer being project-level nodes, the middle layer being sub-project-level nodes, and the bottom layer being task-level nodes. Each level of node is mapped hierarchically through parent-child association fields and stored in the project schedule database, generating a unique model version number.

[0068] The S300 includes:

[0069] S301. Establish a dynamic comparison mechanism between actual progress data and the progress benchmark system. Obtain the actual start time and actual progress percentage of each task in real time through the project data interface. Align the actual progress data with the planned time parameters of the corresponding tasks in the progress benchmark system and calculate the time deviation ΔT and the progress completion rate deviation ΔP.

[0070] S302. A tiered threshold method is used to identify tasks with significant deviations. Basic thresholds [ΔT_base, ΔP_base] are set, where ΔT_base is the basic threshold for time deviation, determined by analyzing the normal time fluctuation range (the difference between actual and planned completion times under normal execution conditions) of non-critical path tasks in historical projects, and taking its upper statistical limit at confidence level ρ; ΔP_base is the basic threshold for schedule deviation, set based on project management schedule tolerance standards, with a value range of 5%-10% of the planned schedule; for critical path tasks, a path criticality coefficient α is defined, where α = min(λ, (Total task float / Total project duration); where λ is a preset minimum sensitivity coefficient, 0 < λ < 1; for critical path tasks, the basic threshold is scaled by α to obtain the critical path judgment threshold [α × ΔT_base, α × ΔP_base]; establish a mapping relationship between task path type and judgment threshold, non-critical path tasks use the basic judgment threshold [ΔT_base, ΔP_base]; critical path tasks use the scaled judgment threshold [α × ΔT_base, α × ΔP_base]; when the |ΔT| and |ΔP| of a task both exceed the judgment threshold of its corresponding path type, it is marked as a preliminary candidate task;

[0071] S303. Construct a deviation impact assessment model. The specific implementation process includes:

[0072] Establish an evaluation system encompassing three dimensions: path criticality, deviation propagation risk, and resource dependency complexity.

[0073] Path criticality K: Used to quantify the criticality of a task in the schedule network. It is calculated as the reciprocal of the total float of the task, i.e., K = 1 / total float of the task. When the total float is 0, the total project duration is used as the denominator.

[0074] Deviation propagation risk R: Used to assess the potential impact of the task schedule deviation on subsequent tasks. It is calculated by analyzing the topology of the task in the dependency network. The formula is R = number of direct successor tasks + γ × number of successor tasks on the critical path, where γ is the critical path impact weighting factor. γ>1 is determined based on the statistical regularity of deviation propagation in historical project data and is used to amplify the propagation impact of successor tasks on the critical path.

[0075] Resource dependency complexity C: This measures the degree to which a task depends on scarce resources. It is calculated by multiplying the resource scarcity index by the resource demand. The resource scarcity index Sq represents the current scarcity of resource type q, and it is calculated as Sq = 1 - (current available quantity of resource type q / current total demand of resource type q). Therefore, C = Σ(Sq × the demand of the task for resource q).

[0076] Establish the comprehensive influence weight calculation formula: W=β1×K+β2×R+β3×C, where the weight coefficients β1, β2, and β3 are determined by the analytic hierarchy process and satisfy β1+β2+β3=1; based on the calculated weight value W, all preliminary candidate tasks are sorted in descending order, and the top M tasks are selected to generate the final list of tasks to be scheduled.

[0077] The S400 includes:

[0078] Historical project data was collected to construct a training sample set. The sample features included three dimensions: dynamic progress features (time deviation, progress completion deviation rate, and deviation duration in days); task dependency features (total task float, number of subsequent critical tasks, and total number of subsequent dependent tasks); and resource supply features (scarcity of required resource types, time required for resource allocation, and quantity of functionally equivalent resources). A gradient boosting decision tree algorithm was employed, using the dynamic progress features, task dependency features, and resource supply features as input features. The algorithm was based on post-scheduling deviations. The urgency of resource requirements for tasks in historical scheduling scenarios is assessed by weighted comprehensive evaluation of the correction effect, resource utilization efficiency, and delay time of subsequent tasks on the critical path. A comprehensive score threshold range (≥0.85 for urgent, 0.7-0.85 for high, 0.5-0.7 for medium, and <0.5 for low) is set to classify tasks into four levels: urgent, high, medium, and low, which serve as the model's label. The model is trained by dividing the samples through k-fold cross-validation. After training, the contribution of each feature is determined by feature importance analysis. A screening threshold is set to retain high-contribution features and delete low-importance features. The model is retrained based on the optimized feature combination, and finally, a classification model that can output the probability distribution of tasks at the four resource requirement urgency levels of urgent, high, medium, and low is constructed.

