Engineering cost project cost control strategy optimization method and system based on reinforcement learning

Abstract

The invention provides an engineering cost project cost control strategy optimization method and system based on reinforcement learning, and relates to the technical field of engineering cost, and the method comprises the steps: 2, executing resource distribution and cost control at a current construction stage according to an initial cost control strategy, and obtaining the actual cost, progress and quality data; step 3, integrating actual cost, progress and quality data, constructing a multi-dimensional data distribution model, and forming an abstract decision space; fitting the discrete data into a continuous decision curved surface, calculating the curvature of a local area, and analyzing the structure change characteristics of the decision curved surface; and carrying out region division in the abstract decision space based on the structure change features, mapping original data to corresponding partitions, extracting features in each partition, and generating strategy correction parameters. According to the invention, dynamic adaptive optimization of the engineering cost project cost control strategy is realized, and the accuracy of engineering project cost control is improved.

Description

Technical Field

[0001] This invention relates to the field of engineering cost technology, and in particular to a method and system for optimizing cost control strategies for engineering cost projects based on reinforcement learning. Background Technology

[0002] In the field of engineering project cost control, traditional management methods mostly rely on static budget preparation and phased cost accounting, which makes it difficult to respond in a timely manner to the dynamic changes in the environment during construction. Especially when faced with uncertainties such as fluctuations in building material market prices, frequent design changes, and project delays, existing methods have some limitations, leading to deviations between cost control and the actual construction status.

[0003] For example, in the construction project of a certain urban rail transit station, a detailed cost control plan was initially formulated based on historical data. However, when construction entered the main structure stage, the prices of major building materials continued to rise, and due to geological conditions being more complex than expected, some engineering designs needed to be changed, resulting in local delays. At this time, traditional cost control methods mostly rely on manual experience to adjust the budget, which can only make ex-post corrections to single cost indicators and is difficult to coordinate the dynamic balance between cost, schedule, and quality simultaneously. When increasing manpower to catch up with the schedule, quality risks and resource allocation efficiency may be ignored, which may lead to higher rectification costs in later stages. This adjustment method based on discrete events and manual judgment lacks continuous analysis and overall optimization of multi-dimensional construction data, which can easily lead to the one-sidedness and lag of control strategies, thereby affecting the achievement of the overall cost target of the project. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for optimizing cost control strategies for engineering cost projects based on reinforcement learning, so as to realize dynamic adaptive optimization of cost control strategies for engineering cost projects and improve the accuracy of cost control for engineering projects.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a method for optimizing cost control strategies in engineering cost projects based on reinforcement learning, the method comprising: Step 1: Obtain the initial budget, real-time construction environment status, and past cost control records for the project to build an initial cost control strategy; Step 2: Based on the initial cost control strategy, implement resource allocation and cost control in the current construction phase to obtain actual cost, schedule, and quality data; Step 3: Integrate actual cost, schedule and quality data to construct a multi-dimensional data distribution model and form an abstract decision space; fit discrete data into a continuous decision surface, calculate the curvature of local regions, and analyze the structural change characteristics of the decision surface; divide regions within the abstract decision space based on structural change characteristics, map the original data to the corresponding partitions, extract the inherent characteristics of each partition, and generate strategy correction parameters. Step 4: Based on the strategy correction parameters, and combined with actual cost, schedule and quality data, normalize the multi-dimensional data points and map them to a unit hypersphere; quantify the difference by calculating the shortest arc length between data points, and perform difference analysis and multi-objective collaborative evaluation to obtain the comprehensive cost control evaluation result; Step 5: Based on the comprehensive cost control evaluation results and strategy correction parameters, optimize the current strategy using a reinforcement learning model to generate an optimized cost control strategy. Step 6: Apply the optimized cost control strategy to the next construction phase and monitor material price fluctuations, design changes, and construction delays in real time to achieve dynamic optimization control.

[0006] Secondly, a reinforcement learning-based system for optimizing cost control strategies in engineering cost projects includes: The module is used to obtain the initial budget, real-time construction environment status and past cost control records of the project, and to build the initial cost control strategy. The execution module is used to perform resource allocation and cost control in the current construction phase according to the initial cost control strategy, and to obtain actual cost, schedule and quality data; The analysis module integrates actual cost, schedule, and quality data to construct a multi-dimensional data distribution model and form an abstract decision space. It fits discrete data into a continuous decision surface, calculates the curvature of local regions, and analyzes the structural change characteristics of the decision surface. Based on the structural change characteristics, it divides the region within the abstract decision space, maps the original data to the corresponding partition, extracts the intrinsic characteristics of each partition, and generates strategy correction parameters. The evaluation module is used to normalize and map multidimensional data points to a unit hypersphere based on strategy correction parameters and actual cost, schedule and quality data; it quantifies the difference by calculating the shortest arc length between data points, and performs difference analysis and multi-objective collaborative evaluation to obtain a comprehensive cost control evaluation result. The optimization module is used to optimize the current strategy based on the comprehensive cost control evaluation results and strategy correction parameters, using a reinforcement learning model to generate an optimized cost control strategy. The feedback module is used to apply the optimized cost control strategy to the next construction phase and monitor building material price fluctuations, design changes and construction delays in real time to achieve dynamic optimization control.

[0007] The above-described solution of the present invention has at least the following beneficial effects: By integrating initial budgets, real-time construction environment conditions, and past cost control records to construct an initial cost control strategy, the initial strategy fully integrates historical experience with the current actual scenario, improving the strategy's relevance and operability, and reducing the risk of cost runaway in the initial stage. Secondly, resource allocation and cost control are executed simultaneously in the current construction phase, and actual cost, progress, and quality data are collected in real time to achieve real-time linkage between the control process and data feedback, ensuring that resource allocation always matches current construction needs and avoiding resource waste or insufficient supply. Thirdly, by integrating actual data to construct a multi-dimensional data distribution structure, discrete data is fitted into a continuous decision surface. Combined with curvature analysis and regional division, strategy correction parameters are extracted, providing a quantitative basis for strategy adjustment and realizing dynamic iterative optimization of the cost control strategy. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the method for optimizing engineering cost control strategies based on reinforcement learning, as provided in an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of a reinforcement learning-based engineering cost control strategy optimization system provided by an embodiment of the present invention. Detailed Implementation

[0010] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0011] like Figure 1 As shown, embodiments of the present invention propose a method for optimizing cost control strategies for engineering cost projects based on reinforcement learning. The method includes the following steps: Step 1: Obtain the initial budget, real-time construction environment status, and past cost control records for the project to build an initial cost control strategy; Step 2: Based on the initial cost control strategy, implement resource allocation and cost control in the current construction phase to obtain actual cost, schedule, and quality data; Step 3: Integrate actual cost, schedule and quality data to construct a multi-dimensional data distribution model and form an abstract decision space; fit discrete data into a continuous decision surface, calculate the curvature of local regions, and analyze the structural change characteristics of the decision surface; divide regions within the abstract decision space based on structural change characteristics, map the original data to the corresponding partitions, extract the inherent characteristics of each partition, and generate strategy correction parameters. Step 4: Based on the strategy correction parameters, and combined with actual cost, schedule and quality data, normalize the multi-dimensional data points and map them to a unit hypersphere; quantify the difference by calculating the shortest arc length between data points, and perform difference analysis and multi-objective collaborative evaluation to obtain the comprehensive cost control evaluation result; Step 5: Based on the comprehensive cost control evaluation results and strategy correction parameters, optimize the current strategy using a reinforcement learning model to generate an optimized cost control strategy. Step 6: Apply the optimized cost control strategy to the next construction phase and monitor material price fluctuations, design changes, and construction delays in real time to achieve dynamic optimization control.

[0012] In this embodiment of the invention, an initial cost control strategy is constructed by integrating the initial budget, real-time construction environment status, and past cost control records. This ensures that the initial strategy fully integrates historical experience with the current actual scenario, enhancing its relevance and operability, and reducing the risk of cost runaway in the initial stage. Secondly, resource allocation and cost control are executed simultaneously during the current construction phase, and actual cost, progress, and quality data are collected in real time. This achieves real-time linkage between the control process and data feedback, ensuring that resource allocation always matches current construction needs and avoiding resource waste or insufficient supply. Thirdly, by integrating actual data to construct a multi-dimensional data distribution structure, discrete data is fitted into a continuous decision surface. Combined with curvature analysis and regional division, strategy correction parameters are extracted, providing a quantitative basis for strategy adjustment and realizing dynamic iterative optimization of the cost control strategy.

[0013] In a preferred embodiment of the present invention, step 1 above, which involves obtaining the initial budget, real-time construction environment status, and past cost control records of the project, and constructing an initial cost control strategy, may include: In this embodiment of the invention, core budget-related data are extracted from the project's preliminary filing documents, bidding documents, winning bid notices, and draft construction organization designs. This includes, but is not limited to, detailed bid quotations, bill of quantities, itemized pricing tables, fee standard explanations and provisional sums, and provisional estimates for specialized engineering projects. The budget is then broken down by project section and item, determining the specific amounts for labor, materials, machinery, measures fees, and regulatory fees for each item, forming a hierarchical budget list, and marking the calculation basis for each item's budget. Deviations between the bidding control price in the bidding documents and the winning bid budget are compared to verify whether key processes have been omitted in the budget preparation. For items with excessively high provisional sums or abnormal unit prices, the budget is reviewed jointly by cost estimators and construction managers to correct budget deviations and form the final confirmed initial budget benchmark. A site survey is conducted by construction, technical, and cost estimators to record site conditions, existing facilities, and determine construction constraints. Real-time supplier quotations and building material market information are used to verify the budget. We obtained current labor unit prices, major building material market prices and supply cycles, and machinery rental unit prices from channels such as announcements from Taiwan and local housing and construction departments. Simultaneously, we collected information on weather warnings, local policy adjustments, and fluctuations in logistics and transportation costs. The collected information was categorized and recorded according to static conditions and dynamic variables, with update cycles marked. From the enterprise project management database, we screened completed / under-construction projects similar in type, scale, and process to extract core records, including cost deviation data for each item, resource consumption ledgers, cost control measure implementation records, and the achievement of final cost targets. We removed abnormal data caused by special unforeseen circumstances from historical projects, retaining only valuable routine records. We focused on identifying frequently occurring cost deviations in similar projects and the effectiveness of corresponding countermeasures. We summarized cost control patterns from historical projects, such as cost proportions at different construction stages, optimal resource allocation ratios, and common risk points, forming a historical data reference report to provide a basis for initial strategy formulation.

[0014] Based on the initial budget details and historical data, establish the basic allocation ratios for labor, materials, and machinery at each construction stage, and determine the upper limit for the usage of each resource and its arrival time to avoid resource idleness or shortage. Based on the initial budget, set cost warning lines and stop-loss lines for each sub-project, such as an 8% overrun of labor costs triggering a warning and a 15% overrun triggering a stop-loss. Clarify the response process after a warning and the emergency measures corresponding to the stop-loss line. Combining the real-time construction environment and historical experience, set the constraints on cost for progress and quality, such as the foundation construction period not exceeding 10% of the budget period and the concrete strength qualification rate needing to reach over 95%, and determine the priority adjustment principles when these three factors are unbalanced. For labor, determine the shift schedule and attendance calculation standards for each trade, and limit overtime frequency to control additional costs. For materials, establish a procurement and acceptance process. The plan includes: process and inventory management solutions; for machinery, optimizing shift scheduling to improve equipment utilization and reduce idle shift costs; establishing cost expenditure approval processes and authorizing at different levels based on amount; setting regular accounting cycles and standardizing accounting forms and data reporting paths; embedding simplified risk response plans into the strategy based on real-time construction environment variables and historical risk points, such as reserving a certain percentage of price adjustment reserves for fluctuations in building material prices; for site constraints, pre-setting temporary facility optimization plans; organizing cost, construction, technical, and supervision personnel to conduct strategy reviews to verify whether the strategy matches the initial budget, construction environment, and historical experience, focusing on reviewing the rationality of resource allocation, the scientific nature of early warning thresholds, and the feasibility of risk contingency plans; collecting feedback from all parties, iteratively adjusting the strategy content, and forming a final executable initial cost control strategy.

