Engineering cost control method and device based on deep learning

Abstract

The invention discloses an engineering cost control method and device based on deep learning. The method comprises the steps of obtaining historical cost data, current construction progress data and engineering budget data of an engineering project; based on the historical cost data, the current construction progress data and the project budget data, constructing a multi-dimensional state representation containing an accumulated cost deviation, a progress deviation and a budget constraint, and obtaining a current project cost control state vector; inputting the current engineering cost control state vector into a trained deep learning network, and outputting engineering regulation and control parameters; according to the engineering regulation and control parameters, the resource input amount and the construction rhythm of the engineering project are jointly adjusted, and a construction adjustment plan is generated; and determining a current cost control execution scheme of the engineering project based on the construction adjustment plan. According to the invention, the accuracy of engineering cost regulation and control can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an engineering cost control method and apparatus based on deep learning. Background Technology

[0002] Engineering cost control is a core management aspect that ensures high-quality completion of engineering projects within budget, directly impacting the project's economic benefits. Currently, the industry's commonly used engineering cost control methods primarily rely on experience-driven management models and prediction models based on traditional statistics. Specifically, the experience-driven model typically involves project managers using spreadsheets and other tools to perform cost accounting, variance analysis, and resource allocation based on historical project data and their own experience. This approach is highly subjective, inefficient, and struggles to cope with the dynamic coupling of multiple factors in complex engineering projects. While prediction models based on traditional statistics, such as regression analysis or time series models, can reveal some linear patterns, their prediction accuracy and decision-making effectiveness are often limited when dealing with the nonlinear, high-dimensional, and strong time-series dependencies between engineering costs, schedules, and resources, making it difficult to provide forward-looking and dynamic control strategies.

[0003] Therefore, existing engineering cost control methods suffer from inaccurate control effects. Summary of the Invention

[0004] This invention provides a method and apparatus for engineering cost control based on deep learning, which can effectively improve the accuracy of engineering cost control.

[0005] One embodiment of the present invention provides a deep learning-based engineering cost control method, comprising: Obtain historical cost data, current construction progress data, and project budget data for the engineering project; Based on the historical cost data, current construction progress data, and project budget data, a multi-dimensional state representation including cumulative cost deviation, schedule deviation, and budget constraints is constructed to obtain the current project cost control state vector. The current engineering cost control state vector is input into the trained deep learning network, which outputs engineering control parameters. Based on the engineering control parameters, the resource input and construction pace of the engineering project are jointly adjusted to generate a construction adjustment plan. Based on the aforementioned construction adjustment plan, the current cost control implementation scheme for the project is determined.

[0006] As an improvement to the above scheme, the step of constructing a multi-dimensional state representation based on the historical cost data, current construction progress data, and project budget data, including cumulative cost deviation, schedule deviation, and budget constraints, to obtain the current project cost control state vector, includes the following sub-steps: Based on the historical cost data and the project budget data, the cumulative cost deviation is calculated to obtain the cost deviation value; Based on the current construction progress data and the predefined planned progress data, the progress deviation is calculated to obtain the progress deviation value; Based on the project budget data, budget constraints are extracted to obtain the budget constraint vector; The cost deviation value, the schedule deviation value, and the budget constraint vector are concatenated and normalized to obtain the current project cost control status vector.

[0007] As an improvement to the above scheme, the step of inputting the current engineering cost control state vector into a trained deep learning network and outputting engineering control parameters includes the following sub-steps: The current engineering cost control state vector is input into the input layer of the trained deep learning network to perform feature transformation and obtain the hidden feature representation. The hidden feature representation is nonlinearly transformed through multiple hidden layers of the deep learning network to obtain a high-level feature representation; The high-level feature representation is passed through the output layer of the deep learning network and mapped to the engineering control parameters.

[0008] As an improvement to the above solution, the step of jointly adjusting the resource input and construction rhythm of the project based on the engineering control parameters to generate a construction adjustment plan includes the following sub-steps: The engineering control parameters were analyzed to obtain the resource input adjustment amount and the construction rhythm adjustment amount. Based on the aforementioned resource input adjustment amount, the adjustment and allocation scheme for each resource type of the project is determined, and the resource adjustment scheme is obtained; Based on the aforementioned construction rhythm adjustment amount, a construction procedure time adjustment plan for the project is determined, resulting in a schedule adjustment plan. The resource adjustment plan and the schedule adjustment plan are integrated to generate the construction adjustment plan.

[0009] As an improvement to the above solution, determining the current cost control implementation plan for the project based on the construction adjustment plan includes the following sub-steps: The construction adjustment plan is broken down into execution units corresponding to each sub-project, and the resource input and construction start and end time of each execution unit are clearly defined. Based on the resource input and construction start and end time of each execution unit, and in conjunction with the project budget data, set the cost consumption limit and progress assessment node for each execution unit; By integrating the cost consumption caps, progress assessment nodes, and corresponding resource adjustment plans and progress adjustment plans of all execution units, and clarifying the responsible parties of each execution unit, the current cost control execution plan is obtained.

[0010] Another embodiment of the present invention provides an engineering cost control device based on deep learning, comprising: The acquisition module is used to acquire historical cost data, current construction progress data, and project budget data for engineering projects. The construction module is used to construct a multi-dimensional state representation containing cumulative cost deviation, schedule deviation and budget constraints based on the historical cost data, current construction progress data and project budget data, so as to obtain the current project cost control state vector; The output module is used to input the current engineering cost control state vector into the trained deep learning network and output engineering control parameters. The generation module is used to jointly adjust the resource input and construction rhythm of the project based on the engineering control parameters, and generate a construction adjustment plan. The determination module is used to determine the current cost control implementation plan for the project based on the construction adjustment plan.

[0011] As an improvement to the above solution, the building module is specifically used for: Based on the historical cost data and the project budget data, the cumulative cost deviation is calculated to obtain the cost deviation value; Based on the current construction progress data and the predefined planned progress data, the progress deviation is calculated to obtain the progress deviation value; Based on the project budget data, budget constraints are extracted to obtain the budget constraint vector; The cost deviation value, the schedule deviation value, and the budget constraint vector are concatenated and normalized to obtain the current project cost control status vector.

