Multi-dimensional building engineering cost dynamic evaluation method and system

Abstract

The invention provides a multi-dimensional building engineering cost dynamic evaluation method and system, and the method comprises the steps: obtaining engineering design basic data, building material price data and construction cost data of a building project, and obtaining standardized multi-source cost monitoring data; obtaining a cost evaluation model; wherein a feature extraction layer of the cost evaluation model is embedded into a multi-source data feature weighted fusion module, and through a dynamic weight algorithm, fusion weights are distributed according to influence degrees of various data on the cost; a bifunction cost calibration mechanism composed of a cost deviation correction function and a dynamic adjustment function is embedded in front of a model output layer, and the bifunction cost calibration mechanism corrects the evaluation deviation in real time based on the building material price fluctuation amplitude and the construction cost change; and inputting the standardized multi-source cost monitoring data into the cost evaluation model, and finally outputting a dynamic cost evaluation value and a cost change trend of the construction project. According to the method and the device, a dynamic adjustment mechanism is added, and the evaluation accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a multi-dimensional dynamic evaluation method and system for construction project costs. Background Technology

[0002] In the field of construction engineering, as the construction industry develops towards intelligence and refinement, traditional engineering cost assessment methods are gradually revealing their inability to adapt to the needs of complex projects, specifically in the following aspects: First, there is a lack of dynamic adjustment mechanisms. Building material prices fluctuate in real time due to factors such as market supply and demand, regional differences, and transportation costs. Construction costs also change dynamically due to factors such as adjustments in labor wages, changes in machinery rental prices, and deviations in the construction period. Traditional assessment methods are mostly based on static data (such as building material quotations at the beginning of the project and fixed labor unit prices), lacking a real-time correction mechanism for dynamic data changes. When building material prices or construction costs fluctuate significantly, the assessment results will be significantly biased, making it difficult to meet the needs of cost control throughout the entire project lifecycle.

[0003] Secondly, the accuracy of model feature extraction and calibration is limited. Some existing technologies attempt to introduce neural network models for cost assessment, but most adopt general-purpose network frameworks (such as basic convolutional neural networks and fully connected neural networks) without optimizing the model structure for the characteristics of engineering cost data. For example, the feature extraction layer only extracts features from a single type of data and does not establish cross-dimensional feature correlations between multi-source data; there is a lack of targeted bias calibration mechanisms before the output layer, and errors are reduced only through iterative model training, which cannot accurately correct systematic biases caused by fluctuations in building material prices and changes in construction costs, making it difficult to further improve the assessment accuracy. Summary of the Invention

[0004] The main objective of this invention is to provide a multi-dimensional dynamic cost assessment method and system for construction projects, aiming to overcome the shortcomings of current cost assessment methods, such as the lack of dynamic adjustment mechanisms and low assessment accuracy.

[0005] To achieve the above objectives, this invention provides a multi-dimensional dynamic cost assessment method for construction projects, comprising the following steps: Obtain basic engineering design data, building material price data, and construction cost data for building projects to obtain standardized multi-source cost monitoring data; A cost assessment model based on a convolutional neural network is obtained; wherein, the feature extraction layer of the cost assessment model embeds a multi-source data feature weighted fusion module, and allocates fusion weights according to the degree of influence of various types of data on the cost through a dynamic weight algorithm; a dual-function cost calibration mechanism consisting of a cost deviation correction function and a dynamic adjustment function is embedded before the model output layer, and the dual-function cost calibration mechanism corrects the assessment deviation in real time based on the fluctuation range of building material prices and changes in construction costs; Standardized multi-source cost monitoring data is input into the cost assessment model. After the multi-source data is weighted and fused through the feature extraction layer, the assessment deviation is corrected by the dual-function cost calibration mechanism, and finally the dynamic cost assessment value and cost change trend of the building project are output.

[0006] Furthermore, the basic engineering design data includes the building's main structural type, component size parameters, spatial layout planning data, and decoration and finishing standard data.

[0007] Furthermore, standardized multi-source cost monitoring data is input into the cost assessment model. After weighted fusion of the multi-source data through a feature extraction layer, the assessment bias is corrected by a dual-function cost calibration mechanism. Finally, the dynamic cost assessment value and cost change trend of the construction project are output, including: Standardized multi-source cost monitoring data are converted into feature matrices adapted to convolutional neural network inputs according to the classification dimensions of engineering design basic data, building material price data, and construction cost data. Each feature matrix is ​​input into the feature extraction layer of the cost assessment model, and the fusion weight of each feature matrix is ​​calculated and allocated in real time through the dynamic weight algorithm of the multi-source data feature weight fusion module. Multi-source data features are weighted and fused based on the assigned fusion weights to obtain a fused feature vector; The dual-function cost calibration mechanism integrates feature vectors. First, it calculates the deviation correction value based on the fluctuation range of building material prices in the building material price data using the cost deviation correction function. Then, it calculates the dynamic adjustment value by combining the changes in construction costs in the construction cost data using the dynamic adjustment function, and evaluates the deviation correction. The output layer outputs the dynamic cost assessment value and cost change trend curve of the building project.

[0008] Furthermore, the feature matrix dimension of the engineering design basic data is set to 256×256, and the feature matrix dimension of the building material price data and construction cost data is set to 128×128.

[0009] Furthermore, the fusion weights of each feature matrix are calculated and assigned in real time, including: Based on the feature matrix of basic engineering design data, two core influencing factors were extracted: building structure complexity coefficient and component quantity ratio. Based on the feature matrix of building material price data, two core influencing factors were extracted: main material price volatility and category scarcity coefficient. Based on the feature matrix of construction cost data, two core influencing factors were extracted: labor cost ratio and machinery usage frequency. Based on historical data of construction projects, the correlation between six core influencing factors and cost assessment errors was analyzed, and weights were assigned to each core influencing factor. The three types of feature matrices are used as query vector, key vector, and value vector, respectively. After linear transformation, the basic attention score is obtained through the attention score calculation model. Then, the initial fusion weight is obtained by combining the weights of the core influence factors corresponding to each feature matrix. The initial fusion weights are dynamically adjusted based on the fluctuation range of building material prices and the rate of change in construction costs in the current batch of data; the dynamically adjusted fusion weights are then assigned to the corresponding feature matrices.

[0010] Furthermore, using the cost deviation correction function, a deviation correction value is calculated based on the fluctuation range of building material prices in the building material price data, including: Extract the price fluctuation range of various main materials from building material price data; classify the main materials into core main materials and auxiliary main materials according to their impact on construction project costs; The price fluctuation ranges of core main materials and auxiliary main materials are respectively adapted with corresponding correction functions; among them, the correction functions include linear deviation correction functions and piecewise nonlinear deviation correction functions; Based on the adapted correction function, the single-type deviation correction value of the core main material and the auxiliary main material are calculated separately; the single-type deviation correction values ​​of all main materials are weighted and summed to obtain the deviation correction value.

[0011] Furthermore, a dynamic adjustment value is calculated using a dynamic adjustment function, incorporating changes in construction costs from the construction cost data, and an evaluation bias correction is performed, including: Extract changes in construction costs from construction cost data, including changes in labor costs and machinery costs; The corresponding dynamic adjustment function is determined based on the direction and magnitude of changes in construction costs; the dynamic adjustment function includes proportional dynamic adjustment function, piecewise proportional adjustment function, and threshold dynamic adjustment function; Based on the dynamic adjustment function corresponding to each change in construction cost, the corresponding classification adjustment value is calculated, and the dynamic adjustment value is obtained by merging the calculations. The deviation correction is evaluated by combining the aforementioned deviation correction value and the dynamic adjustment value.

