Engineering dynamic cost management method based on earned value and machine learning bias analysis

Abstract

The application discloses an engineering dynamic cost management method based on earned value and machine learning bias analysis and relates to the technical field of engineering dynamic cost management. The application obtains a planned work budget cost, a completed work budget cost and an actual generated cost based on an engineering building information model and construction site sensing data; then, an earned value bias index time sequence containing a cost performance index and a progress performance index is calculated, which is respectively sent to two modeling paths together with engineering structural features to obtain time sequence bias prediction values and structural bias prediction values; a dynamic performance correction factor is obtained through meta-learner stacking fusion to replace a fixed performance reciprocal factor; and a completion estimated value is calculated according to the actual generated cost, the dynamic performance correction factor and the difference between a total work budget value and the completed work budget cost, and a construction resource scheduling control signal is output. The application takes into account bias time sequence evolution and structural nonlinear influence, so that the scheduling signal can be dynamically adjusted according to the engineering execution state.

Description

Technical Field

[0001] This application relates to the technical field of dynamic cost management for engineering projects, and in particular to a method for dynamic cost management for engineering projects based on earned value and machine learning bias analysis. Background Technology

[0002] In the field of construction cost management, especially in large-scale construction projects with long construction cycles, numerous professional systems, and complex execution environments, the level of cost deviation control throughout the entire life cycle directly affects the efficiency of investment utilization and the rationality of construction resource allocation. Earned value management (EVM), as a classic method for measuring the joint deviation of cost and schedule, is widely used in the dynamic cost control of construction projects. It quantifies the project execution status through cost performance indices and schedule performance indices, and extrapolates the estimated value of completion from these indices for subsequent construction resource allocation and investment control.

[0003] In related technologies, the classic earned value formula is usually used to extrapolate the estimated value of completion. That is, based on the actual costs incurred, the estimated value of completion is obtained by multiplying the difference between the total budgeted value of completion and the budgeted cost of the work completed by a preset performance adjustment factor.

[0004] However, in actual engineering applications, there is often a large deviation between the estimated completion value and the final actual completion cost of the project, and this deviation tends to accumulate and increase in the later stages of project execution; the construction resource scheduling instructions generated accordingly are difficult to match with the actual working conditions of the project, thus affecting the effectiveness of investment control. Summary of the Invention

[0005] This application provides a dynamic cost management method for engineering projects based on earned value and machine learning deviation analysis, which is used to improve the consistency between the estimated value of completion and the actual execution status of the project within the earned value management framework.

[0006] In a first aspect, this application provides a method for dynamic cost management of engineering projects based on earned value and machine learning bias analysis, the method comprising: Based on the engineering building information model and construction site perception data, the planned work budget cost, the completed work budget cost and the actual cost incurred are obtained. Based on the planned work budget cost, the completed work budget cost, and the actual cost incurred, a time series of earned value deviation index, including cost performance index and schedule performance index, is calculated. The earned value deviation index time series and engineering structure characteristics are modeled separately to obtain the time series deviation prediction value and the structure deviation prediction value; Based on the meta-learner, the predicted values ​​of the time series deviation and the predicted values ​​of the structural deviation are stacked and fused to obtain the dynamic performance correction factor. Based on the actual costs incurred, the dynamic performance correction factor, and the difference between the total budget for completion and the budget cost of the completed work, a completion estimate is calculated, and a construction resource scheduling control signal is output based on the completion estimate.

[0007] In the above embodiments, this application first obtains the planned work budget cost, the completed work budget cost, and the actual cost based on the engineering building information model and construction site perception data, thus constructing the data foundation for earned value analysis. Then, it calculates the earned value deviation index time series, which includes the cost performance index and the schedule performance index, and feeds this time series and the engineering structured features into two independent modeling paths to obtain the time series deviation prediction value and the structure deviation prediction value. Subsequently, a meta-learner stacks and fuses the two prediction values ​​to obtain a dynamic performance correction factor, replacing the original fixed performance inverse factor. Finally, it combines the actual cost, the dynamic performance correction factor, and the difference between the total completed budget and the completed work budget cost to calculate the estimated completion value and output a construction resource scheduling control signal. This application not only improves the ability of the estimated completion value to jointly capture the time series evolution of deviations and the structural nonlinear effects, but also enables the output construction resource scheduling control signal to dynamically adjust according to the project execution status, improving the responsiveness of the cost control closed loop.

[0008] In some embodiments, obtaining the planned work budget cost based on the engineering building information model and construction site perception data includes: Based on the aforementioned construction information model, the quantities of work are obtained; Construct a bidirectional mapping matrix between work breakdown structure nodes and cost breakdown structure items, and retrieve the comprehensive unit price of the bill of quantities from the pre-set cost sub-library; wherein, the pre-set cost sub-library includes a cleanroom grade unit cost sub-library, a radiation protection material cost sub-library, a medical gas terminal cost sub-library, a large medical equipment supporting civil engineering renovation cost sub-library, and a decoration material substitution cost sub-library under infection control requirements. Based on the project quantity, the comprehensive unit price in the bill of quantities, and the two-way mapping matrix, the planned work budget cost is obtained.