[0012] As an improvement to the above solution, the output module is specifically used for: The current engineering cost control state vector is input into the input layer of the trained deep learning network to perform feature transformation and obtain the hidden feature representation. The hidden feature representation is nonlinearly transformed through multiple hidden layers of the deep learning network to obtain a high-level feature representation; The high-level feature representation is passed through the output layer of the deep learning network and mapped to the engineering control parameters.

[0013] As an improvement to the above solution, the generation module is specifically used for: The engineering control parameters were analyzed to obtain the resource input adjustment amount and the construction rhythm adjustment amount. Based on the aforementioned resource input adjustment amount, the adjustment and allocation scheme for each resource type of the project is determined, and the resource adjustment scheme is obtained; Based on the aforementioned construction rhythm adjustment amount, a construction procedure time adjustment plan for the project is determined, resulting in a schedule adjustment plan. The resource adjustment plan and the schedule adjustment plan are integrated to generate the construction adjustment plan.

[0014] As an improvement to the above solution, the determining module is specifically used for: The construction adjustment plan is broken down into execution units corresponding to each sub-project, and the resource input and construction start and end time of each execution unit are clearly defined. Based on the resource input and construction start and end time of each execution unit, and in conjunction with the project budget data, set the cost consumption limit and progress assessment node for each execution unit; By integrating the cost consumption caps, progress assessment nodes, and corresponding resource adjustment plans and progress adjustment plans of all execution units, and clarifying the responsible parties of each execution unit, the current cost control execution plan is obtained.

[0015] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: First, historical cost data, current construction progress data, and project budget data are acquired to provide data support for subsequently characterizing the project cost status. Then, based on these three types of data, a multi-dimensional state representation including cumulative cost deviation, schedule deviation, and budget constraints is constructed to obtain the current project cost control state vector. This state vector can simultaneously cover historical cost deviations, current schedule differences, and budget limitations, achieving a centralized presentation of multi-dimensional project information. Next, this state vector is input into a trained deep learning network. Leveraging the deep learning network's ability to process multi-dimensional nonlinear correlated data, it outputs project control parameters that match the current project status. Subsequently, based on these control parameters, the project's resource input and construction rhythm are jointly adjusted to avoid adaptation conflicts arising from individual adjustments to resource input and construction rhythm, generating a construction adjustment plan. Finally, based on this construction adjustment plan, the current cost control execution scheme is determined, transforming control measures into implementable content. Therefore, the embodiments of the present invention integrate historical cost data, current construction progress data, and project budget data to form a multi-dimensional state representation, and utilize the project control parameters output by the deep learning network to achieve joint adjustment of resource input and construction rhythm. This solves the problem that existing project cost control methods are unable to integrate multi-dimensional data to form a device state representation and cannot coordinate the control of resources and construction rhythm, thereby improving the adaptability of cost control schemes to actual construction needs and effectively improving the accuracy of project cost control. Attached Figure Description

[0016] Figure 1 is a flowchart illustrating an engineering cost control method based on deep learning according to an embodiment of the present invention. Figure 2 is a schematic diagram of the structure of an engineering cost control device based on deep learning provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Referring to Figure 1, it is a flowchart illustrating an engineering cost control method based on deep learning according to an embodiment of the present invention. The engineering cost control method based on deep learning includes: S10: Obtain historical cost data, current construction progress data, and project budget data for the engineering project; S11. Based on the historical cost data, current construction progress data and project budget data, construct a multi-dimensional state representation that includes cumulative cost deviation, schedule deviation and budget constraints, and obtain the current project cost control state vector. S12, input the current engineering cost control state vector into the trained deep learning network, and output the engineering control parameters; S13, Based on the engineering control parameters, the resource input and construction rhythm of the engineering project are jointly adjusted to generate a construction adjustment plan; S14. Based on the construction adjustment plan, determine the current cost control implementation plan for the project.

[0019] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: First, historical cost data, current construction progress data, and project budget data are acquired to provide data support for subsequently characterizing the project cost status. Then, based on these three types of data, a multi-dimensional state representation including cumulative cost deviation, schedule deviation, and budget constraints is constructed to obtain the current project cost control state vector. This state vector can simultaneously cover historical cost deviations, current schedule differences, and budget limitations, achieving a centralized presentation of multi-dimensional project information. Next, this state vector is input into a trained deep learning network. Leveraging the deep learning network's ability to process multi-dimensional nonlinear correlated data, it outputs project control parameters that match the current project status. Subsequently, based on these control parameters, the project's resource input and construction rhythm are jointly adjusted to avoid adaptation conflicts arising from individual adjustments to resource input and construction rhythm, generating a construction adjustment plan. Finally, based on this construction adjustment plan, the current cost control execution scheme is determined, transforming control measures into implementable content. Therefore, the embodiments of the present invention integrate historical cost data, current construction progress data, and project budget data to form a multi-dimensional state representation, and utilize the project control parameters output by the deep learning network to achieve joint adjustment of resource input and construction rhythm. This solves the problem that existing project cost control methods are unable to integrate multi-dimensional data to form a device state representation and cannot coordinate the control of resources and construction rhythm, thereby improving the adaptability of cost control schemes to actual construction needs and effectively improving the accuracy of project cost control.

