Intelligent monitoring and early warning system and method for project progress and cost

Abstract

The invention discloses an intelligent monitoring and early warning system and method for project progress and cost, and relates to the technical field of constructional engineering management. The streaming data fusion module adopts an Apache Flink engine and applies a dynamic time warping algorithm to carry out time alignment on multi-source asynchronous data to realize semantic unification; the space-time diagram neural network prediction module constructs a space-time association diagram, integrates global features such as weather and supply chain fluctuation, and adaptively learns an influence weight by using a graph attention mechanism; the self-adaptive early warning module adopts a Bayesian online learning framework, an early warning threshold value is dynamically adjusted according to a historical false alarm rate and a missing report rate, the method comprises the steps of data acquisition, fusion, mapping, joint prediction and self-adaptive early warning, the problem of data islands is solved, complex space-time association is accurately captured through a space-time diagram neural network, and real-time early warning is achieved. Joint prediction of progress and cost risk is realized, false report and missing report are reduced, and real-time performance, accuracy and decision-making efficiency of engineering management and control are improved.

Description

Technical Field

[0001] This invention relates to the field of construction project management technology, and in particular to an intelligent monitoring and early warning system and method for project progress and cost. Background Technology

[0002] In the field of construction engineering management, especially in the construction supervision of large-scale building projects, achieving refined and real-time joint control of progress and cost is a core requirement for improving project efficiency and reducing the risk of cost overruns. With the development of digital technologies such as BIM and the Internet of Things, project management has gradually shifted from the traditional paper-based ledgers and manual inspection mode to dynamic monitoring based on information systems. Such systems aim to integrate multi-dimensional data to collaboratively monitor and scientifically decide on the progress and cost of engineering projects, ensuring that projects are completed on time within budget. Existing technologies include a gas risk intelligent monitoring method and system based on multi-source data (Publication No.: CN121146499A). This method includes: acquiring static attribute data, dynamic sensor data, and future event data of the gas pipeline network to construct a dynamic attribute map; inputting the dynamic attribute map into a spatiotemporal graph neural network model to predict the physical failure probability of each node; calculating the time-varying failure impact of each node based on future event data and cost data; calculating the risk value of the node by combining the physical failure probability and the time-varying failure impact; and generating maintenance decisions based on the risk value distribution by optimizing the model. The system includes: a data acquisition module, a failure rate prediction module, a time-varying failure handling module, a value at risk (VAT) calculation module, and a maintenance decision generation module. This invention constructs a time-varying influence function and portfolio model to proactively quantify risk and optimize maintenance decisions, solving the problem of assessment lag and improving the efficiency and safety of pipeline network operation and maintenance.

[0003] However, existing technical solutions have the following drawbacks: existing systems mostly rely on single or limited types of data sources, and the data formats and collection cycles of different systems are inconsistent, forming "data silos"; current early warning models are mostly based on the critical path method or simple regression analysis. These methods treat processes, resources and costs as static or independent variables, and cannot effectively characterize the complex spatiotemporal dynamic relationships and nonlinear mutual influences between them; at the same time, the early warning thresholds of existing systems are mostly based on static rules, and once set, they cannot be optimized and dynamically adjusted online according to the actual progress of the project and historical early warning feedback.

[0004] In response to the aforementioned technologies, a solution is proposed. Summary of the Invention

[0005] The purpose of this application is to provide an intelligent monitoring and early warning system and method for engineering progress and cost to solve the technical problems of existing technologies, such as barriers in data acquisition and fusion, and insufficient dynamic adaptability of prediction models and analysis methods.

[0006] This application provides an intelligent monitoring and early warning system and method for project progress and cost, employing the following technical solution: An intelligent monitoring and early warning system for project progress and cost, comprising: Edge computing data acquisition module: Deployed at the construction site, it is used to collect progress, resource and environmental data through various IoT sensing devices; Streaming data fusion module: connected to the edge computing data acquisition module, used for time alignment and semantic unification of multi-source asynchronous data; Spatiotemporal graph neural network prediction module: connected to the streaming data fusion module, used to construct a spatiotemporal correlation graph based on the fused data and perform joint prediction; Adaptive early warning module: connected to the spatiotemporal graph neural network prediction module, used to dynamically trigger early warning signals based on the prediction results.

