An engineering cost real-time evaluation and early warning system based on multi-source data fusion and dynamic learning

Abstract

This application discloses a real-time engineering cost assessment and early warning system based on multi-source data fusion and dynamic learning, comprising: a multi-source data acquisition module for collecting construction process data, historical project data, market data, and design drawing data; a preprocessing module for preprocessing and fusing the collected multi-source data to form a multi-dimensional feature vector with a unified timestamp; a hybrid neural network assessment module for inputting the multi-dimensional feature vector into a pre-constructed hybrid neural network model for cost assessment; a dynamic optimization learning module for online learning and dynamic optimization of the hybrid neural network assessment module based on a transfer learning strategy; a blockchain storage module for storing key data generated during the assessment process; and a progress monitoring and early warning module for real-time monitoring of construction progress, calculating progress deviations, and triggering corresponding early warnings. This system achieves accurate assessment, real-time early warning, and dynamic optimization of engineering costs.

Description

Technical Field

[0001] This invention relates to the field of engineering cost analysis technology, specifically to a real-time engineering cost assessment and early warning system based on multi-source data fusion and dynamic learning. Background Technology

[0002] With the deepening development of the digital economy, engineering cost management is transforming towards digitalization and intelligence, especially in the construction industry. However, traditional engineering cost systems have many limitations: First, existing engineering cost systems mainly rely on historical project data and quota standards, lacking comprehensive consideration of real-time construction data, equipment operation and maintenance data, and market fluctuation data. This leads to a significant deviation between the evaluation results and the actual project situation. Moreover, they are difficult to handle complex nonlinear relationships during evaluation and are lagging behind the impact of dynamic factors such as market fluctuations and changes in construction conditions.

[0003] Secondly, traditional systems are mostly post-project assessments, unable to provide real-time cost evaluation and early warning during project implementation. Cost engineers typically rely on static drawings and historical data for estimations, making it difficult to reflect cost changes as the project progresses. Once established, traditional cost evaluation models are difficult to dynamically adjust based on project progress and market changes. When faced with new materials, new processes, or abnormal market fluctuations, the accuracy of the evaluation decreases significantly.

[0004] Furthermore, errors or even tampering may occur during the collection, transmission, and storage of engineering data, affecting the credibility of cost assessment results. Moreover, the lack of effective risk identification and early warning mechanisms makes it difficult to prevent problems such as cost overruns. Summary of the Invention

[0005] In order to solve the above-mentioned technical problems, this application proposes the following technical solution:

[0006] This application provides a real-time engineering cost assessment and early warning system based on multi-source data fusion and dynamic learning, including:

[0007] The multi-source data acquisition module is used to collect construction process data, historical project data, market data, and design drawing data that are required for engineering cost assessment.

[0008] The preprocessing module is used to preprocess and fuse the collected multi-source data to form a multi-dimensional feature vector with a unified timestamp.

[0009] A hybrid neural network evaluation module is used to input the multidimensional feature vector into a pre-built hybrid neural network model for cost evaluation;

[0010] The dynamic optimization learning module performs online learning and dynamic optimization processing on the hybrid neural network evaluation module based on a transfer learning strategy.

[0011] The blockchain evidence storage module is used to perform hash calculations and distributed storage on key data generated during the evaluation process.

[0012] The progress monitoring and early warning module is used to monitor the construction progress in real time through image recognition technology, calculate progress deviations, and trigger corresponding early warnings.

[0013] In one possible implementation, the step of preprocessing and fusing the collected multi-source data to form a multi-dimensional feature vector with a unified timestamp includes:

[0014] Data cleaning and standardization preprocessing are performed on the collected historical engineering data, real-time construction data, market data, and design drawing data.

[0015] Preprocessed data from different sources are aligned and synchronized using a unified timestamp;

[0016] A feature fusion algorithm is used to fuse aligned multi-source data to generate a unified multi-dimensional feature vector.

[0017] In one possible implementation, the hybrid neural network model adopts an architecture that combines convolutional neural networks and long short-term memory networks, including: an input layer, the output of which is connected to the input of a convolutional layer, the output of which is connected to the input of a pooling layer, the output of which is connected to the input of an LSTM layer, the output of which is connected to the input of a fully connected layer, and the output of which is connected to an output layer.