[0020] As an example, in this embodiment, historical cost data is preferentially extracted from the enterprise project management ERP device or historical project archive, focusing on completed project data of the same type as the current project (such as residential and municipal engineering), specifically including details such as labor (working hours and unit prices for each type of work), materials (purchase quantity and unit price of main / auxiliary materials), machinery (equipment rental / usage fees), and management fees. At the same time, project data without abnormal interference (such as extreme material price increases or policy changes) in the past 3-5 years are screened, and outliers exceeding three times the standard deviation of the mean are removed to form a structured historical cost dataset. Current construction progress data is obtained through "real-time on-site collection + multi-party verification": actual working hours and completed work volume (such as concrete pouring volume and rebar binding tonnage) are collected through IoT devices on the construction site (such as tower crane load monitoring and worker electronic check-in devices); the completion status of the process (such as whether it has passed acceptance) is supplemented by the daily supervision report and the weekly construction team report, and the on-site construction images are compared with the progress report to correct data deviations (such as avoiding false reporting of work volume), and finally a real-time updated current construction progress data table is formed. The project budget data is extracted from the project's early approval documents and cost estimation software. It includes the total budget amount confirmed by the client, the budget percentage of each sub-item (such as foundation and main structure), the upper limit of resource unit price (such as the budget unit price of steel), and the fee rate standard (such as the management fee rate). At the same time, the budget data is split into "sub-item - cost item" and aligned with the classification dimensions of historical cost data and current progress data to ensure the consistency of subsequent data processing.

[0021] As an example, the construction of a multi-dimensional state representation including cumulative cost deviation, schedule deviation, and budget constraints based on the historical cost data, current construction progress data, and project budget data to obtain the current project cost control state vector includes the following sub-steps: Based on the historical cost data and the project budget data, the cumulative cost deviation is calculated to obtain the cost deviation value; Based on the current construction progress data and the predefined planned progress data, the progress deviation is calculated to obtain the progress deviation value; Based on the project budget data, budget constraints are extracted to obtain the budget constraint vector; The cost deviation value, the schedule deviation value, and the budget constraint vector are concatenated and normalized to obtain the current project cost control status vector.

[0022] In this embodiment, the cumulative cost deviation is first calculated based on historical cost data and project budget data. By integrating the current project's cumulative actual cost with the cumulative planned budget cost and the deviation experience of similar historical projects, the degree of deviation of the current project's cost from the budget is accurately reflected. Next, the schedule deviation is calculated based on the current construction progress data and predefined planned progress data. By introducing the importance weight of each process to weight and summarize the deviation of each process, the difference between the overall progress and the plan is accurately reflected. Then, key constraint elements such as the total budget limit, the budget ratio limit of sub-projects, and the unit price limit of major resources are extracted from the project budget data to form a budget constraint vector, which comprehensively covers the multi-dimensional restrictions of the budget on cost control. Finally, the cost deviation value, schedule deviation value and budget constraint vector are concatenated, and the data of each dimension are normalized according to the industry standard range to obtain the current project cost control status vector. This eliminates the difference in the dimensions of data of different dimensions, ensures that the data can be adapted to the feature processing of the subsequent deep learning network, and improves the accuracy of the output of subsequent project control parameters.

[0023] In this embodiment, a dynamic weighting factor is introduced to address the problem of excessive interference from long-term deviations on the current control state in traditional algorithms. Specifically, the actual costs (covering all resource consumption such as labor, materials, and machinery) for each accounting period from the start of the project to the current node are first extracted from historical cost data and recorded as the actual cost for each accounting period. ( (This is the accounting period number, such as by week or by month); then, the planned workload and budgeted cost for each accounting period are matched from the project budget data. (That is, the budget amount corresponding to the planned workload for this period). Then, an improved weighted cumulative algorithm is used to calculate the cumulative cost deviation, as shown in the formula: , This is the final cumulative cost deviation value. For the first The dynamic weighting factor for each accounting cycle (ranging from 0 to 1) is determined by the proportion of the workload in that cycle to the total project workload and the time urgency—the higher the proportion of workload and the closer the cycle is to the current milestone. The larger the value, the stronger the impact of recent key cycle cost fluctuations on the current deviation judgment, ensuring that the calculation results are more in line with the current cost control needs.

[0024] Secondly, the schedule deviation is calculated based on the current construction progress data and the predefined planned progress data to obtain the schedule deviation value. Traditional schedule deviation calculation only focuses on the difference in the amount of work, ignoring the impact of time urgency on cost control. This step improves upon this by incorporating the time dimension into the deviation calculation logic. In practice, the actual completed amount of each sub-item (such as foundation engineering, main structure engineering, etc.) is first counted from the current construction progress data, and then the sum is compared with the total project amount to obtain the current percentage of the total actual completed amount. Then, extract the planned quantities of each sub-item of the project that should be completed at the current node from the predefined planned schedule data, and similarly calculate the percentage of the total planned completed project quantity. At the same time, the time influence factor is introduced. (The value ranges from 1 to 2), and its value is dynamically adjusted according to the remaining proportion of the total project duration—the shorter the remaining duration, the better. The closer the value is to 2, the better it reflects the corrective effect of the urgency of the project on schedule deviations. The final formula for calculating schedule deviations is: , The resulting schedule deviation value quantifies the difference in the amount of work completed and reflects the potential impact of time factors on cost control, making the schedule representation more consistent with the actual control scenario of the project.

[0025] Next, budget constraints are extracted from the project budget data to obtain a budget constraint vector. This step requires extracting the core constraint information from the budget in a structured manner to ensure that the constraint vector comprehensively covers the boundary conditions of cost control. Specifically, first, the upper limit of the total project budget (i.e., the maximum total cost that can be invested in the entire project) is extracted from the project budget data as the first element of the constraint vector; then, according to the construction sequence of each sub-project, the budget proportion constraint of each sub-project is extracted sequentially (i.e., the cost expenditure of this sub-project must not exceed the limited proportion of the total budget, such as the main structure project not exceeding 45% of the total budget), as the middle element of the vector; finally, according to the construction stage (e.g., the preliminary preparation stage, the main construction stage, the completion and acceptance stage, etc.), the budget expenditure limit of each stage (i.e., the amount that the cumulative cost expenditure within this stage must not exceed) is extracted as the subsequent elements of the vector. Through the above structured extraction, the scattered budget constraint information is integrated into a budget constraint vector with clear dimensions, providing clear constraint dimension data for the subsequent construction of multi-dimensional state representation.