[0007] By adopting the above technical solutions, the edge computing data acquisition module, through the deployment of edge smart gateways integrating various IoT sensing devices at the construction site, achieves efficient and low-latency acquisition of progress, resource, and environmental data, providing a solid data foundation for subsequent analysis. The streaming data fusion module utilizes an advanced stream processing engine and dynamic time warping algorithm to solve the time series alignment problem caused by differences in sampling frequencies of multi-source asynchronous data. Combined with ontology-based knowledge graphs for semantic unification, it effectively eliminates data silos and forms a high-quality, highly consistent fused data stream. Based on this fused data, the spatiotemporal graph neural network prediction module abstracts complex construction elements into a spatiotemporal relationship graph containing entities such as processes, resources, and costs, as well as their interrelationships. Through the inherent mechanism of graph neural networks, it captures the complex spatial dependencies and dynamic temporal evolution patterns between these entities, thereby achieving deep joint prediction of progress and cost risks. Ultimately, the adaptive early warning module dynamically optimizes the early warning threshold based on the prediction results and uses an online learning mechanism, achieving self-adjustment and continuous improvement of the early warning strategy. These interconnected modules work collaboratively to achieve more real-time, accurate, and adaptive joint monitoring and early warning of project progress and costs, effectively overcoming the shortcomings of existing technologies such as difficulties in data fusion, static and rigid models, and fixed early warning thresholds.

[0008] The edge computing data acquisition module includes an edge smart gateway that supports LoRa or NB-IoT communication protocols. The edge smart gateway integrates an RFID reader, a UWB positioning base station, and a smart camera. The RFID reader is used to track the status of materials, the UWB positioning base station is used to obtain the location information of personnel and equipment, and the smart camera is equipped with a YOLOv8 model to automatically identify the progress of construction procedures.

[0009] By adopting the above technical solutions, the edge computing data acquisition module achieves low-power and long-distance data transmission through an edge smart gateway supporting LoRa or NB-IoT communication protocols, ensuring stable communication in complex construction site environments. This gateway integrates an RFID reader to automatically track material status using radio frequency identification technology, monitoring inventory and flow in real time. The UWB positioning base station uses ultra-wideband technology to obtain high-precision location information of personnel and equipment, improving resource scheduling efficiency. The YOLOv8 model inside the smart camera automatically identifies construction progress based on deep learning algorithms, reducing manual intervention and improving identification accuracy. The collaborative work of these components enables comprehensive data acquisition covering progress, resources, and environment dimensions, providing a real-time, multi-source, and high-quality data foundation for subsequent streaming data fusion and spatiotemporal neural network prediction, thereby enhancing the real-time performance and reliability of the entire system for project progress and cost management.

[0010] The streaming data fusion module uses the Apache Flink streaming engine. It employs a dynamic time warping algorithm to time-align asynchronous data streams with different sampling frequencies, where the difference in sampling frequencies is no greater than 10 times. Then, it uses a knowledge graph based on the OWL ontology to perform semantic disambiguation on the multi-source data.

[0011] By adopting the above technical solution, the streaming data fusion module utilizes the high throughput and low latency characteristics of the Apache Flink streaming engine to process multi-source asynchronous data in real time. Its core lies in first applying a dynamic time warping algorithm to precisely align asynchronous data streams with sampling frequency differences within a factor of 10, effectively eliminating time matching errors caused by asynchronous sensor sampling times and establishing a unified time benchmark for subsequent analysis. Then, a knowledge graph based on the OWL ontology is used to perform semantic disambiguation and unification on data from different devices, constructing a standardized data semantic model that is understandable to machines. This fundamentally solves the problem of difficulty in fusion of multi-source heterogeneous data from construction sites due to inconsistent semantic descriptions. This technical solution, through a progressive processing flow of time alignment followed by semantic fusion, ensures that the data flowing into the subsequent spatiotemporal graph neural network prediction module has high consistency and accuracy in both time and semantic dimensions, laying a reliable data foundation for accurate joint prediction of project progress and cost.

[0012] The spatiotemporal graph constructed by the spatiotemporal graph neural network prediction module includes nodes and edges. The nodes include nodes representing process tasks, nodes representing material inventory, nodes representing personnel attendance, and nodes representing equipment status. The edges are used to represent process dependencies, resource allocation relationships, and cost relationships between nodes. The spatiotemporal graph neural network prediction module also inputs weather data, supply chain fluctuation index, and design change frequency as global features into the spatiotemporal correlation graph.