[0015] In a preferred embodiment of the present invention, step 2 above, which involves performing resource allocation and cost control in the current construction phase according to the initial cost control strategy to obtain actual cost, schedule, and quality data, may include: In this embodiment of the invention, step 220 involves obtaining the specific types, quantities, and time requirements of labor, materials, and machinery resources for the current construction stage based on the initial cost control strategy, and forming an initial resource allocation plan. Specifically, this includes: the construction manager, in conjunction with cost estimators and technicians, determining the detailed requirements of labor, materials, and machinery resources one by one and integrating them into a plan, based on the resource allocation benchmarks, budget amounts for each item, and construction schedule constraints in the initial cost control strategy, combined with the specific process requirements of the current construction stage. Regarding labor resources, the specific types are first determined, covering trades suitable for main structure construction and complex geological treatment, such as steelworkers, concrete workers, formwork workers, scaffolders, and foundation pit excavators; then the specific quantities are calculated, and the construction manager determines the quantities based on the current... The quantities of rebar tying, concrete pouring, and formwork erection for each stage were calculated. Based on the initial cost control strategy's daily labor workload standard, the required number of workers was calculated by multiplying the daily workload of each trade by the planned number of construction days. An additional 3 to 5 reserve personnel were also allocated to handle adjustments to the work process due to complex geological conditions. Finally, time requirements were determined, with the arrival times of each trade arranged according to the sequence of work processes and material delivery dates. For example, rebar workers needed to arrive one day before the arrival of rebar materials, and concrete workers needed to arrive on the day the formwork was erected, ensuring seamless transitions with subsequent processes. Regarding material resources, specific types were determined: HRB400E grade rebar suitable for the main structure of the rail transit system, C30 and C40 grade concrete, waterproof membrane, scaffolding steel pipes, etc. Calculation tools were used. When determining the quantity of materials, the cost estimator calculates it based on the current stage of the bill of quantities, using the formula: quantity of each material × (1 + material loss rate agreed in the initial cost control strategy). For example, the quantity of reinforcing steel is calculated as the quantity of reinforcing steel binding × (1 + reinforcing steel loss rate), and the quantity of concrete is calculated as the quantity of concrete pouring × (1 + concrete loss rate). Considering the risk of fluctuations in building material prices, a small amount of spare materials is reserved. Time requirements are determined in conjunction with construction progress milestones. For example, reinforcing steel materials need to arrive 3 days before the start of the reinforcing steel binding process, and concrete materials need to arrive in batches according to pouring times. The arrival time for each batch is communicated to the supplier 1 day in advance to avoid delays in construction due to material delays. Regarding machinery resources, specific types include tower cranes, concrete pump trucks, excavators, and reinforcing steel... Equipment such as cutting machines and welding machines are suitable for the main structure construction and foundation pit treatment. The quantity is calculated by the construction supervisor in conjunction with the machinery manager, based on the current construction intensity divided by the output of a single machine per shift. For example, the number of concrete pump trucks is calculated by dividing the daily concrete pouring volume by the daily output of a single concrete pump truck per shift, and the number of excavators is calculated by dividing the remaining excavation volume of the foundation pit by the daily output of a single excavator per shift. At the same time, an appropriate number of auxiliary machines are also required. The timing requirements are coordinated with the arrival of materials and the progress of the work. For example, excavators need to arrive on the same day as the foundation pit excavation work begins, and tower cranes need to be installed and debugged before the arrival of steel bars and formwork materials to ensure that resources are synchronized with the work process. The specific types, quantities, and timing requirements of the above-mentioned labor, materials, and machinery are compiled into a booklet to form an initial resource allocation plan.

[0016] Step 221: Based on the initial resource allocation plan, execute on-site resource scheduling and cost control, generating resource execution records including actual resource consumption and corresponding cost expenditures. Specifically, this includes: For labor scheduling, the labor administrator organizes personnel to their posts according to the job types, quantities, and arrival times determined in the plan. Daily attendance is checked, and daily labor costs are calculated based on the actual attendance multiplied by the daily wage standard for each job type. Unplanned personnel additions are strictly prohibited. In cases where complex geological conditions lead to rework of some procedures, temporary deployment of mobile personnel is only permitted after joint confirmation by the construction supervisor and cost specialist. The reason for temporary deployment, the number of personnel, and working hours are recorded, and the corresponding labor costs are calculated simultaneously. For material scheduling, the material administrator contacts suppliers according to the time and quantity agreed upon in the plan. Upon material arrival, the material administrator, together with quality inspection personnel, verifies the specifications, quantity, and quality certificates. After acceptance, the material procurement cost is calculated based on the actual quantity received multiplied by the purchase price. Daily statistics on actual material usage are compiled, calculated as: Daily usage = Daily arrival quantity - Daily inventory. The quantity is calculated based on the previous day's inventory, and material loss is recorded simultaneously. For any loss exceeding the initial strategy's agreed-upon rate, the cause must be investigated and confirmed by both the material manager and the cost specialist, and then included in the cost expenditure record. Regarding machinery scheduling, the machinery manager arranges machinery entry and debugging according to the plan, recording the actual number of machine shifts used daily. Machinery rental fees are calculated based on the actual number of shifts multiplied by the corresponding machine shift rental price. Simultaneously, machinery maintenance costs are tallied, summed as a single maintenance consumable cost plus labor service fees. The machinery scheduling sequence is optimized to avoid idle machinery. For situations where machinery is idle due to design changes, its use is adjusted promptly, and the idle time and cost losses are recorded. Regarding cost expenditure control, the cost specialist reviews each expenditure according to the initial budget and resource allocation plan. Expenditures under 50,000 yuan are paid after approval by the project manager; expenditures of 50,000 yuan or more require review and approval by the company's cost management department before payment. All expenditures must be documented with supporting documentation. The cost specialist summarizes the actual consumption of labor, materials, and machinery and the corresponding cost expenditures each day. A periodic summary is conducted weekly to form a detailed resource execution record. The record must include the resource type, planned quantity, actual quantity, difference quantity, corresponding cost, and reason for the difference. The record is then filed after being signed and confirmed by all parties.

[0017] Step 222: Based on resource execution records, collect real-time cost flow data, project progress completion data, and key construction quality indicator detection data during the construction process. Specifically, this includes: simultaneously collecting three types of core data based on resource execution records to ensure data timeliness and accuracy. Cost flow data collection is the responsibility of cost estimators, who liaise daily with finance personnel and resource administrators to collect labor wage payment slips, material purchase invoices, machinery rental agreements and payment vouchers, maintenance and repair expense slips, etc., categorizing them into individual cost items plus comprehensive expenses. Individual cost items are further categorized into actual labor costs + actual material costs + actual machinery costs. In summary, comprehensive costs include site management fees, temporary facility fees, etc., which are recorded item by item based on the actual amount incurred in the current period. Simultaneously, the cost expenditure data is compared with the resource execution records to verify consistency. For incomplete invoices or discrepancies in amounts, timely verification and correction are conducted with the relevant personnel to ensure that every cost transaction is traceable. The data collection of project progress completion is the responsibility of the construction statistician, who measures the actual completion status of each sub-item at the construction site daily. Combined with the construction progress plan corresponding to the initial resource allocation scheme, the daily and cumulative progress completion percentages are calculated by dividing the actual completed amount by the total amount of work for that sub-item. For example, for the rebar tying sub-item, the percentage is calculated as the actual length of rebar tied on that day divided by the total length of rebar tied in that stage. The overall length is calculated based on the progress, with concrete pouring progress calculated as the actual daily pouring volume divided by the total concrete pouring volume for that stage. For delays caused by complex geology or fluctuating building material prices, detailed records of the reasons for and duration of the delays are kept, along with annotations on their impact on subsequent procedures. Daily progress data is entered into the construction log and correlated with resource consumption data in the resource execution record. Data collection for key construction quality indicators is the responsibility of quality inspection personnel. Based on the construction specifications for rail transit engineering and the quality constraints in the initial cost control strategy, on-site testing of core indicators is conducted. For example, concrete strength is measured by randomly selecting three samples from each pouring batch and sending them to… Designated testing institutions will conduct tests. The thickness of the concrete cover for reinforcing bars will be measured on-site at 10% of the components in each batch, with 5 testing points selected for each group of components. The settlement of the foundation pit will be measured 3 times a day at regular intervals, with 5 monitoring points recorded each time. The test results will be recorded in a graded manner. Qualified data will be archived directly, while unqualified data will require the construction team to be notified immediately for rectification. At the same time, the rectification measures, rectification time, and re-inspection results after rectification will be recorded to ensure that the quality data corresponds synchronously with the construction progress and cost data, forming a complete data chain. The three types of data will be summarized and checked daily to ensure that the data logic is consistent and to avoid the disconnect between cost overruns and schedule delays, or schedules being ahead of schedule but quality not meeting standards.

[0018] Through on-site multi-position collaborative scheduling, real-time inspection and control, and simultaneous review of cost expenditures, dynamic control of resource consumption and cost expenditures can be achieved.

[0019] In a preferred embodiment of the present invention, step 3 above integrates actual cost, schedule, and quality data to construct a multi-dimensional data distribution model, forming an abstract decision space; fits discrete data into a continuous decision surface, calculates the curvature of local regions, and analyzes the structural change characteristics of the decision surface; based on the structural change characteristics, divides the abstract decision space into regions, maps the original data to the corresponding partitions, extracts the intrinsic features of each partition, and generates strategy correction parameters, which may include: In this embodiment of the invention, step 330 involves standardizing the actual cost data, schedule data, and quality data, integrating the standardized three-dimensional data into a unified format multi-dimensional vector, where each dimension corresponds to a quantitative indicator of cost, schedule, and quality. Specifically, this includes: using the multi-dimensional dataset of the main structure construction of rail transit stations collected in step 222 as the core object, operating according to the principles of dimensional processing and unified integration, and refining the processing throughout the entire process based on the actual project procedures and data characteristics; regarding cost data standardization, the actual cost flow data is broken down according to core procedures, determining the labor, material, and machinery costs for the three main procedures of rebar tying, formwork installation, and concrete pouring, and then further specifying the costs for each... For each itemized cost, a standardized value range is defined. Taking the cost of steel reinforcement materials in the steel reinforcement binding process as an example, firstly, all steel reinforcement procurement expenditure data collected during the current main structure construction phase is extracted. The minimum and maximum actual expenditure amounts in this itemized cost are verified and selected as the upper and lower limits for standardization. Subsequently, the steel reinforcement material costs are standardized batch by batch. The conversion logic is: the actual expenditure amount of a single steel reinforcement purchase minus the minimum actual expenditure amount of this item, and the result is divided by the difference between the maximum and minimum actual expenditure amounts of this item. If the single actual expenditure amount equals the minimum expenditure amount, the standardized value is 0; if it equals the maximum expenditure amount, the standardized value is 1; if it exceeds the maximum expenditure amount, it is still counted as 1; if it is lower... For the minimum expenditure, it is still calculated as 0. Similarly, the labor cost and machinery cost of the rebar tying process, and the cost of each item in the formwork installation and concrete pouring processes are standardized and converted separately. Then, the average value of the standardized cost of all items in each process is calculated. The calculation method is to add the standardized cost of each item in the process, divide the sum by the quantity of the item cost, and use this average value as the standardized cost dimension value of the corresponding process to ensure coverage of all item costs and avoid distortion of single item data. In terms of schedule data standardization, the schedule plan of each process set in the initial resource allocation plan is used as the benchmark. The daily planned completion quantity and stage planned completion quantity of each process of rebar tying, formwork installation and concrete pouring are broken down, and then the schedule of each process is defined. For the standardized value range, taking the concrete pouring process as an example, the minimum planned completion quantity is the amount of foundation pouring that must be completed each day, and the maximum planned completion quantity is the upper limit of the pouring quantity after the optimized schedule. Then, the standardized value of the progress is calculated daily. The calculation logic is to subtract the minimum planned completion quantity from the actual concrete pouring quantity of the day, and divide the result by the difference between the maximum planned completion quantity and the minimum planned completion quantity of the day. If the actual pouring quantity of the day exceeds the maximum planned completion quantity, the standardized value is counted as 1; if it is lower than the minimum planned completion quantity, the standardized value is counted as 0. At the same time, the reason for the progress delay is marked. The same logic is used to complete the standardized conversion for the formwork installation and rebar tying processes. The progress standardized value of each process is updated daily to ensure that it is in line with the real-time construction progress.Regarding the standardization of quality data, indicators are categorized into acceptable range indicators and deviation indicators, both based on construction specifications and design requirements. For acceptable range indicators, taking the tensile strength of reinforcing steel as an example, the design requirement for an acceptable range is 400-500 MPa, which serves as the upper and lower limits for standardization. The calculation logic is: subtract 400 MPa from the actual measured value of a single sample, and divide the result by the difference between 500 MPa and 400 MPa. If the measured value is lower than 400 MPa (unacceptable), the standardized value is 0; if it is higher than 500 MPa (better than the standard), the standardized value is 1. For deviation indicators, taking the thickness of the reinforcing steel protective layer as an example, the design allowable deviation range is ±5 mm, meaning the minimum allowable deviation is -5 mm and the maximum allowable deviation is 5 mm. The calculation logic is: 1 minus (actual deviation)... The absolute value of the difference is subtracted from the absolute value of the minimum allowable deviation, and the result is divided by the difference between the absolute values ​​of the maximum and minimum allowable deviations. The closer the actual deviation value is to 0, the closer the standardized value is to 1. If the actual deviation value exceeds the allowable range, the standardized value is counted as 0. The quality standardized value for each process is the average of the standardized values ​​of all test indicators for that process. After standardizing the three types of data, a three-dimensional multi-dimensional vector is generated in the format of one vector per day for each process. The three dimensions of each vector correspond to the cost standardized value, schedule standardized value, and quality standardized value for that process on that day. At the same time, the vector association information is labeled, including the construction date, specific process location, corresponding resource execution record number, and whether there are any abnormal factors, to ensure that each vector can be traced back to the original construction data and site conditions.