[0026] Finally, the cost deviation value, schedule deviation value, and budget constraint vector are concatenated and normalized to obtain the current project cost control status vector. During the concatenation process, the scalar form is first... and Each element is converted into a single-element vector, and then concatenated with the budget constraint vector in a fixed order of "cost variance - schedule variance - budget constraint" to form a multi-dimensional original state vector. Because the data units of different dimensions differ significantly (e.g., cost variance is monetary data, schedule variance is proportional data), differential normalization processing is required: and Normalization using mean-standard deviation is given by the following formula: (in This is the average of historical deviation data for similar engineering projects. (where the standard deviation is the historical data), mapping it to the interval [-1, 1]; maximum normalization is applied to the budget constraint vector, as shown in the formula. (in The historical maximum value corresponding to the constraint type (such as the historical maximum value of the total budget limit, the maximum limit value of the item proportion constraint, etc.) is mapped to the interval [0,1]. After concatenation and normalization, the resulting current engineering cost control state vector eliminates the difference in dimensions and retains the relative characteristics of each dimension of data. It can be directly used as input data for subsequent deep learning networks to ensure that the network can accurately capture the multidimensional state information of engineering cost control.

[0027] As an example of the above scheme, the step of inputting the current engineering cost control state vector into a trained deep learning network and outputting engineering control parameters includes the following sub-steps: The current engineering cost control state vector is input into the input layer of the trained deep learning network to perform feature transformation and obtain the hidden feature representation. The hidden feature representation is nonlinearly transformed through multiple hidden layers of the deep learning network to obtain a high-level feature representation; The high-level feature representation is passed through the output layer of the deep learning network and mapped to the engineering control parameters.

[0028] In this embodiment, the current engineering cost control state vector is first input into the input layer of a trained deep learning network for feature transformation. Through linear mapping, multi-dimensional state information is converted into a hidden feature representation that the network can process, achieving adaptation of the original state data to the feature space. Next, the hidden feature representation undergoes nonlinear transformation through multiple hidden layers. A hierarchical structure containing activation functions is used to capture the nonlinear correlation and deep interaction between cost deviations, schedule deviations, and budget constraints, resulting in a high-level feature representation that reflects the essential characteristics of the state. Then, this high-level feature representation is mapped to engineering control parameters through the output layer. Linear transformation and dimensional adjustment ensure that the output parameters directly correspond to the adjustment needs of resource input and construction rhythm. Therefore, this embodiment, through hierarchical feature processing and mapping of the deep learning network, achieves accurate conversion from engineering cost control state to control parameters, improving the adaptability of control parameters to the actual engineering state and the rationality of adjustment schemes.

[0029] In this embodiment, firstly, the current engineering cost control state vector is input into the input layer of a trained deep learning network to perform feature transformation to obtain hidden feature representations. Here, the current engineering cost control state vector is a multi-dimensional vector after concatenation and normalization, whose dimensions cover all feature dimensions of cumulative cost deviation, schedule deviation, and budget constraints. Traditional input layers only use simple linear transformations (such as y=Wx+b), which cannot adapt to the differentiated importance of different features in engineering data. This step improves upon this by introducing an engineering feature adaptation weight matrix. Specifically, the input layer first performs dimensionality verification on the state vector to ensure that it matches the number of neurons in the network input layer; then, it performs an improved feature transformation operation, the formula of which is: , The output is the hidden feature representation. The input is the current engineering cost control state vector. The weight matrix for the input layer is dynamically adjusted based on the importance of engineering features, such as the weight of the budget constraint dimension being higher than that of the non-critical deviation dimension. This is the input layer bias vector. An engineering feature adaptation factor (with a value ranging from 0.8 to 1.2, derived from the historical feature contribution statistics of similar engineering projects, used to correct the feature transformation amplitude under different engineering scenarios) is introduced. This improved transformation strengthens the expression of features strongly related to cost control in the state vector, weakens irrelevant noise interference, and lays the foundation for subsequent advanced feature extraction.

[0030] Secondly, the hidden feature representations are nonlinearly transformed through multiple hidden layers of the deep learning network to obtain high-level feature representations. Traditional deep learning hidden layers often use fixed structures and single activation functions, which can easily lead to insufficient extraction of cross-dimensional features in engineering data (such as the correlation between cost deviation and construction rhythm). This step improves upon this by using dynamic inter-layer fusion and a combination of activation functions. In practice, the deep learning network is set with 3-5 hidden layers (the number of layers is determined by the complexity of the project; more complex projects have more layers to extract deeper correlated features). The first hidden layer is the output of the input layer. As input, a cross-layer feature fusion mechanism is introduced between hidden layers. Taking the k-th hidden layer (k≥2) as an example, its nonlinear transformation formula is: , For the first The feature representations output by each hidden layer (the final output of the 5th hidden layer is the high-level feature representation). For the first The weight matrix of the layer, For the first Output features of layer -1 The cross-layer fusion weights (with values ​​ranging from 0.3 to 0.5, used to retain effective features from the first two layers to avoid gradient vanishing) For the first Layer bias vector, To activate the gating weight matrix, the ELU function is used to handle positive and negative deviation features in the engineering data (such as the differentiated expression of cost overruns and savings), and the Sigmoid function serves as the activation gating to filter feature dimensions that are key to cost control (such as prioritizing budget-related features when budget constraints are tight). Through this multi-layered nonlinear transformation, deep correlation features between engineering costs, resources, and schedules can be extracted from the hidden features, forming a high-level feature representation that reflects the core needs of cost control.