[0009] In the above embodiments, this application first extracts the engineering quantities based on the engineering building information model, then constructs a two-way mapping matrix between the work breakdown structure nodes and the cost breakdown structure items, and obtains the comprehensive unit price of the bill of quantities from the cleanroom level unit cost sub-library, the radiation protection material cost sub-library, the medical gas terminal cost sub-library, the large medical equipment supporting civil engineering renovation cost sub-library, and the infection control decoration material substitution cost sub-library. Then, the planned work budget cost is obtained by aggregating and summarizing according to the mapping matrix; this enables the reuse and retrieval of special cost data and improves the consistency of the planned work budget cost preparation.

[0010] In some embodiments, the work breakdown structure is hierarchically classified by unit, floor, functional zone, professional system, and process. The functional zone includes at least one of the following: operating room, intensive care unit, interventional surgery unit, nuclear medicine unit, supply unit, and clean ward unit. The process of obtaining the planned work budget cost based on the project quantity, the comprehensive unit price in the bill of quantities, and the two-way mapping matrix includes: Obtain correction coefficients that characterize the reduced efficiency of non-stop construction and the additional requirements for infection control; The planned work budget cost is obtained based on the project quantity, the comprehensive unit price in the bill of quantities, the two-way mapping matrix, and the correction coefficient.

[0011] In the above embodiments, this application classifies the work breakdown structure into five levels: individual units, floors, functional zones, professional systems, and processes. The functional zones are further refined to units such as operating rooms, intensive care units, interventional surgery units, nuclear medicine units, supply units, and clean wards. Correction coefficients characterizing the reduced efficiency of non-stop construction and the additional requirements for infection control are obtained. These correction coefficients, along with the quantities, the comprehensive unit price in the bill of quantities, and the two-way mapping matrix, are used in the calculation of the planned work budget cost. This allows the planned work budget cost to reflect the scenario characteristics of medical engineering projects and reduces the baseline deviation caused by the omission of scenario-specific cost items.

[0012] In some embodiments, modeling the time series of the earned value deviation index to obtain a predicted time series deviation value includes: Obtain exogenous disturbance variables, which include at least two of the following: material price index, climate impact coefficient, and labor supply index; The earned value deviation index time series and the exogenous disturbance variable are input together into a long short-term memory network to obtain the time series deviation prediction value.

[0013] In the above embodiments, the time-series prediction method that relies solely on the earned value index (EVI) sequence itself is insufficient to reflect the driving effect of the external environment on the evolution of cost deviations. For example, during a period of material price increases or work stoppages caused by extreme weather, the EVI sequence itself may not show abnormalities, but potential deviations are already accumulating. This application introduces at least two exogenous disturbance variables, including the material price index, climate impact coefficient, and labor supply index, into the time-series modeling stage. These variables, along with the EVI time series, are input into a long short-term memory network for modeling to obtain the predicted time-series deviation value. This not only enables the time-series channel to perceive the driving effect of external factors on cost deviations in advance but also improves the robustness of the predicted time-series deviation value under scenarios of price fluctuations and tight labor supply.

[0014] In some embodiments, a gradient boosting decision tree is used to model the structural features of an engineering project to obtain predicted values ​​of structural deviations.

[0015] In the above embodiments, this application uses a gradient boosting decision tree to model the structured features of the engineering project, and gradually compresses the residuals through iterative fitting of multiple rounds of weak learners to obtain the predicted value of the structural deviation; thus improving the ability of the predicted value of the structural deviation to express the coupling of high-dimensional nonlinear features.

[0016] In some embodiments, the engineering structure features include: Cleanliness level parameters and radiation protection level parameters characterizing the physical properties of engineering projects; The cumulative change rate and design refinement rate characterize the evolution of the design process; Historical performance values ​​of construction contractors and frequency of user intervention are used to characterize the intensity of external disturbances.

[0017] In the above embodiments, this application refines the structural features of the project into three dimensions: cleanliness level parameters and radiation protection level parameters that characterize physical properties; cumulative change rate and design deepening rate that characterize the evolution of the design process; and historical performance value of the construction contractor and frequency of user intervention that characterize the intensity of external interference. This improves the adaptability of the predicted structural deviation value to different engineering deviation scenarios.