[0020] Step 331 involves mapping the multidimensional vectors to a predefined abstract decision space, forming a set of data points representing the project status. The distribution pattern of these data points within the decision space is then analyzed to construct a multidimensional data distribution structure reflecting the dynamic relationships between cost, schedule, and quality. Specifically, this includes: first, determining that the predefined abstract decision space is a three-dimensional space. The three coordinate axes of this space correspond to the standardized cost dimension, standardized schedule dimension, and standardized quality dimension obtained in step 330, respectively. The value range of each coordinate axis is set to 0 to 1, perfectly matching the range of standardized data values. Any point within the space uniquely corresponds to a standardized combination of cost, schedule, and quality states, which can be intuitively represented. The project control status of a specific process and time period is identified. Then, all the multi-dimensional vectors generated in step 330 are mapped one by one to this three-dimensional space. The mapping is performed in order according to the process location and date labeled on the vectors. Each vector corresponds to an independent data point in the space. Simultaneously, complete association information is added to each data point, including the corresponding construction period, core process and location, resource consumption deviation, cost deviation, schedule deviation, quality inspection results, and details of abnormal factors. After mapping, distribution pattern analysis is conducted. First, a method of area-by-area investigation and time-segmented comparison is used to identify the clustered and discrete areas of the data points. When investigating clustered areas, they are divided according to the coordinate axis value range into 0-0.2 and 0.2-0. The five intervals 0.4, 0.4-0.6, 0.6-0.8, and 0.8-1.0 are used to examine the distribution of data points in combinations of these intervals. For example, check for clusters of standardized cost values ​​(0.3-0.5), standardized schedule values ​​(0.7-0.9), and standardized quality values ​​(0.8-1.0). These clusters represent a controllable cost, ahead-of-time schedule, and excellent quality management. The process characteristics corresponding to the data points within these clusters are also recorded. When examining discrete data points, focus on independent data points that deviate from all clustered areas. Verify their correlation information one by one and analyze the reasons for the dispersion. For example, a data point with a standardized cost value of 0.85 (high cost) and a standardized schedule value of 0.35 (slow schedule). A quality standardization value of 0.9 (quality qualified) corresponds to a review of resource execution records and on-site construction logs to confirm whether a significant increase in building material prices led to cost overruns in steel reinforcement procurement, thereby affecting construction progress. The type and degree of abnormal factors corresponding to this discrete point are also noted. Subsequently, for clustered areas and discrete data points, the dynamic correlation between cost, schedule, and quality is analyzed in depth. For clustered areas, the changing trends of the three factors for data points within that area are statistically analyzed. For example, when the cost standardization value increases from 0.3 to 0.5, does the schedule standardization value increase from 0.7 to 0.9, indicating that increased machinery and labor input leads to increased costs while accelerating construction progress, and does the quality standardization value remain at 0?For discrete data points, the analysis examines the interference of anomalies on the correlation between the three elements. For example, when complex geological conditions lead to design changes, the corresponding data points often exhibit characteristics of increased cost standardization values, decreased schedule standardization values, and maintained quality standardization values. This means that design changes require additional labor and materials, resulting in cost overruns and schedule delays, but quality is strictly controlled according to the new design requirements, and no problems arise. In conjunction with the core needs of the main structure construction of rail transit stations, the analysis focuses on the impact patterns of three key anomalies—fluctuations in building material prices, changes in geological conditions, and design changes—on the correlation between the three elements, forming a multi-dimensional data distribution structure. This structure is presented in written form, defining the coordinate range of each cluster area, the corresponding correlation pattern of the three elements, the applicable construction scenario, the coordinate location of discrete data points, the reasons for dispersion, the impact paths of anomalies, and the core patterns of the three element offsets caused by different anomalies. This ensures that the structure reflects the on-site construction status and the synergistic relationship between the three elements.

[0021] Step 332: Based on the multi-dimensional data distribution structure, and utilizing the statistical distribution characteristics of the data point set, an abstract decision space defined by the multi-dimensional data distribution structure is ultimately generated. Specifically, this includes: first, systematically extracting the statistical distribution characteristics of the data point set; second, classifying and processing data points according to clustered regions and discrete data points based on the multi-dimensional data distribution structure throughout the process; third, for each clustered region, first counting the total number of data points within the region, and then calculating the standardized average cost, standardized average schedule, and standardized average quality of all data points within the region. The calculation method involves summing the standardized values ​​of all data points in the corresponding dimension within the region. Divide by the total number of data points in the region; simultaneously calculate the fluctuation range of the standardized values ​​for each dimension, i.e., the maximum value minus the minimum value of the corresponding dimension's standardized value in the region, and record the reasons for the fluctuations; for example, in the cost-schedule and quality balanced cluster region, with a total of 50 data points, the cost standardized average is 0.42, with a fluctuation range of 0.08 (maximum value 0.47, minimum value 0.39), the schedule standardized average is 0.78, with a fluctuation range of 0.12 (maximum value 0.85, minimum value 0.73), and the quality standardized average is 0.86, with a fluctuation range of 0.07 (maximum value 0.90, minimum value 0.83), and the reasons for the fluctuations are as follows: Since all differences are normal process connections and there are no abnormal factors interfering; for discrete data points, their relative distances to each cluster area are statistically analyzed. For example, if a discrete point is closest to the balanced cluster area, only the cost dimension value deviates from the upper limit of the area by 0.3, while the progress and quality dimension values ​​are within the range of the area. At the same time, the number and deviation of discrete data points caused by different abnormal factors are statistically analyzed. Subsequently, in combination with the control constraints of the construction of the main structure of the rail transit station, each cluster area is screened and optimized, and cluster areas that exceed the constraints are eliminated, such as areas where the cost standardization average value exceeds 0.7 and areas where the quality standardization average value is lower than 0.6. Retain feasible clustering areas that meet management and control requirements, and determine the core adaptation scenarios for each feasible area. For example, balanced clustering areas are adapted to regular construction periods without abnormal factors, while progress-priority clustering areas are adapted to construction periods where schedules need to be caught up. At the same time, incorporate the abnormal factors, offset patterns, and response boundaries corresponding to discrete data points into the decision space definition, and determine the allowable offset range of data points when different abnormal factors occur. For example, if the cost dimension value offset due to the rise in building material prices does not exceed 0.2, it can still be reversed by optimizing resource allocation. If it exceeds 0.2, the initial cost strategy and the control direction after the offset need to be adjusted. If the offset of the progress dimension value is less than 0.3. The catch-up can be achieved by extending the operation time without incurring additional costs. Finally, by integrating the statistical characteristics of feasible cluster areas, rules for handling abnormal factors, and the correlation patterns of the three elements, a formal abstract decision space is generated. This decision space is still a three-dimensional space, with feasible cluster areas as the core feasible decision domain, and the statistical characteristics and applicable scenarios of each area are labeled. The offset range of discrete data points is used as the anomaly warning boundary, and the anomaly types and response directions corresponding to different boundaries are labeled. Simultaneously, the interpretation rules for any data point within the space are determined: if a data point is within a feasible area, it represents a normal control status; if it exceeds the feasible area but does not reach the warning boundary, it represents a slight deviation requiring fine-tuning of resource allocation; if it exceeds the warning boundary, it represents a serious deviation requiring strategy correction. This decision space is entirely defined by a multi-dimensional data distribution structure, and all rules and features originate from actual on-site construction data, adapting to the dynamic control needs of the main structure construction of rail transit stations.

[0022] Step 333: Based on the discrete data points of cost, schedule, and quality indicators distributed within the abstract decision space, a surface fitting algorithm is used to fit the discrete data points, forming a continuous decision surface. Specifically, this includes: using the abstract decision space generated in step 332 as a carrier, locking all discrete data points, including routine construction data points and abnormal data points caused by fluctuations in building material prices, geological changes, or design modifications; using a cubic spline interpolation algorithm for fitting, ensuring a complete match with the process characteristics and data features of the main structure construction of rail transit stations; firstly, data point preprocessing is performed, classifying all discrete data points by construction process and construction time period to form subdivided data groups, such as the rebar tying process - daily morning - routine construction group, and the concrete pouring process - daily afternoon - building material price fluctuation impact group, ensuring the fitting results accurately match different construction scenarios; extreme abnormal data points are screened, defined as data points that deviate excessively from the mean of the three-dimensional standardized values ​​of most data points in the same group, and these data points are directly removed to avoid interfering with the fitting accuracy; finally, weights are assigned to the retained data points, with routine construction data points uniformly weighted at 1, and data points affected by abnormal factors also weighted at 1.2. The weighting is based on the fact that abnormal scenarios have a more significant impact on cost control, and their contribution to the fitted surface needs to be increased to ensure that the surface can truly reflect the correlation trend of the three elements of abnormal scenarios. Subsequently, the fitting is performed segment by segment according to the logic of cubic spline interpolation algorithm. Taking the construction progress time axis of each process as the clue, the discrete data points in the same subdivided data group are arranged in chronological order. An independent fitting curve is constructed between two adjacent data points. During fitting, it is necessary to ensure that each curve segment accurately passes through the two endpoint data points, and the tangent slope of the two adjacent curve segments at the connection point is consistent. That is, first determine the slope of the previous curve segment at the connection point, and then adjust the initial slope of the next curve segment to make it exactly the same as the slope of the previous segment, so as to avoid discontinuities such as breaks and sharp corners in the surface. For example, in the conventional group of rebar binding process, the discrete data points of two consecutive mornings of rebar binding of No. 1 foundation pit are taken to construct a smooth curve between the two points. The values ​​of each point in the middle of the curve gradually change according to the values ​​of the two endpoints, and the slope is flat. After establishing the principle of a smooth transition, the segmented curves for all time periods and locations within the process are pieced together one by one to form process-level fitted surface segments. Next, surface integration and verification are carried out, integrating the fitted surface segments from the three major processes of rebar tying, formwork installation, and concrete pouring. Secondary fitting adjustments are made to the data points at the process transitions to ensure a smooth transition between different process surface segments without obvious protrusions or depressions. After fitting, 10% of the original data points are randomly sampled, and the differences between the actual coordinates of the data points and the coordinates of the corresponding surface positions are compared. If the differences are too large, the weights of the corresponding abnormal data points are readjusted, and fitting is performed again until the deviations of all sampled data points from the surface are within a reasonable range. Finally, a continuous and smooth decision surface covering the entire abstract decision space is formed. This surface can completely replicate the dynamic correlation trends of cost, schedule, and quality as the construction progresses and the scenario changes, clearly reflecting the degree of influence of different abnormal factors on the correlation relationships of the three factors.

[0023] Step 334: Calculate the geometric curvature of each local location on the decision surface, and identify the structurally abrupt and gradual curvature regions based on the numerical distribution of the geometric curvature. Specifically, this includes the following steps: delineating local regions point by point, manually calculating curvature, and statistically distributing the curvature by region. First, delineate the local regions. Taking any data point on the decision surface as the center, select 10 adjacent data points based on the principle of closest spatial distance to form a local region. Each local region corresponds to a small construction scene segment within the abstract decision space. The selection of 10 adjacent points is based on the principle of covering the associated data around the central data point while avoiding the distortion of curvature calculation due to an excessively large region. Subsequently, the geometric curvature was calculated region by region. The first step was to determine the direction of the normal vector of the central data point. Using the spatial coordinates of the central data point and its 10 neighboring data points, the direction of each connecting line was manually marked. This, combined with the concavity / convexity trend of the decision surface at the central data point, determined the normal vector (the direction perpendicular to the tangent plane of the surface). The second step was to calculate the angle between each connecting line and the normal vector. The 10 angle values ​​were added together to obtain the total angle, which was then divided by 10 (the number of neighboring data points) to obtain the average angle. The third step was to calculate the distance between the 10 neighboring data points. The distance was calculated between every two neighboring points, resulting in 9 sets of distance values. These 9 sets of distance values ​​were then added together to obtain the total distance. The fourth step is to calculate the curvature value. This is done by dividing the average angle by the average distance. A larger average angle and a smaller average distance result in a larger curvature value, indicating more pronounced surface undulations in the local area and more significant changes in the relationship between the three elements in the construction scenario. Conversely, a smaller average angle and a larger average distance result in a smaller curvature value, indicating a smoother surface in the local area and a more stable relationship between the three elements in the construction scenario. After calculating the curvature of all local areas, all curvature values ​​are arranged in ascending order to define the curvature distribution range. First, the range where most curvature values ​​are concentrated is defined as the smooth range. The curvature within this range... The value fluctuation range is narrow, and the curvature value difference between adjacent areas is small, corresponding to a conventional construction scenario where there is no abnormal interference and the three elements change in a fixed proportion. Then, areas where the curvature value suddenly exceeds the upper limit of the smooth range and the curvature value difference with the adjacent areas is identified as structural mutation areas. These areas correspond to scenarios where the relationship between the three elements changes drastically, such as a sharp increase in building material prices leading to a sudden increase in costs while the progress remains stable, or a sudden change in geological conditions leading to fluctuations in both costs and progress. At the same time, the correlation information of the original data points corresponding to each structural mutation area is checked one by one, and the types of abnormal factors and the impact scenarios corresponding to the area are marked, forming a correspondence ledger of local area-curvature value-abnormal factor.