[0012] This invention also provides a multi-dimensional dynamic cost assessment system for construction projects, comprising: The monitoring module is used to acquire basic engineering design data, building material price data, and construction cost data for building projects, and obtain standardized multi-source cost monitoring data. The acquisition module is used to acquire a cost assessment model based on a convolutional neural network framework. The feature extraction layer of the cost assessment model embeds a multi-source data feature weighted fusion module, which allocates fusion weights according to the degree of influence of various data on the cost through a dynamic weighting algorithm. A dual-function cost calibration mechanism, consisting of a cost deviation correction function and a dynamic adjustment function, is embedded before the model output layer. This dual-function cost calibration mechanism corrects the assessment deviation in real time based on the fluctuation range of building material prices and changes in construction costs. The evaluation module is used to input standardized multi-source cost monitoring data into the cost evaluation model. After the multi-source data is weighted and fused through the feature extraction layer, the evaluation deviation is corrected by the dual-function cost calibration mechanism, and finally the dynamic cost evaluation value and cost change trend of the building project are output.

[0013] The present invention provides a multi-dimensional dynamic cost assessment method and system for construction projects, comprising: acquiring basic engineering design data, building material price data, and construction cost data of a construction project to obtain standardized multi-source cost monitoring data; acquiring a cost assessment model based on a convolutional neural network framework; wherein, the feature extraction layer of the cost assessment model embeds a multi-source data feature weighted fusion module, and allocates fusion weights according to the degree of influence of various types of data on the cost through a dynamic weight algorithm; embedding a dual-function cost calibration mechanism consisting of a cost deviation correction function and a dynamic adjustment function before the model output layer, the dual-function cost calibration mechanism correcting the assessment deviation in real time based on the fluctuation range of building material prices and changes in construction costs; inputting the standardized multi-source cost monitoring data into the cost assessment model, completing the multi-source data weighted fusion through the feature extraction layer, correcting the assessment deviation through the dual-function cost calibration mechanism, and finally outputting the dynamic cost assessment value and cost change trend of the construction project. In this invention, a dual-function cost calibration mechanism consisting of a cost deviation correction function and a dynamic adjustment function is embedded before the model output layer. The dual-function cost calibration mechanism corrects the evaluation deviation in real time based on the fluctuation range of building material prices and changes in construction costs. It can dynamically correct the evaluation results according to the real-time fluctuation range of building material prices and changes in construction costs, which not only adds a dynamic adjustment mechanism but also improves the accuracy of the evaluation. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the steps of a multi-dimensional dynamic cost assessment method for building construction projects in one embodiment of the present invention; Figure 2 This is a structural block diagram of a multi-dimensional dynamic cost evaluation system for building construction projects according to an embodiment of the present invention; Figure 3This is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0015] The implementation, functional features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] It is particularly important to note that all technical steps, algorithm applications, and parameter settings in the technical solution of this application have clear technical objectives and application value. They do not utilize complex steps and algorithmic formulas to achieve simple functions. To provide detailed explanations of each step and avoid ambiguity, some conventional algorithms are used for illustration. However, this does not mean that the algorithms and technical features listed herein are the only way to implement the technical solution of this application, nor is it intended to limit the scope of protection of this application. This application is not a combination or stacking of the listed algorithms and technical features; its essence is to exemplify the implementation methods of this application to fully explain it. It does not pursue formal complexity by adding meaningless technical steps, nor does it involve the accumulation of technologies divorced from practical needs; it conforms to the conventional logic of technical improvement and design.

[0018] Reference Figure 1 One embodiment of the present invention provides a multi-dimensional dynamic cost assessment method for construction projects, comprising the following steps: Step S1: Obtain basic engineering design data, building material price data, and construction cost data for the construction project to obtain standardized multi-source cost monitoring data; Step S2: Obtain a cost assessment model based on a convolutional neural network framework; wherein, the feature extraction layer of the cost assessment model embeds a multi-source data feature weighted fusion module, and allocates fusion weights according to the degree of influence of various types of data on the cost through a dynamic weight algorithm; a dual-function cost calibration mechanism consisting of a cost deviation correction function and a dynamic adjustment function is embedded before the model output layer, and the dual-function cost calibration mechanism corrects the assessment deviation in real time based on the fluctuation range of building material prices and changes in construction costs; Step S3: Input standardized multi-source cost monitoring data into the cost assessment model. After the multi-source data is weighted and fused through the feature extraction layer, the assessment deviation is corrected by the dual-function cost calibration mechanism, and finally the dynamic cost assessment value and cost change trend of the building project are output.

[0019] In this embodiment, as described in step S1 above, the core objective is to provide comprehensive, standardized, and directly usable basic data for subsequent cost assessment, addressing the problems of fragmented data sources, inconsistent formats, and varying data quality in traditional cost assessments. During the data acquisition phase, three types of core data need to be acquired: First, basic engineering design data, which serves as the benchmark for cost assessment. This data specifically covers the main structural type of the building project (e.g., frame structure, shear wall structure), component dimensional parameters (e.g., beam and column cross-sectional dimensions, floor slab thickness), spatial layout planning (e.g., building area, floor height), and decoration and finishing standards (e.g., external wall insulation level, interior decoration materials). This type of data typically originates from professional technical documents such as the project's construction drawings and preliminary design estimates. Second, building material price data, which is a dynamic influencing factor in cost assessment. This data specifically includes various main materials (steel, concrete, cement) and auxiliary materials (sand, gravel, bricks) required for the project. The data includes: 1) Real-time market transaction unit price of building materials (such as roofing tiles and waterproofing materials), price ranges of different suppliers, discount coefficients corresponding to bulk purchases of building materials, and recent price fluctuation trends of building materials. The data sources are building material trading platforms, offline supplier price lists, and industry price monitoring reports. 2) Construction cost data. This type of data is practical data related to cost assessment on-site. Specifically, it involves the unit price of labor for each construction process (such as the daily wage standard for steelworkers and concrete workers), machinery rental costs (such as the unit price per shift for tower cranes and excavators), process time (such as the construction cycle of the main structure and the decoration construction cycle), and on-site management costs. This data is mainly extracted from the construction unit's cost accounting ledger and project construction organization design plan.

[0020] After data collection is completed, standardization processing is required, which includes three aspects: First, format unification, which unifies the units of measurement, time dimensions, and data formats of data from different sources; second, outlier removal, which uses statistical methods to identify and remove abnormal data, such as exorbitant building material prices that deviate from the normal range or construction records that deviate significantly from the time required for routine procedures, to ensure data authenticity; and third, missing value completion, which involves reasonably completing missing items in some data by combining historical data from similar projects, industry averages, or professional algorithms, ultimately forming standardized multi-source cost monitoring data with a well-structured, complete, and uniform format, laying the data foundation for subsequent model input.

[0021] As described in step S2 above, by optimizing the structure of the general convolutional neural network, the technical pain points of insufficient fusion of multi-source data features and inability to accurately correct evaluation bias in traditional models are solved.

[0022] First, the model is based on a convolutional neural network (CNN) framework, leveraging the powerful local feature extraction capabilities of CNNs to effectively uncover hidden correlation features in cost data. Building upon this framework, two core improvements are made to address the specific needs of cost assessment scenarios: The first improvement involves embedding a multi-source data feature weighting fusion module into the feature extraction layer. Traditional convolutional neural network feature extraction layers often extract features from single types of data, failing to reflect the differentiated impact of different cost data on the assessment results. This weighted fusion module, however, incorporates a dynamic weighting algorithm. Its core logic is to first construct a data impact assessment system based on professional knowledge and historical case data in the field of construction engineering cost, clarifying the baseline weights of three types of data—basic engineering design data, building material price data, and construction cost data—on the cost at different project stages (such as the design and construction stages). Then, combined with the specific characteristics of real-time collected data, the baseline weights are dynamically adjusted. For example, during the construction stage, when building material prices fluctuate significantly, the fusion weight of building material price data is automatically increased; during the design stage, when engineering design parameters change, the fusion weight of basic engineering design data is automatically increased. This achieves differentiated feature fusion of multi-source data, making the extracted features more aligned with the actual needs of cost assessment.