[0079] By extracting relevant features of the tasks to be scheduled and inputting them into a trained classification model, the probability distribution of tasks at four levels (urgent, high, medium, and low) is obtained. The level corresponding to the highest probability is selected as the final urgency level of the task, and differentiated resource allocation rules are formulated based on the urgency level to construct a complete demand description framework: First, a resource demand list is generated based on task requirements and workload, including the type, technical specifications, and quantity requirements of the required resources; then, time constraint parameters are determined based on the task time parameters in the schedule baseline system, including the expected start time, the latest start time, and the expected duration; then, the urgency level is directly used as a quantified scheduling priority indicator; finally, based on the resource status awareness network, functionally equivalent backup resource types are found for critical resources to form resource alternatives; finally, all the above elements are integrated into a unified structured data format to generate a complete scheduling demand description.

[0080] The S500 includes:

[0081] S501: The cloud platform receives a list of tasks to be scheduled and their corresponding scheduling requirements from S400. For each task to be scheduled, it retrieves candidate resources that meet the conditions within the current time and task time window from the resource status database to form an initial candidate resource set. The selection conditions are: the resource type matches the primary or backup type required by the task; the resource is currently idle or can be made available within the required time window based on its task progress; and the expected idle period of the resource can fully cover the time window required for task execution.

[0082] S502. Using resource allocation scheme A as the decision variable, where Aij represents whether task i is allocated to resource j, Aij=1 indicates allocation, Aij=0 indicates no allocation, and each task can only be allocated to resources in its initial candidate resource set; define three core optimization objectives to construct a multi-objective optimization function:

[0083] The first optimization objective F1(A) is the total scheduling distance, which is the sum of the path distances for all allocated resources to move from their current location to their corresponding task location. The path distance is calculated using the spatial Euclidean distance between the real-time location coordinates of the resources and the coordinates of the task location.

[0084] The second optimization objective F2(A) is the total workload imbalance, which is the sum of squared deviations between the actual working time of each resource and the average working time of all resources. The actual working time is the sum of the cumulative time of the tasks already executed by the resource and the expected execution time of the tasks allocated this time. The average working time is the arithmetic mean of the actual working times of all resources involved in the allocation.

[0085] The third optimization objective, F3(A), is a negative total type fit. The type fit is set based on the functional matching degree between resources and tasks: when the resource type is completely consistent with the primary type required by the task, the fit is 1.0; when the resource type is a backup type required by the task and has similar functions, the fit range is 0.6-0.9. The specific value is determined by a preset functional similarity evaluation matrix. This matrix compares the core functional parameters of the resource (such as the equipment's operating accuracy, power, and applicable working conditions; the personnel's skill level and certification type) with the functional parameters required by the task, calculating the parameter matching rate = (number of matched core parameters / total number of core parameters) × 100%. The following rules are set: when the parameter matching rate is ≥90%, the fit is 0.9; when the matching rate is 70%-89%, the fit is 0.7; when the matching rate is 50%-69%, the fit is 0.6; resources with a parameter matching rate <50% are not included in the backup type range; the total type fit is the sum of the type fits of all allocated resources.

[0086] The final multi-objective optimization function is set as: Minimize F(A)=[F1(A),F2(A),-F3(A)]; where Minimize F(A) means that by optimizing resource allocation scheme A, the total scheduling distance F1(A), the total workload imbalance F2(A), and the negative total type fit -F3(A) are minimized, thereby achieving the comprehensive optimization of resource scheduling in three dimensions: spatial efficiency, load balance, and functional matching.

[0087] S503. Using integer encoding, each chromosome corresponds to a complete resource allocation scheme. The chromosome length is equal to the number of tasks to be scheduled, m. The value of the i-th gene position is the resource number j, which belongs to the initial candidate resource set of task i, indicating that task i is allocated to resource j. Randomly generate an initial population of size N. For each task i, randomly select a resource allocation from its initial candidate resource set.