[0018] In some embodiments, after obtaining the predicted values ​​of time series deviation and structural deviation, the method further includes: Based on the Shapley additive interpretation value, the contribution of each dimension of the engineering structured features to the predicted value of the structural deviation is calculated, and a preset number of deviation causal features are determined in descending order of the contribution. Using the aforementioned deviation cause characteristics as query nodes, corrective measures are retrieved from a pre-set knowledge graph of deviation cause-corrective measures association. The step of outputting a construction resource scheduling control signal based on the estimated completion value includes: outputting a construction resource scheduling control signal based on the estimated completion value and the corrective measures.

[0019] In the above embodiments, this application calculates the contribution of each dimension of the structural features in the engineering structured features to the predicted value of structural deviation based on the Shapley additive interpretation value, determines a preset number of deviation cause features in descending order of contribution, and uses them as query nodes to retrieve corrective measures from a preset knowledge graph of deviation cause-corrective measures association. Then, the corrective measures and the estimated completion value are used together as the basis for generating the construction resource scheduling control signal; this expands the construction resource scheduling control signal from a single numerical warning to an executable handling guide, improving the connection capability of the cost control closed loop.

[0020] In a second aspect, embodiments of this application provide an engineering dynamic cost management system based on earned value and machine learning bias analysis, comprising: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a system, cause the system to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, causes the processor to perform a method as described in any possible implementation of the first aspect.

[0023] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. This application uses parallel modeling of the time-series channel and the structural channel, and then uses a meta-learner to stack and fuse the outputs of the two channels to obtain a dynamic performance correction factor, which replaces the performance inverse factor with a fixed structure in the traditional earned value formula. This transforms the extrapolation relationship of the estimated value of completion from a linear mapping to a nonlinear mapping that dynamically adjusts with the project execution status. This improves the ability of the estimated value of completion to simultaneously reflect the nonlinear coupling effect of the time-series evolution of deviations and structural characteristics, and also makes the mid-to-late stage cost prediction more responsive to changes in project status. 2. In this application, exogenous disturbance variables such as material price index, climate impact coefficient, and labor supply index are fed together with the earned value deviation index time series into a long short-term memory network for modeling in the time series channel. This enables the time series deviation prediction value to capture the potential driving force of external environment on cost deviation before the earned value series itself shows abnormality, thus improving the robustness of time series prediction in scenarios such as price fluctuations, extreme weather and tight labor supply. 3. This application quantifies the contribution of structured features by using Shapley additive interpretation values. The deviation cause features selected in descending order of contribution drive the retrieval of corrective measures in the associated knowledge graph. The corrective measures and the estimated completion value are then mapped to the construction resource scheduling and control signal, which expands the cost early warning signal from a simple numerical alarm to a scheduling instruction that can be traced back to the cause and can be executed, thus improving the efficiency of the connection between various links in cost control. Attached Figure Description

[0024] Figure 1This is a flowchart illustrating a dynamic cost management method for engineering projects based on earned value and machine learning bias analysis in an embodiment of this application. Figure 2 This is a schematic diagram of a cost acquisition process in an embodiment of this application; Figure 3 This is a schematic diagram of a process for stacking and fusing meta-learners in an embodiment of this application; Figure 4 This is a schematic diagram of the physical device structure of an engineering dynamic cost management system based on earned value and machine learning deviation analysis in the embodiments of this application. Detailed Implementation

[0025] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0026] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0027] To facilitate understanding, the application scenarios of the embodiments of this application are described below.

[0028] In the field of dynamic cost management for engineering projects, with numerous professional systems, long execution cycles, and diverse external environmental disturbances in large-scale construction projects, the level of cost deviation control throughout the entire life cycle is directly related to the efficiency of investment utilization and the rationality of construction resource allocation.

[0029] In related technologies, earned value management methods are typically used in conjunction with a fixed performance inverse factor to linearly extrapolate the estimated cost of completion. However, this approach struggles to simultaneously characterize the evolution of cost deviations over time and the nonlinear coupling effects of structural features such as engineering physical properties, design process evolution, and external interference on the deviations. Once changes accumulate or external price fluctuations occur during the project execution phase, the estimated cost of completion deviates significantly from the actual cost of completion, making it difficult for downstream construction resource scheduling instructions to match the actual working conditions.

[0030] This application is primarily applied to cost management scenarios in engineering projects, such as medical buildings, where numerous professional systems exist and the execution environment is sensitive. In these scenarios, special requirements such as infection control, cleanliness levels, and radiation protection further exacerbate the high-dimensional nonlinear characteristics of cost deviations. To address these technical problems, this application provides a dynamic cost management method for engineering projects based on earned value and machine learning deviation analysis. The following description, using an embodiment and in conjunction with the accompanying drawings, illustrates a dynamic cost management method for engineering projects based on earned value and machine learning deviation analysis in this application.

[0031] Please see Figure 1 The method provided in this application includes the following steps: S101. Based on the engineering building information model and construction site perception data, obtain the planned work budget cost, the completed work budget cost, and the actual cost incurred.

