Prediction data evaluation and analysis method and system based on construction engineering cost

Abstract

The invention discloses a prediction data evaluation analysis method and system based on construction engineering cost, and relates to the technical field of intelligent data analysis, and the method comprises the steps: collecting multi-source heterogeneous data of construction engineering cost, carrying out the fusion through a multi-head attention mechanism, and generating a comprehensive cost feature vector; constructing a variational Bayesian neural network, outputting a cost prediction value and probability distribution according to the comprehensive cost feature vector, and calculating a confidence interval of the cost prediction value; and combining the risk thermodynamic diagram, the cost prediction value and the BIM model of the constructional engineering through MR interaction to generate a cost visualization view, and according to the change of the actual construction condition, adjusting the cost visualization view in real time through interaction with the cost visualization view, and evaluating the cost prediction value. According to the method, the network weight is parameterized through Gaussian variational distribution, point estimation of a traditional neural network is replaced, and quantification of prediction uncertainty is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent data analysis, and in particular to a method and system for evaluating and analyzing forecast data based on construction project costs. Background Art

[0002] In recent years, with the continuous expansion and significant increase in the complexity of construction projects, the importance of construction cost forecasting in project management has become increasingly prominent. Traditional construction cost forecasting methods primarily rely on static models such as empirical estimation, quota methods, or regression analysis. Their core is linear deduction based on historical data and manual experience.

[0003] With the improvement of the informatization level of the construction industry, especially the development of the Internet of Things (IoT), big data analysis and artificial intelligence technology, construction cost forecasting has gradually evolved towards dynamic and intelligent directions. To improve the prediction accuracy and adaptability, researchers have gradually introduced machine learning and deep learning technologies, such as support vector machines (SVM), random forests (RF), long short-term memory networks (LSTM) and convolutional neural networks (CNN), to achieve modeling and analysis of multi-source information such as historical cost data, material price fluctuations and construction progress. Although the above technologies have improved the accuracy and automation of engineering cost forecasting to a certain extent, there are still many bottlenecks and challenges in practical applications. For example, the prediction model only outputs a single value and cannot quantify the confidence interval and probability distribution of the prediction results, making it difficult to meet the needs of risk assessment and uncertainty quantification in engineering practice. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for evaluating and analyzing forecast data based on construction project cost to solve the problem of difficulty in quantifying the uncertainty of construction project cost forecast results.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a method for evaluating and analyzing forecast data based on construction project cost, which comprises: Collect multi-source heterogeneous data on construction project costs and fuse them through a multi-head attention mechanism to generate a comprehensive cost feature vector; Construct a variational Bayesian neural network to output the cost prediction value and probability distribution based on the comprehensive cost feature vector, and calculate the confidence interval of the cost prediction value; Generate a risk heat map based on the probability distribution and confidence interval of the cost forecast value; Through MR interaction, the risk heat map, cost forecast value and BIM model of the construction project are combined to generate a cost visualization view. According to the actual changes in construction conditions, the cost visualization view is interacted with, the cost visualization view is adjusted in real time, and the cost forecast value is evaluated.

[0007] As a preferred solution of the forecast data evaluation and analysis method based on construction project cost of the present invention, the steps of generating the comprehensive cost feature vector are as follows: Collect multi-source heterogeneous data through IoT devices, public databases, and web crawler tools; Multi-source heterogeneous data are recorded in a unified timestamp format and stored as a structured table to obtain the original cost data set; Clean and standardize the original cost data set to generate a standard cost data set; The standardized cost dataset is fused through a multi-head attention mechanism to generate a preliminary cost feature matrix. The preliminary cost feature matrix is ​​reduced in dimension through variational autoencoder to generate a comprehensive cost feature vector.

[0008] As a preferred solution of the forecast data evaluation and analysis method based on construction project cost according to the present invention, the specific steps of constructing a variational Bayesian neural network are as follows: Initialize the variational Bayesian neural network structure to an input layer and several fully connected layers; Assign Gaussian variational distribution to the weight parameters of each fully connected layer of the variational Bayesian neural network and initialize them; The variational Bayesian neural network is trained by minimizing the loss function.