[0024] Step 335: Based on the natural boundaries between structurally abrupt regions and smooth regions, the abstract decision space is divided into several partitions with different structural characteristics. Specifically, this includes: using the curvature distribution ledger formed in Step 334 as the core basis, accurately identifying natural boundaries, tracking the transition process of curvature values ​​from smooth intervals to abrupt intervals or from abrupt intervals to smooth intervals point by point, defining the region where curvature values ​​change continuously and are located between smooth and abrupt intervals as the transition zone, and the midline of the transition zone as the natural boundary of the partition. The boundary delineation must ensure that it conforms to the surface structure change law, does not sever data points under the same correlation mode, and does not merge abrupt regions caused by different abnormal factors. Subsequently, following the principle of independent partitioning of smooth areas and partitioning of abrupt change areas according to anomaly type, the abstract decision space was divided into several dedicated partitions. Each partition clearly labels its structural characteristics, applicable construction scenarios, corresponding anomaly factors, and boundary range, represented by three-dimensional standardized value ranges. The specific partitions are as follows: First, the cost-schedule-quality equilibrium zone, corresponding to a smooth area with stable curvature values. The boundary range is where the standardized values ​​of cost, schedule, and quality are all in the middle range, and the curvature values ​​are within the smooth range. This zone is suitable for conventional construction scenarios without anomalies. Its structural characteristics include smooth curvature, stable correlation between the three elements, moderate cost increase, synchronized schedule improvement, and excellent quality maintenance. Second, the cost-sensitive zone... - The stable progress zone corresponds to the structural abrupt change region caused by fluctuations in building material prices. Its boundary range is characterized by a wide fluctuation range for standardized cost values ​​and a narrow fluctuation range for standardized progress and quality values, with curvature values ​​in the abrupt change range. This zone is suitable for scenarios involving rising / falling building material prices. Its structural characteristics are that the surface fluctuates dramatically in the cost dimension and smoothly in the progress and quality dimensions, with significant cost fluctuations and relatively stable progress and quality. The third zone is the cost-progress dual fluctuation zone, corresponding to the structural abrupt change region caused by changes in geological conditions. Its boundary range is characterized by a wide fluctuation range for standardized cost and progress values ​​and a narrow fluctuation range for standardized quality values, with curvature values ​​in the abrupt change range. This zone is suitable for geologically complex scenarios. Its structural characteristics are that the surface fluctuates dramatically in both the cost and progress dimensions. The four zones are: 1) a drastic fluctuation in the degree of change, 2) a smooth quality dimension, 3) cost overruns, 4) a full-dimensional adjustment zone, 5) a structural mutation zone caused by design changes, 6) a boundary range where the standardized values ​​of cost, schedule, and quality all fluctuate to a certain extent, the curvature value is in a mutation range, 7) a scenario for adapting to design scheme adjustments, 8) a structural feature of full-dimensional surface fluctuations, re-adaptation of the three elements' relationship, 9) cost increases, schedule rearrangement, and quality standard adjustments; 9) after partitioning, a written abstract decision space partitioning diagram is drawn, marking the boundary coordinates, structural features, adaptation scenarios, and abnormal factors of each partition, and determining the connection rules between each partition to ensure that there are no omissions, overlaps, or mismatches in the partitions.

[0025] Step 336: Map the original data points in the multidimensional vector form of the abstract decision space to the corresponding partitions one by one, based on the spatial coordinates of each data point. Specifically, this includes: first, extracting the spatial coordinates of all original data points in the multidimensional vector form, i.e., the cost-standardized value, schedule-standardized value, and quality-standardized value corresponding to each data point, registering them one by one, and marking the association information and corresponding resource execution record number for each data point; then, verifying the coordinate assignment point by point, comparing the three-dimensional standardized value of a single data point with the boundary range of each partition marked on the partition diagram to determine which partition the data point falls within; for example, a certain original data point corresponds to the conventional construction of the No. 1 foundation pit rebar binding process, with a cost-standardized value of 0.42, a schedule-standardized value of 0.76, and a quality-standardized value of 0.87. According to the partition diagram, the three-dimensional coordinates all fall within the boundary range of the cost-schedule-quality equilibrium zone, i.e., mapping the data point to this partition and marking the partition name in the ledger; a certain original data point corresponds to the No. 2 beam concrete pouring process, which is affected by rising building material prices... Cost overruns occurred, with standardized cost values ​​of 0.83, standardized schedule values ​​of 0.72, and standardized quality values ​​of 0.85. The coordinates fall within the cost-sensitive, schedule-stable zone, and the data is mapped to that zone. The anomaly is identified as fluctuations in building material prices. For data points falling on the zone boundaries, the principle of proximity and data density is applied. First, it's determined which zone's center the boundary point is closer to. Then, the data point density between two adjacent zones is calculated, and the boundary point is assigned to the closer zone with the denser data points. The boundary point's attributes and the basis for assignment are noted in the ledger to avoid ambiguity. During the mapping process, one data point is randomly checked after every 10 mappings to ensure error-free mapping. After all mappings are completed, a comprehensive review is conducted: the number of data points is counted by zone, and the associated scenarios of each zone's data points are checked to ensure consistency with the zone's adaptation scenarios. Data points with mapping errors are corrected one by one, and missing association information is supplemented. Finally, a complete ledger is formed, consisting of zone name, data point list, association information, and resource execution record number, with each data point within a zone corresponding to a specific construction scenario.

[0026] Step 337 involves analyzing all data points mapped within each partition to identify the correlation patterns between cost, schedule, and quality reflected in the data points, and extracting unique statistical characteristics and dominant change patterns within each partition. Specifically, this includes: first, analyzing the cost-schedule-quality equilibrium zone, statistically analyzing the standardized cost values ​​of all data points within that partition, then statistically analyzing the standardized schedule and quality values, and calculating the average value (the sum of all data point values ​​in the same dimension divided by the total number of data points in that partition) and fluctuation range (the maximum value minus the minimum value in that dimension). By analyzing the synchronous change trends of the three dimensions point by point, the correlation pattern is identified as moderate cost growth, synchronous schedule improvement, and stable and excellent quality. This means that through reasonable allocation of human and mechanical resources, costs increase slightly according to the initial plan, schedule progresses according to nodes or even slightly ahead of schedule, and quality is consistently maintained at or above the acceptable level without any abnormal factors interfering. The extracted statistical characteristics are that the standardized values ​​of the three dimensions are concentrated in the middle range, with narrow fluctuation ranges, high data point clustering, and no obvious discrete points. The dominant change pattern is that the standardized cost value increases by a small interval (e.g., 0.01-0.01).02) The progress standardization value increased synchronously within the corresponding small range, while the quality standardization value remained basically within a fixed range without significant fluctuations, and the trend was consistent with the initial resource allocation plan. Subsequently, the cost-sensitive-progress-stable zone was analyzed, focusing on cost fluctuations. Combined with the data points related to building material price fluctuations, the correlation pattern was identified as drastic cost fluctuations and relatively stable progress and quality dimensions. That is, rising building material prices led to cost overruns, resulting in a significant increase in the cost standardization value, or falling prices led to cost savings, resulting in a significant decrease in the cost standardization value. However, by optimizing manual scheduling and improving mechanical operation efficiency, the progress was not significantly affected, and quality was strictly managed according to construction specifications. Controlled and without fluctuations; the extracted statistical characteristics are that the standardized value of cost has a wide fluctuation range and a significant difference between the maximum and minimum values, while the fluctuation range of schedule and quality dimensions is narrow and the values ​​are concentrated; the dominant change pattern is that the greater the increase in building material prices, the more significant the increase in the standardized value of cost, while the values ​​of schedule and quality dimensions remain within a fixed range, unaffected by cost fluctuations, and cost fluctuations can be offset by cost savings in subsequent processes; further analysis of the cost-schedule dual fluctuation zone reveals the linkage between cost and schedule changes, combined with details of geological condition changes, identifying the correlation pattern as rising costs, delayed schedules, and acceptable quality, i.e., complex geological conditions increase construction difficulty, requiring additional manpower. Supplementary geological surveys and adjustments to procedures, along with the additional procurement of waterproofing and reinforcement materials, led to cost overruns and delays in process connections and schedule. However, quality control was strengthened to meet the requirements of the new geological conditions, maintaining a consistently acceptable level. The extracted statistical characteristics show strong synchronicity in the fluctuations of standardized values ​​for cost and schedule, both exhibiting inverse changes, while the standardized value for quality remained stable within the acceptable range. The dominant trend was that the higher the geological complexity, the greater the increase in standardized cost and the greater the decrease in standardized schedule, while the quality value consistently remained above the acceptable level with minimal fluctuations. Finally, the analysis of the full-dimensional adjustment area revealed a three-dimensional linkage, and by combining this with design change details, the correlation pattern was identified as cost. The schedule and quality were all adaptively adjusted. This means that after a design change, additional labor and material inputs were required, leading to increased costs. The schedule needed to be replanned, and quality standards were adjusted according to the new design requirements, resulting in changes across all three dimensions. The extracted statistical characteristics showed that the standardized values ​​of the three dimensions all fluctuated to some extent, with a consistent direction of fluctuation, and quickly stabilized after adjustment. The dominant change pattern was that the larger the adjustment magnitude of the design change, the larger the adjustment magnitude of the standardized values ​​of the three dimensions. After the adjustment, the three elements quickly formed a new synergistic relationship, the fluctuation range narrowed, and they tended to stabilize. After the analysis of each partition was completed, a detailed partition analysis ledger was formed, determining the association patterns, statistical characteristics, dominant patterns, and suitable scenarios for each partition.

[0027] Step 338: Based on the statistical characteristics and dominant change patterns of the data distribution, generate targeted adjustment parameters for the strategy. Specifically, this includes: generating specific, implementable strategy adjustment parameters for each zone based on the analysis results of each zone. These parameters are aligned with the construction needs of the main structure of the rail transit system, directly guiding on-site management adjustments. For cost-schedule-quality equilibrium zones, resource allocation optimization parameters are generated based on the synergistic stability of these three elements. Regarding manpower, the final staffing ratio for each trade is determined to avoid redundancy or shortages, while setting a daily effective working time limit to prevent fatigue-related work from affecting efficiency. Regarding materials, based on the low-loss characteristics of the zone statistics, the upper limit of the loss rate for various building materials is slightly reduced from the initial plan, while also determining the loss management... Control responsibility; in terms of machinery, optimize the allocation of work periods, improve machinery utilization, set a minimum daily working time for machinery, and reduce idle time; the core objective of the parameters is to maintain the current equilibrium state, slightly improve resource utilization efficiency, and control a slight increase in costs; for the cost-sensitive-stable progress zone, based on the pattern of cost fluctuations and stable progress and quality, generate cost control correction parameters; in terms of material procurement, set a warning threshold for building material price fluctuations, based on the cost fluctuation range of regional statistics, and determine that when the increase in building material prices reaches the warning threshold, two adjustment measures will be immediately initiated: first, negotiate bulk purchases with suppliers to obtain price discounts; second, replace the same specifications with lower-priced alternative materials while meeting quality standards, and clarify the quality inspection of the alternative materials. Requirements: Regarding cost allocation, determine the allocation ratio for each item of overspending to avoid excessive pressure from any single item. Additionally, set cost-saving targets for subsequent processes to offset some of the overspending. The core objective of the parameters is to control cost fluctuations and maintain stable schedule and quality. For areas with both cost and schedule fluctuations, generate process and resource adjustment parameters based on the dual-dimensional fluctuation patterns caused by geological influences. Regarding processes, determine the buffer time for process connections in geologically complex areas to prevent further delays, while optimizing the process sequence to reduce interference. Regarding resources, set upper limits for additional labor and materials, clarify the conditions for using backup machinery, and establish an approval process for additional costs. Furthermore, set a schedule catch-up plan, but ensure quality inspection during the catch-up process. The frequency must not be reduced; the core objective of the parameters is to mitigate fluctuations in both cost and schedule, ensuring consistent quality; for the full-dimensional adjustment zone, based on the pattern of adaptive adjustment across three dimensions, comprehensive adjustment parameters are generated. Regarding cost, the upper limit of additional cost after design changes is determined, the breakdown of additional costs is detailed, and the approval process is defined; regarding schedule, the timelines for each process node are replanned, the upper limit of resource input for catching up on schedule is determined, and a buffer period for schedule adjustments is set to avoid excessive catching up affecting quality; regarding quality, the testing frequency and indicator thresholds under the new design standards are set, and rectification deadlines are determined; the core objective of the parameters is to quickly adapt to design changes, promote a new synergistic and stable relationship among cost, schedule, and quality, and avoid control imbalances after adjustments.All strategy modification parameters were compiled into a written document, clearly indicating the corresponding partition, applicable scenario, adjustment basis, and responsible position.

[0028] By transforming scattered discrete data into continuous decision surfaces through surface fitting, the dynamic correlation trends of cost, schedule, and quality can be fully captured, ensuring the overall cost and schedule targets of the project are met.