[0031] Finally, the high-level feature representation is mapped to engineering control parameters through the output layer of the deep learning network. Traditional output layers only output results through linear mapping, which can easily lead to output parameters exceeding the practically feasible range of the project (e.g., negative resource input adjustments or excessively large adjustments to construction rhythm). This step improves the constraint mapping by incorporating the physical meaning of the engineering control parameters. The number of neurons in the output layer matches the dimensionality of the engineering control parameters, covering all dimensions of resource input adjustments (e.g., the adjustment ratios of labor, materials, and machinery) and construction rhythm adjustments (e.g., the ratio of process time compression / extension). Specifically, the high-level feature representation is first dimensionally compressed to match the number of neurons in the output layer; then, the improved constraint mapping operation is performed, using the following formula: , This is the final output vector of engineering control parameters. This is a high-level feature representation output from the hidden layer. This is the output layer weight matrix. This is the output layer bias vector. This is the engineering constraint mapping function. Its core function is to limit the mapping result to the practically feasible range of the project: for adjustments to resource input, Ensure its value is within [-0.2, 0.3] (i.e., a decrease of 20% to an increase of 30%, set based on project resource reserves and market supply capacity); regarding the adjustment amount of construction pace, Ensure its value is within [-0.15, 0.2] (i.e., a 15% reduction to a 20% extension of the construction period, set according to the total construction period requirements and process logic). Through this constraint mapping, the final output engineering control parameters retain the inference accuracy of the deep learning network while conforming to the physical constraints of engineering implementation, and can be directly used for joint adjustment of subsequent resource input and construction rhythm.

[0032] As an example of the above scheme, the step of jointly adjusting the resource input and construction rhythm of the project based on the engineering control parameters to generate a construction adjustment plan includes the following sub-steps: The engineering control parameters were analyzed to obtain the resource input adjustment amount and the construction rhythm adjustment amount. Based on the aforementioned resource input adjustment amount, the adjustment and allocation scheme for each resource type of the project is determined, and the resource adjustment scheme is obtained; Based on the aforementioned construction rhythm adjustment amount, a construction procedure time adjustment plan for the project is determined, resulting in a schedule adjustment plan. The resource adjustment plan and the schedule adjustment plan are integrated to generate the construction adjustment plan.

[0033] In this embodiment, the engineering control parameters are first analyzed to separate the resource input adjustment amount and the construction rhythm adjustment amount, clarifying the specific numerical range and direction of the two types of adjustments. Then, based on the resource input adjustment amount, adjustment ratios and specific values ​​are allocated according to resource type (such as labor, materials, and machinery) to form a resource adjustment plan, ensuring that changes in resource input are adapted to cost deviations. Next, based on the construction rhythm adjustment amount, the start and end times and durations of each process are adjusted according to the logical relationship of the processes, forming a schedule adjustment plan to match changes in construction rhythm with schedule deviations. Finally, the resource adjustment plan and the schedule adjustment plan are integrated to ensure that the time nodes of resource adjustments and the process arrangements of schedule adjustments are coordinated, generating a construction adjustment plan. Therefore, this embodiment, by analyzing control parameters step by step, formulating resource and schedule adjustment plans separately, and then integrating them collaboratively, achieves the linkage adjustment of resource input and construction rhythm, improving the pertinence and feasibility of the construction adjustment plan for engineering cost control needs.

[0034] The process begins by analyzing engineering control parameters to obtain adjustments in resource input and construction schedule. These engineering control parameters are vectors output by the deep learning network, containing both resource and schedule adjustment information. However, the original output is fused and requires structured analysis for separation. Specifically, based on a pre-defined dimensional division rule for the parameter vector (determined during network training and corresponding to the engineering control objective), the first half of the vector is identified as parameters related to resource input adjustments, and the second half as parameters related to construction schedule adjustments. Then, the resource parameters are normalized and inverted (based on the inverse operation of the normalization mentioned above) to restore them to the adjustment ratios in the actual engineering scenario (e.g., the proportion of increased or decreased resource input), yielding the resource input adjustment amount. Similarly, the schedule parameters are inverted to obtain the construction schedule adjustment amount (e.g., the proportion of compressed or extended work procedures). A validity check is introduced during the analysis process. If a certain adjustment amount exceeds the pre-defined reasonable range for the engineering (e.g., the absolute value of the resource adjustment ratio is too large), a secondary correction is triggered (truncating or scaling based on historical adjustment experience of similar projects) to ensure the feasibility of subsequent adjustments.

[0035] Secondly, based on the adjusted resource input, the allocation scheme for each resource type in the project is determined, resulting in a resource adjustment plan. Traditional resource adjustments often use an average allocation method, ignoring the differences in cost sensitivity among different resource types. This step improves upon this by using a resource sensitivity-weighted allocation algorithm. Specifically, the core resource types of the project are first identified, including labor (classified by job type), materials (classified by main and auxiliary materials), and machinery (classified by large equipment and small tools). Then, the cost sensitivity coefficient for each resource type is calculated using the following formula: , For the first Cost sensitivity coefficient of a resource type (the larger the value, the more significant the impact of changes in the input of that resource on the total cost). This is the partial derivative of the total cost with respect to the amount of resource input (derived from historical cost data). This represents the current amount of resources being invested. This represents the total cost of the current project. Based on... The resource input adjustment will be allocated to each resource type according to the following formula: , For the first Adjustments to resource types, Adjusting the total amount of resources invested This is the supply elasticity coefficient of the resource (reflecting market supply capacity; a larger value indicates greater ease of adjustment). This represents the total number of resource types. This algorithm allows for more reasonable adjustment ratios for resources with high sensitivity and supply elasticity (e.g., when the price of main materials fluctuates greatly but supply is sufficient, the adjustment range can be appropriately increased), ultimately forming a resource adjustment plan that includes the specific adjustment amount, direction (increase or decrease), and implementation priority for each resource type.