[0018] In one possible implementation, the multidimensional feature vector is input into a pre-built hybrid neural network model for cost evaluation, including:

[0019] The multidimensional feature vector is input into the input layer of the hybrid neural network model;

[0020] Local features and nonlinear relationships in the feature vector are extracted by three convolutional layers, where the convolutional kernel sizes are 3×3, 5×5, and 3×3, and the activation function is ReLU.

[0021] After dimensionality reduction of the features extracted by the convolutional layer using the max pooling layer, the temporal dependencies in the features are modeled by the LSTM layer to capture the dynamic trends between market fluctuations, project progress and cost.

[0022] The fully connected layer performs global fusion and nonlinear transformation of the local features extracted by the convolutional layer and the temporal trends captured by the LSTM.

[0023] Finally, the cost assessment results and corresponding confidence levels are output at the output layer.

[0024] In one possible implementation, the online learning and dynamic optimization of the hybrid neural network evaluation module based on the transfer learning strategy includes:

[0025] The hybrid neural network model is pre-trained using large-scale historical engineering project data to obtain the base model;

[0026] To address the distributional differences between the target new project data and the source domain data, a domain adaptation loss function is introduced to reduce inter-domain differences and adapt the model features to the target domain.

[0027] Using partially labeled data from the target domain, fine-tun the parameters of the domain-adapted model at a lower learning rate;

[0028] The real-time generated engineering data is input into the fine-tuned model to achieve dynamic calculation of feature indicators.

[0029] In one possible implementation, the domain adaptation loss function is calculated as follows:

[0030]

[0031] in, Indicates domain adaptation loss; The characteristic mapping function maps the input to the reproducing kernel Hilbert space (RKHS). For source domain samples, Represents the target domain sample; , These are the number of samples in the source domain and the target domain, respectively. It is the norm of the RKHS space.

[0032] In one possible implementation, the formula for fine-tuning the parameters of the domain-adapted model with a lower learning rate using partially labeled data from the target domain is as follows:

[0033]

[0034]

[0035]

[0036]

[0037] in, Represents the total loss function; For mission losses, Domain Adaptation Loss (MMD); This indicates a loss of consistency. , , For the loss weight coefficients, satisfying ; Indicates the number of labeled samples in the target domain; , This represents the features and labels of the target domain samples, with MMD being the maximum mean difference. For feature mapping function, For historical engineering projects, For the feature vector of the target domain sample, Indicates the model in parameters Below is the input The predicted output, Indicates the model in parameters Below is the input The predicted output.

[0038] In one possible implementation, the formula for dynamically calculating feature indicators by inputting real-time generated engineering data into the fine-tuned model is as follows:

[0039]

[0040] in, Based on the basic cost assessment value, For the first Quantity of various materials Indicates the first Real-time prices of various materials Indicates the first Number of workers in each process Indicates the first Labor wages in each stage Indicates the first Equipment usage time in the process Indicates the first Equipment shift cost This represents the adjustment factor.

[0041] In one possible implementation, image recognition technology is used to monitor construction progress in real time, calculate progress deviations, and trigger corresponding early warnings, including:

[0042] Images of the construction site are collected using a network of cameras or drones deployed on-site, and then the images are denoised, enhanced, and sized.

[0043] The YOLOv5 target detection algorithm is used to identify, detect and locate construction machinery, personnel, materials and structural components in the preprocessed image.

[0044] Based on the identified construction elements, calculate the actual completed work volume and progress percentage;

[0045] The actual completed work volume is compared with the planned progress to obtain the progress deviation value, and the current deviation level is determined according to the preset deviation level classification standard.

[0046] Based on the determined deviation level, the corresponding early warning signal is automatically triggered, and an early warning notification is generated.

[0047] Compared with the prior art, the beneficial effects of this application are as follows:

[0048] This application achieves the digitalization, visualization, and intelligentization of engineering cost management through real-time data acquisition, multi-dimensional data analysis, intelligent model evaluation, dynamic deep learning, blockchain-based trusted storage, and visualized early warning display. By employing a hybrid neural network model and multi-source data fusion, the evaluation accuracy is improved by more than 30% compared to traditional methods. Simultaneously, it enables real-time evaluation and dynamic updates of engineering costs, with a response time in the sub-second range. A dynamic optimization mechanism based on transfer learning allows the model to quickly adapt to new environments and requirements.

