A construction resource dynamic scheduling optimization method based on big data analysis

Abstract

This invention discloses a dynamic scheduling optimization method for construction resources based on big data analysis, belonging to the field of big data analysis. The method includes: collecting and preprocessing construction resource data to generate a standardized construction resource dataset and a construction event stream; correcting the construction event stream based on timestamps, missing data markers, jitter markers, and source markers to generate a reliable construction event stream; extracting resource events and calculating resource friction to generate a construction resource friction propagation graph and construction disturbance breakpoint data; inputting an improved SCINet model to generate a scheduling executable prediction vector; constructing individual construction scheduling parameters based on the scheduling executable prediction vector; generating scheduling feedback prediction results using the walrus optimization algorithm; filtering, updating, and repairing the individual scheduling parameters to generate a target construction resource scheduling scheme. This invention achieves dynamic scheduling optimization of construction resources, improving resource utilization and the stability of the scheduling scheme's on-site execution.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, and in particular to a method for dynamic scheduling and optimization of construction resources based on big data analytics. Background Technology

[0002] Construction site resource scheduling involves data on construction personnel, machinery and equipment, material inventory, transport vehicles, work areas, construction tasks, process progress, environmental monitoring, and safety management. These data sources are scattered, collected at varying frequencies, and exhibit significant differences in quality. When on-site scheduling relies on manual aggregation and experience-based judgment, problems such as delayed resource status updates, inaccurate construction task scheduling, and mismatches between material supply and machinery allocation are prone to occur.

[0003] Existing construction resource scheduling methods are mostly based on BIM models, schedules, resource demand tables, or resource-constrained project scheduling models, which can complete resource quantity statistics, task schedule matching, and partial resource reallocation. However, existing methods are insufficient in expressing the reliability of collected data, resource flow obstructions, work site blockages, process handover delays, transportation arrival deviations, and material transfer obstructions. They tend to mistake nominally idle resources for available resources, and the resulting scheduling results still result in personnel waiting, machinery running idle, material supply shortages, work site conflicts, and delays in key processes when executed on the construction site.

[0004] Therefore, how to provide a dynamic scheduling and optimization method for construction resources based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a dynamic scheduling optimization method for construction resources based on big data analysis. This invention utilizes a trusted event stream, an improved SCINet model, and a walrus optimization algorithm to achieve construction resource status prediction, scheduling scheme generation, and backfilling and repair. It has the advantages of high data reliability, fast scheduling response, and strong on-site executability.

[0006] A method for dynamic scheduling and optimization of construction resources based on big data analysis according to an embodiment of the present invention includes: Collect and preprocess construction resource data to generate a standardized construction resource dataset. Perform event extraction on the standardized construction resource dataset to generate a construction event stream. The credibility score, low-credibility event removal, and adjacent event correction are performed on the construction event flow to generate a credible construction event flow. Extract resource occupation events, release events, movement events, material arrival events, transportation delay events, work surface closure events, and process handover events from the trusted construction event flow; calculate resource friction and generate a construction resource friction propagation diagram and construction disturbance breakpoint data; An improved SCINet model containing a task resource operation surface displacement encoder, a process resource interactive convolutional block, and a process breakpoint memory unit is constructed. The model is then fed with a reliable construction event flow, a construction resource friction propagation graph, and construction disturbance breakpoint data to generate a scheduling executable prediction vector. A set of individual construction scheduling parameters is constructed based on the scheduling executable prediction vector, the construction resource friction propagation graph, and the standardized construction resource dataset. The Walrus Optimization Algorithm is used to perform perturbation migration, cluster leader navigation update and task priority reverse pruning on the individual set of construction scheduling parameters to generate updated individual scheduling parameters and restore them to candidate construction event recharge sequences. The recharge improves the SCINet model to generate scheduling recharge prediction results. Based on the scheduling backfeed prediction results, individuals with updated scheduling parameters are selected. The failed tasks, failed resources, failed work surfaces, and failed time windows of the eliminated individuals are written into the process breakpoint memory unit. Hard constraint filtering and local scheduling repair are performed on the retained individuals to generate the target construction resource scheduling scheme.

[0007] Optionally, the construction resource data includes construction personnel data, machinery and equipment data, material inventory data, transport vehicle data, work area occupancy data, construction task data, environmental monitoring data, and safety management data.

[0008] Optionally, the preprocessing includes time alignment, data cleaning, and normalization.

[0009] Optionally, generating a trusted construction event stream includes: Read the acquisition end timestamp, missing marker, jitter marker, and source marker of each construction event in the construction event stream, and generate an event credibility evaluation sequence according to the event occurrence time; The time consistency value of adjacent construction events is calculated based on the timestamp of the acquisition terminal, and the number of consecutive missing items of the same resource number and the number of time jitters of the same source tag are counted based on the missing tag and the jitter tag. An event credibility score is generated based on the time consistency value, the number of consecutive missing events, the number of time jitters, and the source tag. Construction events with credibility scores below the credibility threshold are marked as low credibility events. Delete low-reliability events, fill in the status gaps based on resource status changes within adjacent time windows for the same resource number, and reorder them according to event occurrence time, resource number, and construction task number to generate a reliable construction event flow.

[0010] Optionally, generating the construction resource friction propagation map and construction disturbance breakpoint data includes: Read the event occurrence time, resource number, construction task number, work surface number, material batch number, transport vehicle number, event type and event associated location from the trusted construction event stream, and extract resource occupation events, resource release events, resource movement events, material arrival events, transportation delay events, work surface closure events and process handover events; Based on resource movement events, resource release events, material arrival events, transportation delay events, work area closure events, and process handover events, calculate equipment movement friction, shift change friction, material transfer friction, transportation channel friction, work area congestion friction, and process handover friction respectively. Set the construction task number, resource number, work surface number, material batch number, transport vehicle number, and process handover number as graph nodes, and set the resource occupation relationship, resource release relationship, resource movement relationship, material supply relationship, transportation arrival relationship, work surface closure relationship, and process handover relationship as graph edges, and write the corresponding resource friction into the graph edge attributes to generate a construction resource friction propagation graph; Mark the locations where resource flow is obstructed according to the time sequence of occurrence of resource friction under the same construction task number, and generate construction disturbance breakpoint data.

[0011] Optionally, the generation of the scheduled executable prediction vector includes: An improved SCINet model is constructed, which includes a task resource work surface displacement encoder, a process resource interactive convolutional block, and a process breakpoint memory unit. Read the event occurrence time, resource number, construction task number, work surface number, material batch number, transport vehicle number and event type from the trusted construction event stream, and generate a construction event input sequence according to the construction task number and event occurrence time; The task resource work surface displacement encoder reads the construction event input sequence and the construction resource friction propagation map, extracts the resource number corresponding to the current position, the construction task number corresponding to the target work surface number, the resource movement relationship corresponding to the movement duration, and the resource friction in the map edge attributes, and generates the task resource work surface displacement coding sequence. The process resource interaction convolutional block establishes interaction channels according to the construction task number, resource number, work surface number, material batch number, and transport vehicle number. It performs convolutional interaction on resource occupation relationship, resource release relationship, material supply relationship, transportation arrival relationship, work surface closure relationship, and process handover relationship to generate process resource interaction feature sequence. The process breakpoint memory unit reads the construction disturbance breakpoint data and the process resource interaction feature sequence, and writes the location of the resource flow obstruction, construction task number, resource number, work surface number and occurrence time into the breakpoint memory slot to generate process breakpoint memory features. Based on the process resource interaction feature sequence and process breakpoint memory features, the task start-up intensity, resource reachability intensity, material support intensity, work surface acceptance intensity and task release time are generated, and combined to generate a scheduling executable prediction vector.

[0012] Optionally, the set of individual construction scheduling parameters includes: Read the task start-up intensity, resource reachability intensity, material support intensity, work surface acceptance intensity and task release time from the scheduled executable prediction vector; read the construction task number, planned start time, planned completion time and resource requirement from the construction task data. Read the graph node and graph edge attributes in the construction resource friction propagation graph, and extract the allocable work teams, allocable machinery, supplyable material batches, arrangeable transport vehicles, and acceptable work surfaces; Discrete and continuous coding segments are established according to the construction task number. The task sequence, work group number, machinery number, material batch number, transport vehicle number and work surface number are written into the discrete coding segment. The task start time, task completion time, resource allocation time and material arrival time are written into the continuous coding segment to generate an individual set of construction scheduling parameters.