[0013] By adopting the above technical solution, the spatiotemporal graph neural network prediction module constructs a spatiotemporal relationship graph containing multiple nodes such as process tasks, material inventory, personnel attendance, and equipment status. It uses edges to accurately represent the complex relationships between nodes, including process dependencies, resource allocation, and cost correlations, thereby abstracting the engineering system into a topological structure with rich semantic information. Its innovative key feature lies in the module's innovative use of external dynamic factors such as weather data, supply chain fluctuation indices, and design change frequency as global features input into the graph. This design principle allows the model to overcome the limitation of only analyzing internal entity relationships, thus indirectly establishing the implicit correlation between external environmental fluctuations and changes in the status of internal nodes within a unified graph structure. This enables the spatiotemporal graph neural network to simultaneously consider the comprehensive impact of process logic, resource constraints, and the external environment when performing joint schedule and cost predictions, significantly enhancing the understanding and modeling ability of complex nonlinear interactions in engineering systems, ultimately improving the comprehensiveness and accuracy of schedule deviation and cost overrun risk predictions.

[0014] The adaptive early warning module adopts a Bayesian online learning framework. The adaptive early warning module is configured to dynamically adjust the early warning thresholds for schedule deviations and cost overruns based on feedback from the false alarm rate and false negative rate of historical early warnings. The initial threshold of the early warning threshold is set to ±10% of the baseline value.

[0015] By adopting the above technical solution, the core of the adaptive early warning module lies in the application of the Bayesian online learning framework. This framework can continuously optimize and adjust the early warning thresholds for schedule deviation and cost overrun in a probabilistic update manner based on the historical early warning feedback information generated during system operation, namely the real-time data of false alarm rate and false alarm rate. Its design principle lies in treating the setting of the early warning threshold as a dynamic learning process. By combining prior knowledge with newly observed evidence using Bayes' theorem, the threshold can adapt to the actual progress of the project and changes in the external environment, rather than remaining fixed. This mechanism enables the early warning system to learn autonomously from past judgment experience, gradually reducing the occurrence of false alarms and missed alarms. The initial threshold setting of ±10% of the baseline value provides a reasonable starting point for the system. This technical solution effectively improves the accuracy and reliability of the early warning signal, enabling the early warning module to have self-evolution capabilities, and ultimately achieving more intelligent and accurate early warning of project progress and cost risks.

[0016] A method for intelligent monitoring and early warning of project progress and cost includes the following steps: S1. Collect streaming data through edge devices deployed at the construction site; S2. Perform time alignment and semantic fusion on the collected multi-source asynchronous data; S3. Based on the fused data, construct a spatiotemporal relationship diagram of process-resource-cost; S4. Use a spatiotemporal graph neural network model to jointly predict schedule deviations and cost overrun risks; S5. Apply an adaptive threshold mechanism to dynamically trigger early warnings based on prediction results and feedback learning.

[0017] By adopting the above technical solution, multi-source streaming data from the construction site is first collected through edge devices. Then, time alignment and semantic fusion technologies are used to provide high-quality input for subsequent analysis. The key step is to construct a spatiotemporal correlation graph that integrates processes, resources, costs, and their complex relationships. A spatiotemporal graph neural network model is then used to jointly predict progress and risks on this graph. This design principle enables the model to simultaneously capture the spatial dependencies and temporal dynamic evolution of various elements in the engineering system, breaking through the limitations of traditional methods that analyze progress and costs in isolation. Finally, by applying an adaptive threshold mechanism based on Bayesian online learning, the early warning behavior can be dynamically optimized based on historical feedback. Through the close connection of the above steps, this method achieves a more comprehensive, accurate, and adaptive early warning capability for project progress and cost risks.

[0018] In step S2, the time alignment of multi-source asynchronous data specifically involves using a dynamic time warping algorithm to process data streams with sampling frequency differences within 10 times.