[0049] This application employs blockchain technology to construct a system data traceability strategy, ensuring the immutability and traceability of the assessment data. Simultaneously, through image recognition and real-time monitoring, it enables early warning of deviations in project progress. The system has good versatility and scalability, and can be widely applied to cost assessment of various construction projects. Attached Figure Description

[0050] Figure 1 A schematic diagram of an overall engineering cost real-time assessment and early warning system based on multi-source data fusion and dynamic learning provided for embodiments of this application;

[0051] Figure 2 This is a structural diagram of a hybrid neural network model provided in an embodiment of this application;

[0052] Figure 3 A diagram illustrating the feature index calculation system provided in this application embodiment;

[0053] Figure 4 This is a flowchart of the dynamic optimization process for transfer learning provided in an embodiment of this application;

[0054] Figure 5 A flowchart of dynamic cost assessment provided for embodiments of this application;

[0055] Figure 6 A diagram illustrating the progress monitoring and early warning mechanism provided in this application embodiment;

[0056] Figure 7 This is a system performance comparison and evaluation chart provided for an embodiment of this application. Detailed Implementation

[0057] The present solution will now be described in conjunction with the accompanying drawings and specific embodiments.

[0058] Figure 1 A schematic diagram of a real-time engineering cost assessment and early warning system based on multi-source data fusion and dynamic learning, provided for an embodiment of this application, is shown below. Figure 1 This application provides a real-time engineering cost assessment and early warning system based on multi-source data fusion and dynamic learning, including: a multi-source data acquisition module, a preprocessing module, a hybrid neural network assessment module, a dynamic optimization learning module, a blockchain notarization module, and a progress monitoring and early warning module.

[0059] The multi-source data acquisition module is used to collect construction process data, historical project data, market data, and design drawing data that are required for engineering cost assessment. Additionally, it performs real-time monitoring of the construction process, collecting, summarizing, and organizing data on construction progress, material consumption, and manpower input.

[0060] The preprocessing module is used to preprocess and fuse the collected multi-source data to form a multi-dimensional feature vector with a unified timestamp. In this embodiment, the collected historical engineering data, real-time construction data, market data, and design drawing data undergo data cleaning and standardization preprocessing. Data cleaning includes handling missing values, outliers, and duplicate data. Normalization uses the Min-Max standardization method, and the formula for standardizing positive indicators is:

[0061]

[0062] in, X represents the standardized value; X represents the original value. This represents the minimum value in the dataset; This represents the maximum value in the dataset. This calculation formula normalizes feature values ​​of different dimensions and ranges to the [0,1] interval, eliminating the influence of dimensions, accelerating neural network convergence, and improving model stability.

[0063] For negative indicators, inverse standardization is used, and the calculation formula is as follows: .

[0064] Data augmentation is achieved by adding noise and time-series transformations to expand the training data. Preprocessed data from different sources are aligned and synchronized according to a unified timestamp. A feature fusion algorithm is then used to fuse the aligned multi-source data to generate a unified multi-dimensional feature vector.

[0065] The hybrid neural network evaluation module is used to input multidimensional feature vectors into a pre-built hybrid neural network model for cost evaluation.

[0066] In this embodiment, the hybrid neural network model adopts an architecture combining convolutional neural networks and long short-term memory networks to achieve multi-source data fusion, intelligent assessment, and dynamic tracking of cost estimates. A schematic diagram of the hybrid neural network model is shown below. Figure 2 As shown, the model includes: an input layer, the output of which is connected to the input of a convolutional layer; the output of the convolutional layer is connected to the input of a pooling layer; the output of the pooling layer is connected to the input of an LSTM layer; the output of the LSTM layer is connected to the input of a fully connected layer; and the output of the fully connected layer is connected to the output layer. The input layer receives multi-dimensional feature vectors, including material prices, labor costs, machinery costs, and management costs. The convolutional layer uses a three-layer convolutional structure to extract local features and nonlinear correlations from the input feature vector. Its structural equation expression is:

[0067]

[0068]

[0069]

[0070] in, This represents the input feature matrix with dimension . Where B is the batch size, C is the number of channels, H is the height, and W is the width; Let be the weight matrix of the i-th convolutional kernel, with dimension . ,in Number of output channels Input the number of channels. The height and width of the convolution kernel; This represents the bias vector of the i-th layer, with dimension . , Represents the convolution operator; ReLU is the modified linear unit activation function. , This is the output feature map of the i-th layer.