[0013] Optionally, the generation of scheduling recharge prediction results includes: Read the discrete and continuous coded segments from the individual set of construction scheduling parameters, and read the task start-up intensity, resource reachability intensity, material support intensity, work surface acceptance intensity, and task release time from the scheduling executable prediction vector; The walrus optimization algorithm forms a disturbance migration direction based on the task start-up intensity, resource reachability intensity, material support intensity, and work surface acceptance intensity. It migrates the team number, machine number, material batch number, transport vehicle number, and work surface number in the discrete coding segment to the corresponding node of the graph edge with a smaller resource friction value in the construction resource friction propagation graph. It also adjusts the task start time, resource allocation time, and material arrival time in the continuous coding segment to generate individual migration scheduling parameters. The migration scheduling parameter individuals are clustered according to the work surface number and construction task number. Based on the task start-up intensity and resource reach intensity under the same work surface number, a leading scheduling parameter individual is selected. The task sequence, resource allocation time and material arrival time in the leading scheduling parameter individual are then transferred to other migration scheduling parameter individuals in the same cluster to generate navigation scheduling parameter individuals. The navigation scheduling parameter individuals are read from back to front according to the task completion time in the continuous coding segment. The pruning position is determined in ascending order of task startability intensity, resource availability intensity, material support intensity, and work surface acceptance intensity. The task sequence and resource allocation content corresponding to the pruning position are deleted from the navigation scheduling parameter individuals. Based on the task release time, the construction task number, team number, machinery number, material batch number, transport vehicle number, work surface number, and task start time are re-inserted to generate an updated scheduling parameter individual. The updated scheduling parameters are restored to candidate task start events, candidate resource occupancy events, candidate resource release events, candidate material arrival events, candidate transportation arrival events, candidate work surface occupancy events, and candidate process handover events. A candidate construction event reinjection sequence is generated and input into the improved SCINet model to generate scheduling reinjection prediction results.

[0014] Optionally, the generation of the target construction resource scheduling scheme includes: Read the task start-up intensity, resource availability intensity, material support intensity, work surface acceptance intensity, and task release time from the scheduling and recharge prediction results, and read the construction task number, work group number, machinery number, material batch number, transport vehicle number, work surface number, task start time, and task completion time from the individual updated scheduling parameters. Based on the task's workability intensity, resource availability intensity, material support intensity, and work surface acceptance intensity, generate resource unavailability markers, material unavailable markers, and work surface unacceptability markers, and then filter and update the scheduling parameters of individuals, marking the individuals with updated scheduling parameters as eliminated individuals. Extract failed tasks, failed resources, failed work surfaces, and failed time windows from the eliminated individuals, and write the failed tasks, failed resources, failed work surfaces, failed time windows, unreachable resource markers, unavailable material markers, and unacceptable work surface markers into the process breakpoint memory unit; Based on safety management and environmental monitoring data, hard constraint filtering is performed on retained individuals. For construction task numbers that have blocked work surfaces, unreachable resources, and unavailable materials after filtering, local scheduling and repair are performed to generate individuals with repair scheduling parameters. The repaired scheduling parameters are restored to individual task scheduling results, personnel allocation results, machinery allocation results, material supply results, transportation arrangement results, and work area arrangement results, generating a target construction resource scheduling plan.

[0015] The beneficial effects of this invention are: This invention proposes a dynamic scheduling optimization method for construction resources based on big data analysis. By collecting data on construction personnel, machinery and equipment, material inventory, transport vehicles, work area occupancy, construction tasks, process progress, environmental monitoring, and safety management, a standardized construction resource dataset and construction event flow are generated. Based on the timestamps of the data collection endpoints, missing data markers, jitter markers, and source markers, a credibility score is performed, low-credibility events are removed, and adjacent events are corrected to generate a reliable construction event flow. Compared to manually compiling resource ledgers, adjusting single schedule plans, and directly using on-site data for scheduling, this invention reduces the interference of data missingness, collection jitter, source differences, and time asynchrony on the judgment of construction resource status, improving the accuracy of construction resource status identification and the reliability of scheduling input data.

[0016] This invention extracts resource occupancy, release, movement, material arrival, transportation delay, work surface closure, and process handover events from a reliable construction event stream. It calculates resource friction and generates a construction resource friction propagation map and construction disturbance breakpoint data. An improved SCINet model, incorporating task resource work surface displacement encoders, process resource interactive convolutional blocks, and process breakpoint memory units, generates a scheduling-executable prediction vector. The invention then employs a walrus optimization algorithm to perform disturbance migration, cluster leader navigation updates, task priority reverse pruning, and refeedback predictions. Compared to a simple concatenation of conventional resource-constrained scheduling, static BIM schedule optimization, and prediction models with optimization algorithms, this invention can express the implicit obstacles that resources are nominally idle but cannot be deployed due to movement, transportation, work surface closure, and process handover. This reduces personnel waiting time, machine idleness, material supply disruptions, and work surface conflicts, improving the on-site executability of dynamic construction resource scheduling schemes and the continuous construction capability of key processes. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They explain the invention together with the embodiments of the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a construction resource dynamic scheduling optimization method based on big data analysis proposed in this invention; Figure 2 This is a schematic diagram of the improved SCINet model structure for a construction resource dynamic scheduling optimization method based on big data analysis proposed in this invention. Figure 3 This is a schematic diagram of the walrus optimization algorithm and the recharge prediction closed-loop scheduling of a construction resource dynamic scheduling optimization method based on big data analysis proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figure 1 , Figure 2 and Figure 3 A method for dynamic scheduling and optimization of construction resources based on big data analysis includes: Collect and preprocess construction resource data to generate a standardized construction resource dataset. Perform event extraction on the standardized construction resource dataset to generate a construction event stream. The credibility score, low-credibility event removal, and adjacent event correction are performed on the construction event flow to generate a credible construction event flow. Extract resource occupation events, release events, movement events, material arrival events, transportation delay events, work surface closure events, and process handover events from the trusted construction event flow; calculate resource friction and generate a construction resource friction propagation diagram and construction disturbance breakpoint data; An improved SCINet model containing a task resource operation surface displacement encoder, a process resource interactive convolutional block, and a process breakpoint memory unit is constructed. The model is then fed with a reliable construction event flow, a construction resource friction propagation graph, and construction disturbance breakpoint data to generate a scheduling executable prediction vector. A set of individual construction scheduling parameters is constructed based on the scheduling executable prediction vector, the construction resource friction propagation graph, and the standardized construction resource dataset. The Walrus Optimization Algorithm is used to perform perturbation migration, cluster leader navigation update and task priority reverse pruning on the individual set of construction scheduling parameters to generate updated individual scheduling parameters and restore them to candidate construction event recharge sequences. The recharge improves the SCINet model to generate scheduling recharge prediction results. Based on the scheduling backfeed prediction results, individuals with updated scheduling parameters are selected. The failed tasks, failed resources, failed work surfaces, and failed time windows of the eliminated individuals are written into the process breakpoint memory unit. Hard constraint filtering and local scheduling repair are performed on the retained individuals to generate the target construction resource scheduling scheme.

[0020] In this embodiment, the construction resource data includes construction personnel data, machinery and equipment data, material inventory data, transport vehicle data, work area occupancy data, construction task data, environmental monitoring data, and safety management data.

[0021] In this embodiment, the preprocessing includes time alignment, data cleaning, and standardization.

[0022] In this embodiment, generating a reliable construction event flow includes: Read the acquisition timestamp, missing marker, jitter marker, and source marker for each construction event in the construction event stream, and generate an event credibility evaluation sequence according to the event occurrence time. Specifically, generating the event credibility evaluation sequence according to the event occurrence time is as follows: Read the event number, event occurrence time, resource number, construction task number, data acquisition timestamp, missing flag, jitter flag, and source flag for each construction event in the construction event stream. A missing flag of 1 indicates that the field is missing, and a missing flag of 0 indicates that the field is complete. A jitter flag of 1 indicates that the difference between the data acquisition timestamp and the event occurrence time exceeds 2 seconds, and a jitter flag of 0 indicates that the difference does not exceed 2 seconds. Sort the events by occurrence time from earliest to latest, and sort them by resource number and construction task number within the same event occurrence time. Generate an event credibility evaluation sequence with the event number as the row index and the data acquisition timestamp, missing flag, jitter flag, and source flag as column fields. The temporal consistency value of adjacent construction events is calculated based on the timestamps of the data acquisition end. The number of consecutive missing items for the same resource ID and the number of time jitters for the same source tag are counted based on missing and jitter tags. The time consistency value of adjacent construction events is calculated based on the timestamps from the data acquisition terminal, specifically as follows: Read the collection timestamps of two adjacent construction events under the same resource number in the event credibility evaluation sequence. Subtract the previous collection timestamp from the latter collection timestamp to obtain the actual time interval. Read the collection period corresponding to the same source tag in the standardized construction resource dataset. Subtract the actual time interval from the collection period and take the absolute value to obtain the time interval deviation. Divide the time interval deviation by the collection period to obtain the deviation ratio. Subtract the deviation ratio from 1 to obtain the time consistency value. A time consistency value less than 0 is recorded as 0, and a time consistency value greater than 1 is recorded as 1. The number of consecutive missing values ​​is as follows: Read the missing markers of the same resource number in the sequence of events whose credibility is to be evaluated, arranged by the time of the event occurrence. Starting from the first construction event, scan one by one. When a construction event with a missing marker of 1 is encountered, increment the consecutive missing count by 1. When a construction event with a missing marker of 0 is encountered, reset the consecutive missing count to 0. Write the consecutive missing count corresponding to the end of the scan of each construction event into the consecutive missing count field. The number of time jitters is as follows: Read the jitter markers arranged by the time of occurrence under the same source marker in the event credibility evaluation sequence, set a sliding count window with a length of 5 construction events, count the number of construction events with jitter marker 1 in each sliding count window, and write the statistical results into the time jitter count field corresponding to the last construction event in the sliding count window; An event credibility score is generated based on time consistency value, number of consecutive missing events, number of time jitters, and source marker. Construction events with credibility scores below the credibility threshold are marked as low-credibility events. An event credibility score is generated based on temporal consistency, consecutive missing count, temporal jitter count, and source marker, specifically: The time consistency value is used as the time consistency sub-score. The consecutive missing number is divided by 5 to obtain the consecutive missing ratio. The consecutive missing ratio is subtracted from 1 to obtain the data integrity sub-score. The data integrity sub-score is recorded as 0 when it is less than 0. The time jitter number is divided by 5 to obtain the jitter ratio. The jitter ratio is subtracted from 1 to obtain the acquisition stability sub-score. The acquisition stability sub-score is recorded as 0 when it is less than 0. The acquisition source type corresponding to the source tag is mapped to the source reliability sub-score. Smart sensor acquisition is recorded as 1, construction management platform automatic recording is recorded as 0.9, positioning device recording is recorded as 0.85, and manual entry is recorded as 0.7. The event credibility score is obtained by multiplying the time consistency sub-score by 0.35, the data integrity sub-score by 0.25, the acquisition stability sub-score by 0.25, and the source reliability sub-score by 0.15 and then adding them together. The weights are set according to the principle that time synchronization has the greatest impact, field integrity has the same impact as acquisition stability, and source reliability is set as an auxiliary verification principle. The credibility threshold is as follows: The credibility threshold is set to 0.75. 0.75 corresponds to a lower limit for the combination of time consistency sub-score not lower than 0.7, data integrity sub-score not lower than 0.8, collection stability sub-score not lower than 0.8, and source reliability sub-score not lower than 0.7. Construction events with a credibility score lower than 0.75 are marked as low credibility events, and construction events with a credibility score not lower than 0.75 are retained as candidate credibility events. Delete low-reliability events, fill in the status gaps based on resource status changes within adjacent time windows for the same resource number, and reorder them according to event occurrence time, resource number, and construction task number to generate a reliable construction event flow.