[0032] Among them, the construction information model refers to a three-dimensional digital model that carries information on the geometric components, material properties and bill of quantities of the project; the construction site perception data refers to the original progress and cost data collected by the site progress reporting terminal, UAV oblique photography device, laser scanning device and infrastructure finance and supplier settlement interface.

[0033] Specifically, refer to Figure 2 The system first reads the bill of quantities for each component from the building information model to obtain the planned quantities for the current period. Then, it aggregates the budgeted costs of the planned work according to the correspondence between the work breakdown structure and the cost breakdown structure. Next, it uploads the measured quantities of the completed components through the on-site progress reporting terminal, and calculates the budgeted costs of the completed work based on the comprehensive unit price of the corresponding bill of quantities, combined with the verification results of UAV oblique photography and laser scanning. Finally, it retrieves the contract payments, material purchases, labor and machinery costs incurred in this period from the infrastructure finance system and the supplier settlement interface, and summarizes them to obtain the actual costs incurred.

[0034] It should be noted that the generation method of the planned work budget cost includes: First, the system obtains the quantity of work based on the component family attributes of the building information model; then, it constructs a two-way mapping matrix between work breakdown structure nodes and cost breakdown structure items. Each row in this two-way mapping matrix corresponds to a work breakdown structure node, and each column corresponds to a cost breakdown structure item. The matrix elements represent the proportion of the quantity of work collected under that item for that node; next, it retrieves the comprehensive unit price of the bill of quantities from the pre-set cost sub-library. This pre-set cost sub-library includes a cleanroom grade unit cost sub-library, a radiation protection material cost sub-library, a medical gas terminal cost sub-library, a large medical equipment supporting civil engineering renovation cost sub-library, and a decoration material substitution cost sub-library under infection control requirements. Each sub-library stores the corresponding comprehensive unit price using functional unit type, material specifications, and cleanroom or protection level as index keys; finally, the planned work budget cost is obtained by weighting the quantity of work, the comprehensive unit price of the bill of quantities, and the two-way mapping matrix by item.

[0035] Furthermore, for medical engineering scenarios, the work breakdown structure is classified into five levels: individual units, floors, functional zones, professional systems, and processes. The functional zones include at least one of the following: operating room, intensive care unit, interventional surgery unit, nuclear medicine unit, supply unit, and clean ward unit. After obtaining the quantities, the comprehensive unit price of the bill of quantities, and the two-way mapping matrix, the system further obtains correction coefficients that characterize the efficiency reduction and infection control additional requirements of non-stop construction. These correction coefficients can be obtained by reverse statistical analysis of the proportion of reduced working hours and the proportion of infection control additional costs in similar historical projects, and their values ​​are usually between 1.05 and 1.30. Then, the planned work budget cost is obtained by aggregating according to the formula BCWS=ΣQi·Pi·αi, where Qi is the planned quantity of the i-th work, Pi is the comprehensive unit price of the i-th work, and αi is the corresponding correction coefficient.

[0036] Optionally, in non-medical projects or implementation scenarios that do not involve the above-mentioned special scenario constraints, the pre-set cost sub-database can be replaced with a general building component comprehensive unit price database or a quarterly unit price database published by a local cost information station, and the correction coefficient can be set to 1.

[0037] S102. Based on the planned work budget cost, the completed work budget cost, and the actual cost incurred, calculate the earned value deviation index time series, which includes the cost performance index and the schedule performance index.

[0038] Among them, the cost performance index represents the ratio of the budgeted cost of completed work to the actual cost incurred, and the schedule performance index represents the ratio of the budgeted cost of completed work to the budgeted cost of planned work.

[0039] After obtaining three basic cost data points, the system performs this step to record the cost deviation as a quantitative index on a rolling time axis.

[0040] Specifically, the system first retrieves the planned work budget cost, completed work budget cost, and actual cost from the database according to a preset sampling period (e.g., daily granularity); then calculates the cost performance index and schedule performance index for the day according to CPI=BCWP / ACWP and SPI=BCWP / BCWS; next, it writes the daily index into the earned value deviation index sequence table using the timestamp as the index; finally, it extracts the index values ​​of the most recent N sampling periods using a sliding window method to form the earned value deviation index time series.

[0041] S103. Model the earned value deviation index time series and engineering structure characteristics respectively to obtain the predicted values ​​of time series deviation and structural deviation.

[0042] Among them, the time series deviation prediction value refers to the quantitative value of future periodic cost deviation obtained by extrapolation based on the earned value index sequence; the structural deviation prediction value refers to the quantitative value of completion deviation rate obtained by mapping the engineering structured features; the engineering structured features refer to the multi-dimensional numerical description of the physical properties of the project, the evolution of the design process, and the intensity of external disturbances.