[0023] The second improvement involves embedding a dual-function cost calibration mechanism before the model output layer. This mechanism, comprised of a cost deviation correction function and a dynamic adjustment function, is the core module ensuring the accuracy of the assessment. The cost deviation correction function corrects for assessment deviations caused by fluctuations in building material prices. Its input parameter is the price fluctuation range in the building material price data, which is converted into a corresponding deviation correction value through pre-defined function logic. The dynamic adjustment function corrects for deviations caused by changes in construction costs, using changes in labor and machinery costs in the construction cost data as input to calculate the corresponding dynamic adjustment value. The synergistic effect of these two functions allows for accurate calibration of the model's initial assessment results based on real-time changes in building material and construction costs, addressing the problem of traditional models relying solely on data fitting and failing to specifically correct systematic deviations.

[0024] Through the two core improvements mentioned above, a dedicated cost assessment model was finally formed that can achieve intelligent fusion of multi-source data and accurate correction of deviations, providing technical support for subsequent dynamic assessments.

[0025] As described in step S3 above, the core is to input standardized data into the improved model, and through feature processing and deviation calibration of the model, output accurate dynamic cost assessment results to achieve a full-dimensional, real-time assessment of construction project costs.

[0026] First, the standardized multi-source cost monitoring data obtained in step S1 is converted into feature matrices adapted to the input format of a convolutional neural network, according to the classification dimensions of engineering design basic data, building material price data, and construction cost data, respectively, thus completing the model adaptation conversion of the data. Then, the converted feature matrices are input into the cost assessment model. They are first processed by the multi-source data feature weighting and fusion module of the feature extraction layer, and then differentiated feature fusion is completed through a dynamic weighting algorithm, generating a fused feature vector that simultaneously contains the core influences of the three types of data. This vector can comprehensively reflect the multi-dimensional influencing factors of project cost.

[0027] Next, the fused feature vectors undergo a dual-function cost calibration mechanism for bias correction: First, the cost bias correction function reads the price fluctuation range from the building material price data and calculates the corresponding bias correction value to compensate for the initial assessment bias caused by building material price fluctuations. Then, a dynamic adjustment function, combined with changes in labor and machinery costs from the construction cost data, calculates a dynamic adjustment value to perform a secondary calibration for assessment biases caused by cost changes during the construction phase. After this dual-function collaborative correction, the model can effectively offset the assessment errors caused by dynamic changes in building material prices and construction costs, ensuring the accuracy of the assessment results.

[0028] Finally, based on the corrected assessment data, the model output layer simultaneously outputs two core results: first, a dynamic cost assessment value for the construction project, which accurately reflects the actual cost level at the current stage of the project and can be iterated in real time according to data updates; second, a cost change trend, which generates a cost change curve by performing time-series analysis on the assessment results of multiple batches of data, intuitively presenting the fluctuation trend of project costs and providing data support for project investment decisions and cost control. The entire process realizes a closed loop from multi-source data input to accurate assessment result output, effectively solving the problems of low accuracy and poor dynamism in traditional cost assessment methods.

[0029] In one embodiment, the basic engineering design data includes the building structure type, component size parameters, spatial layout planning data, and decoration and renovation standard data.

[0030] In one embodiment, standardized multi-source cost monitoring data is input into the cost assessment model. After weighted fusion of the multi-source data through a feature extraction layer, the assessment bias is corrected by a dual-function cost calibration mechanism, and finally, the dynamic cost assessment value and cost change trend of the construction project are output, including: Standardized multi-source cost monitoring data are converted into feature matrices adapted to convolutional neural network inputs according to the classification dimensions of engineering design basic data, building material price data, and construction cost data. Each feature matrix is ​​input into the feature extraction layer of the cost assessment model, and the fusion weight of each feature matrix is ​​calculated and allocated in real time through the dynamic weight algorithm of the multi-source data feature weight fusion module. Multi-source data features are weighted and fused based on the assigned fusion weights to obtain a fused feature vector; The dual-function cost calibration mechanism integrates feature vectors. First, it calculates the deviation correction value based on the fluctuation range of building material prices in the building material price data using the cost deviation correction function. Then, it calculates the dynamic adjustment value by combining the changes in construction costs in the construction cost data using the dynamic adjustment function, and evaluates the deviation correction. The output layer outputs the dynamic cost assessment value and cost change trend curve of the building project.

[0031] In this embodiment, since convolutional neural networks can only effectively process matrix-based numerical features, and although the standardized multi-source cost monitoring data generated in step S1 has achieved data quality and format unification, it is still categorized structured business data and cannot be directly input into the model. Therefore, feature matrix transformation is required according to the classification dimensions of engineering design basic data, building material price data, and construction cost data: For engineering design basic data, qualitative and quantitative parameters such as building structure type and component size need to be converted into numerical matrices according to preset encoding rules. For example, the frame structure is encoded into a specific numerical vector, and parameters such as beam and column cross-sectional dimensions are mapped proportionally to standard numerical ranges to form an engineering design dimension feature matrix; for building material price data, the real-time unit price and fluctuation range of various building materials need to be arranged according to the category dimension to construct a building material price dimension feature matrix; for construction cost data, indicators such as labor unit price, machinery rental fee, and process time need to be integrated to form a construction cost dimension feature matrix. The dimensions of the three types of feature matrices need to match the input interface of the convolutional neural network feature extraction layer to ensure that the data can be efficiently read and processed by the model.

[0032] Once various feature matrices are input into the feature extraction layer of the cost assessment model, the multi-source data feature weighted fusion module will activate its built-in dynamic weighting algorithm to calculate and allocate weights. First, the algorithm retrieves prior knowledge and historical case data from the field of construction engineering cost estimation to establish a basic weight allocation benchmark, clarifying the initial weight proportions of engineering design basic data, building material price data, and construction cost data in typical project scenarios. Second, it dynamically adjusts the weights based on the actual characteristics of the current batch of data. For example, if the fluctuation range of building material prices exceeds a preset threshold, the fusion weight of the building material price data feature matrix is ​​automatically increased; if the project is in the design optimization stage and engineering design parameters change frequently, the weight of the engineering design basic data feature matrix is ​​increased. Finally, it completes the real-time allocation of fusion weights for various feature matrices, ensuring that the sum of all dimension weights is 1, providing a quantitative basis for subsequent weighted fusion of multi-source data features and guaranteeing the accuracy and adaptability of feature extraction.

[0033] After obtaining the assigned weights for various feature matrices, the model performs feature fusion according to the core logic of weighted summation: based on each feature matrix, the eigenvalues ​​within each matrix are multiplied by their corresponding fusion weights. The weighted feature matrices are then dimensionally aligned and numerically integrated to generate a fused feature vector. This fused feature vector simultaneously encompasses the baseline constraint features of engineering design data, the dynamic fluctuation features of building material price data, and the on-site practical correlation features of construction cost data. Its feature dimensions are effectively expanded, and the importance of different dimensional features is distinguished by weights. Compared to single-dimensional features, this fused feature vector can more comprehensively and accurately characterize the core cost influencing factors of a construction project, providing high-quality feature support for subsequent cost assessment.

[0034] The fused feature vector first inputs a cost deviation correction function for the first round of deviation correction: It extracts the price fluctuation range of various main building materials from the building material price data, combines this with the cost proportion of different building materials, and calculates the corresponding deviation correction value through a preset function logic. For example, when steel prices rise sharply, a positive deviation correction value is generated to compensate for the initial assessment value being too low due to rising building material prices. After completing the first round of correction, the fused feature vector enters a dynamic adjustment function for the second round of deviation correction: it reads core indicators such as changes in labor costs and machinery costs from the construction cost data, combines the cost proportion of each process with the construction progress, and calculates a dynamic adjustment value. If the current labor unit price increases or machinery rental costs increase, a corresponding adjustment value is generated to correct the assessment deviation caused by changes in construction costs. Through the collaborative correction of the two functions, the model can effectively eliminate the systematic errors caused by dynamic changes in building material and construction costs, significantly improving the fit between the assessment results and the actual cost.