[0009] As a preferred solution of the forecast data evaluation and analysis method based on construction project cost described in the present invention, wherein: Gaussian variational distribution of weight parameters of each fully connected layer is allocated, the specific steps are as follows: The weight parameters include the weight matrix and bias vector of each fully connected layer of the variational Bayesian neural network; Based on variational inference, Gaussian variational distribution is assigned to the weight matrix and bias term of each fully connected layer to generate the mean weight matrix, standard deviation weight matrix, mean bias vector and standard deviation bias vector; Use the Xavier initialization method to initialize the mean weight matrix and standard deviation weight matrix; The mean bias vector and standard deviation bias vector are initialized using uniform distribution.

[0010] As a preferred solution of the forecast data evaluation and analysis method based on construction project cost according to the present invention, the specific steps of training the variational Bayesian neural network are as follows: Based on the marginal likelihood lower bound, define the loss function of the variational Bayesian neural network; Based on the loss function, the partial derivative of the loss function is obtained by back propagation for each weight parameter's Gaussian variational distribution; Use the Adam optimizer to update the weight parameters and the Gaussian variational distribution of the weight parameters.

[0011] As a preferred solution of the forecast data evaluation and analysis method based on construction project cost of the present invention, wherein: the output cost forecast value and probability distribution are specifically performed as follows: The comprehensive cost feature vector is input into the trained variational Bayesian neural network, and the mean branch and standard deviation branch of the last fully connected layer respectively output the mean and standard deviation of the cost prediction value; Output the mean of the cost forecast values ​​as the cost forecast value; Based on the mean and standard deviation of the cost prediction value, the Gaussian distribution assumption is adopted to calculate the probability distribution density function value, and Monte Carlo sampling is used to generate the numerical representation of the probability distribution.

[0012] As a preferred solution of the method for evaluating and analyzing the forecast data based on the construction project cost of the present invention, the specific steps of generating the cost visualization view are as follows: Build BIM models for construction projects; Based on the mapping relationship between the record number of multi-source heterogeneous data and the construction number in the BIM model, the cost prediction value is associated with the BIM model of the construction project; Extracting the spatial coordinate system of the BIM model of the construction project, calculating the bounding box of the BIM model of the construction project, and defining the center of the BIM model of the construction project as the center of the bounding box; Overlay the risk heat map onto the horizontal plane of the BIM model of the building project in the form of a semi-transparent grid; The height is fixed to the center of the building to generate a cost visualization view.

[0013] In a second aspect, the present invention provides a forecast data evaluation and analysis system based on construction project cost, comprising a feature vector module, a cost prediction module, a risk map module and an interactive evaluation module. The feature vector module is used to collect multi-source heterogeneous data of construction project costs and fuse them through a multi-head attention mechanism to generate a comprehensive cost feature vector; The cost prediction module is used to construct a variational Bayesian neural network, output a cost prediction value and probability distribution based on the comprehensive cost feature vector, and calculate the confidence interval of the cost prediction value; The risk map module is used to generate a risk heat map based on the probability distribution and confidence interval of the cost forecast value; The interactive evaluation module is used to combine the risk heat map, cost prediction value and BIM model of the construction project through MR interaction to generate a cost visualization view. According to changes in actual construction conditions, the cost visualization view is adjusted in real time by interacting with the cost visualization view and the cost prediction value is evaluated.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the method for evaluating and analyzing predictive data based on construction project cost as described in the first aspect of the present invention.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for evaluating and analyzing predictive data based on construction project cost as described in the first aspect of the present invention.

[0016] The beneficial effects of the present invention are as follows: the present invention parameterizes network weights through Gaussian variational distribution, replacing the point estimation of traditional neural networks, thereby realizing the quantification of prediction uncertainty, calculating risk scores through the weighted ratio of standard deviation to confidence interval width, realizing multi-dimensional risk aggregation, and the intuitive display of heat maps can be used for resource priority allocation. The feature vectors are adjusted through HoloLens 2 gesture interaction to realize dynamic scene deduction. Users can observe the adjustment effects in real time, assisting in rapid decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 The figure is a flow chart of the evaluation and analysis method based on the forecast data of the construction project cost.