[0019] By adopting the above technical solution, step S2 explicitly specifies the use of the dynamic time warping algorithm to process multi-source asynchronous data streams with sampling frequency differences within 10 times. Its core principle lies in the fact that the algorithm can flexibly perform optimal path matching for data sequences with different sampling times and frequencies through nonlinear dynamic programming, thereby achieving local stretching or compression of the time axis to complete accurate alignment. This innovative design effectively overcomes the information distortion problem that may be caused by traditional linear interpolation or simple resampling methods when processing construction site data with different rhythms and change rates. By limiting the frequency difference to within 10 times, this solution ensures alignment accuracy while taking into account the computational efficiency of the algorithm, laying an accurate and consistent time benchmark for subsequent semantic fusion and joint prediction based on spatiotemporal correlation graphs, and ultimately improving the reliability and timeliness of the entire system's prediction of progress and cost risks.

[0020] In step S4, the joint prediction uses a graph attention mechanism to adaptively learn the weights of the impact of weather changes and supply chain fluctuations on different process nodes.

[0021] By adopting the above technical solution, the graph attention mechanism is creatively applied to the joint prediction process of the spatiotemporal graph neural network model in step S4. The core principle is that this mechanism enables the model to automatically calculate and assign the importance weights of different neighboring nodes and global features to the target node during information propagation and aggregation. Specifically, in this solution, the model dynamically evaluates the differentiated impact of external global features such as weather changes and supply chain fluctuations on the state of each process node through the graph attention mechanism, rather than treating them equally. For example, the impact weight of rainstorms on outdoor pouring processes is greater than that on indoor installation processes. This ability to adaptively learn weights allows the model to more finely characterize the complex propagation paths and impact levels of external environmental disturbances within the engineering system, thereby significantly improving the accuracy and interpretability of the joint prediction of schedule deviations and cost overrun risks, and overcoming the limitations of traditional methods that use static weights or manual experience allocation.

[0022] In summary, this application includes at least one of the following beneficial technical effects: 1. By integrating various IoT devices such as RFID, UWB positioning, and smart cameras through an edge computing gateway, low-latency data collection of multiple elements at the construction site, including personnel, machinery, materials, methods, and environment, was achieved. Furthermore, through a streaming processing engine and dynamic time warping algorithm, these asynchronous and high-frequency heterogeneous data streams were effectively aligned and merged, solving the data silo problem and laying a high-quality data foundation for accurate early warning. 2. By combining graph attention mechanisms and long short-term memory networks, the model can automatically learn the spatial dependencies and temporal evolution patterns between nodes without relying on manually preset fixed rules or weights. This enables the model to adaptively capture the impact of dynamic factors such as weather changes and supply chain fluctuations on different construction stages, significantly improving the accuracy and robustness of predicting schedule deviations and cost overrun risks. 3. By introducing an adaptive threshold early warning mechanism based on Bayesian online learning, dynamic optimization of the early warning threshold is achieved. This mechanism can continuously adjust the early warning threshold based on feedback from historical early warning results, enabling the early warning system to have self-learning and optimization capabilities, thereby effectively reducing the probability of false alarms. 4. By combining the fine-grained risk tracing capabilities provided by the ST-GNN model, the system can link early warning information to specific processes, resource bottlenecks, or dynamic interference factors, providing managers with more targeted decision support and significantly improving the efficiency and accuracy of response from risk perception to intervention actions. Attached Figure Description

[0023] Figure 1 This is the system architecture diagram of the present invention.

[0024] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0025] The following is in conjunction with the appendix Figure 1 -Appendix Figure 2 This application will be described in further detail below.

[0026] This application discloses an intelligent monitoring and early warning system and method for engineering progress and cost.

[0027] An intelligent monitoring and early warning system for project progress and cost, comprising: Edge computing data acquisition module: Deployed at the construction site, it is used to collect progress, resource and environmental data through various IoT sensing devices; Streaming data fusion module: connected to the edge computing data acquisition module, used for time alignment and semantic unification of multi-source asynchronous data; Spatiotemporal graph neural network prediction module: connected to the streaming data fusion module, used to construct a spatiotemporal correlation graph based on the fused data and perform joint prediction; Adaptive early warning module: connected to the spatiotemporal graph neural network prediction module, used to dynamically trigger early warning signals based on the prediction results.