[0029] In a preferred embodiment of the present invention, step 4 above, based on strategy correction parameters and combined with actual cost, schedule, and quality data, normalizes multidimensional data points and maps them to a unit hypersphere; by calculating the shortest arc length between data points to quantify the difference, and performing difference analysis and multi-objective collaborative evaluation, a comprehensive cost control evaluation result is obtained, which may include: In this embodiment of the invention, step 440 involves acquiring the actual cost, progress, and quality data of the engineering cost project within a preset period, aligning and recombining them according to the time sequence to form new multidimensional data points based on each monitoring time point. Specifically, this includes: first, determining the preset period and monitoring time points; the preset period is set to 7 days to align with weekly progress review requirements; setting two core monitoring time points daily, namely 10:00 AM and 4:00 PM, while also marking special periods such as holidays and severe weather to ensure data coverage of the entire construction scenario; subsequently, data acquisition is carried out synchronously by different positions, fully linking the resource execution records and partition ledgers mentioned earlier, extracting the actual cost data within the preset period from the cost flow ledger, and splitting it according to the monitoring time point to specific data points. The data covers all work processes and stages, including labor costs, material procurement, machinery rental, and temporary expenses. Each data point is labeled with the corresponding monitoring time point, work process location, and resource execution record number. Invoices and payment vouchers are also verified to ensure the accuracy and traceability of cost data. Actual progress data for the corresponding period is extracted from progress reports and on-site construction logs. The cumulative completed amount is calculated according to the monitoring time point. For rebar tying, the data is calculated based on the actual weight of the tied rebars; for formwork installation, it is calculated based on the actual area laid; and for concrete pouring, it is calculated based on the actual volume of the poured components. The progress data for each monitoring time point corresponds to the amount of work completed before that time point, and factors affecting progress are also noted. Actual quality data for the corresponding period is extracted from the quality inspection log. Data is organized according to monitoring time points down to specific testing batches. Material testing includes the results of testing the tensile strength of reinforcing bars and the slump of concrete for each batch. Process testing includes the results of testing the spacing of reinforcing bars, the flatness of formwork, and the density of concrete. The quality data for each monitoring time point corresponds to the testing results of the batches of processes completed before that time point, and the names of the testing personnel and on-site supervisors are marked to ensure that the quality data is verifiable. After the data acquisition is completed, it is aligned point by point according to the time sequence. The cost data, progress data, and quality data of the same monitoring time point and the same process part are matched one by one and recombined into new three-dimensional multi-dimensional data points. Each data point corresponds to a specific time point of a specific year, month, and day, and includes three core dimensions. The cumulative data for that part at that time point is... The data includes actual cost (the sum of all sub-item costs), cumulative actual progress (the cumulative value of the corresponding process work), and cumulative quality inspection pass rate (the number of qualified inspection batches divided by the total number of inspection batches), along with complete annotation of related information. For example, the data point for the rebar binding of the No. 1 foundation pit beam at 10:00 AM on [Date] in 202X is as follows: the cost dimension is the sum of the cost of rebar materials, rebar worker wages, and rebar processing machinery rental costs for the rebar binding of this part before this time point; the progress dimension is the total weight of the rebar already bound in this part; and the quality dimension is the pass rate of the two batches of rebar binding processes that have been completed in this part. This forms a new multi-dimensional data point set with monitoring time points as units, process parts as carriers, and complete and traceable data.

[0030] Step 441: Based on the feature partition adjustment coefficients defined in the strategy correction parameters, normalize the new multidimensional data points and convert all values ​​to a unified scaling range to generate standard multidimensional data points. Specifically, this includes: first, retrieving the strategy correction parameter ledger generated in step 338, extracting the dimension adjustment coefficients corresponding to each feature partition. The coefficients are set strictly to match the fluctuation characteristics of the partition data to ensure that the core control dimensions are highlighted after normalization. The specific coefficient standards are as follows: for the cost-schedule-quality balanced zone, all three dimension adjustment coefficients are 1.0; for the cost-sensitive-schedule stable zone, the cost dimension adjustment coefficient is 1.2, and the schedule and quality dimension adjustment coefficients are both 1.0; for the cost-schedule dual-fluctuation zone, the cost and schedule dimension adjustment coefficients are both 1.1, and the quality dimension adjustment coefficient is 1.0. The coefficient is 1.0; the adjustment coefficients for all three dimensions of the full-dimensional adjustment area are 1.1; then, the new multidimensional data points are normalized dimensionally, uniformly converted to a scale range of 0-1. Cost dimension normalization first calculates the maximum and minimum values ​​of the cost dimension for all new multidimensional data points within a preset period. The maximum value is the maximum cumulative cost at each monitoring time point and each process location within the period, and the minimum value is the minimum value of the corresponding dimension within the period, while excluding extreme abnormal cost values; then, the minimum cost value is subtracted from the cost value of a single data point, the result is divided by the difference between the maximum and minimum cost values, and then multiplied by the cost adjustment coefficient of the corresponding feature partition to calculate the preliminary normalization value; if the preliminary normalization value is greater than 1, it is counted as 1; if it is less than 0, it is counted as 0; for example... A certain data point belongs to the cost-sensitive, stable-schedule zone, corresponding to the concrete pouring process of the column section in foundation pit No. 1, with a cost of 62,000 yuan. Within the preset period, the maximum cost of the concrete pouring process is 98,000 yuan, and the minimum is 34,000 yuan. The initial calculation process is: 62,000 yuan minus 34,000 yuan, resulting in 28,000 yuan; 28,000 yuan divided by the difference between 98,000 yuan and 34,000 yuan (64,000 yuan) yields 0.4375; 0.4375 multiplied by the cost adjustment coefficient of 1.2 yields 0.525, and the final normalized cost value is 0.525. Similarly, for schedule dimension normalization, first, the maximum and minimum values ​​of the schedule dimension (the maximum cumulative progress completion amount for each monitoring time point and each process part within the preset period) of all new multi-dimensional data points are calculated. Exclude extreme abnormal progress values; then subtract the minimum progress value from the progress value of a single data point, divide the result by the difference between the maximum and minimum progress values, and multiply by the progress adjustment coefficient of the corresponding feature partition for that data point. If the value exceeds the range of 0-1, the boundary value is used. For example, if a data point belongs to the cost-schedule dual fluctuation zone, corresponding to the steel reinforcement binding process at the bottom of Pit 1, the progress value is 280 kg. Within the preset period, the maximum progress value of this process is 520 kg and the minimum progress value is 120 kg. The calculation process is 280 kg minus 120 kg, resulting in 160 kg; 160 kg divided by the difference between 520 kg and 120 kg, 400 kg, yields 0.4; 0.4 multiplied by the progress adjustment coefficient 1.1 yields 0.44, and the progress normalization value is 0.44; Quality dimension normalization is based on the quality inspection pass rate. It is calculated by multiplying the quality pass rate of a single data point by the quality adjustment coefficient of the corresponding feature zone. If the result is greater than 1, it is taken as 1; if it is less than 0, it is taken as 0. For example, if a data point belongs to the full-dimensional adjustment zone, corresponding to the No. 2 beam formwork installation process, the quality pass rate is 92% (2 out of 3 batches were qualified, and 1 batch was qualified after rectification; pass rate = 3 ÷ 3 × 100%). The calculation process is 92% multiplied by the quality adjustment coefficient 1.1, resulting in 1.012. Since the result is greater than 1, the final quality normalization value is taken as 1.0. After all dimensions are processed, the cost, schedule, and quality normalization values ​​of each monitoring time point and each process part are combined to generate standard multidimensional data points.

[0031] Step 442: Map the standard multidimensional data points to a high-dimensional feature space through a preset coordinate transformation relationship, so that each standard multidimensional data point is projected and positioned on the surface of a unit hypersphere in the high-dimensional space. Specifically, this includes: determining that the preset high-dimensional feature space is a three-dimensional extended space, the unit hypersphere is a three-dimensional sphere with a fixed radius of 1, and the center of the sphere is set as the origin (0, 0, 0). The core principle of the coordinate transformation relationship is to maintain the relative proportion of each dimension, and to adjust the coordinates of the data points so that the straight-line distance to the center of the sphere is exactly 1, ensuring that all data points are comparable within the same spherical space. Then, perform coordinate transformation mapping on each standard multidimensional data point one by one. The first step is to calculate the square of the three-dimensional normalized value of a single standard data point. The first step is to multiply the normalized cost value by itself, then multiply the normalized schedule value by itself, and finally multiply the normalized quality value by itself. The sum is the three-dimensional numerical sum of squares. The second step is to calculate the square root of this sum, which is the initial straight-line distance from the data point to the center of the sphere. For example, if the sum of squares is 1.5329, its square root is 1.238. The third step is to divide the normalized value of each dimension by this initial distance to obtain the mapped spherical coordinates. That is, mapped cost coordinates = normalized cost value ÷ initial distance, mapped schedule coordinates = normalized schedule value ÷ initial distance, and mapped quality coordinates = normalized quality value ÷ initial distance. This transformation ensures that the distance from the data point to the center of the sphere is 1, exactly falling on the surface of the unit hypersphere. For example, if a standard data point has a normalized value of (0.525, 0.44, 1.0), the first step is to calculate the sum of squares: multiplying 0.525 by itself gives 0.2756, multiplying 0.44 by itself gives 0.1936, and multiplying 1.0 by itself gives 1.0. The sum of these three is 0.2756 + 0.1936 + 1.0 = 1.4692. The second step is to calculate the initial distance, which is the square root of 1.4692 (approximately 1.212). The third step is to calculate the mapped coordinates: cost coordinate = 0.525 ÷ 1.212 ≈ 0.433, schedule coordinate = 0.44 ÷ 1.212 ≈ 0.363, and quality coordinate = 1.0 ÷ 1.212 ≈ 0.825. Finally, this data point is mapped... The data is mapped to spherical coordinates (0.433, 0.363, 0.825). During the mapping process, after mapping 10 data points, the coordinate positions are marked using a written spherical diagram, and the distance from the data point to the center of the sphere is measured with a ruler. If the distance deviates by 1, the calculation results of the sum of squares and the initial distance are rechecked, and the mapped coordinates are corrected. At the same time, combined with the feature partition attributes mentioned above, the standard data points of different partitions are labeled with text to distinguish them. The cost-schedule-quality balanced zone is labeled as regular, the cost-sensitive-schedule stable zone is labeled as cost warning, the cost-schedule dual fluctuation zone is labeled as dual fluctuation, and the full-dimensional adjustment zone is labeled as full adjustment. Finally, a positioning ledger of all standard data points on the unit hypersphere is formed.

[0032] Step 443: On the surface of the unit hypersphere, for the current data point representing the actual execution state of the current cycle, select a reference data point representing the target or baseline state, calculate the shortest arc length on the hypersphere surface, and use the geometric distance value as a quantitative difference measure between the current state and the target state in multidimensional space; specifically, the current data point is the standard data point of the last monitoring time point of the preset cycle, and representative data points of the core processes within the cycle are selected to represent the final state of actual execution within the cycle, while also labeling the characteristic partition attributes and the impact of abnormal factors of the data point; the reference data points are divided into two categories, one being the target reference point, Based on the expected targets set by the initial cost control strategy, namely the cost ceiling, schedule nodes, and quality compliance rate preset in the initial plan, the spherical coordinates obtained after normalization in step 441 and mapping in step 442 are used to ensure that the target reference point is consistent with the feature partition to which the current data point belongs. Another type is high-quality benchmark points, which are spherical coordinate points obtained by selecting high-quality project data from the main structure construction, process, and geological conditions of similar rail transit stations, and processing them through the same process, as auxiliary references. Then, the shortest arc length of the unit hyperspherical surface is calculated. The first step is to determine the coordinates of the current data point and the reference data point on the sphere. The first step is to mark two points and the center of the sphere on a written spherical diagram, forming a triangle (the two points and the center of the sphere constitute the triangle). The second step is to calculate the central angle formed by the two points and the center of the sphere. This is done by multiplying the corresponding dimensions of the two points' coordinates and determining the angle. Then, multiply the current data point by the cost coordinate, schedule coordinate, and quality coordinate of the reference data point, and add these three products. The closer the result is to 1, the smaller the angle between the lines connecting the two points and the center of the sphere; the closer the result is to -1, the larger the angle. The third step is to calculate the shortest arc length. Since the unit hypersphere radius is 1, the value of the shortest arc length is equal to the value of the central angle (when the radius is 1, the arc length = central angle). Arc length is a quantitative measure of the difference between the current state and the target state. The longer the arc length, the greater the difference between the actual execution and the target / baseline state; the shorter the arc length, the smaller the difference. For example, the sum of the coordinate products of the current data point (0.433, 0.363, 0.825) and the target reference point (0.45, 0.4, 0.8) is approximately 0.195 + 0.145 + 0.66 = 1.0, corresponding to a very small central angle and a very short shortest arc length, indicating that the difference between the actual execution and the target state is small. If the sum of the coordinate products of the current data point and the target reference point is 0...5. A larger central angle and longer arc length indicate a significant difference. Simultaneously, for the high-quality benchmark, calculate the arc length using the same logic and compare the current data point with the target reference point and the high-quality benchmark. If the difference with the target reference point is large but the difference with the high-quality benchmark is small, the initial target setting may be too stringent and needs adjustment based on actual construction conditions. If the difference with both is large, the cause of the deviation needs to be thoroughly investigated. After calculation, record the arc length value, central angle size, difference level, and preliminary analysis results of the difference in the logbook, while also noting the basis for selecting the target reference point and the high-quality benchmark.