[0071] This formula describes the forward propagation process of a convolutional neural network. It extracts spatial features from the input data through three convolutional layers, with a ReLU activation function introduced after each convolutional layer to introduce non-linearity. The convolutional kernel sizes are 3×3, 5×5, and 3×3, extracting features from low to high levels layer by layer.

[0072] The features extracted by the convolutional layer are reduced in dimensionality using a max pooling layer with a 2×2 pooling window.

[0073] The LSTM layer contains 128 hidden units and is used to handle temporal dependencies, capturing the dynamic trends between market fluctuations, project progress, and costs. Its mathematical expression is:

[0074]

[0075]

[0076]

[0077]

[0078]

[0079]

[0080] Among them, The input vector representing time step t; Indicates the hidden state of the previous time step; This represents the state of the tuple at the previous time step; This indicates the output of the forget gate, controlling which information is discarded from the tuple state; This indicates the input gate output, controlling which new information is stored in the element's state; The candidate state contains new information that may be stored in the tuple state; Indicates the state of the tuple at the current time step; This indicates the output gate output, controlling which information is output from the initial state; It represents the hidden state of the current time step, and is also the output of that time step; This represents the Sigmoid function. The output range is (0,1). It is the hyperbolic tangent function. The output range is (1,1). This indicates element-wise multiplication (Hadamard product). , , , Represents the weight matrix; , , , The bias vector is used. The LSTM unit controls the flow of information through three gating mechanisms (forget gate, input gate, and output gate), effectively solving the gradient vanishing problem of traditional RNNs and capturing long-term temporal dependencies in engineering cost data.

[0081] The fully connected layer adopts a three-layer structure with 256, 128 and 64 neurons respectively. The local features extracted by the convolutional layer and the temporal trends captured by LSTM are globally fused and nonlinearly transformed through the fully connected layer. Finally, the cost assessment result and the corresponding confidence level are output in the output layer.

[0082] A multi-source data fusion strategy employs the Kalman filter algorithm to fuse data from multiple sensor sources, thereby improving data accuracy. The fusion formula is as follows:

[0083]

[0084] in, The state estimate at time k. Let k be the prior state estimate. For Kalman gain, Let k be the observation value at time k. This represents the observation matrix.

[0085] During the model training phase, forward propagation calculates the predicted output through a hybrid neural network. The weighted combined loss function is calculated using the following formula:

[0086]

[0087]

[0088]

[0089]

[0090] in, Represents the total loss function; For mean square error loss, Indicates the mean absolute error loss. Characterizing regularization loss, , , The loss weighting coefficients satisfy the following conditions: By combining the advantages of various loss functions, MSE is more sensitive to large errors, MAE is more robust to small errors, and regularization prevents overfitting. The model's generalization ability is improved by balancing the various losses through weighting coefficients.

[0091] See Figure 3 This application establishes an engineering cost evaluation index system based on characteristic index combination rules. These characteristic index combination rules include rules for combining indicators such as material, labor, machinery usage, and management costs. The main calculation formula is as follows:

[0092] The formula for calculating the combination of material cost indicators is as follows:

[0093]

[0094] in, This represents the total cost of materials; The quantity of the i-th material (unit: tons, cubic meters, pieces, etc.); The unit price of the i-th material (unit: yuan / unit); Let be the adjustment coefficient for the i-th material, taking into account factors such as transportation costs, loss rate, and taxes; n is the total number of material types. This formula is mainly used to calculate the total cost of all materials in an engineering project, taking into account quantity, unit price, and various adjustment factors, and is a fundamental component of project cost.

[0095] The formula for calculating the combination of labor cost indicators is as follows:

[0096]

[0097] in, This represents the total cost of labor. The working hours (in hours or man-days) for job type j. The wage rate for job type j (unit: yuan / hour, yuan / workday); is the adjustment coefficient for job type j, taking into account factors such as overtime pay, social security, and benefits; m is the total number of job types. This formula is mainly used to calculate the total labor cost required for an engineering project, taking into account the working hours, wage rates, and related adjustment factors for different job types.

[0098] The formula for calculating the combination of mechanical performance indicators is as follows:

[0099]

[0100] in, This represents the total cost of using the machinery; It is the usage time of the kth type of machine (unit: shift, hour); This is the rental rate for the kth type of machinery (unit: yuan / shift, yuan / hour). is the adjustment factor for the k-th type of machinery, taking into account fuel costs, maintenance costs, depreciation costs, etc.; p is the total number of machinery types. This calculation formula is mainly used to calculate the total cost of using machinery and equipment in engineering projects, taking into account usage time, rental rates, and related adjustment factors.