[0043] Specifically, refer to Figure 3 The system first inputs the earned value deviation index time series into a pre-trained time series prediction sub-model to obtain the time series deviation prediction value; then it reads the engineering structured feature vector of the current project from the engineering data asset library; next, it inputs the structured feature vector into a pre-trained structure prediction sub-model to obtain the structure deviation prediction value; finally, it temporarily stores the two prediction values ​​in the form of key-value pairs in the fusion buffer for subsequent meta-learner calls.

[0044] It should be noted that the time series prediction sub-model in this embodiment preferably adopts a Long Short-Term Memory (LSTM) network. Its network structure includes an input layer, two stacked LSM units, a fully connected layer, and an output layer. The time step of the input layer matches the sliding window length N. During the training phase, the earned value deviation index sequence of historical projects is used as the training sample, and the actual deviation amount of several subsequent periods is used as the label. The mean squared error loss function is used to iteratively update the network parameters until convergence. To enhance the ability to perceive external environmental disturbances, the system further acquires exogenous disturbance variables during time series modeling. These exogenous disturbance variables include at least two of the following: material price index, climate impact coefficient, and labor supply index. The material price index is retrieved monthly from the cost information release platform, the climate impact coefficient is normalized from the meteorological data interface based on the number of hours of work stoppage and reduced efficiency, and the labor supply index is normalized from the labor real-name system based on the number of days on duty. The system concatenates the earned value deviation index time series and the exogenous disturbance variables along the feature dimension and inputs them together into the LSM network to obtain the predicted time series deviation value.

[0045] In this embodiment, the structural prediction sub-model is preferably implemented using a gradient boosting decision tree, such as XGBoost or LightGBM. It uses an additive model with multiple rounds of weak learners to gradually fit the residuals. During the training phase, the structured feature vectors of historical projects are used as training samples, and the final completion deviation rate is used as the label. The second-order Taylor expansion loss function is used to iteratively optimize the leaf node weights until the specified number of trees is completed.

[0046] In this embodiment, the structural features of the project include cleanliness level parameters and radiation protection level parameters characterizing the physical properties of the project. The cleanliness level parameters are encoded as integers from Level I to Level IV, and the radiation protection level parameters are mapped to integer values ​​according to the lead equivalent grade of the shielding material. The features also include the cumulative change rate and design refinement rate characterizing the evolution of the design process. The cumulative change rate is the ratio of the amount of design changes to the amount of contract work, and the design refinement rate is the ratio of the number of refined design drawings to the total number of construction drawings. Furthermore, the features also include the historical performance value of the construction contractor and the frequency of user intervention in refining the design process, characterizing the intensity of external interference. The historical performance value of the construction contractor is obtained by reverse normalization of the contractor's cost deviation rate in similar past projects, and the frequency of user intervention in refining the design process is the number of times the user department raises refining opinions within a unit period.

[0047] In some embodiments, for baseline reconstruction events such as cleanliness level adjustment, radiation protection level upgrade, medical functional zoning adjustment, or significant additions or deletions to the cost breakdown structure set during the mid-term of project execution, if the earned value deviation index time series before and after the change is directly input into the long short-term memory network as a whole, the distribution drift of the time series deviation prediction value will be caused by the abrupt change in the statistical distribution before and after the sequence. To this end, the system monitors the version iteration logs of the Building Information Model (BIM) and the addition and deletion records of the cost breakdown structure (WIS) subject set. When the change in the total quantity of work or the weighted average of the comprehensive unit price in the bill of quantities exceeds a preset threshold within the same sampling period, a baseline reconstruction event is determined to have occurred in that sampling period, and a change breakpoint is marked on the earned value deviation index time series. Then, taking the change breakpoint as the boundary, the subsequences before and after the change breakpoint are normalized according to the mean and variance within their respective windows, and an event indicator bit is inserted at the concatenation position of the two normalized subsequences. The event indicator bit takes the first value in the sampling period before the change breakpoint and the second value in the sampling period after the change breakpoint. Finally, the earned value deviation index time series, exogenous disturbance variables, and event indicator bits, which have been segmented, normalized, and concatenated, are concatenated on the feature dimension and input together into the Long Short-Term Memory (LSTM) network to obtain the time series deviation prediction value. In this way, even if a major phase change occurs in the middle of the project execution, the Long Short-Term Memory Network can still explicitly perceive the location of the distribution breakpoint by using event indicator bits, so that the time series deviation prediction value can maintain the continuity across the change breakpoint; the dynamic performance correction factor obtained by the downstream meta-learner can also quickly follow the new execution status in the short period after the change, avoiding the deviation of the completion estimate from the old distribution inertia of the historical segment in the middle.