[0023] In this embodiment, generating the construction resource friction propagation map and construction disturbance breakpoint data includes: Read the event occurrence time, resource number, construction task number, work surface number, material batch number, transport vehicle number, event type and event associated location from the trusted construction event stream, and extract resource occupation events, resource release events, resource movement events, material arrival events, transportation delay events, work surface closure events and process handover events; Based on resource movement events, resource release events, material arrival events, transportation delay events, work area closure events, and process handover events, calculate equipment movement friction, shift change friction, material transfer friction, transportation channel friction, work area congestion friction, and process handover friction, respectively. The friction caused by equipment movement is specifically as follows: Read the resource starting position and target work surface number corresponding to the same resource number in the resource movement event table. The resource starting position is given by the event association position. The center coordinate of the work surface corresponding to the target work surface number is read from the work surface occupancy data in the standardized construction resource dataset. Calculate the difference between the horizontal coordinate and the vertical coordinate of the resource starting position and the work surface center coordinate in the construction plane coordinate system. Square the two differences respectively, add them together and take the square root to obtain the resource movement distance. Read the movement duration in the resource movement event. Divide the resource movement distance by 50 to obtain the distance friction component. Divide the movement duration by 30 to obtain the time friction component. Record the distance friction component as 1 when it is greater than 1. Record the time friction component as 1 when it is greater than 1. Add the distance friction component and the time friction component and divide by 2 to generate the equipment movement friction. The friction between work teams is specifically as follows: Read adjacent construction events corresponding to the same resource number from the resource release event table and the resource occupation event table. Subtract the event occurrence time of the previous resource release event from the event occurrence time of the next resource occupation event to obtain the shift changeover time. Read the work surface number of the previous resource release event and the work surface number of the next resource occupation event. Read the center coordinates of the two work surfaces from the work surface occupation data in the standardized construction resource dataset. Calculate the planar distance between the center coordinates of the two work surfaces to obtain the changeover distance. Divide the shift changeover time by 20 to obtain the shift changeover time friction component. Divide the changeover distance by 30 to obtain the changeover distance friction component. Record both friction components as 1 if they are greater than 1. Add the shift changeover time friction component and the changeover distance friction component and divide by 2 to generate the shift changeover friction. Friction during material transport, specifically: Read the material batch number, event associated location, and target work surface number from the material arrival event table. The event associated location is used as the material storage location. Read the center coordinates of the work surface corresponding to the target work surface number from the work surface occupancy data in the standardized construction resource dataset. Calculate the planar distance between the material storage location and the center coordinates of the work surface to obtain the material transfer distance. Read the material inventory data in the standardized construction resource dataset, extract the material quantity and planned demand quantity corresponding to the material batch number, subtract the material quantity from the planned demand quantity to obtain the material shortage quantity. If the material shortage quantity is less than 0, record it as 0. Divide the material transfer distance by 40 to obtain the transfer distance friction component. Divide the material shortage quantity by the planned demand quantity to obtain the material shortage friction component. If both friction components are greater than 1, record them as 1. Add the transfer distance friction component and the material shortage friction component and divide by 2 to generate the material transfer friction. Friction along transportation routes, specifically: Read the transport vehicle number, planned arrival time, actual arrival time, and event-related location from the transport delay event table. Subtract the planned arrival time from the actual arrival time to obtain the transport delay time. Record the transport delay time as 0 if it is less than 0. Read the transport route length and passage status corresponding to the transport vehicle number from the transport vehicle data in the standardized construction resource dataset. Divide the transport delay time by 30 to obtain the delay friction component. Divide the transport route length by 1000 to obtain the path friction component. Record the passage status as 0 if it is unobstructed, 0.5 if it is slow, and 1 if it is closed. Record all three components as 1 if they are greater than 1. Add the delay friction component, path friction component, and passage status component and divide by 3 to generate the transport passage friction. The congestion and friction at the work surface are specifically as follows: Read the work surface number, closure start time, and closure duration from the work surface closure event table. Read the number of occupied work teams, the number of occupied machines, the number of work teams that can be accommodated, and the number of machines that can be accommodated for the same work surface number from the work surface occupancy data in the standardized construction resource dataset. Divide the number of occupied work teams by the number of work teams that can be accommodated to obtain the work team occupancy component. Divide the number of occupied machines by the number of machines that can be accommodated to obtain the machine occupancy component. Divide the closure duration by 60 to obtain the closure time component. Record all three components greater than 1 as 1. Add the work team occupancy component, machine occupancy component, and closure time component and divide by 3 to generate the work surface congestion friction. Friction during process handover, specifically: Read the preceding construction task number, the following construction task number, and the handover duration from the process handover event table. The preceding construction task number is given by the upstream task field corresponding to the construction task number in the process handover event, and the following construction task number is given by the downstream task field corresponding to the construction task number in the process handover event. Read the completion ratio and quality acceptance status corresponding to the preceding construction task number from the process progress data in the standardized construction resource dataset. Subtract the completion ratio from 1 to get the completion gap component. Record the quality acceptance status as passed as 0, the quality acceptance status as pending acceptance as 0.5, and the quality acceptance status as failed as 1. Divide the handover duration by 30 to get the handover time component. Record all three components greater than 1 as 1. Add the completion gap component, the quality acceptance component, and the handover time component and divide by 3 to generate the process handover friction. The construction task number, resource number, work surface number, material batch number, transport vehicle number, and process handover number are set as graph nodes. Resource occupancy relationships, resource release relationships, resource movement relationships, material supply relationships, transport arrival relationships, work surface closure relationships, and process handover relationships are set as graph edges. Corresponding resource friction is written into the graph edge attributes to generate a construction resource friction propagation graph. Specifically, generating the construction resource friction propagation graph involves: Read the trusted construction event stream, extract the construction task number and generate a construction task node table after deduplication, extract the resource number and generate a resource node table, extract the work surface number and generate a work surface node table, extract the material batch number and generate a material batch node table, extract the transport vehicle number and generate a transport node table, read the preceding construction task number and the following construction task number from the process handover event table, concatenate the preceding construction task number and the following construction task number according to the event occurrence time to generate the process handover number, deduplicate them to generate the process handover node table, arrange the 6 node tables according to node type, node number, associated construction task number, and associated event occurrence time to generate the graph nodes of the construction resource friction propagation diagram; Read the resource occupancy event table and establish resource occupancy edges between construction task nodes and resource nodes. Read the resource release event table and establish resource release edges between resource nodes and construction task nodes. Read the resource movement event table and establish resource movement edges between resource nodes and work surface nodes. Read the material arrival event table and establish material supply edges between material batch nodes and construction task nodes. Read the transportation delay event table and establish transportation arrival edges between transportation nodes and material batch nodes. Read the work surface closure event table and establish work surface closure edges between work surface nodes and construction task nodes. Read the process handover event table and establish process handover edges for the preceding construction task nodes, process handover nodes, and subsequent construction task nodes in sequence. Write the event occurrence time, event number, and resource friction corresponding to each graph edge into the graph edge attributes. According to the chronological order of resource friction occurrence under the same construction task number, the locations where resource flow is obstructed are marked, and construction disturbance breakpoint data is generated. Specifically, the generation of construction disturbance breakpoint data is as follows: Read the resource flow obstruction segment, determine the breakpoint type according to the graph edge type corresponding to the maximum resource friction, the resource movement edge corresponds to the equipment movement breakpoint, the resource release edge and resource occupation edge correspond to the shift change breakpoint, the material supply edge corresponds to the material transfer breakpoint, the transport arrival edge corresponds to the transport channel breakpoint, the work surface closure edge corresponds to the work surface congestion breakpoint, and the process handover edge corresponds to the process handover breakpoint. Arrange the breakpoint number, breakpoint type, construction task number, resource number, work surface number, material batch number, transport vehicle number, start time, end time, maximum resource friction and associated graph edge number into a table to generate construction disturbance breakpoint data.