[0101] The management fee indicator combination rules and calculation formula are as follows:

[0102]

[0103] in, Indicates the total management fee; Management fees are typically a percentage (e.g., 5%-15%). By convention, management fees are usually calculated as a percentage of direct costs (materials, labor, machinery) to cover indirect costs such as project management, office expenses, and coordination.

[0104] The rules for calculating characteristic index values ​​include:

[0105] Engineering quantity calculation rules: Engineering quantity calculation rules based on design drawings and construction specifications.

[0106] Price determination rules: Price determination methods and rules refer to market prices and historical data.

[0107] Rate determination rules: The rules for determining rates, fees, etc., based on industry standards and contractual agreements.

[0108] Adjustment coefficient rules: Adjustment calculation coefficient rules determined by taking into account factors such as region, season, and market fluctuations.

[0109] To achieve high-quality measurement and dynamic calculation of engineering cost characteristic indicators, this embodiment combines project information and multi-source data, implements an evaluation model to provide input features, and constructs a specific calculation formula for calculating characteristic indicator values ​​based on characteristic indicator value calculation rules, project information, and characteristic indicators. This formula employs a multiple linear regression method, trained on historical data to ensure the dynamism and accuracy of the calculation, thereby supporting real-time assessment and decision-making regarding engineering costs.

[0110] The specific calculation formula for calculating the feature index value based on the feature index value calculation rules, project information, and feature index is as follows:

[0111]

[0112]

[0113] in, For the i-th engineering quantity parameter, such as earthwork volume or building area, For the i-th price parameter (such as material unit price, labor rate). Let i be the i-th environmental parameter (such as season or region coefficient). Let be the i-th market parameter (such as the price index or policy impact). The weighting coefficients are calculated using the entropy weighting method. , Let i be the entropy value of the i-th index. The Sigmoid function is a feature transformation function used for normalization and nonlinear mapping.

[0114] The dynamic optimization learning module performs online learning and dynamic optimization of the hybrid neural network evaluation module based on the transfer learning strategy.

[0115] See Figure 4 In this embodiment, a hybrid neural network model is pre-trained using large-scale historical engineering project data to obtain a base model. The source domain pre-trained model is as follows:

[0116]

[0117] in, This represents the optimal model parameters in the source domain; This is the model parameter vector; It is the number of samples in the source domain; It is the feature vector of the i-th sample in the source domain; The label (actual cost) of the i-th sample in the source domain; This indicates that the model behaves differently with respect to the input under parameter θ. The predicted output; L represents the loss function, which measures the difference between the predicted value and the true value; It is a regularization term to prevent overfitting; is the regularization coefficient, which controls the strength of regularization. This expression represents training the model on the source domain (historical engineering data) and obtaining initial model parameters that can accurately predict costs by minimizing the prediction loss and the regularization term.

[0118] Feature extractor construction:

[0119] in, These are the parameters for the feature extraction layer.

[0120] By introducing a domain adaptation loss function, inter-domain differences are reduced, allowing model features to adapt to the target domain. The formula for calculating the domain adaptation loss function is as follows:

[0121]

[0122] in, Indicates domain adaptation loss; The characteristic mapping function maps the input to the reproducing kernel Hilbert space (RKHS). For source domain samples, Represents the target domain sample; , These are the number of samples in the source domain and the target domain, respectively. It is the norm of the RKHS space. Let Hilbert space be the regenerated kernel. The above equation uses the maximum mean difference (MMD) to measure the difference between the feature distributions of the source and target domains. By minimizing the MMD loss, the model learns feature representations that are insensitive to changes in the domain.

[0123] Using partially labeled data from the target domain, the parameters of the domain-adapted model are fine-tuned at a low learning rate.

[0124] The formula for calculating the loss function using a weighted combination of multi-objective optimization is as follows:

[0125]

[0126]

[0127]

[0128]

[0129] in, Represents the total loss function; To mitigate task loss, ensure the model's prediction accuracy in the target domain; Domain Adaptation Loss (MMD); This represents consistency loss, which improves model robustness and encourages the model to produce consistent outputs for different enhanced versions of the same input. , , For the loss weight coefficients, satisfying ; Indicates the number of labeled samples in the target domain; , This represents the features and labels of the target domain samples. This multi-objective loss function balances task performance, domain adaptability, and model robustness, enabling the model to perform well on new items (target domain).