[0048] Optionally, the time-series prediction sub-model can also be replaced with a model that has sequence modeling capabilities, such as a gated recurrent unit network or a time-series convolutional network; the structural prediction sub-model can also be replaced with a model that has nonlinear fitting capabilities, such as a random forest, a deep neural network, or a support vector regression; the specific dimensions of the engineering structural features can be flexibly expanded or tailored according to the project type. For example, in non-medical projects, the cleanliness level parameters and radiation protection level parameters can be replaced with fire protection level parameters or seismic resistance level parameters. This application does not impose any restrictions on this.

[0049] S104. Based on the meta-learner, the predicted values ​​of time series deviation and structural deviation are stacked and fused to obtain the dynamic performance correction factor.

[0050] Among them, the meta-learner refers to the upper-level learner that takes the output of the base learner as input and further learns the optimal fusion relationship; stacked fusion refers to the integration method that takes the output of the base learner as the feature vector and maps it to the final output by the meta-learner; the dynamic performance correction factor refers to the nonlinear correction scalar that is dynamically adjusted according to the project execution status and is used to extrapolate the cost of the remaining workload.

[0051] Specifically, after the system reads the timing deviation prediction value and the structural deviation prediction value from the fusion buffer, it concatenates the two prediction values ​​into a two-dimensional input vector in a preset order; then it sends the input vector into a pre-trained meta-learner for forward inference; and the output layer of the meta-learner provides a dynamic performance correction factor.

[0052] It should be noted that in this embodiment, the meta-learner includes an input layer, a hidden layer, and an output layer. The input layer has 2 nodes (corresponding to two base predictions), and the output layer has 1 node (corresponding to the dynamic performance correction factor). The hidden layer uses the ReLU activation function, and the output layer uses Sigmoid followed by linear recalibration to ensure that the output is a positive real number. During the training phase, the predicted values ​​of the base learner on the historical project validation set are used as input samples, and the actual completion performance factor (i.e., the ratio of actual remaining cost to actual remaining budget) of the corresponding historical project is used as the label. The mean squared error loss function is used to iteratively optimize the network parameters. Compared with the fixed 1 / CPI or 1 / (CPI·SPI) correction factor in the traditional earned value formula, the dynamic performance correction factor can adaptively adjust the extrapolation ratio according to the current project's temporal evolution and structured feature state, thereby improving the consistency of the completion estimate with the actual project execution state.

[0053] In some embodiments, the predicted values ​​of time series deviation and the predicted values ​​of structural deviation may diverge—for example, the time series channel extrapolates the deviation based on the recent earned value deviation index time series and obtains a positive deviation, while the structural channel obtains a negative deviation based on the engineering structured feature mapping, or the two values ​​are in the same direction but have significantly different magnitudes. This application first standardizes the predicted values ​​of temporal deviation and structural deviation based on the output standard deviations of the temporal prediction sub-model and the structural prediction sub-model on their respective historical training sets, obtaining temporal standard scores and structural standard scores. Then, it calculates the absolute value of the difference between the temporal standard scores and the structural standard scores as the deviation index. When the deviation index does not exceed a preset threshold, the meta-learner outputs a dynamic performance correction factor according to the conventional stacked fusion path. When the deviation index exceeds the preset threshold, it determines that there is a mechanistic deviation between the two paths. Before splicing them into a two-dimensional input vector, a deviation indicator bit is added and made active. After receiving the active deviation indicator bit, the meta-learner switches to the conservative fusion branch, that is, it adds a constraint to its output layer that takes the more pessimistic side of the temporal deviation predicted value and the structural deviation predicted value as the lower limit of the dynamic performance correction factor, so as to avoid the dynamic performance correction factor being compressed by averaging. Correspondingly, during sampling periods when the deviation indicator is active, the construction resource scheduling control signal additionally carries a prediction uncertainty level marker. This allows the labor scheduling terminal, material procurement ordering interface, and machinery dispatching interface to take tiered responses based on the prediction uncertainty level. For example, under high uncertainty levels, centralized procurement of bulk materials can be postponed, and the validity period of labor scheduling instructions can be shortened. This avoids stacking and fusion distortion caused by the opposite output direction of the base learner, thereby improving the robustness of the cost control closed loop under conditions where the causes of deviations are unclear.

[0054] S105. Based on the actual costs incurred, the dynamic performance correction factor, and the difference between the total budget for completion and the budget cost of the completed work, calculate the estimated value for completion, and output the construction resource scheduling control signal based on the estimated value for completion.

[0055] Among them, the total budget for completion refers to the total budget agreed upon in the project approval baseline that covers the entire scope of work; the estimated cost of completion refers to the predicted total cost of the project at the current execution status; and the construction resource scheduling control signal refers to the structured electronic signal used to issue adjustment instructions to the labor scheduling terminal, the material procurement ordering interface, and the machinery and equipment dispatching interface.