[0024] In this embodiment, generating the scheduled executable prediction vector includes: An improved SCINet model is constructed, which includes a task resource job plane displacement encoder, a process resource interactive convolutional block, and a process breakpoint memory unit. Specifically, the construction of the improved SCINet model involves: This paper retains the recursive binary sampling structure, SCI-Block interaction structure, convolutional feature extraction structure, subsequence interaction update structure, residual connection path, sequence rearrangement output framework, and prediction projection framework of the traditional SCINet model. A construction event input adaptation region is set at the traditional input sequence position. The original input structure, which receives a single equally spaced time series, is transformed into an input structure that receives a reliable construction event stream, a construction resource friction propagation map, and construction disturbance breakpoint data. A task resource work surface displacement encoder is set at the traditional odd-even sampling position. The original sampling structure, which splits the time series according to odd and even positions, is transformed into a structure that generates a task resource work surface displacement coding sequence based on the construction task number, resource number, work surface number, and event occurrence time. The construction semantic coding structure sets up a process resource interaction convolution block at the traditional SCI-Block position, transforming the original structure that only performs interactive convolution between two sampling sub-sequences into a construction resource coupling structure that performs convolution interaction between the task channel, resource channel, work surface channel, material channel, and transportation channel. At the traditional hidden state transmission path position, a process breakpoint memory unit is set up, transforming the original hidden state structure that only transmits the current convolution feature into a breakpoint memory structure that reads construction disturbance breakpoint data and retains the position of resource flow obstruction. At the traditional prediction projection framework position, a prediction readout layer is set up, transforming the original time series prediction output structure into a task start-up strength, resource reachability strength, material support strength, work surface acceptance strength, and task release time output structure. The task resource operation surface displacement encoder includes: Construction event reading unit: Reads the event occurrence time, resource number, construction task number, work surface number, material batch number, transport vehicle number, event type, and event associated location from the trusted construction event stream, groups them according to construction task number, and arranges them from earliest to latest according to the event occurrence time within each construction task number group to generate a construction event input sequence; Displacement field extraction unit: Read the resource movement edge in the construction resource friction propagation diagram, extract the resource starting position, target work surface number, movement duration and resource friction corresponding to the resource movement edge, and read the work surface center coordinates corresponding to the target work surface number from the work surface occupancy data in the standardized construction resource dataset; Displacement coding generation unit: Subtract the horizontal coordinate of the center coordinate of the working surface from the horizontal coordinate of the resource's starting position to obtain the horizontal displacement; subtract the vertical coordinate of the center coordinate of the working surface from the vertical coordinate of the resource's starting position to obtain the vertical displacement; arrange the horizontal displacement, vertical displacement, movement duration, and resource friction into a 4-dimensional displacement coding row. The 4-dimensional displacement coding row is used to express the spatial offset, time consumption, and frictional resistance when the resource flows from the current position to the target working surface. Sequence Output Unit: Concatenates the 4D displacement code row with the resource number, construction task number, work surface number, material batch number, transport vehicle number and the integer index corresponding to the event type to generate a single event code row. Arrange all event code rows according to the construction task number and the event occurrence time to generate the task resource work surface displacement code sequence. Process resource interaction convolutional blocks include: Interactive channel establishment unit: Reads the task resource work surface displacement coding sequence, writes the event coding line with the same construction task number into the task channel, writes the event coding line with the same resource number into the resource channel, writes the event coding line with the same work surface number into the work surface channel, writes the event coding line with the same material batch number into the material channel, and writes the event coding line with the same transport vehicle number into the transport channel. Each channel is arranged from earliest to latest according to the event occurrence time, and fills the position of missing event coding line with a 0 value line; Convolutional window construction area: Convolutional windows with a length of 3 event coding lines are set in the task channel, resource channel, work surface channel, material channel and transportation channel respectively. The length of the convolutional window is set to 3, which specifically corresponds to the local changes of the same construction task in the three event stages of adjacent resource arrival, operation execution and resource release. Channel convolution calculation area: Read the horizontal displacement, vertical displacement, movement duration and resource friction in each convolution window, calculate the average, maximum and last values ​​of the same field in the three event coding lines, and concatenate the average, maximum and last values ​​to form the channel convolution features. The channel convolution features are arranged in a fixed order of task channel, resource channel, work surface channel, material channel and transportation channel. The graph-edge interaction calculation unit reads the resource occupation edge, resource release edge, material supply edge, transportation arrival edge, work surface closure edge and process handover edge connected to the same construction task number in the construction resource friction propagation graph, and splices the resource friction in each graph edge attribute to the channel convolution feature in the order of edge type to generate a process resource interaction feature sequence. The process breakpoint memory unit includes: Breakpoint data reading unit: Reads the breakpoint number, breakpoint type, construction task number, resource number, work surface number, material batch number, transport vehicle number, start time, end time, maximum resource friction, and associated graph edge number from the construction disturbance breakpoint data; Breakpoint memory slot establishment unit: Breakpoint memory slots are established according to the construction task number. Each breakpoint memory slot is set with 5 memory positions. The 5 memory positions correspond to the top 5 construction disturbance breakpoints under the same construction task number, sorted from largest to smallest in terms of maximum resource friction. Positions with fewer than 5 construction disturbance breakpoints are filled with a value of 0. The 5 memory positions are used to retain the record of the biggest obstruction that has the greatest impact on scheduling in the same construction task. Breakpoint memory writing unit: Write the breakpoint type index, resource number index, work surface number index, start time, end time, maximum resource friction and associated graph edge number of each construction disturbance breakpoint into the breakpoint memory slot. Sort more than 5 construction disturbance breakpoints under the same construction task number in descending order of maximum resource friction and delete the construction disturbance breakpoints that are 6th in the sort and beyond. Breakpoint Feature Generation Unit: Reads the breakpoint memory slot, counts the number of breakpoints under each construction task number, reads the maximum resource friction, reads the earliest start time and the latest end time, subtracts the earliest start time from the latest end time to get the breakpoint duration, and arranges the number of breakpoints, maximum resource friction and breakpoint duration in columns to generate process breakpoint memory features. Predictive readout layer, including: Read input unit: Read the resource occupancy features, resource release features, material supply features, transportation arrival features, work surface closure features and process handover features in the process resource interaction feature sequence; read the number of breakpoints, maximum resource friction and breakpoint duration in the process breakpoint memory features. Strength Calculation Unit: The average of resource release characteristics, material supply characteristics, and process handover characteristics is subtracted from the maximum resource friction to generate the task start-up strength. The average of equipment movement friction and transportation channel friction is subtracted from 1 to generate the resource reachable strength. The material supply characteristics are subtracted from the material transfer friction to generate the material support strength. The work surface congestion friction is subtracted from 1 to generate the work surface acceptance strength. Strength values ​​less than 0 are recorded as 0, and strength values ​​greater than 1 are recorded as 1. Time Readout Unit: Reads the event occurrence time of the last resource release event corresponding to the construction task number, adds the breakpoint duration to the event occurrence time of the last resource release event, and generates the task release time; Vector output unit: Arranges the task startability strength, resource availability strength, material support strength, work surface acceptance strength, and task release time according to the construction task number to generate a scheduling executable prediction vector; In the improved SCINet model, the trusted construction event stream enters the construction event reading area to generate the construction event input sequence. The construction event input sequence and the construction resource friction propagation graph enter the task resource work surface displacement encoder to generate the task resource work surface displacement encoding sequence. The task resource work surface displacement encoding sequence enters the process resource interaction convolution block to generate the process resource interaction feature sequence. The construction disturbance breakpoint data and the process resource interaction feature sequence enter the process breakpoint memory unit to generate the process breakpoint memory feature. The process resource interaction feature sequence and the process breakpoint memory feature enter the prediction readout layer to generate the scheduling executable prediction vector. Read the event occurrence time, resource number, construction task number, work surface number, material batch number, transport vehicle number and event type from the trusted construction event stream, and generate a construction event input sequence according to the construction task number and event occurrence time; The task resource work surface displacement encoder reads the construction event input sequence and the construction resource friction propagation map, extracts the resource number corresponding to the current position, the construction task number corresponding to the target work surface number, the resource movement relationship corresponding to the movement duration, and the resource friction in the map edge attributes, and generates a task resource work surface displacement encoding sequence. Specifically, the generation of the task resource work surface displacement encoding sequence is as follows: Read the resource number, construction task number, and work surface number from the construction event input sequence. Read the resource starting position, target work surface number, movement duration, and resource friction corresponding to the resource movement edge in the construction resource friction propagation diagram. Read the work surface center coordinates corresponding to the target work surface number from the work surface occupancy data in the standardized construction resource dataset. Subtract the horizontal coordinate of the work surface center coordinate from the horizontal coordinate of the resource starting position to obtain the horizontal displacement. Subtract the vertical coordinate of the work surface center coordinate from the vertical coordinate of the resource starting position to obtain the vertical displacement. Arrange the horizontal displacement, vertical displacement, movement duration, and resource friction into displacement code rows by column. Arrange all displacement code rows by construction task number and event occurrence time to generate the task resource work surface displacement code sequence. The process resource interaction convolutional block establishes interaction channels based on construction task number, resource number, work surface number, material batch number, and transport vehicle number. It performs convolutional interactions on resource occupancy relationships, resource release relationships, material supply relationships, transportation arrival relationships, work surface closure relationships, and process handover relationships, generating a process resource interaction feature sequence, where: The process resource interaction convolutional block establishes interaction channels according to the construction task number, resource number, work surface number, material batch number, and transport vehicle number. Specifically, it reads the task resource work surface displacement code sequence, writes the displacement code row with the same construction task number into the task channel, writes the displacement code row with the same resource number into the resource channel, writes the displacement code row with the same work surface number into the work surface channel, writes the displacement code row with the same material batch number into the material channel, and writes the displacement code row with the same transport vehicle number into the transport channel. Each interaction channel is arranged from earliest to latest according to the event occurrence time. The position of missing displacement code row is filled with a 0 value row, generating the task channel matrix, resource channel matrix, work surface channel matrix, material channel matrix, and transport channel matrix. Convolutional interaction, specifically: Read the task channel matrix, resource channel matrix, work surface channel matrix, material channel matrix, and transportation channel matrix. Use three consecutive event feature rows as one convolution window. Read the lateral displacement, longitudinal displacement, movement duration, and resource friction from each convolution window. Calculate the average, maximum, and last values ​​of the same field in the three event feature rows. Concatenate the average, maximum, and last values ​​to form the channel convolution feature. Read the graph edge attributes connected to the same construction task number in the construction resource friction propagation graph. Concatenate the resource friction corresponding to resource occupation relationship, resource release relationship, material supply relationship, transportation arrival relationship, work surface closure relationship, and process handover relationship after the channel convolution feature to generate a process resource interaction feature sequence. The process breakpoint memory unit reads construction disturbance breakpoint data and process resource interaction feature sequences, writes the resource flow obstruction location, construction task number, resource number, work surface number, and occurrence time into the breakpoint memory slot, and generates process breakpoint memory features, including: The location where resource flow is obstructed, the construction task number, the resource number, the work surface number, and the time of occurrence are written into the breakpoint memory slot, specifically: Read the breakpoint number, breakpoint type, construction task number, resource number, work surface number, start time, end time, maximum resource friction, and associated graph edge number from the construction disturbance breakpoint data. Establish breakpoint memory slots according to the construction task number. Each breakpoint memory slot contains a breakpoint type index, a resource number index, a work surface number index, a start time, an end time, maximum resource friction, and associated graph edge number. Write the top 5 breakpoints with the largest resource friction under the same construction task number in descending order into the breakpoint memory slots. Fill the positions with fewer than 5 breakpoints with a value of 0 to generate a breakpoint memory slot matrix. The generation process breakpoint memory features are as follows: Read the breakpoint memory slot matrix, calculate the number of breakpoints, maximum resource friction, earliest start time, and latest end time for each construction task number, subtract the earliest start time from the latest end time to get the breakpoint duration, and arrange the number of breakpoints, maximum resource friction, and breakpoint duration by column to generate process breakpoint memory features. Based on the process resource interaction feature sequence and process breakpoint memory features, the task startability intensity, resource availability intensity, material support intensity, work surface acceptance intensity, and task release time are generated, and then combined to generate a scheduling executable prediction vector. Specifically, the combined generation of the scheduling executable prediction vector is as follows: The scheduling executable prediction vector is generated as follows: Resource occupancy features, resource release features, material supply features, transportation arrival features, work surface closure features, and process handover features are read from the process resource interaction feature sequence. The number of breakpoints, maximum resource friction, and breakpoint duration are read from the process breakpoint memory features. The task startability intensity is obtained by averaging the resource release features, material supply features, and process handover features, and then subtracting the maximum resource friction. The resource reachability intensity is obtained by subtracting the average of equipment movement friction and transportation channel friction from 1. The material support intensity is obtained by subtracting material transfer friction from the material supply features. The work surface acceptance intensity is obtained by subtracting work surface congestion friction from 1. The task release time is obtained by adding the breakpoint duration to the event occurrence time of the last resource release event corresponding to the construction task number. Values ​​of task startability intensity, resource reachability intensity, material support intensity, and work surface acceptance intensity less than 0 are recorded as 0, and values ​​greater than 1 are recorded as 1. The task startability intensity, resource reachability intensity, material support intensity, work surface acceptance intensity, and task release time are arranged according to the construction task number to generate the scheduling executable prediction vector.