[0035] After the fused feature vectors complete the dual-function bias correction, the model output layer generates two core deliverables based on the corrected assessment data: First, a dynamic cost assessment value for the construction project. This value is an accurate cost figure after multi-source data fusion and bias correction, directly reflecting the actual cost level at the current stage of the project and supporting real-time iteration with batch data updates. Second, a cost change trend curve. The model retrieves multiple batches of historical assessment data, sorts and fits the cost assessment values ​​according to the time dimension, generating a visualized cost change trend curve. This curve clearly presents the fluctuation pattern and trend of project costs, helping project managers predict cost change trends. Both deliverables can be simultaneously output to the terminal interface, providing comprehensive and accurate data support for project investment decisions, dynamic cost control, and risk warning.

[0036] In one embodiment, the feature matrix dimension of the engineering design basic data is set to 256×256, and the feature matrix dimension of the building material price data and the construction cost data is set to 128×128.

[0037] In one embodiment, the fusion weights of each feature matrix are calculated and assigned in real time, including: Based on the feature matrix of basic engineering design data, two core influencing factors were extracted: building structure complexity coefficient and component quantity ratio. Based on the feature matrix of building material price data, two core influencing factors were extracted: main material price volatility and category scarcity coefficient. Based on the feature matrix of construction cost data, two core influencing factors were extracted: labor cost ratio and machinery usage frequency. Based on historical data of construction projects, the correlation between six core influencing factors and cost assessment errors was analyzed, and weights were assigned to each core influencing factor. The three types of feature matrices are used as query vector, key vector, and value vector, respectively. After linear transformation, the basic attention score is obtained through the attention score calculation model. Then, the initial fusion weight is obtained by combining the weights of the core influence factors corresponding to each feature matrix. The initial fusion weights are dynamically adjusted based on the fluctuation range of building material prices and the rate of change in construction costs in the current batch of data; the dynamically adjusted fusion weights are then assigned to the corresponding feature matrices.

[0038] In this embodiment, the basic data feature matrix of engineering design is the benchmark data for cost assessment. Therefore, it is necessary to extract two core influencing factors: the building structure complexity coefficient and the component quantity ratio. The building structure complexity coefficient needs to be comprehensively quantified in combination with the project's structural type (e.g., the complexity of a frame-shear wall structure is higher than that of a pure frame structure), node connection method, seismic resistance level, and other indicators. The component quantity ratio is calculated by the ratio of the number of core load-bearing components (e.g., load-bearing columns and main beams) to the total number of components in the project, thereby reflecting the construction difficulty of the engineering structure and the basic cost base.

[0039] For the building material price data feature matrix, which is dynamic fluctuation data for cost assessment, two core influencing factors need to be extracted: the volatility of main material prices and the category scarcity coefficient. The volatility of main material prices is the ratio of the difference between the current market unit price of core main materials such as steel and concrete and the average price of the past three months to the average price, reflecting the short-term fluctuation range of building material prices. The category scarcity coefficient is assigned a value based on factors such as the supply and demand data of the building material market, the supplier's inventory preparation cycle, and the difficulty of transportation, reflecting the ease of purchasing main materials and the potential risk of price increases.

[0040] For the construction cost data feature matrix, which represents on-site practical data for cost assessment, two core influencing factors need to be extracted: the proportion of labor costs and the frequency of machinery use. The proportion of labor costs is the ratio of total project labor expenses to total construction costs, reflecting the impact of labor costs on the overall cost. The frequency of machinery use is the ratio of the actual number of shifts for large machinery such as tower cranes and excavators to the planned number of shifts, reflecting the actual consumption of machinery rental costs. Through these operations, six core influencing factors strongly correlated with cost assessment can be precisely extracted from the three feature matrices, laying the analytical foundation for subsequent weight allocation.

[0041] Next, historical data from large-scale construction projects is retrieved. This data must cover cost assessment cases of different project types (residential, commercial, and industrial buildings), different regions, and different construction periods. It must also include the original data of the six core influencing factors for each case, along with the corresponding cost assessment error values ​​(i.e., the deviation rate between the assessed value and the actual cost). Subsequently, professional data analysis algorithms (such as Pearson correlation coefficient analysis and gradient-enhanced feature importance analysis) are used to quantitatively analyze the correlation between each core influencing factor and the cost assessment error. For example, if the data analysis results show that the absolute value of the correlation coefficient between the price volatility of main materials and the cost assessment error is the highest, it indicates that this factor has the greatest impact on the assessment accuracy and should be assigned a higher weight. If the correlation between the frequency of machinery use and the assessment error is weak, a relatively lower weight should be assigned. After completing the single-factor weight allocation, the weights need to be integrated according to the category of the feature matrix. The sum of the factor weights corresponding to the three types of feature matrices—engineering design basic data, building material price data, and construction cost data—must be clarified to provide core coefficient support for the subsequent calculation of the initial fusion weights.

[0042] Furthermore, the three types of feature matrices are assigned roles as the query vector (Q), key vector (K), and value vector (V) in the attention mechanism: the engineering design basic data feature matrix is ​​typically used as the query vector to anchor the baseline direction of cost assessment; the building material price data feature matrix is ​​used as the key vector to match dynamically fluctuating cost influencing factors; and the construction cost data feature matrix is ​​used as the value vector to provide supplementary cost features at the on-site practical level. Next, a linear transformation is performed on the three vectors, mapping them to the same feature dimension space through a pre-defined learnable transformation matrix, eliminating dimensional differences between different feature matrices. After the linear transformation, the model inputs the attention score, and through vector dot product operations combined with a scaling mechanism, a basic attention score is obtained. This score quantifies the potential correlation strength between the three types of feature matrices. Finally, the basic attention score is weighted and fused with the weights of the core influencing factors corresponding to each feature matrix: taking the basic data feature matrix of engineering design as an example, the weights of the building structure complexity coefficient and the proportion of the number of components should be incorporated into the basic attention score of the matrix according to a preset ratio. Similarly, the weight integration of the other two types of matrices is completed, and the initial fusion weights that take into account both data correlation characteristics and industry professional attributes are finally obtained.

[0043] Set dynamic adjustment trigger and quantification rules: For the fluctuation range of building material prices, if the fluctuation range of the main material price in the current period exceeds the preset threshold of the historical average (such as 10%), the weight of the building material price data feature matrix will be increased according to the excess ratio of the fluctuation range, so as to strengthen the impact of dynamic price factors on cost assessment; For the change rate of construction costs, if the change rate of labor costs or machinery usage costs exceeds the planned range, the weight of the construction cost data feature matrix will be adjusted synchronously according to the size of the cost change rate. For example, if the proportion of labor costs increases by 8% year-on-year, the weight of the construction cost data feature matrix will be appropriately increased to ensure that on-site cost changes are fully considered.

[0044] After completing the dynamic calibration of the initial fusion weights, the adjusted fusion weights are assigned to the corresponding feature matrices: the weights of the engineering design basic data feature matrix ensure the accuracy of the cost assessment benchmark; the weights of the building material price data feature matrix respond to market price fluctuations; and the weights of the construction cost data feature matrix match changes in actual on-site costs. The three types of weights have their own focus and work together to provide a scientific quantitative basis for the accurate fusion of multi-source data features in the future.

[0045] In one embodiment, a deviation correction value is calculated based on the fluctuation range of building material prices in the building material price data using a cost deviation correction function, including: Extract the price fluctuation range of various main materials from building material price data; classify the main materials into core main materials and auxiliary main materials according to their impact on construction project costs; The price fluctuation ranges of core main materials and auxiliary main materials are respectively adapted with corresponding correction functions; among them, the correction functions include linear deviation correction functions and piecewise nonlinear deviation correction functions; Based on the adapted correction function, the single-type deviation correction value of the core main material and the auxiliary main material are calculated separately; the single-type deviation correction values ​​of all main materials are weighted and summed to obtain the deviation correction value.

[0046] In this embodiment, the first step is to extract the price fluctuation range of main materials: all categories of main materials required for project construction are selected from the building material price data. For each category of main material, its actual purchase price in the current period and the corresponding benchmark price are retrieved (the benchmark price is usually the contract price during the project bidding stage or the average market price in the past 6 months). Through the calculation logic of (actual price in the current period - benchmark price) / benchmark price, the price fluctuation range of a single category of main material is obtained. At the same time, a corresponding list of main material categories and fluctuation ranges is formed to ensure that the fluctuation data is traceable and quantifiable.