[0033] Step 444 involves performing dimensional analysis on the quantitative difference measure to determine the specific direction and magnitude of deviation between the current execution effect and the expected goal in the three main dimensions of cost, schedule, and quality, thus obtaining the dimensional deviation analysis results. Specifically, this includes: breaking down the quantitative difference dimension by dimension, combining the data point coordinates on the unit hypersphere, the original data, and the feature partition attributes to determine the deviation direction, magnitude, and cause of each dimension; simultaneously analyzing the correlation between dimensions; first, extracting the cost dimension spherical coordinates of the current data point and the target reference point, and comparing their magnitudes; if the current cost coordinate is greater than the target reference point cost coordinate, it indicates that the deviation direction is cost. Overspending; if the cost is less than the target reference point cost coordinate, the deviation direction is considered cost saving; the deviation range is calculated by dividing the absolute value of the difference between the current cost coordinate and the target reference point cost coordinate by the target reference point cost coordinate. The larger the result, the larger the deviation range. Simultaneously, in conjunction with the original cost data, verify the actual overspending / saving amount, link it to specific sub-cost items, and investigate the cause of the deviation; for example, if the current data point cost coordinate is 0.433 and the target reference point cost coordinate is 0.4, the absolute value of the difference is 0.033, the deviation range = 0.033 ÷ 0.4 = 0.0825, the deviation direction is slight overspending, which, in conjunction with the original data, is verified to be caused by a slight increase in the purchase price of steel reinforcement materials, without any other reason. Other abnormal factors; progress dimension analysis compares the current data point with the target reference point's progress dimension spherical coordinates. If the current coordinate is greater than the target coordinate, the deviation direction indicates progress is ahead; if it is less than the target coordinate, the deviation direction indicates progress is behind. The deviation magnitude is calculated by dividing the absolute value of the difference between the current progress coordinate and the target reference point's progress coordinate by the target reference point's progress coordinate. Simultaneously, combined with the original progress data, the actual ahead / behind workload and corresponding duration are calculated to analyze the reasons for the deviation. For example, if the current data point's progress coordinate is 0.363 and the target reference point's progress coordinate is 0.4, the absolute value of the difference is 0.037, and the deviation magnitude = 0.037 ÷ 0.4 = 0.0925. The initial delay was slightly delayed, which was verified to be due to minor water seepage in the geology causing a one-hour suspension of the rebar tying process; no other interference was observed. Quality dimension analysis compared the current data point with the target reference point's spherical coordinates. If the current coordinates were less than the target coordinates, the deviation direction indicated substandard quality; if they were greater than or equal to the target coordinates, the deviation direction indicated acceptable / better quality. The deviation magnitude was only calculated when quality was substandard, calculated by dividing the absolute value of the difference between the target reference point's quality coordinates and the current quality coordinates by the target reference point's quality coordinates. This was combined with quality inspection records to determine the specific substandard indicators and analyze the causes; for example, the current data point's quality coordinates were 0.825, and the target reference point's quality coordinates were 0.8. The current coordinates are greater than the target coordinates, the deviation direction is quality better than the target, and there is no deviation magnitude. Verification shows that all batches of rebar tying work passed inspection, and the rebar spacing deviations were all controlled within the lower limit of the allowable range, indicating excellent control effectiveness. Subsequently, the analysis results from various dimensions were integrated to determine the correlation between dimensions, such as whether there is a correlation between slight cost overruns and slight schedule delays, and whether quality exceeding the target depends on additional costs. Finally, a ledger of dimensional deviation analysis results was formed.

[0034] Step 445: Based on the preset adjustment coefficients for the collaborative goals of cost, schedule, and quality, a comprehensive calculation is performed on the deviation magnitude of each dimension to evaluate the overall collaborative performance under the constraints of multiple goals (cost, schedule, and quality), resulting in a multi-goal collaborative evaluation value. Specifically, this includes: first, determining the preset adjustment coefficients for the collaborative goals, with a total coefficient of 1.0. The specific allocation is as follows: Quality dimension adjustment coefficient 0.4 (quality directly relates to structural safety and is a core control goal with the highest weight); Cost dimension adjustment coefficient 0.35 (fluctuations in building material prices significantly impact cost and require key control); Schedule dimension adjustment coefficient 0.25 (schedule can be caught up through subsequent process optimizations and has a relatively low weight). Then, a comprehensive calculation is performed. The first step involves adjusting the deviation magnitude of each dimension... Standardized correction involves setting correction rules based on the direction of deviation. If the deviation is positive, the deviation magnitude is calculated as a negative value; if the deviation is negative, the deviation magnitude is calculated as a positive value; if there is no deviation magnitude, the corrected deviation magnitude is counted as 0. For example, a slight cost overrun with a deviation of 0.0825 results in a correction of +0.0825; a slight schedule delay with a deviation of 0.0925 results in a correction of +0.0925; and a quality exceeding the target has no deviation magnitude and is corrected to 0. The second step is to calculate the weighted deviation value for each dimension by multiplying the corrected deviation magnitude for each dimension by the corresponding collaborative target adjustment coefficient. That is, the cost weighted deviation value = cost corrected deviation magnitude × 0.35, and the schedule weighted deviation value = schedule corrected deviation magnitude × 0.35. 0.25, Quality-weighted deviation = Deviation after quality correction × 0.4, calculate the weight of each dimension's impact on overall collaborative performance; for example, Cost-weighted deviation = 0.0825 × 0.35 ≈ 0.0289, Schedule-weighted deviation = 0.0925 × 0.25 ≈ 0.0231, Quality-weighted deviation = 0 × 0.4 = 0; Third step, calculate the multi-objective collaborative evaluation value, add the weighted deviation values ​​of the three dimensions, and the sum is the collaborative evaluation value. The closer the evaluation value is to 0, the better the overall collaborative performance and the more it conforms to multi-objective constraints; A positive and larger evaluation value indicates poorer collaborative performance and an imbalance in multi-dimensional control; A negative evaluation value indicates that the overall performance is better than the target and the collaborative effect is good; for example The sum of the above data is 0.0289 + 0.0231 + 0 = 0.052, resulting in a collaborative evaluation value of 0.052, which is close to 0, indicating good overall collaborative performance. Simultaneously, based on the previously mentioned feature partitioning attributes, the collaborative evaluation value is recalibrated. The calibration rules are: Cost-Schedule-Quality Balanced Zone Evaluation Value × 1.0 (no calibration required); Cost-Sensitive-Stable Zone Evaluation Value × 1.05 (strengthening the impact of the cost dimension); Cost-Schedule Dual Fluctuation Zone Evaluation Value × 1.1 (strengthening the impact of both cost and schedule dimensions); All-Dimensional Adjustment Zone Evaluation Value × 1.15 (strengthening the impact of all dimensions, adapting to adjustment needs). For example, if the current data point belongs to the Cost-Sensitive-Stable Zone, the calibrated collaborative evaluation value would be 0.052 × 1.05 ≈ 0.0546, after calibration, remains close to 0, indicating good synergy performance. After calculation, the synergy evaluation value, correction process, weighted calculation details, basis for secondary calibration, and synergy level are recorded. The corresponding standards for synergy level are: evaluation value less than 0.05 is excellent, 0.05-0.1 is good, 0.1-0.2 is average, and greater than 0.2 is poor.

[0035] Step 446 involves integrating and judging the dimensional deviation analysis results with the multi-objective collaborative evaluation values ​​to ultimately generate a comprehensive cost control evaluation result. Specifically, this includes: conducting integrated analysis based on the principles of dimensional decomposition and overall collaboration to generate a feasible comprehensive evaluation result; firstly, integrating the multi-objective collaborative evaluation value with the average deviation of each dimension, i.e., the comprehensive evaluation baseline value = multi-objective collaborative evaluation value + (sum of absolute values ​​of deviation of each dimension ÷ 3). The closer the baseline value is to 0, the better the overall performance; subsequently, a graded judgment is performed, and judgment criteria are defined based on the construction control requirements of the main structure of rail transit. A baseline value less than 0.05 is judged as excellent, indicating that the actual execution deviates little from the target and multi-dimensional collaboration is good; a baseline value between 0.05 and... A baseline value between 0.1 and 0.2 is considered "good," indicating a slight deviation but overall controllable coordination. A baseline value between 0.1 and 0.2 is considered "average," indicating a significant deviation requiring targeted adjustments. A baseline value greater than 0.2 is considered "poor," indicating a severe deviation, imbalance in coordination, and a need for comprehensive rectification. Simultaneously, based on the dimensional deviation analysis results, specific rectification directions are determined. For example, if the assessment is "good" and the deviation stems from a slight cost overrun, cost control parameters need to be adjusted. If the assessment is "average" and the deviation stems from both schedule delays and cost overruns, process and resource adjustment parameters need to be optimized. If the assessment is "poor," and there is no deviation in the quality dimension but a severe imbalance between cost and schedule, strategy parameters need to be revised, coordination goals adjusted, and a written comprehensive cost control assessment result ultimately generated.

[0036] By aligning and reorganizing data according to time series, quantifying differences, and decomposing dimensions for analysis, the deviations and related impacts of each dimension of cost, schedule, and quality can be presented intuitively.

[0037] In a preferred embodiment of the present invention, step 5 above, which optimizes the current strategy using a reinforcement learning model based on the comprehensive cost control evaluation results and strategy correction parameters to generate an optimized cost control strategy, may include: In this embodiment of the invention, step 550 involves converting the comprehensive cost control assessment result into a penalty signal; simultaneously, the strategy correction parameters are fused with the current environmental status information of the project to form a comprehensive status feature vector; specifically, this includes: First, converting the comprehensive cost control assessment result into a penalty signal involves retrieving the comprehensive cost control assessment result ledger generated in step 446, extracting core assessment indicators, comprehensively judging the level, the comprehensive assessment baseline value, the deviation range of each dimension, and the feature partition attributes, setting penalty signal conversion rules, and setting the baseline penalty value according to the assessment level, adjusting the baseline value and deviation range, calibrating the partition attributes, and setting the baseline penalty value to 0 (no penalty, representing that the control standard is met) when the comprehensive judgment is excellent. The baseline penalty value is set to 1 for a good rating (minor penalty, indicating fine-tuning is needed); 3 for a fair rating (moderate penalty, indicating optimization is needed); and 5 for a poor rating (severe penalty, indicating rectification is needed). Then, adjustments are made based on the comprehensive evaluation baseline value. The further the baseline value deviates from 0, the more the penalty value accumulates. The accumulation rule is: Penalty value = Baseline penalty value + (Comprehensive evaluation baseline value ÷ Corresponding level upper limit) × 2. For example, if a project is rated as fair, with a comprehensive evaluation baseline value of 0.1129, the calculation is 3 + (0.1129 ÷ 0.2) × 2 = 3 + 1.129 = 4.129, yielding the initial penalty value. Then, calibration is performed according to the characteristic zoning attributes: the penalty value for the cost-schedule-quality equilibrium zone is × 1.0, resulting in a final penalty value. The first part calculates the penalty signal by setting a threshold of 1.05 for the sensitive-schedule stable zone (strengthened cost deviation penalty), 1.1 for the cost-schedule dual-fluctuation zone (strengthened dual-dimensional deviation penalty), and 1.15 for the full-dimensional adjustment zone (strengthened full-dimensional deviation penalty). After calibration, the final penalty signal is obtained and labeled with specific values. The conversion process and calculation basis are recorded simultaneously. The second part involves fusing strategy correction parameters with environmental status information to form a comprehensive status feature vector. The strategy correction parameters from step 338 are extracted and categorized by feature partitions, covering resource allocation optimization parameters, cost control correction parameters, process and resource adjustment parameters, and comprehensive adjustment parameters. Each parameter is labeled with its corresponding partition and applicable scenario. The current environmental status information of the project is collected, specifically including the status of construction processes. The system considers the status, real-time status of resources, current status of abnormal factors, external environment, and schedule pressure. Then, it integrates these elements according to the principles of dimensional alignment, numerical normalization, and orderly combination, mapping the strategy correction parameters to environmental status information into 12 core dimensions, each corresponding to a specific numerical value. For example, for the cost control dimension, the normalized value of the price warning threshold in the strategy correction parameters is multiplied by the normalized value of the current building material price fluctuation range in the environmental status to obtain the integrated value for that dimension. For the human resources dimension, the normalized value of the human resource allocation ratio in the strategy parameters is added to the normalized value of the ratio of current on-duty personnel to planned personnel, and then divided by 2 to obtain the integrated value. After each of the 12 integrated values ​​is calculated, they are combined in a fixed order to form a comprehensive status feature vector.

[0038] Step 551: Input the penalty signal and the comprehensive state feature vector into the pre-trained reinforcement learning model, calculate the comprehensive state feature vector through the policy network, and generate the decision probability distribution in the current state; specifically, this includes: first, constructing the reinforcement learning model in detail. The reinforcement learning model is a three-layer network structure. The number of nodes in the input layer is set to 12, corresponding to the 12 dimensions of the comprehensive state feature vector. Each node is responsible for receiving the fused value of the corresponding dimension. The node activation rule is to only receive values ​​in the range of 0-1. Values ​​outside the range are corrected according to the boundary value before being input. The hidden layer is set to 2 layers, and the number of nodes in each layer is 24. The number of nodes is set to twice the number of nodes in the input layer to ensure that the correlation between each dimension can be fully explored. The first hidden layer is responsible for the initial processing of the input layer data. In the first layer, each node receives values ​​from all 12 nodes in the input layer. The output value is calculated by multiplying each input value by its corresponding weight. Initial weights are set between 0.05 and 0.1 for similar project data. All results are summed, and a bias value is added, with an initial bias value of 0.02. The second hidden layer receives the output values ​​from the 24 nodes in the first hidden layer and calculates them using the same logic. Weights and bias values ​​are set independently to ensure in-depth feature mapping. The output layer has 8 nodes, corresponding to 8 core decision-making actions (adjusting labor allocation, optimizing material loss rate, extending machine operation time, initiating material price negotiation, increasing process buffer time, strengthening quality inspection, increasing cost investment, and adjusting schedule catch-up plans). Each node corresponds to one type of decision. The reinforcement learning model... In the pre-training process, historical data from three similar rail transit station main structure construction projects were selected as training samples. These samples covered different construction stages, abnormal scenarios, and strategy adjustment cases corresponding to evaluation results, resulting in 1000 valid samples. During training, the sample feature vectors and penalty signals were input into the reinforcement learning model. Through iterative adjustments to the network weights and biases, the deviation between the reinforcement learning model's output decision and the final sample decision was gradually reduced until the deviation was within a reasonable range, i.e., more than 80% of the sample output decisions were consistent with the final decisions. This completed the pre-training, resulting in a pre-trained reinforcement learning model. The final initial weights, biases, and network structure parameters were recorded. After pre-training, data processing personnel combined the penalty signal generated in step 550 with the overall... The combined state feature vector is input into the model. After verification by the input layer nodes, the signal and vector are passed to the hidden layer. The two hidden layers are integrated and calculated step by step according to the set logic. The output layer nodes receive the final output value of the hidden layer. Then, the decision probability distribution is calculated through the policy network. The policy network is the core computing unit of the reinforcement learning model. The probability allocation rule is set, and the decision is based on the comprehensive judgment of the values ​​of each node in the output layer and the penalty signal. First, the sum of the values ​​of all nodes in the output layer is calculated. The initial probability is obtained by dividing the value of each node by the sum. Then, the probability is adjusted in combination with the penalty signal. The larger the penalty signal value, the higher the probability of risk-averse decision-making and the lower the probability of aggressive adjustment decision-making. The adjustment rule is: risk-averse decision probability = initial probability × (1 + penalty signal value × 0).1) The probability of an aggressive adjustment decision is calculated as: Initial Probability × (1 - Penalty Signal Value × 0.1). This ensures that a larger penalty leads to a more conservative decision. For example, with a penalty signal value of 4.129, the initial probability of a risk-averse decision is 0.15, which is adjusted to 0.15 × (1 + 4.129 × 0.1) = 0.15 × 1.4129 ≈ 0.2119; the initial probability of an aggressive adjustment decision is 0.2, which is adjusted to 0.2 × (1 - 4.129 × 0.1) = 0.2 × 0.5871 ≈ 0.1174. After adjusting all decision probabilities, the sum of the probabilities is ensured to be 1, forming the decision probability distribution under the current state.