[0019] Figure 2 This is a module diagram of the forecast data evaluation and analysis system based on the construction project cost.

[0020] Figure 3 Schematic diagram of the variational Bayesian neural network structure.

[0021] Figure 4 Flowchart generated for MR interaction visualization. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0025] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a method for evaluating and analyzing forecast data based on construction project cost, comprising the following steps: S1. Collect multi-source heterogeneous data on construction project costs and fuse them through a multi-head attention mechanism to generate a comprehensive cost feature vector.

[0026] Among them, multi-source heterogeneous data includes real-time construction data, historical cost data and external market data; Collect real-time construction data through IoT devices.

[0027] Specifically, radio frequency identification tags are used to track daily material consumption, such as daily cement usage; infrared sensors are used to record worker attendance data, such as the number of workers on each type of work and the duration of their work; and embedded sensors are used to monitor the operating status of construction equipment, such as the daily operating hours of excavators. Collect historical cost data from public databases.

[0028] Specifically, by accessing industry sharing platforms, we can obtain cost data for similar projects across developers, such as the unit price of materials for each project, the labor costs of various types of work, etc. Use web crawler tools to collect external market data.

[0029] Specifically, the Scrapy framework (a web crawler tool) can be used to extract real-time prices of building materials, such as the daily price of cement, from the building materials market platform. Macroeconomic indicators, such as the annual inflation rate, can also be obtained through public API interfaces. Multi-source heterogeneous data are recorded in a unified timestamp format and stored as a structured table to obtain the original cost data set; Clean the original cost data set, i.e. remove outliers, and standardize it to generate a standard cost data set; Specifically, the anomaly score of each original cost data is calculated according to the anomaly score formula. According to the anomaly score distribution of all original cost data, an anomaly score threshold is set. When the anomaly score of the original cost data is less than the anomaly score threshold, it is determined to be an outlier and removed. The original cost data after the outlier is removed is interpolated. The Z-score normalization method is applied to standardize the original cost data set after outlier removal and interpolation to obtain a standard cost data set. The standardized cost dataset is fused through a multi-head attention mechanism to generate a preliminary cost feature matrix. Specifically, the matrix construction and data splicing method is used to extract the features of each record (standardized, such as cement usage, etc.) from the standardized cost data set, and all features are spliced ​​into a matrix form in the order of record numbers. Each row represents the feature vector of a record, and each column represents a feature. The extracted features are filled into the matrix in a fixed order using numerical mapping to obtain a feature matrix. The feature matrix is ​​converted into a tensor format through matrix operations, and the multi-head attention mechanism is initialized, including setting the number of attention heads, single-head dimension, and total dimension. For example, the number of attention heads is set to 8, the single-head dimension is set to 64, and the total dimension is set to 512. According to the feature matrix, the query matrix, key matrix, and value matrix are calculated through linear transformation in the input layer, and the scaled dot product attention formula is used. Over-scale dot product and softmax normalization are performed to calculate the weight matrix of each attention head, add residual connection, concatenate the weight matrices of all attention heads, and obtain the attention weight matrix. In the fully connected layer, based on the fully connected layer weight matrix and the bias vector, the attention weight matrix is ​​linearly transformed to obtain the intermediate feature matrix. The ReLU activation function is used to judge each element of the intermediate feature matrix. If the element value is greater than 0, the original value is retained; if it is less than or equal to 0, it is set to 0 to enhance the feature expression ability and generate a nonlinear feature matrix. Dropout regularization is applied to the nonlinear feature matrix to prevent overfitting and generate a regularized feature matrix. The regularized feature matrix is ​​layer-normalized in the application layer to generate a preliminary cost feature matrix. The preliminary cost feature matrix is ​​reduced in dimension by using a variational autoencoder to generate a comprehensive cost feature vector. Among them, the variational autoencoder includes an encoder and a decoder. The encoder is used to map the preliminary cost feature matrix to the latent space, output the mean vector and the standard deviation vector, and generate the latent vector by reparameterizing the mean vector and the standard deviation vector. The decoder reconstructs the latent vector by minimizing the reconstruction loss and KL divergence to generate a comprehensive cost feature vector.