[0028] Specifically, the modules are connected sequentially through data transmission interfaces to form a complete data acquisition, processing, analysis, and early warning chain; The edge computing data acquisition module is deployed at the construction site, collecting progress, resource, and environmental data through various IoT sensors. This module includes an edge smart gateway supporting LoRa or NB-IoT communication protocols, integrating three core devices: an RFID reader, a UWB positioning base station, and a smart camera. The RFID reader tracks material status by reading RFID tags attached to materials such as reinforcing bars and formwork, enabling full lifecycle tracking of materials from arrival to installation and ensuring transparent management of material flow. The UWB positioning base station acquires the location information of personnel and equipment. By deploying positioning base stations around the construction area, combined with UWB tags worn by personnel, real-time positioning with an accuracy of approximately 15cm is achieved, providing precise spatial data support for personnel attendance and equipment scheduling. The smart camera is equipped with a YOLOv8 model to automatically identify the progress of construction processes. Through computer vision technology, it monitors the status of key processes such as formwork erection completion rate and concrete pouring progress in real time, avoiding the subjectivity and lag of traditional manual inspections. The edge smart gateway ensures data integrity under network fluctuations through a breakpoint resume mechanism. When the network connection is unstable, the gateway temporarily stores the data in a local cache and automatically uploads it after the network is restored, ensuring the reliability of data transmission. The streaming data fusion module connects to the edge computing data acquisition module to perform time alignment and semantic unification on multi-source asynchronous data. This module uses the Apache Flink stream processing engine and employs a dynamic time warping algorithm to time-align asynchronous data streams with different sampling frequencies, where the difference in sampling frequencies is on the order of tens of times. Specifically, for asynchronous data streams with sampling frequencies such as 10Hz for UWB data, 1Hz for RFID data, and 0.2Hz for video recognition results, the DTW algorithm maps data with different timestamps to a unified time base, eliminating time deviations caused by differences in device sampling frequencies. Subsequently, a knowledge graph based on the OWL ontology is used to perform semantic disambiguation on the multi-source data. By establishing a unified terminology system, semantic consistency of the same concepts in different data sources is ensured. For example, "pouring completed" has the same semantic definition in the schedule and video recognition, resulting in stable disambiguation accuracy. This module also sets data integrity and consistency verification rules to automatically identify and filter invalid data, improving the accuracy of subsequent analysis. The spatiotemporal graph neural network prediction module connects to the streaming data fusion module, constructing a spatiotemporal correlation graph based on the fused data and performing joint predictions. The spatiotemporal correlation graph constructed by this module includes two basic elements: nodes and edges. Nodes represent work processes, material inventory, personnel attendance, and equipment status, each type of node carrying corresponding status information and attribute data. Edges represent the work process dependencies, resource allocation relationships, and cost relationships between nodes, describing the intensity of influence and transmission paths between different elements through edge weights and directions. Work process dependency edges reflect the logical sequence of construction processes, resource allocation relationship edges describe the configuration relationships between personnel, equipment, materials, and specific processes, and cost relationship edges quantify the impact of changes in each element on project costs. This module also inputs weather data, supply chain volatility index, and design change frequency as global features into the spatiotemporal correlation graph. These external factors influence the prediction results of the entire graph network through global feature vectors, enabling the model to comprehensively consider the combined impact of internal construction factors and external environmental changes on project progress and costs. Spatiotemporal graph neural networks propagate information between nodes through message passing mechanisms, learn the dynamic evolution of engineering projects by combining time series features, and output the progress completion status and cost expenditure prediction within future time windows; The adaptive early warning module connects to the spatiotemporal graph neural network prediction module, dynamically triggering early warning signals based on prediction results. This module employs a Bayesian online learning framework, dynamically adjusting the early warning thresholds for schedule deviations and cost overruns based on feedback from historical false alarm and false negative rates. The initial threshold is set at ±10% of the baseline value; that is, an early warning is triggered when the predicted schedule deviation or cost overrun exceeds 10% of the baseline value. As the system accumulates accurate early warning data during operation, the Bayesian learning algorithm automatically optimizes the early warning threshold setting based on the statistical distribution of false alarm and false negative rates, improving the accuracy of early warnings. When the prediction results indicate that the risk of schedule delays or cost overruns exceeds the dynamic threshold, the system immediately sends an early warning signal to project managers, providing risk source identification and suggested measures to achieve proactive risk management. The edge smart gateway uses the EGW-210 model, which supports LoRa / NB-IoT dual-mode communication and provides flexible connectivity solutions in construction environments with different network coverage conditions. The streaming data fusion module uses Apache Flink 1.14 to build a streaming processing cluster. This version has significant advantages in streaming processing performance and stability and can support the real-time processing needs of large-scale concurrent data streams.