[0025] In this embodiment, the construction of the individual set of construction scheduling parameters includes: Read the task start-up intensity, resource reachability intensity, material support intensity, work surface acceptance intensity, and task release time from the scheduled executable prediction vector. Read the construction task number, planned start time, planned completion time, and resource requirement from the construction task data, where: Read the scheduled executable prediction vector, specifically: Read the task start-up intensity, resource reachability intensity, material support intensity, work surface acceptance intensity and task release time from the scheduled executable prediction vector according to the construction task number. Arrange the five fields in a fixed column order to form task prediction rows. Arrange all task prediction rows in ascending order of construction task number to generate a task prediction matrix. Read the construction task data, specifically: Read the construction task number, planned start time, planned completion time, resource requirement quantity, required type of work, required type of machinery, required type of material, and target work area number from the construction task data in the standardized construction resource dataset. Write data with the same construction task number into the same task requirement row to generate a task requirement matrix. The task prediction matrix is ​​related to the task requirement matrix, specifically as follows: Using the construction task number as the matching key, the task start-up intensity, resource availability intensity, material support intensity, work surface acceptance intensity and task release time in the task prediction matrix are written into the corresponding row of the task demand matrix to generate the scheduling task base table. Read the graph node and graph edge attributes in the construction resource friction propagation graph, and extract the allocable work teams, allocable machinery, supplyable material batches, arrangeable transport vehicles, and acceptable work surfaces; Discrete and continuous coding segments are established according to the construction task number. The task sequence, work group number, machinery number, material batch number, transport vehicle number, and work area number are written into the discrete coding segment. The task start time, task completion time, resource allocation time, and material arrival time are written into the continuous coding segment, generating a set of individual construction scheduling parameters. Specifically, the set of individual construction scheduling parameters is generated as follows: The discrete and continuous coding lines corresponding to the same construction task number are concatenated into one individual coding line. All individual coding lines corresponding to the construction task numbers are combined into one construction scheduling parameter individual. The top 5 resource combinations are sorted from smallest to largest according to resource friction, and the top 5 task sequences are sorted from largest to smallest according to task startability intensity. These combinations generate 25 construction scheduling parameter individuals. They are then arranged from 1 to 25 according to individual number to generate a set of construction scheduling parameter individuals.