[0036] Next, based on the adjustment amount of the construction rhythm, the time adjustment plan for the construction procedures of the project is determined, resulting in a schedule adjustment plan. Traditional schedule adjustments often directly scale the time of each procedure proportionally, which can easily disrupt the logical dependencies between procedures (e.g., incomplete preceding procedures affect subsequent procedures). This step introduces a procedure dependency weight adjustment mechanism. Specifically, the construction procedure network of the project is first analyzed (including the preceding and succeeding procedures and logical relationships of each procedure), and the critical path (the sequence of procedures that determines the total project duration) and non-critical paths are identified; then, dependency weights are assigned to each procedure. (Value range 0-1, critical path processes and processes located at nodes that depend on multiple processes) (Higher value); Based on the adjustment amount of the construction rhythm, the time adjustment amount for each process is calculated according to the following formula: , For the first The time adjustment amount for each process step Adjusting the total amount to accommodate the construction schedule. This is the time flexibility coefficient for the process (reflecting the space that the process can be compressed or extended; the larger the value, the greater the flexibility of adjustment). This represents the total number of processes. During the adjustment process, the process logic must be verified simultaneously: if the end time of a preceding process after adjustment is later than the planned start time of a subsequent process, the adjustment amount of the subsequent process will be automatically corrected (appropriately extended or compressed) to ensure smooth process transitions. The final result is a schedule adjustment plan that includes the adjusted start and end times of each process, the critical path update results, and the reasons for the adjustments.

[0037] Finally, the resource adjustment plan and schedule adjustment plan are integrated to generate a construction adjustment plan. Integration is not a simple patchwork; it requires resource-schedule matching verification to ensure synergy between the two. Specifically, a correlation model between resource input and process time is first established (e.g., an increase in labor input for a process can shorten its time). The resource increment in the resource adjustment plan is then verified to support the process compression in the schedule adjustment plan (or whether resource reduction matches process extension). If a mismatch exists (e.g., a process is planned to shorten its time but the corresponding labor input is not increased), coordination and correction are initiated: non-critical resources or non-critical processes are adjusted first, and matching is achieved by fine-tuning the resource allocation ratio or the magnitude of process adjustment. After integration, the adjustment content is categorized according to construction stage (e.g., foundation stage, main structure stage), clarifying the details of resource adjustment, process time adjustment, and their relationship for each stage, ultimately forming a structured construction adjustment plan.

[0038] As an example of the above scheme, determining the current cost control implementation scheme for the project based on the construction adjustment plan includes the following sub-steps: The construction adjustment plan is broken down into execution units corresponding to each sub-project, and the resource input and construction start and end time of each execution unit are clearly defined. Based on the resource input and construction start and end time of each execution unit, and in conjunction with the project budget data, set the cost consumption limit and progress assessment node for each execution unit; By integrating the cost consumption caps, progress assessment nodes, and corresponding resource adjustment plans and progress adjustment plans of all execution units, and clarifying the responsible parties of each execution unit, the current cost control execution plan is obtained.

[0039] In this embodiment, the construction adjustment plan is first decomposed into execution units corresponding to each sub-project, clarifying the resource input and start / end time of each execution unit, thus refining the overall adjustment requirements to the specific execution level of each sub-project. Next, based on the resource input and start / end time of each execution unit, and combined with project budget data, cost consumption caps and progress assessment nodes are set for each execution unit, providing clear cost and progress control standards for the execution process of each unit. Then, the cost consumption caps, progress assessment nodes, and corresponding resource and progress adjustment schemes of all execution units are integrated, clarifying the responsible parties for each execution unit, ensuring that the execution requirements of each unit are coordinated and that responsibility is clearly assigned, resulting in the current cost control execution plan. Therefore, this embodiment, by refining the construction adjustment plan into execution units, setting unit control standards, and integrating responsible parties, realizes the transformation of cost control requirements from planning to specific execution, improving the pertinence and operability of the cost control execution plan.

[0040] In this embodiment, the construction adjustment plan is first decomposed into execution units corresponding to each sub-project, clarifying the resource input and start and end times of each execution unit. The construction adjustment plan here refers to the integrated scheme generated above, which includes resource and schedule adjustments. Its content covers the entire project's adjustment framework and needs to be further broken down into the smallest executable units. Specifically, based on the project's sub-project classification standards (e.g., based on building structure, foundation engineering, main structure engineering, decoration engineering, etc.), the construction adjustment plan is first decomposed into several execution units, each corresponding to an independent sub-project or sub-sub-project (e.g., main structure engineering can be subdivided into steel reinforcement engineering, concrete engineering, etc.). During the decomposition process, it is necessary to ensure that the work scope of each execution unit does not overlap and covers all adjustment content, with clear and traceable boundaries. Subsequently, resource input information for each execution unit was extracted from the construction adjustment plan, including the adjusted input quantities of labor (specific job types and quantities), materials (main and auxiliary materials), and machinery (equipment models and usage duration). Simultaneously, based on the schedule adjustment plan, the start and end times of each execution unit were clearly defined (accurate to specific time periods, and coordinated with the timelines of preceding and following execution units to avoid schedule conflicts). This decomposition transforms the macro-level adjustment plan into micro-level, actionable unit tasks, laying the foundation for setting subsequent control indicators.

[0041] Secondly, based on the resource input and construction start and end times of each execution unit, and combined with the project budget data, cost consumption caps and progress assessment milestones are set for each execution unit. Traditional methods only set cost caps according to budget proportions, ignoring the dynamic relationship between resource input and time. This step improves upon this by using a resource-time coupling model. Specifically, the cost consumption cap is calculated using the following formula: , For the first The upper limit of cost consumption per execution unit The budget amount for the sub-item of the execution unit (extracted from the project budget data). This is the budget fluctuation factor (ranging from 0 to 0.1, dynamically adjusted based on the overall budget tightness of the project; the tighter the budget, the higher the factor). (smaller) For the first The execution unit for the first The amount of resources invested in this type For the first The unit price of this type of resource (determined based on a combination of current market prices and historical cost data). The construction duration of this execution unit (calculated from the start and end times). This represents the total number of resource types. By using the lower of the "budget ceiling" and the "estimated actual resource consumption" logic, the cost ceiling is ensured to neither exceed budget constraints nor exceed actual resource input needs. For progress assessment milestones, they are not simply divided equally by time, but are set in conjunction with the key processes of the execution unit (such as completion of concrete curing, steel reinforcement acceptance, etc.). Each milestone corresponds to a quantifiable completion indicator (such as the percentage of completed work, acceptance of a specific process), and the milestone time must fall within the start and end time range of the execution unit's construction to ensure the accuracy of the assessment.