[0130] The formula for calculating the Elastic Weighted Consolidation (EWC) loss is:

[0131]

[0132] in, Indicates the total EWC loss; This indicates a loss in the data from the new task; The diagonal elements of the Fisher information matrix, representing the parameters. The importance of old tasks; The current parameter value; This represents the optimal parameter values ​​obtained by training on the old task; This represents the EWC regularization strength coefficient. The EWC algorithm prevents the model from forgetting knowledge from old tasks when learning new tasks by adding a quadratic penalty term to the loss function. The Fisher information matrix measures the importance of each parameter to the task; parameters with higher importance are subject to stronger constraints during updates.

[0133] To maintain the model's timeliness, the data stream processing method is to regularly update it with new data.

[0134] The dynamic model update method is as follows:

[0135]

[0136] Random noise is added to enhance the model's dynamic robustness. The model performance monitoring expression is constructed as follows:

[0137]

[0138] When performance degrades beyond a threshold The model recalibration behavior is triggered at certain times.

[0139] In knowledge distillation and model compression, the teacher-student architecture performs lightweight model calculations, and the calculation formula is as follows:

[0140]

[0141]

[0142] in, , These are the logits outputs for the teacher and student models, respectively. This is a temperature parameter used to control the smoothness of the output. L1 and L2 regularization are combined to achieve model compression, thus making the model lightweight.

[0143]

[0144] The transfer learning optimization algorithm is taken as an example of the dynamic optimization algorithm of cost model transfer learning.

[0145] Initial learning rate:

[0146] Domain adaptation phase:

[0147] Task-based learning phase:

[0148]

[0149] Consistent regularization:

[0150]

[0151] Total loss calculation:

[0152] Parameter update:

[0153] Learning rate adjustment:

[0154] Continuous online learning when new data arrives:

[0155]

[0156]

[0157] The formula for dynamically calculating feature indicators by inputting real-time generated engineering data into the fine-tuned model is as follows:

[0158]

[0159] in, Based on the basic cost assessment value, For the first Quantity of various materials Indicates the first Real-time prices of various materials Indicates the first Number of workers in each process Indicates the first Labor wages in each stage Indicates the first Equipment usage time in the process Indicates the first Equipment shift cost This represents the adjustment factor.

[0160] The blockchain-based evidence storage module is used for hash calculation and distributed storage of key data generated during the evaluation process. It includes a data hash calculation unit, a smart contract execution unit, a distributed storage unit, and a timestamp service unit. Integrating blockchain technology enables secure data storage, ensuring the traceability and immutability of the evaluation process.

[0161] The progress monitoring and early warning module is used to monitor the construction progress in real time through image recognition technology, calculate progress deviations, and trigger corresponding early warnings.

[0162] See Figure 5 In this embodiment, image recognition technology is used to achieve real-time monitoring of project progress. The specific analysis process includes image acquisition, target detection, progress calculation, deviation analysis, and early warning triggering.

[0163] Images of the construction site are collected using a network of cameras or drones deployed on-site. These images undergo noise reduction, enhancement, and size normalization. The YOLOv5 object detection algorithm is used to measure object detection confidence. The pre-processed images are then used to identify and locate construction machinery, personnel, materials, and structural components. The calculation formula is as follows:

[0164]

[0165] in, Indicates the probability of an object's existence; IOU represents the intersection-union ratio. The confidence score output by the Sigmoid function; This represents the original output value of the network; This represents the area of ​​the intersection between the predicted bounding box and the ground truth bounding box. This is the area of ​​the union of the predicted bounding box and the ground truth bounding box. The YOLO algorithm treats object detection as a regression problem, and this formula calculates the confidence score of the detection result, taking into account both the probability of object presence and the accuracy of bounding box localization.

[0166] Based on the identified construction elements, calculate the actual completed work volume and progress percentage. The calculation formula is as follows:

[0167]

[0168] The actual completed work volume is compared with the planned progress to obtain the progress deviation value. Based on the preset deviation level classification standard, the current deviation level is determined. The calculation formula is as follows:

[0169]

[0170] in, Indicates the percentage of schedule deviation. Indicates the actual amount of work completed; This indicates the planned amount of work to be completed. It quantifies the deviation between the project progress and the plan; a positive deviation indicates progress is ahead of schedule, and a negative deviation indicates progress is behind schedule. Based on the determined deviation level, it automatically triggers the corresponding warning signal and generates a warning notification.