[0026] In this embodiment, generating the scheduling and backfeed prediction results includes: Read the discrete and continuous coded segments from the individual set of construction scheduling parameters, and read the task start-up intensity, resource reachability intensity, material support intensity, work surface acceptance intensity, and task release time from the scheduling executable prediction vector; The walrus optimization algorithm, based on task startability intensity, resource availability intensity, material support intensity, and work surface acceptance intensity, forms a disturbance migration direction. It migrates the work group number, machine number, material batch number, transport vehicle number, and work surface number in the discrete coding segment to the corresponding node of the graph edge with a smaller resource friction value in the construction resource friction propagation graph. It also adjusts the task start time, resource allocation time, and material arrival time in the continuous coding segment, generating individual migration scheduling parameters, where: The walrus optimization algorithm determines the perturbation migration direction based on the task's startability intensity, resource availability intensity, material support intensity, and work surface acceptance intensity. Specifically: Read the task startability intensity, resource availability intensity, material support intensity, and work surface acceptance intensity corresponding to each construction task number in the individual scheduling matrix to be updated. Compare the four intensity values ​​with 0.75 respectively. When the intensity value is lower than 0.75, generate the corresponding disturbance mark. When the resource availability intensity is lower than 0.75, generate the resource migration direction. When the material support intensity is lower than 0.75, generate the material migration direction. When the work surface acceptance intensity is lower than 0.75, generate the work surface migration direction. When the task startability intensity is lower than 0.75, generate the time migration direction. The migration of work group numbers and machine numbers is as follows: Read the resource nodes and resource movement edges connected to the same construction task number in the construction resource friction propagation graph, extract the resource friction of the resource movement edges, sort the resource nodes according to the resource friction from small to large, select the top 3 resource nodes with matching resource types as candidate migration nodes, replace the team number and machine number in the discrete coding segment with the team number and machine number in the candidate migration node with the resource reachability strength of less than 0.75, and generate the resource migration coding segment. The migration of material batch numbers and transport vehicle numbers is as follows: Read the material batch node, transportation node, material supply edge, and transportation arrival edge connected to the same construction task number in the construction resource friction propagation diagram. Extract the material transfer friction of the material supply edge and the transportation channel friction of the transportation arrival edge respectively. Add the material transfer friction and the transportation channel friction and divide by 2 to get the material transportation comprehensive friction. Sort the material batch node and transportation node according to the material transportation comprehensive friction from small to large. Select the top 3 material batch numbers and transportation vehicle numbers that match the material type and replace the material batch numbers and transportation vehicle numbers with material support strength lower than 0.75 in the discrete coding segment. The work area number migration is as follows: Read the work surface nodes and work surface closing edges connected to the same construction task number in the construction resource friction propagation diagram, extract the work surface congestion friction of the work surface closing edges, read the work surface occupancy data in the standardized construction resource dataset, retain the work surface numbers that are closed but open, have fewer occupied work groups than the number of work groups that can be accommodated, and have fewer occupied machines than the number of machines that can be accommodated, sort them according to the work surface congestion friction from smallest to largest, and select the top 3 work surface numbers to replace the work surface numbers in the discrete coding segment whose work surface acceptance intensity is less than 0.75. Adjusting consecutive coding segments, specifically: Read the task start time, resource allocation time and material arrival time from the continuous coding segment, read the task release time from the scheduled executable prediction vector, rewrite the task start time as the later of the task release time and the original task start time, rewrite the resource allocation time as the task start time minus the movement duration corresponding to the resource movement event, and rewrite the material arrival time as the arrival time corresponding to the selected material batch number, and generate individual migration scheduling parameters. Migration scheduling parameters are clustered according to work area number and construction task number. A leading scheduling parameter is selected based on the task startability intensity and resource availability intensity under the same work area number. The task sequence, resource allocation time, and material arrival time of the leading scheduling parameter are then passed to other migration scheduling parameter individuals within the same cluster to generate navigation scheduling parameter individuals. The migration scheduling parameters are clustered according to the work area number and construction task number, specifically as follows: Read the job surface number and construction task number from the individual migration scheduling parameters, group the coding rows with the same job surface number into the same job surface cluster, create task sub-clusters according to the construction task number within the same job surface cluster, and arrange the job surface cluster number, task sub-cluster number, individual number and coding row number into a cluster index table. The leader scheduling parameter individual is selected based on the task's workability intensity and resource availability intensity under the same work surface number, specifically as follows: Read the task startability intensity and resource reachability intensity of each migration scheduling parameter individual within the same work surface cluster. Add the task startability intensity and resource reachability intensity to obtain the leader evaluation value. Sort the migration scheduling parameter individuals in descending order of leader evaluation value, select the migration scheduling parameter individual ranked first as the leader scheduling parameter individual, and write the leader scheduling parameter individual number into the cluster index table. Generate individual navigation scheduling parameters, specifically: Read the task order, resource allocation time, and material arrival time from the leading scheduling parameter individual; read other migration scheduling parameter individuals within the same work area cluster; locate the coding line in other migration scheduling parameter individuals that has the same construction task number as the leading scheduling parameter individual; replace the task order with the task order in the leading scheduling parameter individual; replace the resource allocation time with the resource allocation time in the leading scheduling parameter individual; replace the material arrival time with the material arrival time in the leading scheduling parameter individual; and generate the navigation scheduling parameter individual. The navigation scheduling parameter individuals are read from back to front according to the task completion time in the continuous coding segment. The pruning positions are determined in ascending order of task startability intensity, resource availability intensity, material support intensity, and work surface acceptance intensity. The task sequence and resource allocation content corresponding to the pruning position are deleted from the navigation scheduling parameter individuals. Based on the task release time, the construction task number, work group number, machinery number, material batch number, transport vehicle number, work surface number, and task start time are re-inserted to generate updated scheduling parameter individuals. Among these: Read individual navigation scheduling parameters from back to front, specifically: Read the task completion time in the individual continuous coding segment of the navigation scheduling parameters, sort the coding rows from late to early according to the task completion time, read the sorted coding rows in sequence, and write the construction task number, team number, machinery number, material batch number, transport vehicle number, work surface number and task start time corresponding to each coding row into the reverse read table. Determine the pruning location, specifically: Read the task start strength, resource availability strength, material support strength and working surface acceptance strength corresponding to each construction task number in the reverse read table. Add the four strength values ​​and divide by 4 to get the trimming evaluation value. Sort the trimming evaluation values ​​from smallest to largest and mark the first 20% of the coded rows as trimming positions. If the number of coded rows is less than 5, select the coded row with the lowest trimming evaluation value as the trimming position. Delete the content corresponding to the trimmed location, specifically: Read the discrete and continuous coding segments in the individual navigation scheduling parameters, remove the task order, team number, machine number, material batch number, transport vehicle number, work area number, task start time, task completion time, resource allocation time and material arrival time from the coding line corresponding to the trimming position, retain the untrimmed coding line, and generate the trimmed remaining coding segment; Generate and update individual scheduling parameters, specifically as follows: Read the construction task number and task release time corresponding to the pruning position. Read the team number, machine number, material batch number, transport vehicle number and work surface number that are connected to the construction task number and are sorted from smallest to largest resource friction in the construction resource friction propagation diagram. Set the task start time to the later of the task release time and the planned start time. Set the task completion time to the task start time plus the difference between the planned completion time and the planned start time. Write the re-inserted discrete field and continuous field into the pruning remaining coding segment to generate an updated scheduling parameter individual. The updated scheduling parameters are restored to candidate task start events, candidate resource occupancy events, candidate resource release events, candidate material arrival events, candidate transportation arrival events, candidate work surface occupancy events, and candidate process handover events. A candidate construction event refeed sequence is generated and input into the improved SCINet model to generate scheduling refeed prediction results, where: Generate a candidate construction event recharge sequence, specifically as follows: Merge the candidate task commencement event table, candidate resource occupancy event table, candidate resource release event table, candidate material arrival event table, candidate transportation arrival event table, candidate work surface occupancy event table, and candidate process handover event table, sort the events from earliest to latest according to their occurrence time, and sort them according to the construction task number and resource number within the same event occurrence time to generate a candidate construction event backflow sequence; The improved SCINet model is used to generate scheduling refeeding prediction results, specifically: The candidate construction event recharge sequence is input into the task resource work surface displacement encoder. A recharge displacement coding sequence is generated according to the resource start position, target work surface number, movement duration and resource friction. The recharge displacement coding sequence is input into the process resource interaction convolution block to generate the recharge process resource interaction feature sequence. The recharge process resource interaction feature sequence and the existing breakpoint memory slot in the process breakpoint memory unit are input into the prediction readout layer to generate the task startability strength, resource reachability strength, material support strength, work surface acceptance strength and task release time after recharge. The five fields are arranged according to the construction task number to generate the scheduling recharge prediction result.

[0027] In this embodiment, generating the target construction resource scheduling scheme includes: Read the task start-up intensity, resource availability intensity, material support intensity, work surface acceptance intensity, and task release time from the scheduling and recharge prediction results, and read the construction task number, work group number, machinery number, material batch number, transport vehicle number, work surface number, task start time, and task completion time from the individual updated scheduling parameters. Based on the task's workability intensity, resource availability intensity, material support intensity, and work surface acceptance intensity, generate resource unavailability markers, material unavailable markers, and work surface unacceptability markers, and then filter and update the scheduling parameters of individuals, marking the individuals with updated scheduling parameters as eliminated individuals. Extract failed tasks, failed resources, failed work surfaces, and failed time windows from the eliminated individuals, and write the failed tasks, failed resources, failed work surfaces, failed time windows, unreachable resource markers, unavailable material markers, and unacceptable work surface markers into the process breakpoint memory unit; Based on safety management and environmental monitoring data, hard constraint filtering is applied to retained individuals. For construction task numbers that exhibit issues such as blocked work surfaces, unreachable resources, and unavailable materials after filtering, local scheduling and repair are performed, generating individuals with repair scheduling parameters. Among these: Hard constraint filtering is applied to retained individuals based on safety management data and environmental monitoring data, specifically as follows: Read the work area number, transport vehicle number, machinery number, task start time and task completion time from the retained individual; read the prohibited work areas, prohibited transport channels, prohibited machinery status and safe operation period from the safety control data; read the stop wind speed mark, rainfall stop mark and insufficient visibility mark from the environmental monitoring data. If the work area number belongs to the prohibited work area, or the transport channel corresponding to the transport vehicle number belongs to the prohibited transport channel, or the prohibited machinery status corresponding to the machinery number is prohibited, or the task start time to task completion time is not within the safe operation period, or the stop wind speed mark is 1, or the rainfall stop mark is 1, or the insufficient visibility mark is 1, then mark the retained individual as failing the hard constraint; otherwise, mark it as passing the hard constraint. For construction tasks that, after filtering, have issues such as blocked work surfaces, unreachable resources, and unavailable materials, partial scheduling and repair will be performed, specifically as follows: Read the retained individuals that fail the hard constraints and the construction task numbers that still have blocked work surfaces, unreachable resources, and unavailable materials after filtering. In the construction resource friction propagation graph, read the work group nodes, machine nodes, material batch nodes, transportation nodes, and work surface nodes connected to the same construction task number. Delete the corresponding nodes of the graph edges with resource friction not less than 0.6. Delete the work surface nodes, transportation nodes, and machine nodes that fail the hard constraints. Sort the remaining graph edges according to the resource friction from smallest to largest, and reselect the work group number, machine number, material batch number, transportation vehicle number, and work surface number that are ranked first in the sorting. Generate and repair individual scheduling parameters, specifically: Read the work group number, machine number, material batch number, transport vehicle number and work surface number of the reselected work group number, machine number, material batch number, transport vehicle number and work surface number of the local scheduling repair; read the task release time in the scheduling backfill prediction result; adjust the task start time to the later of the task release time and the original task start time; adjust the task completion time to the task start time plus the difference between the original task completion time and the original task start time; write the reselected discrete field and the adjusted continuous field into the coding line of the corresponding construction task number to generate individual repair scheduling parameters; The repaired scheduling parameters are restored to task scheduling results, personnel allocation results, machinery allocation results, material supply results, transportation arrangement results, and work area arrangement results, generating a target construction resource scheduling plan. Specifically, generating the target construction resource scheduling plan involves: Read the construction task number, work surface number, task start time and task completion time from the individual repair scheduling parameters, group them according to the work surface number and sort them according to the task start time to generate the work surface arrangement result. Associate the task scheduling result, personnel allocation result, machinery allocation result, material supply result, transportation arrangement result and work surface arrangement result according to the construction task number to generate the target construction resource scheduling plan.