[0088] The fitness function is used to evaluate the merits of each assignment scheme in the population. The fitness function is obtained by weighted summation of three optimization objectives, and the formula is: Fitness = (w1 × Distance Optimization Value) + (w2 × Load Balancing Optimization Value) + (w3 × Fit Optimization Value); where w1, w2, and w3 are weight coefficients, satisfying w1 + w2 +w3=1, and dynamically adjusted according to the distribution of task urgency levels: when urgent and high-urgency tasks account for ≥60%, increase w3 (type fit weight) to 0.4-0.5, w1 (distance weight) to 0.3-0.4, and decrease w2 (load balancing weight) to 0.1-0.2, prioritizing functional matching and rapid response; when urgency levels are mainly medium and low, increase w2 to 0.4-0.5, with w1 and w3 each accounting for 0.25-0.3; distance optimization value = baseline distance / total scheduling distance (F1(A)); load balancing optimization value = 1 / (1 + total workload imbalance (F2(A))); fit optimization value = total type fit (F3(A)) / maximum possible fit; the baseline distance is the sum of the maximum possible distances between all tasks and candidate resources; the maximum possible fit is the sum of fit when all tasks are allocated primary resources;

[0089] The optimization process involves several steps: First, a small number of individuals are randomly selected from the current population. The individuals with the highest fitness are chosen to advance to the next generation. Gene exchange is performed on these selected individuals, with random exchange points chosen to swap gene segments between two parent individuals, generating new offspring. Individuals are then randomly adjusted, with resources allocated to certain tasks replaced by other available resource numbers for the same task according to a preset probability. The individuals with the highest fitness in each generation are retained to advance directly to the next generation, ensuring that the best solution is not lost during optimization. Iterative optimization stops when the number of generations reaches a preset maximum or when the improvement in the best fitness across multiple generations is less than a set threshold. From the final set of optimized solutions, allocation priorities are determined based on the urgency level of each task. Priority is given to resource allocation schemes with the shortest spatial distance and highest type fit for high-urgency tasks. The final output is the resource allocation scheme that maximizes overall scheduling efficiency.

[0090] This invention implements the following: Heterogeneous IoT sensing devices are deployed in the construction area: satellite positioning modules are installed on mobile devices such as concrete pump trucks and construction elevators to collect coordinates and operating speed; status sensors are added to fixed equipment such as rebar cutters to monitor current and failure rate; wireless positioning beacons are provided to construction workers to record their location and work group; all devices collect data every 30 seconds and transmit it to edge computing nodes via a wireless network; the nodes standardize the raw data into feature vectors containing "resource ID, type code, coordinates, working status, core parameters, and timestamp," use threshold detection to remove abnormal data, and linear interpolation to complete missing sequences, finally uploading the data to the cloud platform; the cloud platform stores the data according to resource type, generates a distribution heatmap, and completes the construction of the resource status perception network;

[0091] Import planning data from the project management system, parse the work breakdown structure of 128 tasks, and extract attributes such as task number and duration; establish task dependencies using an adjacency matrix (e.g., electromechanical pre-embedding only after the main structure is poured), and construct a project network diagram; determine the critical path "main structure pouring - electromechanical installation - exterior wall construction" with zero total float by forward and backward calculations using the critical path method, integrate resource requirement data to establish a hierarchical schedule baseline system, verify and store it in the database, and generate version number V1.0;

[0092] Real-time progress data revealed that the 15th floor main structure pouring task actually started 12 hours late and had a completion rate of 30% (planned 50%). A basic threshold was set (8 hours of time, 8% of progress), and the critical path task coefficient was set to 0.6. After scaling, the threshold was exceeded, and the task was marked as a preliminary candidate task. After calculation by the deviation impact assessment model, the task ranked first in comprehensive impact weight and was determined as the final task to be scheduled.