[0030] S2. Construct a variational Bayesian neural network, output the cost prediction value and probability distribution based on the comprehensive cost feature vector, and calculate the confidence interval of the cost prediction value.

[0031] Initialize the variational Bayesian neural network to a three-layer fully connected layer structure, and assign a Gaussian variational distribution to the weight parameters of each fully connected layer, where the weight parameters include a weight matrix and a bias vector; Specifically, the variational Bayesian neural network includes three fully connected layers. The first and second fully connected layers both use the ReLU activation function. The third layer includes two branches, including a mean branch and a standard deviation branch. Based on variational inference, a Gaussian variational distribution is assigned to the weight matrix of each fully connected layer of the variational Bayesian neural network to generate the corresponding mean weight matrix and standard deviation weight matrix. The mean matrix is ​​initialized using the Xavier initialization method, and the standard deviation matrix is ​​initialized to 0.1. A Gaussian variational distribution is assigned to the bias vector of each fully connected layer of the variational Bayesian neural network to generate the corresponding mean bias vector and standard deviation bias vector. The mean bias vector is initialized using a uniform distribution, and the standard deviation bias vector is initialized to a constant value of 0.1. The comprehensive cost feature vector is input into the first fully connected layer of the variational Bayesian neural network. The weight matrix and bias vector of the first fully connected layer are sampled from the Gaussian variational distribution through the reparameterization method to obtain the first layer output. The weight matrix and bias vector of the second fully connected layer are sampled in the same way as the first fully connected layer to obtain the second layer output. The mean branch and standard deviation branch of the third fully connected layer output the mean and standard deviation of the cost prediction value respectively. The calculation formula is as follows: ; ; in, is the mean of the predicted cost values, is the weight matrix of the cost prediction mean branch in the third layer, is the second layer output, is the bias vector of the cost prediction mean branch in the third layer, is the standard deviation of the cost forecast, is the weight matrix of the standard deviation branch of the cost prediction value in the third layer, is the bias vector of the standard deviation branch of the cost prediction value in the layer; Based on the lower bound of marginal likelihood (ELBO), the loss function of the variational Bayesian neural network is defined and calculated as follows: ; ; ; in, is the loss value, Is the likelihood term, which measures the fit between the predicted cost and the actual cost, and its value range is , is the KL divergence, constrained Gaussian variational distribution, with a value range of , For the Gaussian variational distribution The expectation operator under About weight parameters The expected value of Indicates that given the weight parameter Next, create real value The conditional probability density function describes the variational Bayesian neural network based on the weight parameter The degree of match between the generated cost forecast and the actual cost, To create real value, For the weight parameter Gaussian variational distribution of is the weight parameter The prior distribution of , the standard normal distribution, For the The real value of multi-source heterogeneous records, For the The average of the cost prediction values ​​of multi-source heterogeneous records, is the index variable of the number of multi-source heterogeneous data records in the comprehensive cost feature vector, is the number of data records in the comprehensive cost feature vector, For the The standard deviation of the cost prediction value corresponding to the training, For the Weight parameters The mean of the Gaussian variational distribution (mean weight matrix and mean bias vector), For the Weight parameters The standard deviation of the Gaussian variational distribution (standard deviation weight matrix and standard deviation bias vector), is the index variable for the number of weight parameters, is the number of weight parameters; It should be noted that for the weight parameter , refers to the index of each multi-source heterogeneous data record in the comprehensive cost feature vector. For example, when When referring to the daily consumption of cement among the daily consumption of various materials; The Gaussian variational distribution of each weight parameter is used to find the partial derivative of the loss function, calculate the gradient, and use the Adam optimizer to update the weight parameter and its Gaussian variational distribution; After the training is completed, the comprehensive cost feature vector is input into the variational Bayesian neural network, and the mean and standard deviation of the cost prediction value are output. The mean of the cost prediction value is output as the cost prediction value; Based on the mean and standard deviation of the cost forecast values, the Gaussian distribution assumption is adopted to calculate the probability distribution density function, and Monte Carlo sampling is used to generate a numerical representation of the probability distribution; It should be noted that the cost forecast value is not just the total cost of the construction project, but also includes the cost of each process, component, material, machinery and labor of the construction project; Based on the probability distribution of the cost forecast value, the confidence interval of the cost forecast value is calculated as follows: ; in, is the confidence interval, which indicates the confidence range of the cost forecast value. is the confidence coefficient, which corresponds to a 95% confidence level and is based on a standard normal distribution with a value of 1.96. is a volatility correction indicator with a value range of ; It should be noted that when When , we get the upper limit of the confidence interval. , we get the lower limit of the confidence interval.