[0028] Example 1: In a comprehensive construction resource scheduling scenario, the construction site includes 4 construction sections, 36 construction tasks, 10 work teams, 14 pieces of machinery and equipment, 260 batches of materials, 12 transport vehicles, and 22 work areas. Construction resource data was continuously collected over 72 hours, yielding 720 personnel attendance records, 3024 machinery operation records, 1040 material inventory records, 5184 transport vehicle location records, 2376 work area occupancy records, 864 construction task progress records, 3456 environmental monitoring records, and 416 safety management records. The raw data contained issues such as data collection delays, missed material arrival records, and vehicle location inconsistencies. Specifically, records with event timestamp deviations exceeding 2 seconds accounted for 7.8%, records with missing material batch fields accounted for 2.9%, and records with transport vehicle location inconsistencies accounted for 5.6%. The training sample consists of 150 consecutive construction scheduling cycles of the same type of project, containing 98,200 construction events. The validation sample consists of 30 construction scheduling cycles, containing 19,640 construction events. The test sample consists of 72 hours of data from this embodiment.

[0029] After the data enters the processing flow, this invention performs time alignment, data cleaning, and standardization on the construction resource data. Personnel status is standardized as on-duty, off-duty, being transferred, and waiting; machinery status is standardized as available, occupied, under maintenance, and faulty; and material status is standardized as in storage, in transit, awaiting transfer, and short of materials. The processed data generates a standardized construction resource dataset and a construction event flow. The construction event flow includes seven categories of events: resource occupation, resource release, resource movement, material arrival, transportation delay, work site closure, and process handover, forming a total of 24,680 construction events. Each construction event is written with a timestamp, a missing event flag, a jitter flag, and a source flag. In the credibility scoring, time consistency is weighted at 0.35, field integrity at 0.25, acquisition stability at 0.25, source reliability at 0.15, and the credibility threshold at 0.75. After credibility scoring, removal of low-credibility events, and correction of adjacent events, 1,684 low-credibility events are identified, 1,327 status gaps are filled, and a credible construction event flow of 24,323 events is generated.

[0030] In the construction resource friction calculation phase, this invention extracts 3260 resource movement events, 260 material arrival events, 284 transportation delay events, 72 work surface closure events, and 148 process handover events from the reliable construction event flow. Taking the transfer of mechanical equipment M07 from work surface A08 to work surface A13 as an example, the initial resource position coordinates are 128.6m and 46.2m, the target work surface center coordinates are 169.4m and 66.8m, the planar movement distance is 45.7m, the movement duration is 24min, the distance friction component is 0.91, the time friction component is 0.8, and the equipment movement friction is 0.86. Taking material batch B126 as an example, the distance from the material storage location to the target work surface is 52.4m, the planned demand quantity is 12t, the available inventory quantity is 10.9t, the material shortage quantity is 1.1t, the transfer distance friction component is truncated to 1, the material shortage friction component is 0.09, and the material transfer friction is 0.55. After all events are calculated, a construction resource friction propagation graph with 864 graph nodes and 3920 graph edges is generated. Among them, there are 438 high friction edges with a resource friction of not less than 0.6. After merging, 76 construction disturbance breakpoint data are formed, including 21 equipment movement breakpoints, 12 team change breakpoints, 16 material transfer breakpoints, 10 transportation channel breakpoints, 9 work surface congestion breakpoints, and 8 process handover breakpoints.

[0031] The SCINet model is improved to read reliable construction event streams, construction resource friction propagation maps, and construction disturbance breakpoint data. The task resource work surface displacement encoder converts each resource movement event into a 4D displacement encoding line composed of lateral displacement, longitudinal displacement, movement duration, and resource friction. This line is then concatenated with the resource number, construction task number, work surface number, material batch number, transport vehicle number, and event type index to form a 24323×10 task resource work surface displacement encoding sequence. The process resource interaction convolutional block writes the encoding sequence into the task channel, resource channel, work surface channel, material channel, and transport channel. The convolutional window length is three event encoding lines, generating 36 sets of process resource interaction feature sequences. The process breakpoint memory unit retains the top 5 breakpoints with the highest resource friction for each construction task. For construction task T14, it writes the material transfer breakpoint (0.79), the transport channel breakpoint (0.73), and the work surface congestion breakpoint (0.68), with a cumulative breakpoint duration of 76 minutes. The predictive readout layer outputs 36 scheduled executable prediction vectors, of which the task startability strength of T14 is 0.48, the resource reachability strength is 0.61, the material support strength is 0.54, the work surface acceptance strength is 0.66, and the task release time is delayed by 68 minutes compared to the original plan.

[0032] The Walrus Optimization Algorithm constructs a set of individual construction scheduling parameters based on the scheduling executable prediction vector. Each individual consists of discrete and continuous coding segments. The discrete coding segment includes task sequence, work group number, machine number, material batch number, transport vehicle number, and work surface number. The continuous coding segment includes task start time, task completion time, resource allocation time, and material arrival time. Initially, 25 individual construction scheduling parameters are constructed. During the disturbance migration phase, tasks with a friction intensity below 0.75 are read. Material batch T14 is replaced from B126 to B139 (resource friction 0.34), transport vehicle V03 is replaced from V03 to V09 (channel friction 0.31), and work surface A13 is replaced from A13 to A17 (congestion friction 0.26). During the clustering leader navigation phase, 14 clusters are formed according to work surface number and construction task number. During the task priority reverse pruning phase, tasks are read from back to front according to completion time. The 20% of inefficient coding lines after the comprehensive intensity sorting are deleted and re-inserted into low-friction resource combinations. After updating the scheduling parameters and restoring the individual candidate construction event re-feedback sequences, the improved SCINet model was re-inputted. The task startability intensity of T14 increased from 0.48 to 0.78, the material support intensity increased from 0.54 to 0.8, and the work surface acceptance intensity increased from 0.66 to 0.83.

[0033] During the target scheme screening phase, in the scheduling and recharge prediction results, there were 5 individuals with resource availability strength below 0.75, 4 individuals with material support strength below 0.75, and 3 individuals with work surface acceptance strength below 0.75. These individuals were marked as eliminated, and the failed tasks, failed resources, failed work surfaces, and failed time windows were written into the process breakpoint memory unit. The retained individuals continued to be filtered by hard constraints of safety control data and environmental monitoring data, with 2 work surfaces prohibited from entry, 1 transportation channel prohibited, and 1 stop-work wind speed marked. After filtering, 10 individuals were retained. For the local scheduling repair, for T09 with work surface blockage and T21 with unreachable resources, work groups, machinery, material batches, transport vehicles, and work surfaces with resource friction below 0.6 were reselected. Finally, the target construction resource scheduling scheme was generated, including 36 task scheduling results, 10 personnel allocation results, 14 machinery allocation results, 260 batches of material supply records, 12 transportation arrangement results, and 22 work surface arrangement results.

[0034] In the comparative experiment, the traditional method used a BIM schedule combined with a resource-constrained scheduling model. The training sample size was the same: 150 consecutive construction scheduling cycles. The test sample size was the same: 72 hours of construction data. The average scheduling calculation time for the traditional method was 18.6 minutes, while for this invention it was 11.4 minutes. The resource status identification accuracy rate for the traditional method was 82.1%, while for this invention it was 90.6%. The traditional method had 14 material supply interruption events, while this invention had 8. The total idling time for machinery was 39.5 hours for the traditional method, while for this invention it was 29.7 hours. The total waiting time for personnel was 34.2 hours for the traditional method, while for this invention it was 25.8 hours. The number of work surface conflicts was 18 for the traditional method, while for this invention it was 11. The average delay for key processes was 84 minutes for the traditional method, while for this invention it was 56 minutes. The first-time execution pass rate of the scheduling plan on site was 73.8% for the traditional method, while for this invention it was 84.6%. This embodiment demonstrates that the present invention has a reproducible data processing process in terms of reliable processing of construction events, expression of resource friction, prediction of scheduling backflow, and local scheduling repair. It can reduce material waiting, mechanical idling, and work surface conflicts in the resource scheduling of construction sites, and improve the execution stability of the dynamic scheduling scheme for construction resources.