[0047] Subsequently, main materials were categorized based on their impact on the overall construction cost: according to the proportion of each main material in the total project cost and its level of impact on structural safety, main materials were divided into core main materials and auxiliary main materials. Core main materials are those with a high cost proportion and significant impact on structural safety, typically such as steel, ready-mixed concrete, and prestressed steel bars. Price fluctuations in these materials directly determine the overall trend of the project cost. Auxiliary main materials are those with a relatively low cost proportion and are primarily functional, typically such as cement, sand, and waterproof membranes. Their price fluctuations have a relatively limited impact on the overall cost. This classification process allows for differentiated management of main materials of varying importance, laying the foundation for subsequent precise adjustments.

[0048] Two types of basic correction functions are predefined with adaptation rules to clarify their application boundaries: the first is a linear deviation correction function, whose core feature is that the correction value is directly proportional to the fluctuation range of the main material price. It is suitable for scenarios with small fluctuation ranges within a reasonable range, ensuring the stability of the correction under small fluctuations and avoiding correction deviations caused by the nonlinear characteristics of the function. The second is a piecewise nonlinear deviation correction function, which divides the fluctuation range into different intervals according to the fluctuation range and sets a higher correction coefficient for fluctuation intervals exceeding the threshold. It is suitable for scenarios with large fluctuation ranges and can achieve strong targeted correction for large price fluctuations, making up for the deficiency of the linear function in the case of large fluctuations.

[0049] Subsequently, function adaptation was carried out based on the type of main material and the fluctuation range: For core main materials, due to their significant impact on cost, a stricter fluctuation threshold needs to be set. When the price fluctuation range of core main materials is ≤ the preset threshold, a linear deviation correction function is applied to ensure the stability of the correction; when the fluctuation range is > the preset threshold, a piecewise nonlinear deviation correction function is immediately switched to strengthen the correction of large fluctuations. For auxiliary main materials, since their impact on the overall cost is limited, the fluctuation threshold can be appropriately relaxed. The piecewise nonlinear correction function is only activated when the fluctuation range far exceeds the threshold. In normal fluctuation scenarios, the linear deviation correction function is used to balance correction accuracy and computational efficiency.

[0050] Next, the deviation correction value for each type of main material is calculated: based on the correction function adapted in the previous steps, and combined with the cost proportion weight of each type of main material, the single-type correction value for core main materials and auxiliary main materials is calculated separately. For core main materials, the single-type correction value = price fluctuation range of the main material × cost proportion weight of the core main material × corresponding correction function coefficient, where the cost proportion weight of the core main material must be higher than that of auxiliary main materials, and the correction function coefficient will be dynamically adjusted according to the fluctuation range; for auxiliary main materials, the single-type correction value = price fluctuation range of the main material × cost proportion weight of the auxiliary main material × corresponding correction function coefficient, with an overall correction intensity lower than that of core main materials, ensuring that the correction priority matches the importance of the main materials.

[0051] Then, a weighted summation of the total deviation correction value is performed: the individual correction values ​​for all core and auxiliary main materials are aggregated, with differentiated category weights applied during the summation process. The summation weight for the individual correction values ​​of core main materials is no less than 70%, and the summation weight for the individual correction values ​​of auxiliary main materials is no more than 30%, thus highlighting the dominant influence of core main material price fluctuations on the overall cost deviation. Through this weighted summation logic, a total deviation correction value is finally obtained that accurately reflects the overall cost deviation caused by building material price fluctuations. This provides a quantitative basis for the accurate calibration of subsequent assessment results, ensuring that the cost assessment value dynamically adapts to the actual prices in the building material market.

[0052] In one embodiment, a dynamic adjustment value is calculated by using a dynamic adjustment function, incorporating changes in construction costs from the construction cost data, and an evaluation deviation correction is performed, including: Extract changes in construction costs from construction cost data, including changes in labor costs and machinery costs; The corresponding dynamic adjustment function is determined based on the direction and magnitude of changes in construction costs; the dynamic adjustment function includes proportional dynamic adjustment function, piecewise proportional adjustment function, and threshold dynamic adjustment function; Based on the dynamic adjustment function corresponding to each change in construction cost, the corresponding classification adjustment value is calculated, and the dynamic adjustment value is obtained by merging the calculations. The deviation correction is evaluated by combining the aforementioned deviation correction value and the dynamic adjustment value.

[0053] In this embodiment, complete raw cost change data is first extracted from the construction cost data. This data needs to cover the planned and actual values ​​of all dimensions of costs, including labor, machinery, and materials, at each stage of project construction. Subsequently, based on the composition attributes and impact logic of costs, construction cost changes are broken down into two core dimensions: changes in labor costs and changes in machinery costs. Extracting changes in labor costs requires combining the planned unit price of labor for each trade, the actual unit price of labor, the planned number of laborers, and the actual number of laborers. First, the cost change amount for a single trade is calculated (Change amount for a single trade = (Actual unit price - Planned unit price) × Actual number of laborers). Then, the changes for all trades are summed to obtain the overall change in labor costs. Simultaneously, the labor cost change rate is calculated (Labor cost change rate = Overall labor cost change / Planned total labor cost). Extracting changes in machinery costs requires targeting various construction machinery such as tower cranes, excavators, and concrete pump trucks. Combining their planned rental unit price, actual rental unit price, planned number of shifts, and actual number of shifts, the cost change amount for a single machine is calculated (Change amount for a single machine = (Actual rental unit price - Planned rental unit price) × Actual number of shifts). Finally, these are integrated to obtain the overall change in machinery costs and the machinery cost change rate (Machinery cost change rate = Overall machinery cost change / Planned total machinery cost). By breaking down costs along this dimension, we can achieve more refined control over changes in construction costs, laying a data foundation for accurate adjustments later.

[0054] The core characteristics and application boundaries of three types of dynamic adjustment functions are predefined: First, the proportional dynamic adjustment function, whose core logic is that the adjustment value is related to the cost change in a fixed proportion. It is suitable for scenarios where the cost change is small and the direction of change is stable, ensuring the smoothness and controllability of corrections under small changes. Second, the segmented proportional adjustment function, which divides the cost change into multiple intervals and sets different adjustment ratios for different intervals. The larger the change, the higher the corresponding adjustment ratio. It is suitable for scenarios with moderate cost fluctuations and can achieve a step-by-step precise response to the change. Third, the threshold dynamic adjustment function, which sets an upper limit threshold for the cost change. When the cost change exceeds the threshold, the adjustment value no longer increases with the increase of the change. It is suitable for scenarios with large cost fluctuations and can avoid the distortion of evaluation results caused by extreme cost changes.

[0055] Subsequently, function adaptation is performed based on the direction (increase or decrease) and magnitude of changes in construction costs: when the rate of change in labor or machinery costs is ≤ a preset small change threshold, a proportional dynamic adjustment function is applied to ensure that the correction intensity matches the small change; when the rate of change in costs is > a small change threshold and ≤ a medium change threshold, a segmented proportional adjustment function is switched to increase the correction ratio according to the range of changes; when the rate of change in costs is > a medium change threshold, a threshold-based dynamic adjustment function is activated to avoid the risk of over-correction through threshold limits. Simultaneously, differentiated thresholds are set for the different characteristics of labor and machinery costs. For example, labor costs have a more direct impact on the overall project cost and fluctuate relatively stably, so a relatively strict threshold can be set; machinery costs are more volatile due to market supply and demand, so the threshold can be appropriately relaxed, thereby achieving precise function adaptation.

[0056] Next, the classification adjustment values ​​are calculated: based on the dynamic adjustment function adapted in the previous steps, and combined with the cost proportion weights of labor costs and machinery costs, the classification adjustment values ​​corresponding to the two types of costs are calculated separately. For labor costs, the classification adjustment value = change in labor costs × labor cost proportion weight × corresponding dynamic adjustment function coefficient, where the labor cost proportion weight needs to be determined in conjunction with the project type (e.g., labor-intensive projects have a higher weight), and the function coefficient will be dynamically adjusted according to the range of change; for machinery costs, the classification adjustment value = change in machinery costs × machinery cost proportion weight × corresponding dynamic adjustment function coefficient, with an overall weight lower than that of labor costs, ensuring that the adjustment priority matches the degree of cost impact.