[0093] A sample set of 1200 scheduling data points from the past 3 years was collected. An urgency model was trained using a gradient boosting decision tree algorithm and k-fold cross-validation. Core features were retained after feature importance analysis. Inputting the task features yielded an 82% probability of urgency level, thus classifying it as an emergency. Based on this, scheduling requirements were generated: 2 primary and 1 functionally equivalent standby concrete pump trucks were needed, to be in place within 2 hours, operate for 48 hours, and have a conveying height of no less than 100 meters. The cloud platform retrieved 3 candidate pump trucks that met the criteria. A multi-objective optimization function was constructed based on "total scheduling distance, load imbalance, and type suitability," with weights of 0.5, 0.2, and 0.3. An initial population of 50 was generated using integer encoding. After 15 iterations, 2 pump trucks were prioritized for the emergency task: 1 truck located 300 meters from the work site and waiting to start work, and 1 truck idle after 1.5 hours.

[0094] Real-time monitoring after sending dispatch instructions: the 300-meter pump truck is in place within 30 minutes, and another one arrives on time; after 24 hours, the task completion rate reaches 70%, and the progress deviation is recovered; during this period, if one pump truck fails, the system immediately selects a waiting pump truck 800 meters away to ensure the progress of the task and form a dynamic closed loop.

[0095] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. An artificial intelligence-based engineering resource scheduling optimization method, characterized in that: The method includes: S100: By using IoT sensors deployed at construction sites, real-time quantity data and location coordinates of manpower, equipment and pipeline resources for each project are obtained, and a multi-dimensional feature vector containing resource type, working status and performance parameters is established to form a resource status perception network. S200. Construct a schedule baseline system that includes task dependencies and time constraints based on project plan data; S300: By comparing the actual progress data of each project with the progress benchmark system, identify the tasks to be scheduled that have a progress deviation exceeding a preset threshold. S400: Collect historical project data to build a task urgency classification model, extract the progress dynamics, task dependencies and resource supply characteristics of the tasks to be scheduled, input the trained model, generate the resource demand urgency level of each task to be scheduled, and construct a scheduling demand description that includes the required resource type, quantity specifications and time window. S500: Match the scheduling requirement description with the resource status awareness network. Based on the spatial distance between the current location of the resource and the location of the task to be scheduled, the current workload status of the resource, and the type compatibility between the resource and the task, construct a multi-objective optimization function. By solving the multi-objective optimization function, generate a resource allocation scheme that maximizes the overall scheduling efficiency, in which priority is given to matching the resource with the closest spatial distance and the highest type compatibility for high urgency tasks. S600: Based on the generated resource allocation scheme, send scheduling instructions to the corresponding resources and monitor the execution status of the instructions; track the movement trajectory of resource locations and the recovery status of task progress in real time; when a new progress deviation or resource status change is detected, re-trigger the scheduling process to form a closed-loop dynamic optimization mechanism.

2. The artificial intelligence-based engineering resource scheduling optimization method according to claim 1, characterized in that: S100 includes: S101: Deploy heterogeneous IoT sensing devices within the construction site, including installing satellite positioning modules on mobile devices, adding operational status sensors to fixed equipment, and configuring wireless positioning beacons for human resources; each sensing device collects the spatial coordinates, working status, and performance indicators of the resources according to a preset sampling period, and transmits the collected raw data to the edge computing node via a wireless network; S102: Standardize the raw resource data received by the edge computing node and construct a multi-dimensional feature vector V in a unified format; the structure of the feature vector V is defined as [resource unique identifier ID, resource type code, real-time location coordinates, current working status parameters, core performance parameters, data acquisition timestamp]; among which, the working status parameters are converted into numerical variables through a preset encoding system; S103: The edge computing node preprocesses the standardized feature vector V, uses a threshold detection method to identify and remove abnormal data points, and uses an interpolation algorithm to complete the missing data sequence. The processed feature vector V is then uploaded to the cloud platform. The cloud platform stores the resource status in a partitioned database according to resource type, and visualizes the resource distribution status based on spatial positioning data, thus completing the construction of the resource status awareness network.