[0042] Finally, the cost consumption caps, schedule assessment nodes, and corresponding resource adjustment plans and schedule adjustment plans of all execution units are integrated to clarify the responsible parties of each execution unit and obtain the current cost control execution plan. The integration process needs to achieve "horizontal collaboration" and "vertical closed loop": Horizontally, execution units are categorized according to construction stages (such as foundation stage, main structure stage), and it is verified whether the total cost cap of each unit within the same stage does not exceed the budget limit of that stage (extracted from the project budget data), and whether there are conflicts in schedule assessment nodes (such as the node time of a certain unit being later than the end time of the same stage). If there are conflicts, the cost caps or node times of the relevant units are adjusted retrospectively; Vertically, the sequential relationship of execution units is sorted out to ensure that the resource adjustment plan of the subsequent unit can only be initiated after the schedule assessment node of the preceding unit is completed (such as the material input of the concrete project can only be executed according to the adjustment plan after the acceptance of the preceding steel reinforcement project). Simultaneously, a responsible entity is clearly defined for each execution unit. This division of responsibility must be aligned with the project management structure (e.g., construction teams are responsible for specific work processes, while the project department coordinates across teams). The plan must specify each entity's responsibility for controlling the cost limits and schedule milestones for their corresponding unit (e.g., teams bear direct execution responsibility, while the project department bears supervisory responsibility). The final cost control execution plan is presented in a structured document format, including a detailed list of each execution unit, corresponding control indicators (cost limits, schedule milestones), adjustment basis (resource and schedule adjustment plans), and division of responsibilities, achieving a complete closed loop from plan adjustment to implementation.

[0043] Referring to Figure 2, it is a schematic diagram of the structure of an engineering cost control device based on deep learning according to an embodiment of the present invention. The engineering cost control device based on deep learning includes: Module 10 is used to acquire historical cost data, current construction progress data, and project budget data for the project. Module 11 is used to construct a multi-dimensional state representation containing cumulative cost deviation, schedule deviation and budget constraints based on the historical cost data, current construction progress data and project budget data, so as to obtain the current project cost control state vector; Output module 12 is used to input the current engineering cost control state vector into the trained deep learning network and output engineering control parameters; The generation module 13 is used to jointly adjust the resource input and construction rhythm of the project according to the engineering control parameters, and generate a construction adjustment plan. The determination module 14 is used to determine the current cost control implementation plan for the project based on the construction adjustment plan.

[0044] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: First, historical cost data, current construction progress data, and project budget data are acquired to provide data support for subsequently characterizing the project cost status. Then, based on these three types of data, a multi-dimensional state representation including cumulative cost deviation, schedule deviation, and budget constraints is constructed to obtain the current project cost control state vector. This state vector can simultaneously cover historical cost deviations, current schedule differences, and budget limitations, achieving a centralized presentation of multi-dimensional project information. Next, this state vector is input into a trained deep learning network. Leveraging the deep learning network's ability to process multi-dimensional nonlinear correlated data, it outputs project control parameters that match the current project status. Subsequently, based on these control parameters, the project's resource input and construction rhythm are jointly adjusted to avoid adaptation conflicts arising from individual adjustments to resource input and construction rhythm, generating a construction adjustment plan. Finally, based on this construction adjustment plan, the current cost control execution scheme is determined, transforming control measures into implementable content. Therefore, the embodiments of the present invention integrate historical cost data, current construction progress data, and project budget data to form a multi-dimensional state representation, and utilize the project control parameters output by the deep learning network to achieve joint adjustment of resource input and construction rhythm. This solves the problem that existing project cost control methods are unable to integrate multi-dimensional data to form a device state representation and cannot coordinate the control of resources and construction rhythm, thereby improving the adaptability of cost control schemes to actual construction needs and effectively improving the accuracy of project cost control.

[0045] As an example of the above scheme, the building module is specifically used for: Based on the historical cost data and the project budget data, the cumulative cost deviation is calculated to obtain the cost deviation value; Based on the current construction progress data and the predefined planned progress data, the progress deviation is calculated to obtain the progress deviation value; Based on the project budget data, budget constraints are extracted to obtain the budget constraint vector; The cost deviation value, the schedule deviation value, and the budget constraint vector are concatenated and normalized to obtain the current project cost control status vector.

[0046] As an example of the above scheme, the output module is specifically used for: The current engineering cost control state vector is input into the input layer of the trained deep learning network to perform feature transformation and obtain the hidden feature representation. The hidden feature representation is nonlinearly transformed through multiple hidden layers of the deep learning network to obtain a high-level feature representation; The high-level feature representation is passed through the output layer of the deep learning network and mapped to the engineering control parameters.

[0047] As an example of the above scheme, the generation module is specifically used for: The engineering control parameters were analyzed to obtain the resource input adjustment amount and the construction rhythm adjustment amount. Based on the aforementioned resource input adjustment amount, the adjustment and allocation scheme for each resource type of the project is determined, and the resource adjustment scheme is obtained; Based on the aforementioned construction rhythm adjustment amount, a construction procedure time adjustment plan for the project is determined, resulting in a schedule adjustment plan. The resource adjustment plan and the schedule adjustment plan are integrated to generate the construction adjustment plan.

[0048] As an example of the above scheme, the determining module is specifically used for: The construction adjustment plan is broken down into execution units corresponding to each sub-project, and the resource input and construction start and end time of each execution unit are clearly defined. Based on the resource input and construction start and end time of each execution unit, and in conjunction with the project budget data, set the cost consumption limit and progress assessment node for each execution unit; By integrating the cost consumption caps, progress assessment nodes, and corresponding resource adjustment plans and progress adjustment plans of all execution units, and clarifying the responsible parties of each execution unit, the current cost control execution plan is obtained.

[0049] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0050] 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 are also considered to be within the scope of protection of the present invention.