[0057] Subsequently, a combined calculation of the total dynamic adjustment value is performed: the adjustment values ​​for labor costs and machinery costs are weighted and summed according to preset weights, with the combined weight for labor cost adjustments being no less than 50% and the combined weight for machinery cost adjustments being no more than 50%, thus highlighting the core impact of labor cost changes on the overall cost. Simultaneously, the adjustment value's sign is determined based on the direction of cost changes. The adjustment value is positive when costs increase and negative when costs decrease, ensuring that the direction of the adjustment value is consistent with the logic of cost changes. Ultimately, a total dynamic adjustment value that accurately reflects the overall impact of changes in construction costs is generated.

[0058] The process retrieves the total deviation correction value calculated using the cost deviation correction function from the previous process. This value is primarily used to compensate for assessment deviations caused by fluctuations in building material prices. Simultaneously, it retrieves the total dynamic adjustment value generated in this process, which is mainly used to correct assessment deviations caused by changes in construction costs. A collaborative correction logic is then established: when both building material prices and construction costs rise, the correction formula is: Initial Assessment Value + Total Deviation Correction Value - Total Dynamic Adjustment Value. This compensates for the impact of rising building material prices with a positive deviation correction value and deducts the overestimation deviation caused by rising construction costs with a positive dynamic adjustment value. When both building material prices and construction costs fall, the correction formula is: Initial Assessment Value - Total Deviation Correction Value + Total Dynamic Adjustment Value. This offsets the underestimation deviation caused by falling building material prices with a negative deviation correction value and supplements the actual impact of falling construction costs with a negative dynamic adjustment value. For scenarios where building material and construction costs move in opposite directions, a weighted balance is applied based on the absolute values ​​of the two types of correction values ​​to ensure that the correction results simultaneously adapt to changes in both types of factors. This collaborative correction logic enables comprehensive calibration of multi-dimensional cost-influencing factors, ultimately outputting accurate assessment results that closely match the actual project cost, ensuring the scientific rigor and practicality of the cost assessment.

[0059] In one embodiment, obtaining a cost assessment model based on a convolutional neural network framework includes: Obtain the project batch, project type, and construction unit identification number for the construction project; Obtain a preset decoding table, and modify the preset decoding table based on the construction unit identification number to obtain a modified decoding table; Based on the aforementioned change decoding table, the project batch is decoded to obtain the decoding identifier; A corresponding curve segment is generated based on the project type; the identifier characters in the decoded identifier are evenly distributed on the curve segment, and multiple target identifier characters are selected from them based on the spatial relationship between each identifier character and the curve segment to form a target character segment; Match the cost assessment model corresponding to the target character segment in the database.

[0060] In this embodiment, three types of core identification information are first collected: First, the project batch, which is the batch number of the project, usually compiled by the project construction unit or competent authority according to the project approval order. It contains implicit information such as the project construction period and regional sequence, and can be used to distinguish similar projects with different construction cycles and batches. Second, the project type, which needs to clarify the core business attributes of the project, such as residential buildings, commercial complexes, industrial plants, municipal roads, etc. The cost composition and influencing factors of different project types are significantly different, and it is the core classification basis for subsequent model screening. Third, the construction unit identification number, which is the unique identification code of the construction unit. It is usually associated with key information such as the construction unit's qualification level, historical construction performance, and professional technical capabilities, and can be used to match a cost assessment model that is suitable for the construction characteristics of the unit. The accurate collection of these three types of information lays the foundational data foundation for subsequent decoding, identification generation, and model matching.

[0061] The system retrieves a pre-defined decoding table, which is a set of industry-standard project batch decoding rules. This table includes standard batch coding decomposition logic and the correspondence between coded characters and basic project attributes, enabling standardized decoding of project batches. Then, based on the construction unit's identification number, the pre-defined decoding table is modified: first, the system retrieves the unit's historical construction data using its identification number, analyzing its typical construction techniques, cost control habits, cost deviation characteristics, and other core attributes. Based on these attributes, the coding weights and character mapping rules in the pre-defined decoding table are adjusted. For example, if a construction unit has higher cost control accuracy in steel structure projects, the weight of steel structure-related coded characters in the decoding table is increased; if it is prone to cost deviations in the decoration and finishing stage, the mapping association of decoration-related codes is strengthened. Finally, a modified decoding table adapted to the construction unit is generated, giving the decoding rules unit-specific characteristics.

[0062] After the change decoding table is generated, the project batches will be decoded layer by layer according to the customized decoding rules in the table. For example, the string of the project batch is split into segments of a preset length, and then each segment is matched with the mapping rules in the change decoding table to convert basic characters such as numbers and letters in the batch into feature characters with engineering significance. Then, all the converted feature characters are integrated in the order of batch coding to form a set of decoding identifiers with project batch attributes and construction unit characteristics. This identifier can accurately represent the core basic characteristics of the project and is the key link connecting project information and model matching.

[0063] Based on the project type, a corresponding curve segment is generated: This can be done by retrieving historical cost fluctuation data for that project type and generating a curve segment that characterizes the cost characteristics of that type of project, based on dimensions such as cost fluctuation amplitude and the changing trends of key influencing factors. The curvature of the curve segment corresponds to the importance of the cost influencing factors; the larger the curvature, the more critical the corresponding cost influencing factor. Then, all the identifier characters in the decoded identifier are evenly distributed on this curve segment according to an equal spacing rule, so that each character corresponds to a specific position on the curve segment. Next, a selection process is conducted based on the spatial relationship between the identifier characters and the curve segment: priority is given to selecting the identifier characters corresponding to the extreme points of curvature of the curve segment (i.e., the most critical areas of cost influence), while redundant characters in the flat areas of the curve segment are removed. Finally, the selected core characters are integrated according to the spatial order of the curve segment to form a target character segment. This character segment contains both the core characteristics of the project batch and the construction unit, and also fits the cost characteristics of the project type, providing a key basis for accurate model matching.

[0064] A database is pre-built to associate target character segments with cost assessment models. This database stores convolutional neural network models corresponding to different project types, construction companies, and project batches, with each model labeled with a unique feature character segment. Once the target character segment is generated, it is precisely matched against the model labels in the database: first, a search is performed using exact matching rules for the character segment; if a completely matching label exists, the corresponding convolutional neural network cost assessment model is directly retrieved; otherwise, a fuzzy matching is performed based on the similarity of the core characters of the character segment, selecting the model with the highest similarity as the suitable model. This matching logic allows for the rapid acquisition of cost assessment models that highly match the characteristics of the current project, ensuring the accuracy and suitability of subsequent cost assessments.

[0065] In one embodiment, obtaining a cost assessment model based on a convolutional neural network framework includes: Obtain project batches, engineering types, and construction unit identification numbers for construction projects; collect project regional characteristics, structural characteristics, and construction period characteristics; and construct a project feature set. By retrieving historical data from similar projects of the construction unit, the three-dimensional capability weights of the construction unit are calculated from three dimensions: process adaptability, cost control accuracy, and schedule fulfillment rate. The three-dimensional capability weights are mapped to the project feature set in a matrix to generate a unique project-unit adaptation feature map, which intuitively presents the matching degree distribution between project features and construction unit capabilities. A pre-defined modular convolutional neural network model pool is provided. The model pool is divided into pluggable modules according to function, including a basic feature extraction module, a multi-source data fusion module, a deviation calibration module, and an output module. Each module is labeled with an appropriate project feature label. Based on the feature labels of the project-unit adaptation feature map, the matching basic feature extraction module, multi-source data fusion module, deviation calibration module and output module are selected from the model pool and dynamically assembled to form an initial customized model; The regional building material supply and demand coefficient based on the project's geographical characteristics is converted into a price sensitivity coefficient; based on the price sensitivity coefficient, the weight of building material data in the multi-source data fusion module is adjusted; based on the key node time in the construction period characteristics, the time decay coefficient of the deviation calibration module is optimized, and the parameter adaptation and calibration between modules are completed to obtain the cost assessment model. In this embodiment, the first step is to collect basic identification information, including project batches (batch numbers with implicit attributes such as construction period and regional sequence), project types (clearly defining core business formats such as residential, commercial complexes, and industrial plants), and construction unit identification numbers (unique codes associated with the unit's qualifications and performance). Based on this, three types of key characteristic information are further collected: first, project regional characteristics, specifically covering region-specific data such as regional building material supply and demand coefficients, transportation cost benchmarks, and local labor wage guidance prices in the city where the project is located; second, structural characteristics, including structural indicators such as the project's main structural form (frame, shear wall, steel structure, etc.), the proportion of prefabricated components, and seismic resistance level; and third, construction period characteristics, involving construction period control data such as the overall planned construction period, key milestone times at each stage, and construction period flexibility coefficients. Integrating the above basic identification information with the expanded characteristic information forms a project feature set covering all dimensions of the project's attributes, laying a solid data foundation for subsequent model customization.