3. The artificial intelligence-based engineering resource scheduling optimization method according to claim 2, characterized in that: S200 includes: S201. Import project plan data from the project management system, parse the work task decomposition structure, and extract the attribute data of each task, including task number, duration, earliest start time, and latest finish time; establish a logical relationship matrix between tasks according to construction process requirements, and clarify the four types of dependency relationships between tasks: finish-start, start-start, finish-finish, and start-finish. S202. Construct a project network diagram based on task dependencies, and use the critical path method for forward and backward computation: forward computation starts from the project starting point and calculates the earliest start time and earliest finish time of each task in topological order; backward computation starts from the project ending point and calculates the latest start time and latest finish time of each task; the critical path is determined by calculating the total float of the tasks, where the tasks with zero total float constitute the critical path of the project. S203. Integrate resource requirement data with the schedule network to establish a schedule baseline system that includes a task attribute table, a dependency table, a time parameter table, and a resource allocation table. The task attribute table records the basic attributes of tasks, the dependency table stores the logical relationships between tasks in the form of an adjacency matrix, the time parameter table stores the time calculation results of each task, and the resource allocation table records the resource requirement plan for each task in different time periods. The schedule baseline system adopts a hierarchical storage structure, with the top layer being project-level nodes, the middle layer being sub-project-level nodes, and the bottom layer being task-level nodes. Each level of node is mapped hierarchically through parent-child association fields and stored in the project schedule database, generating a unique model version number.

4. The engineering resource scheduling optimization method based on artificial intelligence according to claim 3, characterized in that: The S300 also includes: S301. Establish a dynamic comparison mechanism between actual progress data and the progress benchmark system. Obtain the actual start time and actual progress percentage of each task in real time through the project data interface. Align the actual progress data with the planned time parameters of the corresponding tasks in the progress benchmark system and calculate the time deviation ΔT and the progress completion rate deviation ΔP. S302. A tiered threshold method is used to identify deviation tasks. Basic thresholds [ΔT_base, ΔP_base] are set, where ΔT_base is the basic threshold for time deviation, determined by analyzing the normal time fluctuation range of non-critical path tasks in historical projects and taking its statistical upper limit at confidence level ρ; ΔP_base is the basic threshold for schedule deviation, set based on project management schedule tolerance standards; for critical path tasks, a path criticality coefficient α is defined, where α = min(λ, total task float / total project duration); where λ is a preset minimum... Sensitivity coefficient, 0 < λ < 1; For critical path tasks, the basic threshold is scaled by α to obtain the critical path decision threshold [α × ΔT_base, α × ΔP_base]; A mapping relationship between task path type and decision threshold is established. Non-critical path tasks use the basic decision threshold [ΔT_base, ΔP_base]; critical path tasks use the scaled decision threshold [α × ΔT_base, α × ΔP_base]; When the |ΔT| and |ΔP| of a task both exceed the decision threshold of its corresponding path type, it is marked as a preliminary candidate task.

5. The engineering resource scheduling optimization method based on artificial intelligence according to claim 4, characterized in that: The S300 also includes: S303. Construct a deviation impact assessment model. The specific implementation process includes: Establish an evaluation system encompassing three dimensions: path criticality, deviation propagation risk, and resource dependency complexity. Path criticality K: Used to quantify the criticality of a task in the schedule network. It is calculated as the reciprocal of the total float of the task, i.e., K = 1 / total float of the task. When the total float is 0, the total project duration is used as the denominator. Deviation propagation risk R: Used to assess the potential impact of the task's schedule deviation on subsequent tasks. It is calculated by analyzing the topology of the task in the dependency network. The formula is: R = number of direct successor tasks + γ × number of successor tasks on the critical path, where γ is the critical path impact weighting factor, γ>1, and is determined based on the statistical regularity of deviation propagation in historical project data. Resource dependency complexity C: This measures the degree to which a task depends on scarce resources. It is calculated by multiplying the resource scarcity index by the resource demand. The resource scarcity index Sq represents the current scarcity of resource type q, and it is calculated as Sq = 1 - (current available quantity of resource type q / current total demand of resource type q). Therefore, C = Σ(Sq × the demand of the task for resource q). Establish the comprehensive influence weight calculation formula: W=β1×K+β2×R+β3×C, where the weight coefficients β1, β2, and β3 are determined by the analytic hierarchy process and satisfy β1+β2+β3=1; based on the calculated weight value W, all preliminary candidate tasks are sorted in descending order, and the top M tasks are selected to generate the final list of tasks to be scheduled.