[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for dynamic scheduling and optimization of construction resources based on big data analysis, characterized in that, include: Collect and preprocess construction resource data to generate a standardized construction resource dataset. Perform event extraction on the standardized construction resource dataset to generate a construction event stream. The credibility score, low-credibility event removal, and adjacent event correction are performed on the construction event flow to generate a credible construction event flow. Extract resource occupation events, release events, movement events, material arrival events, transportation delay events, work surface closure events, and process handover events from the trusted construction event flow; calculate resource friction and generate a construction resource friction propagation diagram and construction disturbance breakpoint data; An improved SCINet model containing a task resource operation surface displacement encoder, a process resource interactive convolutional block, and a process breakpoint memory unit is constructed. The model is then fed with a reliable construction event flow, a construction resource friction propagation graph, and construction disturbance breakpoint data to generate a scheduling executable prediction vector. A set of individual construction scheduling parameters is constructed based on the scheduling executable prediction vector, the construction resource friction propagation graph, and the standardized construction resource dataset. The Walrus Optimization Algorithm is used to perform perturbation migration, cluster leader navigation update and task priority reverse pruning on the individual set of construction scheduling parameters to generate updated individual scheduling parameters and restore them to candidate construction event recharge sequences. The recharge improves the SCINet model to generate scheduling recharge prediction results. Based on the scheduling backfeed prediction results, individuals with updated scheduling parameters are selected. The failed tasks, failed resources, failed work surfaces, and failed time windows of the eliminated individuals are written into the process breakpoint memory unit. Hard constraint filtering and local scheduling repair are performed on the retained individuals to generate the target construction resource scheduling scheme.

2. The method for dynamic scheduling and optimization of construction resources based on big data analysis according to claim 1, characterized in that, The construction resource data includes data on construction personnel, machinery and equipment, material inventory, transport vehicles, work site occupancy, construction tasks, environmental monitoring, and safety management.

3. The method for dynamic scheduling and optimization of construction resources based on big data analysis according to claim 1, characterized in that, The preprocessing includes time alignment, data cleaning, and standardization.

4. The method for dynamic scheduling and optimization of construction resources based on big data analysis according to claim 1, characterized in that, The generation of a trusted construction event stream includes: Read the acquisition end timestamp, missing marker, jitter marker, and source marker of each construction event in the construction event stream, and generate an event credibility evaluation sequence according to the event occurrence time; The time consistency value of adjacent construction events is calculated based on the timestamp of the acquisition terminal, and the number of consecutive missing items of the same resource number and the number of time jitters of the same source tag are counted based on the missing tag and the jitter tag. An event credibility score is generated based on the time consistency value, the number of consecutive missing events, the number of time jitters, and the source tag. Construction events with credibility scores below the credibility threshold are marked as low credibility events. Delete low-reliability events, fill in the status gaps based on resource status changes within adjacent time windows for the same resource number, and reorder them according to event occurrence time, resource number, and construction task number to generate a reliable construction event flow.

5. The method for dynamic scheduling and optimization of construction resources based on big data analysis according to claim 1, characterized in that, The generation of the construction resource friction propagation map and construction disturbance breakpoint data includes: Read the event occurrence time, resource number, construction task number, work surface number, material batch number, transport vehicle number, event type and event associated location from the trusted construction event stream, and extract resource occupation events, resource release events, resource movement events, material arrival events, transportation delay events, work surface closure events and process handover events; Based on resource movement events, resource release events, material arrival events, transportation delay events, work area closure events, and process handover events, calculate equipment movement friction, shift change friction, material transfer friction, transportation channel friction, work area congestion friction, and process handover friction respectively. Set the construction task number, resource number, work surface number, material batch number, transport vehicle number, and process handover number as graph nodes, and set the resource occupation relationship, resource release relationship, resource movement relationship, material supply relationship, transportation arrival relationship, work surface closure relationship, and process handover relationship as graph edges, and write the corresponding resource friction into the graph edge attributes to generate a construction resource friction propagation graph; Mark the locations where resource flow is obstructed according to the time sequence of occurrence of resource friction under the same construction task number, and generate construction disturbance breakpoint data.

6. The method for dynamic scheduling and optimization of construction resources based on big data analysis according to claim 1, characterized in that, The generation of the scheduled executable prediction vector includes: An improved SCINet model is constructed, which includes a task resource work surface displacement encoder, a process resource interactive convolutional block, and a process breakpoint memory unit. Read the event occurrence time, resource number, construction task number, work surface number, material batch number, transport vehicle number and event type from the trusted construction event stream, and generate a construction event input sequence according to the construction task number and event occurrence time; The task resource work surface displacement encoder reads the construction event input sequence and the construction resource friction propagation map, extracts the resource number corresponding to the current position, the construction task number corresponding to the target work surface number, the resource movement relationship corresponding to the movement duration, and the resource friction in the map edge attributes, and generates the task resource work surface displacement coding sequence. The process resource interaction convolutional block establishes interaction channels according to the construction task number, resource number, work surface number, material batch number, and transport vehicle number. It performs convolutional interaction on resource occupation relationship, resource release relationship, material supply relationship, transportation arrival relationship, work surface closure relationship, and process handover relationship to generate process resource interaction feature sequence. The process breakpoint memory unit reads the construction disturbance breakpoint data and the process resource interaction feature sequence, and writes the location of the resource flow obstruction, construction task number, resource number, work surface number and occurrence time into the breakpoint memory slot to generate process breakpoint memory features. Based on the process resource interaction feature sequence and process breakpoint memory features, the task start-up intensity, resource reachability intensity, material support intensity, work surface acceptance intensity and task release time are generated, and combined to generate a scheduling executable prediction vector.

7. The method for dynamic scheduling and optimization of construction resources based on big data analysis according to claim 1, characterized in that, The set of individual construction scheduling parameters includes: Read the task start-up intensity, resource reachability intensity, material support intensity, work surface acceptance intensity and task release time from the scheduled executable prediction vector; read the construction task number, planned start time, planned completion time and resource requirement from the construction task data. Read the graph node and graph edge attributes in the construction resource friction propagation graph, and extract the allocable work teams, allocable machinery, supplyable material batches, arrangeable transport vehicles, and acceptable work surfaces; Discrete and continuous coding segments are established according to the construction task number. The task sequence, work group number, machinery number, material batch number, transport vehicle number and work surface number are written into the discrete coding segment. The task start time, task completion time, resource allocation time and material arrival time are written into the continuous coding segment to generate an individual set of construction scheduling parameters.

8. The method for dynamic scheduling and optimization of construction resources based on big data analysis according to claim 1, characterized in that, The generated scheduling and backfeed prediction results include: Read the discrete and continuous coded segments from the individual set of construction scheduling parameters, and read the task start-up intensity, resource reachability intensity, material support intensity, work surface acceptance intensity, and task release time from the scheduling executable prediction vector; The walrus optimization algorithm forms a disturbance migration direction based on the task start-up intensity, resource reachability intensity, material support intensity, and work surface acceptance intensity. It migrates the team number, machine number, material batch number, transport vehicle number, and work surface number in the discrete coding segment to the corresponding node of the graph edge with a smaller resource friction value in the construction resource friction propagation graph. It also adjusts the task start time, resource allocation time, and material arrival time in the continuous coding segment to generate individual migration scheduling parameters. The migration scheduling parameter individuals are clustered according to the work surface number and construction task number. Based on the task start-up intensity and resource reach intensity under the same work surface number, a leading scheduling parameter individual is selected. The task sequence, resource allocation time and material arrival time in the leading scheduling parameter individual are then transferred to other migration scheduling parameter individuals in the same cluster to generate navigation scheduling parameter individuals. The navigation scheduling parameter individuals are read from back to front according to the task completion time in the continuous coding segment. The pruning position is determined in ascending order of task startability intensity, resource availability intensity, material support intensity, and work surface acceptance intensity. The task sequence and resource allocation content corresponding to the pruning position are deleted from the navigation scheduling parameter individuals. Based on the task release time, the construction task number, team number, machinery number, material batch number, transport vehicle number, work surface number, and task start time are re-inserted to generate an updated scheduling parameter individual. The updated scheduling parameters are restored to candidate task start events, candidate resource occupancy events, candidate resource release events, candidate material arrival events, candidate transportation arrival events, candidate work surface occupancy events, and candidate process handover events. A candidate construction event reinjection sequence is generated and input into the improved SCINet model to generate scheduling reinjection prediction results.

9. The method for dynamic scheduling and optimization of construction resources based on big data analysis according to claim 1, characterized in that, The generated target construction resource scheduling scheme includes: Read the task start-up intensity, resource availability intensity, material support intensity, work surface acceptance intensity, and task release time from the scheduling and recharge prediction results, and read the construction task number, work group number, machinery number, material batch number, transport vehicle number, work surface number, task start time, and task completion time from the individual updated scheduling parameters. Based on the task's workability intensity, resource availability intensity, material support intensity, and work surface acceptance intensity, generate resource unavailability markers, material unavailable markers, and work surface unacceptability markers, and then filter and update the scheduling parameters of individuals, marking the individuals with updated scheduling parameters as eliminated individuals. Extract failed tasks, failed resources, failed work surfaces, and failed time windows from the eliminated individuals, and write the failed tasks, failed resources, failed work surfaces, failed time windows, unreachable resource markers, unavailable material markers, and unacceptable work surface markers into the process breakpoint memory unit; Based on safety management and environmental monitoring data, hard constraint filtering is performed on retained individuals. For construction task numbers that have blocked work surfaces, unreachable resources, and unavailable materials after filtering, local scheduling and repair are performed to generate individuals with repair scheduling parameters. The repaired scheduling parameters are restored to individual task scheduling results, personnel allocation results, machinery allocation results, material supply results, transportation arrangement results, and work area arrangement results, generating a target construction resource scheduling plan.

Images