[0032] S3. Generate a risk heat map based on the probability distribution and confidence interval of the cost forecast value.

[0033] Based on the probability distribution of the cost forecast value and the upper and lower limits of the confidence interval, the risk score is calculated by calculating the weighted ratio of the standard deviation of the cost forecast value to the width of the confidence interval. The calculation formula is as follows: ; in, Record risk scores for each piece of multi-source heterogeneous data. is the upper limit of the confidence interval, is the lower limit of the confidence interval; All risk scores are collected and normalized. Using linear color mapping with Matplotlib and based on the Viridis color scheme, the normalized risk scores are mapped to a color gradient to obtain a risk heat map.

[0034] S4. Through MR interaction, the risk heat map, cost forecast value and BIM model of the construction project are combined to generate a cost visualization view. According to the actual changes in construction conditions, the cost visualization view is adjusted in real time by interacting with the cost visualization view, and the cost forecast value is evaluated.

[0035] Build a BIM model of a construction project, including 3D geometric information of the construction project, such as foundation engineering and main structure, as well as metadata, such as component number and material type; Loading the Building Information Model (BIM) through Microsoft HoloLens 2, integrating risk heat maps and cost forecasts with the BIM model of the construction project. Microsoft HoloLens 2 is a head-mounted display device launched by Microsoft that is designed to provide a mixed reality experience. Users can see and interact with 3D holograms. Specifically, based on the mapping relationship between the record numbers of multi-source heterogeneous data and the building numbers in the BIM model of the construction project, the cost forecast value is associated with the BIM model of the construction project. For example, the cost forecast value of 50.5 million yuan for the main material (reinforced concrete) corresponds to component "C001" (beam, slab and column). The spatial coordinate system of the BIM model of the construction project is extracted, the bounding box of the BIM model of the construction project is calculated, the spatial range is determined, and the center of the BIM model of the construction project is defined as the center of the bounding box. The risk heat map is overlaid on the horizontal plane of the BIM model of the construction project in the form of a semi-transparent grid, with the height fixed to the height of the building center, to generate a cost visualization view. Based on actual construction changes, such as construction progress deviations, the cost visualization view can be adjusted in real time through HoloLens 2 gesture recognition; Specifically, a grabbing gesture is defined based on the Mixed Reality Toolkit. For example, pinching the thumbs and index fingers of both hands together for 0.5 seconds with a detection distance of less than 0.05 meters is recognized as a selection operation. The gesture recognition function of Microsoft HoloLens 2 is used to select components in the building information model. An interactive interface is displayed, which floats on the right side of the user's field of view and contains adjustable comprehensive cost feature vectors, such as construction progress deviations and material price fluctuations. Each comprehensive cost feature vector is represented by a slider, and the initial value corresponds to the real-time comprehensive cost feature vector. The user points to the specified comprehensive cost feature vector on the interactive interface, pinches with one hand to confirm, locks the adjustment target, drags with one hand along the horizontal axis, and HoloLens 2 tracks the gesture displacement and maps it to the change in the comprehensive cost feature vector. When both hands are opened, HoloLens 2 detects the release gesture, confirms the adjustment, and updates the real-time comprehensive cost feature vector. Based on the updated comprehensive cost feature vector, the variational Bayesian neural network is input to calculate the new cost visualization view in real time. Analyze the effects of adjustments to cost forecasts and confidence intervals in the updated cost visualization view; Specifically, the rate of change of the new cost forecast value and the original cost forecast value is calculated to obtain the cost change rate, the rate of change of the new confidence interval width and the original confidence interval width is calculated to obtain the width change rate, and the cost change rate threshold is set according to the mean and standard deviation of the new cost forecast value. The width change rate threshold is set according to the mean and standard deviation of the new confidence interval width. When the cost change rate is greater than the cost change rate threshold, it means that this adjustment needs to focus on whether there are any abnormalities in the cost forecast value, such as the cost forecast value differs from the same type and volume of construction projects by more than 25%, otherwise, it is not. When the width change rate is greater than the width change rate threshold, it means that this adjustment significantly increases the uncertainty of the cost forecast value, and the cost forecast value needs to be re-adjusted and re-forecasted, otherwise, it is not.