[0039] Step 552: Based on the penalty signal and the preset policy gradient update algorithm, adjust the internal parameters of the policy network to generate an updated policy network parameter set. Specifically, this includes: first, decomposing the internal parameters of the policy network, including 12×24 sets of weights from the input layer to the first hidden layer, 24×24 sets of weights from the first to the second hidden layer, and 24 bias values ​​for each of the two hidden layers, totaling 12×24+24×24+24+24=960 parameters, with each parameter labeled with its corresponding node relationship; then, based on the penalty signal analysis, the parameter adjustment direction is determined, and the adjustment amplitude... The process involves calculation, parameter updating, and verification. The first step is penalty signal analysis: analyzing penalty signal values ​​and corresponding evaluation results to determine the core direction of parameter adjustment. A penalty signal value greater than 3 (moderate or severe penalty) indicates a significant deviation in current decision-making, requiring adjustments towards strengthening cost control, optimizing schedule adaptation, and maintaining quality priority. Penalty signal values ​​of 1-3 (slight to moderate penalty) suggest adjustments towards fine-tuning resource allocation and strengthening deviation correction. Penalty signal values ​​of 0-1 (no penalty to slight penalty) only require fine-tuning weights and bias values. The second step is to determine the adjustment direction for each parameter, inputting the data to the layer. The weights of the hidden layers are adjusted such that parameters corresponding to the bias dimension are increased, while parameters corresponding to the dominance dimension are stabilized. The bias adjustment direction is consistent with the trend of the corresponding node output value; if the output value is too low, the bias value is increased, and if it is too high, it is decreased. The third step is to calculate the adjustment magnitude, which is calculated as: adjustment magnitude = basic adjustment coefficient × penalty signal value × dimension weight. The basic adjustment coefficient is uniformly set to 0.01, and the dimension weight is set according to the collaborative target adjustment coefficient in step 445. The fourth step is to update the parameters one by one, updating all 960 parameters according to the adjustment direction and magnitude, and simultaneously recording the initial value, adjustment magnitude, adjusted value, and calculation basis of each parameter. The fifth step is to verify the updated parameters, substituting the updated parameters into the reinforcement learning model, re-inputting the penalty signal and comprehensive state feature vector from step 550, calculating the output decision probability distribution, and comparing the distribution difference before and after the update. If the probability of the final decision increases after the update and the simulated value of the penalty signal decreases, the parameter update is valid. If it is invalid, the adjustment magnitude is reversed by 50%, and the update and verification are repeated until the requirements are met. After all parameters are updated and verified, the updated policy network parameter set is formed.

[0040] Step 553: Based on the updated policy network, respond to the state features to generate an optimized cost control strategy. Specifically, this includes: inputting the comprehensive state feature vector generated in step 550 back into the updated policy network; the network re-executes layer-by-layer calculations according to the adjusted parameters, outputting an updated decision probability distribution; then, setting a decision screening threshold (e.g., 0.1) based on actual project needs, eliminating inefficient decisions with probabilities below the threshold, and retaining core decisions with higher probabilities, while ensuring that the screened decisions cover the three core dimensions of resource allocation, cost threshold, and schedule-quality constraints to avoid decision bias; quantifying and transforming the screened decisions to form a feasible optimized cost control strategy; for example, if increasing the steel reinforcement procurement reserve has the highest decision probability, this can be combined with policy adjustment parameters. The material resource adjustment coefficient and the fluctuation range of building material prices are used to determine the adjustment ratio of the purchase quantity (purchase quantity = original planned purchase quantity × (1 + material resource adjustment coefficient × building material price fluctuation range)), and the purchase time node is determined at the same time. If the decision to adjust the labor scheduling method is highly probable, the scheduling adjustment plan is determined by combining the labor resource allocation adjustment coefficient and the construction period delay (changing from a single shift to a two-shift system, the number of workers = the original planned number of workers × (1 + labor resource adjustment coefficient × construction period delay ratio)). At the same time, the cost warning threshold is revised (new threshold = original threshold × (1 + cost warning threshold revision value)), the schedule-quality constraint relationship is clarified, and finally all the quantified decisions are integrated to form an optimized cost control strategy covering resource allocation, cost control, and schedule and quality coordination.

[0041] By quantifying the comprehensive evaluation results into penalty signals and combining them with multi-dimensional parameters to form feature vectors, quantitative driving of strategy optimization is achieved, providing clear data basis for strategy adjustments.

[0042] In a preferred embodiment of the present invention, step 6 above, which applies the optimized cost control strategy to the next construction stage and monitors building material price fluctuations, design changes, and construction delays in real time to achieve dynamic optimization control, may include: In this embodiment of the invention, step 660 involves implementing an optimized cost control strategy in the next construction phase, guiding resource allocation and cost management according to the optimized cost control strategy. Specifically, this includes: For labor resources, calculating the actual number of workers needed based on the labor adjustment coefficient and the project delay ratio given by the strategy. The calculation method is: the original planned number of workers plus the original planned number of workers multiplied by the labor adjustment coefficient and then multiplied by the project delay ratio. Simultaneously, determining the scheduling method, the entry time, work cycle, and assessment standards for each type of work ensures that manpower input matches the progress catch-up requirements without causing redundancy. For material resources, calculating the procurement quantity of various main materials based on the material procurement adjustment ratio and the building material price fluctuation range in the strategy, i.e., the original planned procurement quantity plus the original planned procurement quantity multiplied by the material adjustment ratio and then multiplied by the building material price fluctuation range. Simultaneously determining the procurement time node, supplier selection criteria, and material storage loss control requirements. For new materials involved in design changes, calculating the usage separately and including them in the procurement plan. For machinery resources, calculating the actual machinery operation time based on the machinery operation time adjustment coefficient and the project quantity change ratio given by the strategy, which is the original planned operation time plus the original planned operation time. Multiply the planned work duration by the mechanical adjustment coefficient and then by the percentage change in workload to rationally arrange the equipment entry sequence, work shifts, and maintenance cycles, avoiding idle machinery or insufficient operational capacity. Using the optimized strategy's cost warning threshold as the core, implement full-process control. First, update the cost control ledger, entering the cost thresholds and deviation tolerances for each sub-item of the optimized strategy into the system. Each cost expenditure must be linked to a specific construction stage, implementing a tiered approval process. If a single expenditure exceeds the corresponding stage's threshold by 10% or more, a special audit report must be submitted. Second, establish real-time cost deviation monitoring... The control mechanism calculates the deviation between actual cost and optimized budget daily. The deviation is the actual cost minus the optimized budget. If the deviation is positive and exceeds the tolerance level, unnecessary expenditures in the corresponding stage are immediately suspended, the cause of the overspending is investigated, and countermeasures are taken. If the deviation is negative, the reasons for the savings are analyzed to see if they affect the schedule and quality, ensuring that cost savings do not sacrifice the core objectives of the project. At the same time, the linkage between schedule and quality on cost is tracked. If the schedule is ahead of schedule or the quality compliance rate exceeds expectations, the cost budget allocation for subsequent stages is adjusted according to the strategy requirements, and the savings are reasonably allocated to stages with potential overspending risks.

[0043] Step 661: During the implementation of the optimized cost control strategy, real-time monitoring of the project's internal and external environment is conducted simultaneously to obtain dynamic information, including building material market prices, design changes, and actual construction progress deviations. Specifically, this includes: establishing an hourly, daily, and weekly tiered monitoring mechanism. Hourly monitoring focuses on key building material prices, collecting current transaction prices for various building materials through real-time pricing platforms in the building material market and synchronous feedback from suppliers, while simultaneously recording the time, magnitude, and influencing factors of price changes. Daily monitoring covers construction progress and the site environment, compiling the actual completed work volume at the current stage before the end of each workday, comparing it with the progress plan in the optimized strategy, and calculating progress deviations; simultaneously recording changes in construction site conditions and weather impacts to form a daily monitoring log. Weekly monitoring addresses design changes and policy adjustments, collecting change notices issued by the design unit, summarizing the change content, the amount of work involved, and the impact on construction techniques. The impact of environmental factors is assessed by: simultaneously tracking changes in local environmental protection and safety policies and evaluating their potential impact on construction processes and costs; standardizing and quantifying data after collection to ensure direct use for subsequent analysis; quantifying material price fluctuations as the difference between the current price and the budgeted price in the optimized strategy, then dividing by the optimized budgeted price to obtain the price fluctuation range; quantifying design changes as the changed portion of the work divided by the total work volume in the current stage to obtain the change impact ratio, while also indicating the change type and corresponding cost impact direction; quantifying schedule deviations as the actual completed work volume minus the optimized planned work volume, then dividing by the optimized planned work volume to obtain the schedule deviation ratio; quantifying environmental factors such as weather and site conditions as impact coefficients, where the impact coefficient is the actual impact duration divided by the planned work hours for the day, with a coefficient of 0 when there is no impact; and organizing all quantified data in a fixed order of material price - design change - schedule deviation - environmental impact to form a dynamic information ledger.

[0044] Step 662 involves converting the dynamic information into feature vectors representing the latest project environment state and inputting them into the reinforcement learning model. Combined with the penalty signal regenerated based on the latest project environment state, this drives the reinforcement learning model to perform a new round of optimization and updates to the policy network parameters. Specifically, this includes: first, standardizing the quantized dynamic information from step 661 to eliminate dimensional differences. The standardization method is to divide the quantized value of each dynamic information item by the maximum value of that type of information in similar engineering projects, uniformly mapping all data to the [0, 1] interval. Then, a weight is assigned to each dynamic information item, determined based on its impact on cost control. The fluctuation range of building material prices and the proportion of impact from design changes have the highest weight, followed by the proportion of schedule deviations, while the influence coefficients of weather and site have the lowest weight. All weights sum to 1. Next, each standardized dynamic information item is multiplied by its corresponding weight. All product results are then arranged in a fixed order with the remaining values ​​of the strategy correction parameters involved in step 550 to form a comprehensive state feature vector representing the latest environmental state. Each element of the vector corresponds to a quantified environmental or strategy parameter, comprehensively reflecting the latest dynamics of the current project. Referring to the penalty signal conversion logic of step 550, a new penalty signal is generated based on the cost control effect under the latest environmental state. This signal is first applied to… The current stage's cost, schedule, and quality data are comprehensively evaluated to determine the evaluation level and corresponding basic penalty coefficient. Then, the deviation percentage of each dimension is calculated, i.e., the deviation percentage of a single dimension divided by the sum of the deviation percentages of the three dimensions, to obtain the dimension weight. Finally, the basic penalty coefficient for each level is multiplied by the corresponding dimension weight, and the results for all dimensions are summed to obtain the latest penalty signal. The magnitude of this signal directly reflects the policy's suitability under the current environmental conditions. The latest comprehensive state feature vector and the latest penalty signal are simultaneously input into the reinforcement learning network. The direction of parameter adjustment is determined based on the penalty signal; when the penalty signal is positive, the adjustment direction is to decrease. The parameters that cause the deviation are affected; when the penalty signal is negative, the adjustment direction is to strengthen the influence of the current parameter. Then, the parameter adjustment amount is calculated. The adjustment amount is the latest penalty signal multiplied by the preset learning rate, and then multiplied by the partial derivative of the corresponding parameter of the policy network. The learning rate is kept consistent with step 552 to ensure the stability of parameter adjustment. The connection weights and bias terms of each layer of the policy network are adjusted one by one according to the adjustment amount. At the same time, the constraints of the engineering scenario are followed. If the adjusted parameters exceed the reasonable range, the boundary value is taken as the final parameter. After all parameters are adjusted, a new set of updated policy network parameters is formed to ensure that the network operation logic is adapted to the latest environment state.