6. The engineering resource scheduling optimization method based on artificial intelligence according to claim 5, characterized in that: The S400 includes: Historical project data was collected to construct a training sample set. The sample features included three dimensions: dynamic progress features (time deviation, progress completion deviation rate, and deviation duration in days); task dependency features (total task float, number of subsequent critical tasks, and total number of subsequent dependent tasks); and resource supply features (scarcity of required resource types, time required for resource allocation, and quantity of functionally equivalent resources). A gradient boosting decision tree algorithm was employed, based on the dynamic progress features and task dependency features. The system uses a combination of feature groups and resource supply feature groups as input features. Based on the post-scheduling deviation correction effect, resource utilization efficiency, and the delay time of subsequent tasks on the critical path, it comprehensively evaluates the urgency of resource demand for tasks in historical scheduling scenarios through weighted evaluation, and sets a comprehensive score threshold to classify them into four levels: urgent, high, medium, and low, which serve as the model's label. The model is trained by dividing the samples through k-fold cross-validation. After training, the contribution of each feature is determined through feature importance analysis. A screening threshold is set to retain high-contribution features and delete low-importance features. The model is retrained based on the optimized feature combination, and finally, a classification model that can output the probability distribution of tasks at the four resource demand urgency levels of urgent, high, medium, and low is constructed. By extracting relevant features of the tasks to be scheduled and inputting them into a trained classification model, the probability distribution of tasks at four levels (urgent, high, medium, and low) is obtained. The level corresponding to the highest probability is selected as the final urgency level of the task, and differentiated resource allocation rules are formulated based on the urgency level to construct a complete demand description framework: First, a resource demand list is generated based on task requirements and workload, including the type, technical specifications, and quantity requirements of the required resources; then, time constraint parameters are determined based on the task time parameters in the schedule baseline system, including the expected start time, the latest start time, and the expected duration; then, the urgency level is directly used as a quantified scheduling priority indicator; finally, based on the resource status awareness network, functionally equivalent backup resource types are found for critical resources to form resource alternatives; finally, all the above elements are integrated into a unified structured data format to generate a complete scheduling demand description.

7. The artificial intelligence-based engineering resource scheduling optimization method according to claim 6, characterized in that: The S500 includes: S501: The cloud platform receives a list of tasks to be scheduled and their corresponding scheduling requirements from S400. For each task to be scheduled, it retrieves candidate resources that meet the conditions within the current time and task time window from the resource status database to form an initial candidate resource set. The selection conditions are: the resource type matches the primary or backup type required by the task; the resource is currently idle or can be made available within the required time window based on its task progress; and the expected idle period of the resource can fully cover the time window required for task execution. S502. Using resource allocation scheme A as the decision variable, where Aij represents whether task i is allocated to resource j, Aij=1 indicates allocation, Aij=0 indicates no allocation, and each task can only be allocated to resources in its initial candidate resource set; define three core optimization objectives to construct a multi-objective optimization function: The first optimization objective F1(A) is the total scheduling distance, which is the sum of the path distances for all allocated resources to move from their current location to their corresponding task location. The path distance is calculated using the spatial Euclidean distance between the real-time location coordinates of the resources and the coordinates of the task location. The second optimization objective F2(A) is the total workload imbalance, which is the sum of squared deviations between the actual working time of each resource and the average working time of all resources. The actual working time is the sum of the cumulative time of the tasks already executed by the resource and the expected execution time of the tasks allocated this time. The average working time is the arithmetic mean of the actual working times of all resources involved in the allocation. The third optimization objective, F3(A), is a negative total type fit, which is the sum of the type fit of all allocated resources. The type fit is set according to the functional matching degree between the resource and the task: when the resource type is completely consistent with the primary type required by the task, the fit value is 1.0; when the resource type is a backup type required by the task and functionally equivalent, the fit value ranges from 0.6 to 0.

9. The specific value is determined by a preset functional similarity evaluation matrix. This matrix compares the core functional parameters of the resource with the functional parameters required by the task, calculates the parameter matching rate, and sets the value rule: Parameter matching rate = (Number of matched core parameters / Total number of core parameters) × 100%. The final multi-objective optimization function is set as: Minimize F(A)=[F1(A),F2(A),-F3(A)]; where MinimizeF(A) means that by optimizing resource allocation scheme A, the total scheduling distance F1(A), the total workload imbalance F2(A), and the negative total type fit -F3(A) are minimized, thereby achieving the comprehensive optimization of resource scheduling in three dimensions: spatial efficiency, load balance, and functional matching.