[0056] Specifically, the system first reads the total budgeted value (BAC) for completion from the project baseline database and calculates the budgeted cost of the remaining work (BAC-BCWP). Then, it obtains the estimated value for completion using the formula EAC=ACWP+f_meta·(BAC-BCWP), where EAC is the estimated value for completion (RMB), ACWP is the actual cost incurred (RMB), f_meta is the dynamic performance correction factor, BAC is the total budgeted value for completion (RMB), and BCWP is the budgeted cost of the completed work (RMB). Next, the estimated value for completion is compared with the total budgeted value for completion. If the deviation exceeds a preset threshold range, a scheduling trigger flag is generated. Finally, the system arranges construction resource scheduling control signals according to the scheduling trigger flag and sends them to the corresponding execution terminals.

[0057] Furthermore, to ensure the interpretability and executability of construction resource scheduling control signals, after obtaining the predicted values ​​of time-series deviations and structural deviations, the system can also calculate the contribution of each dimension of the engineering structured features to the predicted structural deviations based on the Shapley additive interpretation value. Specifically: using the structural prediction sub-model as the model to be interpreted, the marginal contributions of each dimension of features under different feature subsets are iterated and weighted by the Shapley formula to obtain the Shapley additive interpretation value of each dimension of features; then, the Shapley additive interpretation values ​​are sorted in descending order, and the top K dimensions (K is a preset number, for example, K is 3) are taken as the deviation causal features; then... The deviation cause characteristics serve as query nodes, and corrective measures are retrieved from a pre-set deviation cause-corrective measure association knowledge graph. This association knowledge graph uses deviation cause characteristics as source nodes, attribution-recommendation relationships as edges, and corrective measures as terminal nodes. For example, if the query node is the price fluctuation of cleanroom air conditioning units, the system retrieves locked-price procurement or alternative brand solutions. If the query node is the frequent secondary refinement by the user department, the system retrieves the design freeze node. Finally, the system compiles construction resource scheduling and control signals based on the completion estimate and corrective measures, so that the scheduling and control signals carry specific handling guidelines while providing early warning levels.

[0058] Optionally, the system can also directly generate blue, yellow, and red three-level scheduling control signals based on the deviation between the estimated completion value and the preset threshold range; in addition to manual sorting, the construction method of the associated knowledge graph can also be automatically generated from the historical engineering rectification archives by rule mining or large model extraction, and this application does not restrict this.

[0059] In this embodiment, the dynamic performance correction factor generation method, which combines dual-channel parallel modeling and stacked meta-learners, replaces the fixed performance reciprocal factor in the traditional earned value formula. Furthermore, it introduces exogenous disturbance variables to participate in time-series modeling and uses Shapley additive explanatory values ​​to drive the retrieval of corrective measures. Therefore, the estimated completion value can simultaneously reflect the time-series evolution of cost deviation and the nonlinear coupling effect of structured characteristics. This effectively solves the technical problems of insufficient accuracy in the estimated completion value and lack of executable handling guidance for early warning signals in traditional earned value management methods, thereby realizing the dynamic matching between construction resource scheduling control signals and the actual execution status of the project.

[0060] The following describes the engineering dynamic cost management system based on earned value and machine learning bias analysis in this application embodiment from the perspective of hardware processing. Please refer to [link to relevant documentation]. Figure 4 This is a schematic diagram of the physical device structure of an engineering dynamic cost management system based on earned value and machine learning bias analysis in the embodiments of this application.

[0061] It should be noted that, Figure 4 The structure of the engineering dynamic cost management system based on earned value and machine learning bias analysis shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0062] like Figure 4 As shown, the system includes a CPU, which can perform various appropriate actions and processes based on a program stored in the ROM or a program loaded into the RAM from a storage portion, such as executing the methods described in the above embodiments. The RAM also stores various programs and data required for system operation. The CPU, ROM, and RAM are interconnected via a bus. I / O interfaces are also connected to the bus.

[0063] The following components are connected to the I / O interface: input sections including the Building Information Modeling (BIM) data access port, construction site progress reporting terminal, UAV oblique photography device, laser scanning device, and infrastructure finance and supplier settlement data access interface; output sections including Liquid Crystal Display (LCD) and speakers; storage sections including hard drives, which also store pre-set cost sub-libraries, deviation cause-correction measure association knowledge graphs, and model parameters of time series prediction sub-models, structural prediction sub-models, and meta-learners; and communication sections including network interface cards such as LAN (Local Area Network) cards and modems, which are also used to send construction resource scheduling control signals to labor dispatch terminals, material procurement order interfaces, and machinery dispatch interfaces. The communication section performs communication processing via a network such as the Internet. Drives are also connected to the I / O interface as needed. Removable media, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on the drive as needed so that computer programs read from them can be installed into the storage section as needed.