[0045] Step 663: Generate a cost control strategy suitable for the new environmental state based on the updated strategy network, and apply the newly generated cost control strategy to the current construction stage to achieve dynamic optimization control. Specifically, this includes: inputting the latest comprehensive state feature vector obtained in step 662 into the updated strategy network, performing layer-by-layer calculations according to the adjusted parameters, that is, after each layer node receives the input value, it multiplies it with the corresponding connection weight, adds a bias term, and passes it to the next layer after being transformed by the activation function, finally outputting a new decision probability distribution; then, setting the decision screening threshold to be consistent with step 553, eliminating inefficient decisions with probabilities below the threshold, retaining core decisions covering resource allocation, cost thresholds, and schedule-quality constraints, quantifying and transforming the core decisions, and generating new, implementable strategies, such as adjusting the material procurement strategy to increase reserves and lock in long-term suppliers in response to the continuous rise in building material prices, with the reserve calculated as the current remaining project quantity multiplied by the application quantity by (1 + the latest price fluctuation range); and addressing design changes. The resulting increase in workload necessitates adjustments to the ratio of labor and machinery inputs. The increase in labor is calculated by multiplying the original quantity by the impact ratio of the design change, and the increase in machinery operation time is calculated by multiplying the original time by the impact ratio of the design change. Simultaneously, the cost warning threshold is revised, with the new threshold being the original threshold multiplied by (1 + the latest deviation ratio). A clear quality baseline is established during the progress catch-up process. The newly generated cost control strategy is quickly synchronized to the site management team and cost management team, replacing the original optimized strategy and applying it to the current construction phase to guide subsequent resource allocation, cost expenditure, and progress control. Meanwhile, the real-time monitoring mechanism from step 661 remains unchanged, continuously collecting dynamic information after the new strategy is implemented. The process of steps 662-663 is repeated to achieve closed-loop control of strategy execution, dynamic monitoring, parameter updates, and strategy iteration. For sudden major uncertainties during construction, the monitoring cycle and parameter update interval can be temporarily shortened to ensure the strategy can quickly respond to extreme scenarios and avoid problems such as cost overruns and schedule delays.

[0046] Through a closed-loop process of real-time monitoring and rapid iteration, the cost control strategy can dynamically adapt to uncertainties such as fluctuations in building material prices and design changes, ensuring that the strategy always matches the current construction environment.

[0047] like Figure 2 As shown, embodiments of the present invention also provide a reinforcement learning-based system for optimizing cost control strategies in engineering cost projects, including: The module is used to obtain the initial budget, real-time construction environment status and past cost control records of the project, and to build the initial cost control strategy. The execution module is used to perform resource allocation and cost control in the current construction phase according to the initial cost control strategy, and to obtain actual cost, schedule and quality data; The analysis module integrates actual cost, schedule, and quality data to construct a multi-dimensional data distribution model and form an abstract decision space. It fits discrete data into a continuous decision surface, calculates the curvature of local regions, and analyzes the structural change characteristics of the decision surface. Based on the structural change characteristics, it divides the region within the abstract decision space, maps the original data to the corresponding partition, extracts the intrinsic characteristics of each partition, and generates strategy correction parameters. The evaluation module is used to normalize and map multidimensional data points to a unit hypersphere based on strategy correction parameters and actual cost, schedule and quality data; it quantifies the difference by calculating the shortest arc length between data points, and performs difference analysis and multi-objective collaborative evaluation to obtain a comprehensive cost control evaluation result. The optimization module is used to optimize the current strategy based on the comprehensive cost control evaluation results and strategy correction parameters, using a reinforcement learning model to generate an optimized cost control strategy. The feedback module is used to apply the optimized cost control strategy to the next construction phase and monitor building material price fluctuations, design changes and construction delays in real time to achieve dynamic optimization control.

[0048] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0049] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for optimizing cost control strategies in engineering cost projects based on reinforcement learning, characterized in that, The method includes: Step 1: Obtain the initial budget, real-time construction environment status, and past cost control records for the project to build an initial cost control strategy; Step 2: Based on the initial cost control strategy, implement resource allocation and cost control in the current construction phase to obtain actual cost, schedule, and quality data; Step 3: Integrate actual cost, schedule and quality data to construct a multi-dimensional data distribution model and form an abstract decision space; fit discrete data into a continuous decision surface, calculate the curvature of local regions, and analyze the structural change characteristics of the decision surface; divide regions within the abstract decision space based on structural change characteristics, map the original data to the corresponding partitions, extract the inherent characteristics of each partition, and generate strategy correction parameters. Step 4: Based on the strategy correction parameters, and combined with actual cost, schedule and quality data, normalize the multi-dimensional data points and map them to a unit hypersphere; quantify the difference by calculating the shortest arc length between data points, and perform difference analysis and multi-objective collaborative evaluation to obtain the comprehensive cost control evaluation result; Step 5: Based on the comprehensive cost control evaluation results and strategy correction parameters, optimize the current strategy using a reinforcement learning model to generate an optimized cost control strategy. Step 6: Apply the optimized cost control strategy to the next construction phase and monitor material price fluctuations, design changes, and construction delays in real time to achieve dynamic optimization control.

2. The method for optimizing cost control strategies for engineering cost projects based on reinforcement learning according to claim 1, characterized in that, Based on the initial cost control strategy, resource allocation and cost control are implemented in the current construction phase to obtain actual cost, schedule, and quality data, including: Based on the initial cost control strategy, the specific types, quantities, and time requirements of labor, materials, and machinery resources for the current construction phase are obtained, and an initial resource allocation plan is formed. Based on the initial resource allocation plan, execute on-site resource scheduling and cost expenditure control, and generate resource execution records including actual resource consumption and corresponding cost expenditures; Based on resource execution records, real-time data is collected on actual cost flow, project progress completion, and key construction quality indicators during the construction process.

3. The method for optimizing cost control strategies for engineering cost projects based on reinforcement learning according to claim 2, characterized in that, By integrating actual cost, schedule, and quality data, a multi-dimensional data distribution model is constructed to form an abstract decision space, including: The actual cost data, schedule data and quality data are standardized, and the standardized three-dimensional data are integrated into a multi-dimensional vector in a unified format, where each dimension corresponds to a quantitative indicator of cost, schedule and quality. Multidimensional vectors are mapped to a pre-defined abstract decision space to form a set of data points representing the project status. The distribution pattern of the data point set in the decision space is analyzed to construct a multidimensional data distribution structure that reflects the dynamic relationship between the three elements of cost, schedule, and quality. Based on the multi-dimensional data distribution structure, and utilizing the statistical distribution characteristics of the data point set, an abstract decision space defined by the multi-dimensional data distribution structure is ultimately generated.

4. The method for optimizing cost control strategies for engineering cost projects based on reinforcement learning according to claim 3, characterized in that, The abstract decision space is used to characterize the feasible decision domain of the three elements of project cost, schedule and quality under specific correlation constraints.

5. The method for optimizing cost control strategies for engineering cost projects based on reinforcement learning according to claim 4, characterized in that, Discrete data is fitted to a continuous decision surface, the curvature of local regions is calculated, and the structural variation characteristics of the decision surface are analyzed. Based on the structural variation characteristics, regions are divided within the abstract decision space, the original data is mapped to the corresponding partitions, the intrinsic features of each partition are extracted, and policy correction parameters are generated, including: Based on discrete data points of cost, schedule and quality indicators distributed within an abstract decision space, a surface fitting algorithm is used to fit the discrete data points to form a continuous decision surface. Calculate the geometric curvature at various local locations on the decision surface, and identify regions of abrupt structural changes and smooth areas of curvature based on the numerical distribution of the geometric curvature. Based on the natural boundaries between regions of structural abrupt change and regions of gradual change, the abstract decision space is divided into several partitions with different structural characteristics. The original data points, in the form of multidimensional vectors based on the abstract decision space, are mapped one by one to the corresponding partitions according to the spatial coordinates of each data point. Analyze all data points mapped within each partition, identify the correlation patterns between cost, schedule, and quality reflected by the data points, and extract the unique statistical characteristics of data distribution and the dominant change patterns within each partition; Based on the statistical characteristics of data distribution and the dominant change patterns, targeted adjustment parameters for the strategy are generated.

6. The method for optimizing cost control strategies for engineering cost projects based on reinforcement learning according to claim 5, characterized in that, Based on the strategy correction parameters, and combined with actual cost, schedule, and quality data, multidimensional data points are normalized and mapped to a unit hypersphere, including: Acquire the actual cost, schedule and quality data of the project within a preset period, align and recombine them according to the time series to form new multidimensional data points with each monitoring time point as the unit; Based on the feature partitioning adjustment coefficients defined in the strategy correction parameters, the new multidimensional data points are normalized and all values ​​are converted to a unified scaling range to generate standard multidimensional data points. Standard multidimensional data points are mapped to a high-dimensional feature space through a preset coordinate transformation relationship, so that each standard multidimensional data point is projected and positioned on the surface of a unit hypersphere in the high-dimensional space.

7. The method for optimizing cost control strategies for engineering cost projects based on reinforcement learning according to claim 6, characterized in that, By calculating the shortest arc length between data points to quantify the difference, and performing difference analysis and multi-objective collaborative evaluation, a comprehensive cost control evaluation result is obtained, including: On the surface of the unit hypersphere, for the current data point representing the actual execution state of the current cycle, select the reference data point representing the target or benchmark state, calculate the shortest arc length on the surface of the hypersphere, and use the geometric distance value as a quantitative measure of the difference between the current state and the target state in multidimensional space. Dimensional analysis is performed on the quantitative difference measurement to determine the specific direction and magnitude of deviation between the current execution effect and the expected goal in the three main dimensions of cost, schedule and quality, so as to obtain the dimensional deviation analysis results. Based on the preset coordination targets of cost, schedule, and quality, the deviation of each dimension is comprehensively calculated and evaluated to assess the overall coordination performance under the constraints of multiple targets of cost, schedule, and quality, and a multi-target coordination evaluation value is obtained. The results of dimensional deviation analysis are integrated with the multi-objective collaborative evaluation values ​​for calculation and judgment, and finally a comprehensive cost control evaluation result is generated.

8. The method for optimizing cost control strategies for engineering cost projects based on reinforcement learning according to claim 7, characterized in that, Based on the comprehensive cost control evaluation results and strategy correction parameters, a reinforcement learning model is used to optimize the current strategy, generating an optimized cost control strategy, including: The comprehensive cost control assessment results are converted into penalty signals; at the same time, the strategy correction parameters are integrated with the current environmental status information of the project to form a comprehensive status feature vector. The penalty signal and the comprehensive state feature vector are input into a pre-trained reinforcement learning model. The comprehensive state feature vector is calculated through the policy network to generate the decision probability distribution in the current state. Based on the penalty signal and the preset policy gradient update algorithm, the internal parameters of the policy network are adjusted to generate an updated policy network parameter set. The updated policy network responds to state features to generate an optimized cost control policy.

9. The method for optimizing cost control strategies for engineering cost projects based on reinforcement learning according to claim 8, characterized in that, The optimized cost control strategy will be applied to the next construction phase, and information on building material price fluctuations, design changes, and schedule delays will be monitored in real time to achieve dynamic optimization control, including: In the next construction phase, the optimized cost control strategy will be implemented, and resource allocation and cost management will be guided by the optimized cost control strategy. During the implementation of the optimized cost control strategy, the internal and external environment of the project is monitored in real time to obtain dynamic information including building material market prices, design changes that have occurred, and deviations in actual construction progress. The dynamic information is converted into a feature vector representing the latest project environment state and input into the reinforcement learning model. Combined with the penalty signal regenerated according to the latest project environment state, the reinforcement learning model is driven to perform a new round of optimization and update of the policy network parameters. The updated strategy network generates cost control strategies suitable for the new environmental conditions, and the newly generated cost control strategies are applied to the current construction phase to achieve dynamic optimization control.

10. A reinforcement learning-based system for optimizing cost control strategies in engineering cost projects, wherein the system implements the method as described in any one of claims 1 to 9, characterized in that, include: The module is used to obtain the initial budget, real-time construction environment status and past cost control records of the project, and to build the initial cost control strategy. The execution module is used to perform resource allocation and cost control in the current construction phase according to the initial cost control strategy, and to obtain actual cost, schedule and quality data; The analysis module integrates actual cost, schedule, and quality data to construct a multi-dimensional data distribution model and form an abstract decision space. It fits discrete data into a continuous decision surface, calculates the curvature of local regions, and analyzes the structural change characteristics of the decision surface. Based on the structural change characteristics, it divides the region within the abstract decision space, maps the original data to the corresponding partition, extracts the intrinsic characteristics of each partition, and generates strategy correction parameters. The evaluation module is used to normalize and map multidimensional data points to a unit hypersphere based on strategy correction parameters and actual cost, schedule and quality data. The difference is quantified by calculating the shortest arc length between data points, and difference analysis and multi-objective collaborative evaluation are performed to obtain the comprehensive evaluation result of cost control. The optimization module is used to optimize the current strategy based on the comprehensive cost control evaluation results and strategy correction parameters, using a reinforcement learning model to generate an optimized cost control strategy. The feedback module is used to apply the optimized cost control strategy to the next construction phase and monitor building material price fluctuations, design changes and construction delays in real time to achieve dynamic optimization control.

Images