[0066] By using the construction unit's identification number, complete historical construction data from at least five similar projects of that unit are retrieved. This data must cover core aspects such as project construction technology selection, cost control records, and schedule fulfillment records. Subsequently, a capability quantification analysis is conducted from three dimensions: first, technology adaptability, which calculates a technology adaptability score by comparing the unit's process solutions from similar projects with the optimal process solution for the project, combined with the cost deviation rate after process implementation; second, cost control accuracy, which quantifies cost control capability by using the deviation rate between actual and planned costs of similar projects as the core indicator, supplemented by the cost deviation correction efficiency; and third, schedule fulfillment rate, which assesses the schedule control level by statistically analyzing the on-time completion rate of key nodes and the overall schedule fulfillment rate of similar projects, combined with the cost impact coefficient of schedule delays. Finally, the entropy weight method is used to assign weights to the scores of the three dimensions. The entropy weight method determines the weight based on the dispersion of each dimension's indicators; higher dispersion indicates a greater impact of that dimension on model adaptability. This ultimately generates three-dimensional capability weights for the construction unit, achieving accurate quantification of the construction unit's comprehensive capabilities.

[0067] The project feature set is quantified and encoded across various dimensions. For example, project type is mapped to a business type coefficient in the 0-1 range, the proportion of prefabricated components is converted into an industrialization degree coefficient, and the regional building material supply and demand coefficient is converted into a basic price sensitivity coefficient, thus completing the numerical processing of the features. Subsequently, a feature-capability mapping matrix is ​​constructed. The rows of the matrix represent the quantified indicators of the project feature set, and the columns represent the three-dimensional capability weights of the construction unit. The matching degree between each project feature and the construction unit's capability is calculated through matrix inner product operations. Finally, the matching degree values ​​are visualized according to feature categories, generating a project-unit adaptation feature map.

[0068] A pre-built modular convolutional neural network model pool is constructed. The core of this pool is based on the cost assessment business process, breaking down the complete convolutional neural network model into four independent and freely combinable pluggable functional modules: First, a basic feature extraction module, containing convolutional units of different kernel sizes (e.g., 3×3, 5×5) and depths to adapt to project feature extraction needs of varying complexity, and labeled with appropriate structural feature tags (e.g., large convolutional kernel module for high-prefabrication projects); second, a multi-source data fusion module, integrating different types of attention mechanisms (e.g., self-attention, cross-attention) to achieve differentiated fusion of multi-source data such as engineering design, building material prices, and construction costs, and labeled with appropriate regional feature tags (e.g., cross-attention module for high-price-sensitive areas); third, a deviation calibration module, containing calibration units with different function combinations (e.g., linear- The system includes four modules: a nonlinear composite function and a time-decay function, used to correct for assessment biases caused by fluctuations in building material prices and changes in construction costs; and an output module with built-in logic for different output results (such as numerical output, trend curve output, and risk warning output), which can match different assessment results requirements and is labeled with appropriate project type tags. All modules are interconnected through standardized interfaces to ensure that the combined model has complete cost assessment capabilities.

[0069] The process involves analyzing the feature tags of the project's unit-adapted feature map and extracting core features with high matching degrees. For example, for a residential project, high-matching features include "high prefabrication rate structure," "regional tight supply and demand of building materials," and "concentration of key nodes." Based on these core feature tags, precise module selection is performed from the modular model pool: for the "high prefabrication rate structure" feature tag, a large convolutional kernel-based feature extraction module is selected; for the "regional tight supply and demand of building materials" feature tag, a multi-source data fusion module with a cross-attention mechanism is selected; for the "concentration of key nodes" feature tag, a deviation calibration module with a time-series decay function is selected; and, combined with the residential project's engineering type tag, an output module supporting numerical and trend curve output is selected. Finally, the four selected modules are dynamically assembled through a standardized interface according to the cost assessment business logic of basic feature extraction, multi-source data fusion, deviation calibration, and output, forming an initial customized model. The functional units of this model fully match the project's core features and the construction unit's capabilities, possessing preliminary dedicated assessment capabilities.

[0070] Next, parameter calibration based on regional characteristics is performed: Regional building material supply and demand coefficients are extracted from the project's regional characteristics and converted into precise price sensitivity coefficients using a preset algorithm. This coefficient quantifies the degree of fluctuation in regional building material prices due to supply and demand. Based on this price sensitivity coefficient, the fusion weights of building material price data in the multi-source data fusion module are adjusted. For example, the higher the price sensitivity coefficient, the greater the fusion weight of building material price data, ensuring the model prioritizes capturing the price characteristics of highly volatile building materials. Subsequently, parameter calibration based on construction period characteristics is performed: Key node times are extracted from the construction period characteristics, and the time distribution density of key nodes is analyzed. Based on the distribution density, the time decay coefficient of the deviation calibration module is optimized. For example, the denser the key nodes, the higher the weight of the time decay coefficient, making the model more sensitive to cost changes near key nodes. After completing these two types of parameter calibrations, the operating logic and parameter thresholds between modules will be fully adapted to the project's regional and construction period characteristics. The initial customized model will be upgraded to a directly applicable, highly specific, and adaptable convolutional neural network cost assessment model, ensuring the accuracy and reliability of subsequent cost assessments.

[0071] In the above embodiments, this application incorporates some existing algorithms and technical features for explanation and description to make the specification more detailed, clear, and complete, thus complying with the provisions of the Patent Law. However, this is not achieved by using a series of complex steps and algorithmic formulas, nor by complicating the technical solution, nor by combining or stacking conventional or simple features. The existing algorithms and technical features listed are for the purpose of disclosing the specific implementation methods of each step of this application (not to limit this application) and to avoid situations where this application cannot be implemented.

[0072] Reference Figure 2Another embodiment of the present invention also provides a multi-dimensional dynamic cost assessment system for construction projects, comprising: The monitoring module is used to acquire basic engineering design data, building material price data, and construction cost data for building projects, and obtain standardized multi-source cost monitoring data. The acquisition module is used to acquire a cost assessment model based on a convolutional neural network framework. The feature extraction layer of the cost assessment model embeds a multi-source data feature weighted fusion module, which allocates fusion weights according to the degree of influence of various data on the cost through a dynamic weighting algorithm. A dual-function cost calibration mechanism, consisting of a cost deviation correction function and a dynamic adjustment function, is embedded before the model output layer. This dual-function cost calibration mechanism corrects the assessment deviation in real time based on the fluctuation range of building material prices and changes in construction costs. The evaluation module is used to input standardized multi-source cost monitoring data into the cost evaluation model. After the multi-source data is weighted and fused through the feature extraction layer, the evaluation deviation is corrected by the dual-function cost calibration mechanism, and finally the dynamic cost evaluation value and cost change trend of the building project are output.

[0073] In this embodiment, the specific implementation of each module in the above system embodiment is described in the above method embodiment, and will not be repeated here.