[0171] See Figure 6 In this embodiment, the deviation levels include normal, slight lag, significant lag, severe lag, slight lead, significant lead, and severe lead. When the absolute value is less than or equal to 5%, the progress is within the normal range. When the absolute value is greater than 5% and less than or equal to 10%, the progress is within a slight range. When the absolute value is greater than 10% and less than or equal to 15%, the progress is within a significant range. When the absolute value is greater than 15%, the progress is in a critical range.

[0172] The application service module includes a visualization unit, an early warning notification unit, a report generation unit, and a progress monitoring unit.

[0173] Taking a commercial complex construction project as an example, this system is used for cost assessment.

[0174] Project Overview: The total construction area is 100,000 square meters, with 2 underground floors and 25 floors above ground. It includes a commercial podium, an office tower, and an underground parking garage. The construction period is 24 months.

[0175] System Deployment and Configuration

[0176] This system can be deployed using a distributed architecture, including data acquisition nodes, computing servers, blockchain nodes, and user terminals. Recommended configurations are as follows: Data acquisition nodes: Intel i7 processor, 16GB RAM, 2TB storage, gigabit network interface; Computing servers: NVIDIA Tesla V100 GPU, 128GB RAM, 10TB SSD storage; Blockchain nodes: Distributed deployment, using the Hyperledger Fabric framework; User terminals: Web browser or mobile app, supporting real-time data visualization. The system collects data through various methods such as API interfaces, sensor networks, and file import. The collected data undergoes preprocessing, specifically including handling missing values, outliers, and duplicate data; extracting meaningful features; constructing feature vectors; and data augmentation through adding noise and time-series transformations to expand the training data.

[0177] The following strategies were adopted during the model training and optimization phases: Algorithm: Adam optimizer, learning rate 0.001; Loss function: weighted combined loss function;

[0178]

[0179] Training strategy: early stopping mechanism, batch size 32, maximum training epochs 200; model validation: 5-fold cross-validation is used to ensure the model's generalization ability.

[0180] System Operation and Maintenance

[0181] The system requires regular maintenance during operation, including performance monitoring (real-time monitoring of system operation status and model performance), model updates (regularly updating the model with new data to maintain its timeliness), data backup (regularly backing up important data to ensure system reliability), and security auditing (regularly conducting security audits to prevent security risks).

[0182] Implementation results:

[0183] Data collection: Collect historical data on similar projects, current market conditions, design drawings, etc.

[0184] Cost assessment: The total estimated cost is 850 million yuan, with a confidence level of 92%.

[0185] Progress monitoring: Real-time monitoring of construction progress using image recognition revealed that the main structure construction was 5% behind schedule;

[0186] Warning Handling: The system triggered a yellow warning, prompting the implementation of expedited work measures;

[0187] Results Verification: The final settlement cost of the project was 860 million yuan, with an assessment error of 1.2%.

[0188] See Figure 7The system performance comparison and evaluation chart provided in this application shows that, compared with traditional systems, the system of this application is far superior in terms of evaluation accuracy, real-time performance, timely early warning, and user satisfaction. The system of this application can be widely used in cost management scenarios of various engineering projects such as building engineering, municipal engineering, and power engineering, providing comprehensive technical support for engineering cost management and helping the digital transformation and intelligent upgrading of the engineering construction industry.