[0064] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by the CPU, it performs the various functions defined in this application.

[0065] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0066] Specifically, the system in this embodiment includes a processor and a memory. The memory stores a computer program, and when the computer program is executed by the processor, it implements the method provided in the above embodiment.

[0067] In another aspect, this application also provides a computer-readable storage medium, which may be included in the system described in the above embodiments; or it may exist independently and not assembled into the system. The storage medium carries one or more computer programs that, when executed by a processor of the system, cause the system to implement the methods provided in the above embodiments.

[0068] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0069] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

Claims

1. A dynamic cost management method for engineering projects based on earned value and machine learning bias analysis, characterized in that, The method includes: Based on the engineering building information model and construction site perception data, the planned work budget cost, the completed work budget cost and the actual cost incurred are obtained. Based on the planned work budget cost, the completed work budget cost, and the actual cost incurred, a time series of earned value deviation index, including cost performance index and schedule performance index, is calculated. The earned value deviation index time series and engineering structure characteristics are modeled separately to obtain the time series deviation prediction value and the structure deviation prediction value; Based on the meta-learner, the predicted values ​​of the time series deviation and the predicted values ​​of the structural deviation are stacked and fused to obtain the dynamic performance correction factor. Based on the actual costs incurred, the dynamic performance correction factor, and the difference between the total budget for completion and the budget cost of the completed work, a completion estimate is calculated, and a construction resource scheduling control signal is output based on the completion estimate.

2. The method according to claim 1, characterized in that, The planned work budget cost, obtained based on the engineering building information model and construction site perception data, includes: Based on the aforementioned construction information model, the quantities of work are obtained; Construct a bidirectional mapping matrix between work breakdown structure nodes and cost breakdown structure items, and retrieve the comprehensive unit price of the bill of quantities from the pre-set cost sub-library; wherein, the pre-set cost sub-library includes a cleanroom grade unit cost sub-library, a radiation protection material cost sub-library, a medical gas terminal cost sub-library, a large medical equipment supporting civil engineering renovation cost sub-library, and a decoration material substitution cost sub-library under infection control requirements. Based on the stated quantities of work, the comprehensive unit price in the bill of quantities, and the two-way mapping matrix, the planned work budget cost is obtained.

3. The method according to claim 2, characterized in that, The work breakdown structure is classified into units, floors, functional zones, professional systems, and procedures. The functional zones include at least one of the following: operating room, intensive care unit, interventional surgery unit, nuclear medicine unit, supply unit, and clean ward unit. The process of obtaining the planned work budget cost based on the project quantity, the comprehensive unit price in the bill of quantities, and the two-way mapping matrix includes: Obtain correction coefficients that characterize the reduced efficiency of non-stop construction and the additional requirements for infection control; The planned work budget cost is obtained based on the project quantity, the comprehensive unit price in the bill of quantities, the two-way mapping matrix, and the correction coefficient.

4. The method according to claim 1, characterized in that, Modeling the time series of the earned value deviation index yields predicted time series deviation values, including: Obtain exogenous disturbance variables, which include at least two of the following: material price index, climate impact coefficient, and labor supply index; The earned value deviation index time series and the exogenous disturbance variable are input together into a long short-term memory network to obtain the time series deviation prediction value.

5. The method according to claim 1, characterized in that, By using gradient boosting decision trees to model the structural features of the project, the predicted values ​​of structural deviations are obtained.

6. The method according to claim 1, characterized in that, The engineering structure features include: Cleanliness level parameters and radiation protection level parameters characterizing the physical properties of engineering projects; The cumulative change rate and design refinement rate characterize the evolution of the design process; Historical performance values ​​of construction contractors and frequency of user intervention are used to characterize the intensity of external disturbances.

7. The method according to claim 1, characterized in that, After obtaining the predicted values ​​of time series deviation and structural deviation, the method further includes: Based on the Shapley additive interpretation value, the contribution of each dimension of the engineering structured features to the predicted value of the structural deviation is calculated, and a preset number of deviation causal features are determined in descending order of the contribution. Using the aforementioned deviation cause characteristics as query nodes, corrective measures are retrieved from a pre-set knowledge graph of deviation cause-corrective measures association. The step of outputting a construction resource scheduling control signal based on the completion estimate includes: outputting a construction resource scheduling control signal based on the completion estimate and the corrective measures.

8. A dynamic engineering cost management system based on earned value and machine learning bias analysis, characterized in that, include: One or more processors and memory; The memory is coupled to the one or more processors, the memory being used to store computer program code, the computer program code including computer instructions, the one or more processors invoking the computer instructions to cause the system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instructions are executed on the system, the system performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, causes the processor to perform the method as described in any one of claims 1-7.

Images