[0036] This embodiment also provides a forecast data evaluation and analysis system based on construction project cost, including: a feature vector module, a cost prediction module, a risk map module and an interactive evaluation module. The feature vector module is used to collect multi-source heterogeneous data on construction project costs and fuse them through a multi-head attention mechanism to generate a comprehensive cost feature vector; the cost prediction module is used to construct a variational Bayesian neural network to output the cost prediction value and probability distribution based on the comprehensive cost feature vector, and calculate the confidence interval of the cost prediction value; the risk map module is used to generate a risk heat map based on the probability distribution and confidence interval of the cost prediction value; the interactive evaluation module is used to combine the risk heat map, cost prediction value and BIM model of the construction project through MR interaction to generate a cost visualization view, and adjust the cost visualization view in real time by interacting with the cost visualization view according to changes in actual construction conditions, and evaluate the cost prediction value.

[0037] This embodiment also provides a computer device, which is suitable for the case of a method for evaluating and analyzing predictive data based on construction project costs, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the method for evaluating and analyzing predictive data based on construction project costs proposed in the above embodiment.

[0038] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0039] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for implementing the forecast data evaluation and analysis based on the construction project cost as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0040] In summary, the present invention uses Gaussian variational distribution to parameterize network weights, replacing the point estimation of traditional neural networks, thereby quantifying prediction uncertainty. The risk score is calculated by the weighted ratio of the standard deviation to the confidence interval width, realizing multi-dimensional risk aggregation. The intuitive display of the heat map can be used for resource priority allocation. The characteristic vector is adjusted through HoloLens 2 gesture interaction, realizing dynamic scene deduction. Users can observe the adjustment effect in real time, assisting in rapid decision-making.

[0041] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for evaluating and analyzing construction project cost prediction data, characterized by: include, Collect multi-source heterogeneous data on construction project costs and fuse them through a multi-head attention mechanism to generate a comprehensive cost feature vector; Construct a variational Bayesian neural network to output the cost prediction value and probability distribution based on the comprehensive cost feature vector, and calculate the confidence interval of the cost prediction value; Generate a risk heat map based on the probability distribution and confidence interval of the cost forecast value; Through MR interaction, the risk heat map, cost forecast value and BIM model of the construction project are combined to generate a cost visualization view. According to the actual changes in construction conditions, the cost visualization view is interacted with, the cost visualization view is adjusted in real time, and the cost forecast value is evaluated.

2. The method for evaluating and analyzing construction project cost prediction data according to claim 1, wherein: The specific steps of generating the comprehensive cost feature vector are as follows: Collect multi-source heterogeneous data through IoT devices, public databases, and web crawler tools; Multi-source heterogeneous data are recorded in a unified timestamp format and stored as a structured table to obtain the original cost data set; Clean and standardize the original cost data set to generate a standard cost data set; The standardized cost dataset is fused through a multi-head attention mechanism to generate a preliminary cost feature matrix. The preliminary cost feature matrix is ​​reduced in dimension through variational autoencoder to generate a comprehensive cost feature vector.