[0189] In this embodiment, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0190] 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, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0191] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. The protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A multi-source data fusion and dynamic learning-based real-time project cost evaluation and early warning system, characterized in that, include: The multi-source data acquisition module is used to collect construction process data, historical project data, market data, and design drawing data that are required for engineering cost assessment. The preprocessing module is used to preprocess and fuse the collected multi-source data to form a multi-dimensional feature vector with a unified timestamp. A hybrid neural network evaluation module is used to input the multidimensional feature vector into a pre-built hybrid neural network model for cost evaluation; The dynamic optimization learning module performs online learning and dynamic optimization processing on the hybrid neural network evaluation module based on a transfer learning strategy, including: The hybrid neural network model is pre-trained using large-scale historical engineering project data to obtain the base model; To address the distributional differences between the target new project data and the source domain data, a domain adaptation loss function is introduced to reduce inter-domain discrepancies, enabling the model features to adapt to the target domain. The calculation formula is as follows: in, Indicates domain adaptation loss; The characteristic mapping function maps the input to the reproducing kernel Hilbert space (RKHS). For source domain samples, Represents the target domain sample; , These are the number of samples in the source domain and the target domain, respectively. It is the norm of the RKHS space; Using partially labeled data from the target domain, the parameters of the domain-adapted model are fine-tuned with a lower learning rate. The calculation formula is as follows: in, Represents the total loss function; For mission losses, Domain Adaptation Loss (MMD); This indicates a loss of consistency. , , For the loss weight coefficients, satisfying ; Indicates the number of labeled samples in the target domain; , This represents the features and labels of the target domain samples. For feature mapping function, For historical engineering projects, For the feature vector of the target domain sample, Indicates the model in parameters Below is the input The predicted output, Indicates the model in parameters Below is the input The predicted output; The real-time generated engineering data is input into the fine-tuned model to achieve dynamic calculation of feature indicators; The blockchain evidence storage module is used to perform hash calculations and distributed storage on key data generated during the evaluation process. The progress monitoring and early warning module is used to monitor construction progress in real time through image recognition technology, calculate progress deviations, and trigger corresponding early warnings, including: Images of the construction site are collected using a network of cameras or drones deployed on-site, and then the images are denoised, enhanced, and sized. The YOLOv5 target detection algorithm is used to identify, detect and locate construction machinery, personnel, materials and structural components in the preprocessed image. Based on the identified construction elements, calculate the actual completed work volume and progress percentage; The actual completed work volume is compared with the planned progress to obtain the progress deviation value, and the current deviation level is determined according to the preset deviation level classification standard. Based on the determined deviation level, the corresponding early warning signal is automatically triggered, and an early warning notification is generated.

2. The multi-source data fusion and dynamic learning based real-time engineering cost evaluation and early warning system according to claim 1, characterized in that, The process of preprocessing and fusing the collected multi-source data to form a multi-dimensional feature vector with a unified timestamp includes: Data cleaning and standardization preprocessing are performed on the collected historical engineering data, real-time construction data, market data, and design drawing data. Preprocessed data from different sources are aligned and synchronized using a unified timestamp; A feature fusion algorithm is used to fuse aligned multi-source data to generate a unified multi-dimensional feature vector.

3. The real-time engineering cost assessment and early warning system based on multi-source data fusion and dynamic learning according to claim 1, characterized in that, The hybrid neural network model adopts an architecture that combines convolutional neural networks and long short-term memory networks, including: an input layer, the output of which is connected to the input of a convolutional layer, the output of which is connected to the input of a pooling layer, the output of which is connected to the input of an LSTM layer, the output of which is connected to the input of a fully connected layer, and the output of which is connected to an output layer.

4. The multi-source data fusion and dynamic learning based real-time engineering cost evaluation and early warning system according to claim 1, characterized in that, The multidimensional feature vectors are input into a pre-built hybrid neural network model for cost evaluation, including: The multidimensional feature vector is input into the input layer of the hybrid neural network model; Local features and nonlinear relationships in the feature vector are extracted by three convolutional layers, where the convolutional kernel sizes are 3×3, 5×5, and 3×3, and the activation function is ReLU. After dimensionality reduction of the features extracted by the convolutional layer using the max pooling layer, the temporal dependencies in the features are modeled by the LSTM layer to capture the dynamic trends between market fluctuations, project progress and cost. The fully connected layer performs global fusion and nonlinear transformation of the local features extracted by the convolutional layer and the temporal trends captured by the LSTM. Finally, the cost assessment results and corresponding confidence levels are output at the output layer.

5. The real-time engineering cost assessment and early warning system based on multi-source data fusion and dynamic learning according to claim 1, characterized in that, The formula for dynamically calculating feature indicators by inputting real-time generated engineering data into the fine-tuned model is as follows: in, Based on the basic cost assessment value, For the first Quantity of various materials Indicates the first Real-time prices of various materials Indicates the first Number of workers in each process Indicates the first Labor wages in each stage Indicates the first Equipment usage time in the process Indicates the first Equipment shift cost This represents the adjustment factor.

Images