[0074] Reference Figure 3 This invention also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 3 As shown, the computer device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the data corresponding to this embodiment. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0075] Those skilled in the art will understand that Figure 3 The structures shown are merely block diagrams of some structures related to the present invention and do not constitute a limitation on the computer devices on which the present invention is applied.

[0076] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0077] In summary, the multi-dimensional dynamic cost assessment method and system for construction projects provided in this embodiment of the invention includes: acquiring basic engineering design data, building material price data, and construction cost data of the construction project to obtain standardized multi-source cost monitoring data; acquiring a cost assessment model based on a convolutional neural network framework; wherein, the feature extraction layer of the cost assessment model embeds a multi-source data feature weighted fusion module, and allocates fusion weights according to the degree of influence of various types of data on the cost through a dynamic weight algorithm; a dual-function cost calibration mechanism consisting of a cost deviation correction function and a dynamic adjustment function is embedded before the model output layer, and the dual-function cost calibration mechanism corrects the assessment deviation in real time based on the fluctuation range of building material prices and changes in construction costs; the standardized multi-source cost monitoring data is input into the cost assessment model, and after the multi-source data weighted fusion is completed through the feature extraction layer, the assessment deviation is corrected by the dual-function cost calibration mechanism, and finally the dynamic cost assessment value and cost change trend of the construction project are output. In this invention, a dual-function cost calibration mechanism consisting of a cost deviation correction function and a dynamic adjustment function is embedded before the model output layer. The dual-function cost calibration mechanism corrects the evaluation deviation in real time based on the fluctuation range of building material prices and changes in construction costs. It can dynamically correct the evaluation results according to the real-time fluctuation range of building material prices and changes in construction costs, which not only adds a dynamic adjustment mechanism but also improves the accuracy of the evaluation.

[0078] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the present invention and embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0079] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0080] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A multi-dimensional dynamic cost assessment method for construction projects, characterized in that, Includes the following steps: Obtain basic engineering design data, building material price data, and construction cost data for building projects to obtain standardized multi-source cost monitoring data; A cost assessment model based on a convolutional neural network is obtained; wherein, the feature extraction layer of the cost assessment model embeds a multi-source data feature weighted fusion module, and allocates fusion weights according to the degree of influence of various types of data on the cost through a dynamic weight algorithm; a dual-function cost calibration mechanism consisting of a cost deviation correction function and a dynamic adjustment function is embedded before the model output layer, and the dual-function cost calibration mechanism corrects the assessment deviation in real time based on the fluctuation range of building material prices and changes in construction costs; Standardized multi-source cost monitoring data is input into the cost assessment model. After the multi-source data is weighted and fused through the feature extraction layer, the assessment deviation is corrected by the dual-function cost calibration mechanism, and finally the dynamic cost assessment value and cost change trend of the building project are output.

2. The multi-dimensional dynamic cost assessment method for building construction projects according to claim 1, characterized in that, The basic engineering design data includes the building's main structural type, component size parameters, spatial layout planning data, and decoration and finishing standard data.

3. The multi-dimensional dynamic cost assessment method for building construction projects according to claim 1, characterized in that, Standardized multi-source cost monitoring data is input into the cost assessment model. After weighted fusion of multi-source data through a feature extraction layer, the assessment bias is corrected by a dual-function cost calibration mechanism. Finally, the dynamic cost assessment value and cost change trend of the construction project are output, including: Standardized multi-source cost monitoring data are converted into feature matrices adapted to convolutional neural network inputs according to the classification dimensions of engineering design basic data, building material price data, and construction cost data. Each feature matrix is ​​input into the feature extraction layer of the cost assessment model, and the fusion weight of each feature matrix is ​​calculated and allocated in real time through the dynamic weight algorithm of the multi-source data feature weight fusion module. Multi-source data features are weighted and fused based on the assigned fusion weights to obtain a fused feature vector; The dual-function cost calibration mechanism integrates feature vectors. First, it calculates the deviation correction value based on the fluctuation range of building material prices in the building material price data using the cost deviation correction function. Then, it calculates the dynamic adjustment value by combining the changes in construction costs in the construction cost data using the dynamic adjustment function, and evaluates the deviation correction. The output layer outputs the dynamic cost assessment value and cost change trend curve of the building project.

4. The multi-dimensional dynamic cost assessment method for building construction projects according to claim 3, characterized in that, in, The feature matrix dimension of the basic engineering design data is set to 256×256, and the feature matrix dimension of the building material price data and construction cost data is set to 128×128.

5. The multi-dimensional dynamic cost assessment method for building construction projects according to claim 3, characterized in that, Real-time calculation and allocation of fusion weights for each feature matrix, including: Based on the feature matrix of basic engineering design data, two core influencing factors were extracted: building structure complexity coefficient and component quantity ratio. Based on the feature matrix of building material price data, two core influencing factors were extracted: main material price volatility and category scarcity coefficient. Based on the feature matrix of construction cost data, two core influencing factors were extracted: labor cost ratio and machinery usage frequency. Based on historical data of construction projects, the correlation between six core influencing factors and cost assessment errors was analyzed, and weights were assigned to each core influencing factor. The three types of feature matrices are used as query vector, key vector, and value vector, respectively. After linear transformation, the basic attention score is obtained through the attention score calculation model. Then, the initial fusion weight is obtained by combining the weights of the core influence factors corresponding to each feature matrix. The initial fusion weights are dynamically adjusted based on the fluctuation range of building material prices and the rate of change in construction costs in the current batch of data; the dynamically adjusted fusion weights are then assigned to the corresponding feature matrices.

6. The multi-dimensional dynamic cost assessment method for building construction projects according to claim 3, characterized in that, The cost deviation correction function is used to calculate the deviation correction value based on the fluctuation range of building material prices in the building material price data, including: Extract the price fluctuation range of various main materials from building material price data; classify the main materials into core main materials and auxiliary main materials according to their impact on construction project costs; The price fluctuation ranges of core main materials and auxiliary main materials are respectively adapted with corresponding correction functions; among them, the correction functions include linear deviation correction functions and piecewise nonlinear deviation correction functions; Based on the adapted correction function, the single-type deviation correction value of the core main material and the auxiliary main material are calculated separately; the single-type deviation correction values ​​of all main materials are weighted and summed to obtain the deviation correction value.

7. The multi-dimensional dynamic cost assessment method for building construction projects according to claim 3, characterized in that, The dynamic adjustment value is calculated using a dynamic adjustment function, taking into account changes in construction costs in the construction cost data, and then the evaluation bias is corrected, including: Extract changes in construction costs from construction cost data, including changes in labor costs and machinery costs; The corresponding dynamic adjustment function is determined based on the direction and magnitude of changes in construction costs; the dynamic adjustment function includes proportional dynamic adjustment function, piecewise proportional adjustment function, and threshold dynamic adjustment function; Based on the dynamic adjustment function corresponding to each change in construction cost, the corresponding classification adjustment value is calculated, and the dynamic adjustment value is obtained by merging the calculations. The deviation correction is evaluated by combining the aforementioned deviation correction value and the dynamic adjustment value.

8. A multi-dimensional dynamic cost evaluation system for construction projects, characterized in that, include: The monitoring module is used to acquire basic engineering design data, building material price data, and construction cost data for building projects, and obtain standardized multi-source cost monitoring data. The acquisition module is used to acquire a cost assessment model based on a convolutional neural network framework. The feature extraction layer of the cost assessment model embeds a multi-source data feature weighted fusion module, which allocates fusion weights according to the degree of influence of various data on the cost through a dynamic weighting algorithm. A dual-function cost calibration mechanism, consisting of a cost deviation correction function and a dynamic adjustment function, is embedded before the model output layer. This dual-function cost calibration mechanism corrects the assessment deviation in real time based on the fluctuation range of building material prices and changes in construction costs. The evaluation module is used to input standardized multi-source cost monitoring data into the cost evaluation model. After the multi-source data is weighted and fused through the feature extraction layer, the evaluation deviation is corrected by the dual-function cost calibration mechanism, and finally the dynamic cost evaluation value and cost change trend of the building project are output.

Images