3. The method for evaluating and analyzing construction project cost prediction data according to claim 1, wherein: The specific steps of constructing the variational Bayesian neural network are as follows: Initialize the variational Bayesian neural network structure to an input layer and several fully connected layers; Assign Gaussian variational distribution to the weight parameters of each fully connected layer of the variational Bayesian neural network and initialize them; The variational Bayesian neural network is trained by minimizing the loss function.

4. The method for evaluating and analyzing construction project cost prediction data according to claim 3, wherein: Assign and initialize the Gaussian variational distribution of weight parameters. The specific steps are as follows: The weight parameters include the weight matrix and bias vector of each fully connected layer of the variational Bayesian neural network; Based on variational inference, Gaussian variational distribution is assigned to the weight matrix and bias term of each fully connected layer to generate the mean weight matrix, standard deviation weight matrix, mean bias vector and standard deviation bias vector; Use the Xavier initialization method to initialize the mean weight matrix and standard deviation weight matrix; The mean bias vector and standard deviation bias vector are initialized using uniform distribution.

5. The method for evaluating and analyzing construction project cost prediction data according to claim 3, wherein: The specific steps of training the variational Bayesian neural network are as follows: Based on the marginal likelihood lower bound, define the loss function of the variational Bayesian neural network; Based on the loss function, the partial derivative of the loss function is obtained by back propagation for each weight parameter's Gaussian variational distribution; Use the Adam optimizer to update the weight parameters and the Gaussian variational distribution of the weight parameters.

6. The method for evaluating and analyzing construction project cost prediction data according to claim 1, wherein: The specific steps of outputting the cost prediction value and probability distribution are as follows: The comprehensive cost feature vector is input into the trained variational Bayesian neural network, and the mean branch and standard deviation branch of the last fully connected layer respectively output the mean and standard deviation of the cost prediction value; Output the mean of the cost forecast values ​​as the cost forecast value; Based on the mean and standard deviation of the cost prediction value, the Gaussian distribution assumption is adopted to calculate the probability distribution density function value, and Monte Carlo sampling is used to generate the numerical representation of the probability distribution.

7. The method for evaluating and analyzing construction project cost prediction data according to claim 1, wherein: The specific steps for generating a cost visualization view are as follows: Build BIM models for construction projects; Based on the mapping relationship between the record number of multi-source heterogeneous data and the construction number in the BIM model, the cost prediction value is associated with the BIM model of the construction project; Extracting the spatial coordinate system of the BIM model of the construction project, calculating the bounding box of the BIM model of the construction project, and defining the center of the BIM model of the construction project as the center of the bounding box; Overlay the risk heat map onto the horizontal plane of the BIM model of the building project in the form of a semi-transparent grid; The height is fixed to the center of the building to generate a cost visualization view.

8. A system for evaluating and analyzing forecast data based on construction project costs, based on the method for evaluating and analyzing forecast data based on construction project costs according to any one of claims 1 to 7, characterized in that: Including feature vector module, cost prediction module, risk map module and interactive evaluation module, The feature vector module is used to collect multi-source heterogeneous data of construction project costs and fuse them through a multi-head attention mechanism to generate a comprehensive cost feature vector; The cost prediction module is used to construct a variational Bayesian neural network, output a cost prediction value and probability distribution based on the comprehensive cost feature vector, and calculate the confidence interval of the cost prediction value; The risk map module is used to generate a risk heat map based on the probability distribution and confidence interval of the cost forecast value; The interactive evaluation module is used to combine the risk heat map, cost prediction value and BIM model of the construction project through MR interaction to generate a cost visualization view. According to changes in actual construction conditions, the cost visualization view is adjusted in real time by interacting with the cost visualization view and the cost prediction value is evaluated.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for evaluating and analyzing construction project cost-based forecast data are implemented as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for evaluating and analyzing construction project cost-based forecast data according to any one of claims 1 to 7 are implemented.

Images