8. The engineering resource scheduling optimization method based on artificial intelligence according to claim 7, characterized in that: The S500 includes: S503. Using integer encoding, each chromosome corresponds to a complete resource allocation scheme. The chromosome length is equal to the number of tasks to be scheduled, m. The value of the i-th gene position is the resource number j, where j belongs to the initial candidate resource set of task i, indicating that task i is assigned to resource j. An initial population of size N is randomly generated. For each task i, a resource allocation is randomly selected from its initial candidate resource set. The fitness function is used to evaluate the merits of each allocation scheme in the population. The fitness function is obtained by weighted summation of three optimization objectives, and the formula is: Fitness = (w1 × distance optimization value) + (w2 × average load) The optimization value is calculated as follows: (w1, w2, and w3) + (w3 × adaptability optimization value); where w1, w2, and w3 are weighting coefficients, satisfying w1 + w2 + w3 = 1, and dynamically adjusted according to the distribution of task urgency levels; distance optimization value = baseline distance / total scheduling distance (F1(A)); load balancing optimization value = 1 / (1 + total workload imbalance (F2(A))); adaptability optimization value = total type adaptability (F3(A)) / maximum possible adaptability; the baseline distance is the sum of the maximum possible distances between all tasks and candidate resources; the maximum possible adaptability is the sum of adaptability when all tasks are allocated primary resources; The optimization process involves several steps: First, a small number of individuals are randomly selected from the current population, and the individual with the best fitness is chosen to advance to the next generation. Then, gene exchange is performed on the selected individuals. A random exchange point is chosen, and gene segments from two parent individuals are exchanged at the point where they are located, generating new offspring. Next, individuals are randomly adjusted, and resources allocated to certain tasks are replaced with other available resource numbers for the same task according to a preset probability. The individual with the highest fitness in each generation is retained to directly advance to the next generation, ensuring that the best solution is not lost during optimization. Iterative optimization stops when the number of generations reaches a preset maximum, or when the improvement in the best fitness across multiple generations is less than a set threshold. Finally, from the final set of optimized solutions, allocation priorities are determined based on the urgency level of each task. The resource allocation solution with the closest spatial distance and highest type fit is prioritized for high-urgency tasks, ultimately outputting the resource allocation solution that maximizes overall scheduling efficiency.

9. An engineering resource scheduling and optimization system based on artificial intelligence, characterized in that: The system includes a resource status perception module, a schedule baseline modeling module, a deviation task identification module, an urgency and requirement generation module, a resource allocation optimization module, and a scheduling closed-loop monitoring module. The resource status perception module acquires real-time quantity data and location coordinates of manpower, equipment and pipeline resources for each project through IoT sensors deployed at the construction site, and establishes a multi-dimensional feature vector containing resource type, working status and performance parameters to form a resource status perception network. The schedule baseline modeling module constructs a schedule baseline system that includes task dependencies and time constraints based on project plan data; The deviation task identification module identifies scheduled tasks whose progress deviation exceeds a preset threshold by comparing the actual progress data of each project with the progress benchmark system. The urgency and demand generation module collects historical project data to build a task urgency classification model, extracts the progress dynamics, task dependencies and resource supply characteristics of the tasks to be scheduled, inputs them into the trained model, generates the resource demand urgency level of each task to be scheduled, and constructs a scheduling demand description that includes the required resource type, quantity specifications and time window. The resource allocation optimization module matches the scheduling requirement description with the resource status awareness network, and constructs a multi-objective optimization function based on the spatial distance between the current location of the resource and the location of the task to be scheduled, the current workload status of the resource, and the type compatibility between the resource and the task. By solving the multi-objective optimization function, a resource allocation scheme that maximizes the overall scheduling efficiency is generated, in which priority is given to matching the resource with the closest spatial distance and the highest type fit for high urgency tasks; The scheduling closed-loop monitoring module sends scheduling instructions to the corresponding resources and monitors the execution status of the instructions according to the generated resource allocation scheme; it tracks the movement trajectory of resource positions and the recovery status of task progress in real time; when a new progress deviation or resource status change is detected, the scheduling process is re-triggered, forming a closed-loop dynamic optimization mechanism.

Images