Claims

1. A deep learning-based engineering cost control method, characterized in that, include: Obtain historical cost data, current construction progress data, and project budget data for the engineering project; Based on the historical cost data, current construction progress data, and project budget data, a multi-dimensional state representation including cumulative cost deviation, schedule deviation, and budget constraints is constructed to obtain the current project cost control state vector. The current engineering cost control state vector is input into the trained deep learning network, which outputs engineering control parameters. Based on the engineering control parameters, the resource input and construction pace of the engineering project are jointly adjusted to generate a construction adjustment plan. Based on the aforementioned construction adjustment plan, the current cost control implementation scheme for the project is determined.

2. The engineering cost control method based on deep learning as described in claim 1, characterized in that, Based on the historical cost data, current construction progress data, and project budget data, a multi-dimensional state representation including cumulative cost deviation, schedule deviation, and budget constraints is constructed to obtain the current project cost control state vector, including the following sub-steps: Based on the historical cost data and the project budget data, the cumulative cost deviation is calculated to obtain the cost deviation value; Based on the current construction progress data and the predefined planned progress data, the progress deviation is calculated to obtain the progress deviation value; Based on the project budget data, budget constraints are extracted to obtain the budget constraint vector; The cost deviation value, the schedule deviation value, and the budget constraint vector are concatenated and normalized to obtain the current project cost control status vector.

3. The engineering cost control method based on deep learning as described in claim 2, characterized in that, The step of inputting the current engineering cost control state vector into a trained deep learning network and outputting engineering control parameters includes the following sub-steps: The current engineering cost control state vector is input into the input layer of the trained deep learning network to perform feature transformation and obtain the hidden feature representation. The hidden feature representation is nonlinearly transformed through multiple hidden layers of the deep learning network to obtain a high-level feature representation; The high-level feature representation is passed through the output layer of the deep learning network and mapped to the engineering control parameters.

4. The engineering cost control method based on deep learning as described in claim 3, characterized in that, The step of jointly adjusting the resource input and construction pace of the project based on the engineering control parameters to generate a construction adjustment plan includes the following sub-steps: The engineering control parameters were analyzed to obtain the resource input adjustment amount and the construction rhythm adjustment amount. Based on the aforementioned resource input adjustment amount, the adjustment and allocation scheme for each resource type of the project is determined, and the resource adjustment scheme is obtained; Based on the aforementioned construction rhythm adjustment amount, a construction procedure time adjustment plan for the project is determined, resulting in a schedule adjustment plan. The resource adjustment plan and the schedule adjustment plan are integrated to generate the construction adjustment plan.

5. The engineering cost control method based on deep learning as described in claim 4, characterized in that, The process of determining the current cost control implementation plan for the project based on the construction adjustment plan includes the following sub-steps: The construction adjustment plan is broken down into execution units corresponding to each sub-project, and the resource input and construction start and end time of each execution unit are clearly defined. Based on the resource input and construction start and end time of each execution unit, and in conjunction with the project budget data, set the cost consumption limit and progress assessment node for each execution unit; By integrating the cost consumption caps, progress assessment nodes, and corresponding resource adjustment plans and progress adjustment plans of all execution units, and clarifying the responsible parties of each execution unit, the current cost control execution plan is obtained.

6. A deep learning-based engineering cost control device, characterized in that, include: The acquisition module is used to acquire historical cost data, current construction progress data, and project budget data for engineering projects. The construction module is used to construct a multi-dimensional state representation containing cumulative cost deviation, schedule deviation and budget constraints based on the historical cost data, current construction progress data and project budget data, so as to obtain the current project cost control state vector; The output module is used to input the current engineering cost control state vector into the trained deep learning network and output engineering control parameters. The generation module is used to jointly adjust the resource input and construction rhythm of the project based on the engineering control parameters, and generate a construction adjustment plan. The determination module is used to determine the current cost control implementation plan for the project based on the construction adjustment plan.

7. The engineering cost control device based on deep learning as described in claim 6, characterized in that, The building module is specifically used for: Based on the historical cost data and the project budget data, the cumulative cost deviation is calculated to obtain the cost deviation value; Based on the current construction progress data and the predefined planned progress data, the progress deviation is calculated to obtain the progress deviation value; Based on the project budget data, budget constraints are extracted to obtain the budget constraint vector; The cost deviation value, the schedule deviation value, and the budget constraint vector are concatenated and normalized to obtain the current project cost control status vector.

8. The engineering cost control device based on deep learning as described in claim 7, characterized in that, The output module is specifically used for: The current engineering cost control state vector is input into the input layer of the trained deep learning network to perform feature transformation and obtain the hidden feature representation. The hidden feature representation is nonlinearly transformed through multiple hidden layers of the deep learning network to obtain a high-level feature representation; The high-level feature representation is passed through the output layer of the deep learning network and mapped to the engineering control parameters.

9. The engineering cost control device based on deep learning as described in claim 8, characterized in that, The generation module is specifically used for: The engineering control parameters were analyzed to obtain the resource input adjustment amount and the construction rhythm adjustment amount. Based on the aforementioned resource input adjustment amount, the adjustment and allocation scheme for each resource type of the project is determined, and the resource adjustment scheme is obtained; Based on the aforementioned construction rhythm adjustment amount, a construction procedure time adjustment plan for the project is determined, resulting in a schedule adjustment plan. The resource adjustment plan and the schedule adjustment plan are integrated to generate the construction adjustment plan.

10. The engineering cost control device based on deep learning as described in claim 9, characterized in that, The determining module is specifically used for: The construction adjustment plan is broken down into execution units corresponding to each sub-project, and the resource input and construction start and end time of each execution unit are clearly defined. Based on the resource input and construction start and end time of each execution unit, and in conjunction with the project budget data, set the cost consumption limit and progress assessment node for each execution unit; By integrating the cost consumption caps, progress assessment nodes, and corresponding resource adjustment plans and progress adjustment plans of all execution units, and clarifying the responsible parties of each execution unit, the current cost control execution plan is obtained.

Images