[0029] The edge computing data acquisition module includes an edge smart gateway that supports LoRa or NB-IoT communication protocols. The edge smart gateway integrates an RFID reader, a UWB positioning base station, and a smart camera. The RFID reader is used to track the status of materials, the UWB positioning base station is used to obtain the location information of personnel and equipment, and the smart camera is equipped with a YOLOv8 model to automatically identify the progress of construction procedures.

[0030] Specifically, the edge computing data acquisition module is implemented by deploying an edge smart gateway of model EGW-210. This gateway supports LoRa and NB-IoT dual-mode communication protocols to adapt to different network coverage conditions at construction sites. The gateway hardware integrates an RFID reader, which reads RFID electronic tags pre-attached to building materials such as steel bars and formwork to collect real-time status data such as the arrival, departure, and installation location of materials, thereby realizing visualized tracking of the entire material flow process. Simultaneously, the integrated UWB positioning base station communicates with UWB tags worn by construction workers and large equipment by deploying base station nodes around the work area to obtain real-time three-dimensional spatial coordinates with an accuracy of approximately 15 centimeters. This provides precise location information for personnel hour statistics and equipment operating efficiency analysis. Furthermore, the gateway integrates a smart camera whose embedded YOLOv8 deep learning model continuously analyzes the real-time video stream, automatically identifying and outputting the completion status and progress percentage of key processes such as formwork erection, rebar tying, and concrete pouring. The expression of the YOLOv8 deep learning model is: in, The input image tensor represents the raw image data collected from the construction site. This represents the set of model parameters, including convolutional layer weights, bias terms, normalization parameters, etc., which are obtained through pre-training on a large-scale dataset. The forward propagation function of YOLOv8 consists of the following core components: Backbone: The feature extraction backbone network, which uses the CSPDarknet architecture to extract multi-scale features; Neck: Feature pyramid network that achieves multi-scale feature fusion through PANet; Head: The detection head, responsible for generating bounding box predictions and classification results; The model output consists of three key components: The geometric parameters of the i-th bounding box are represented by the coordinates of the center point and the width and height of the bounding box, respectively. This represents the confidence score of the i-th detection box, reflecting the probability that a construction target exists within the box; This represents the category probability distribution of the i-th detection box, corresponding to the classification results for different construction procedures; This edge smart gateway also has a breakpoint resume function. When the network connection is unstable, the data is temporarily stored in the local cache and automatically resumed after the network is restored, thereby ensuring the integrity and reliability of the collected progress, resource and environmental data.

[0031] The streaming data fusion module uses the Apache Flink streaming engine. It employs a dynamic time warping algorithm to time-align asynchronous data streams with different sampling frequencies, where the difference in sampling frequencies is no greater than 10 times. Then, it uses a knowledge graph based on the OWL ontology to perform semantic disambiguation on the multi-source data.

[0032] Specifically, the streaming data fusion module uses Apache Flink 1.14 to build a streaming processing cluster to process multi-source asynchronous data streams from the edge computing data acquisition module. This module uses a dynamic time warping algorithm to time-align asynchronous data streams with different sampling frequencies. In specific implementation, for data streams with sampling frequencies within 10 times, such as UWB positioning data at 10Hz, RFID data at 1Hz, and video recognition results at 0.2Hz, the dynamic time warping algorithm calculates the optimal curved path to eliminate nonlinear distortion on the time axis and accurately maps data points arriving at different times to a unified time reference. Subsequently, semantic unification is achieved using a knowledge graph based on the OWL ontology. This knowledge graph predefines entities, attributes, and relationships of core concepts such as progress, resources, and costs within the construction field. Through semantic mapping rules, heterogeneous data, such as material status records from RFID readers, personnel location coordinates from UWB positioning base stations, and process progress descriptions identified by smart cameras, are uniformly converted into standardized semantic representations that conform to the ontology definition. For example, the completion of formwork support in the camera recognition results is semantically associated with the formwork installation process in the schedule plan, thereby eliminating terminological ambiguity and ensuring the consistency of data semantics. This module also sets data integrity verification rules in Flink jobs to automatically filter invalid data with missing key fields or obviously abnormal values. Finally, it outputs a time-synchronized and semantically unified standardized data stream for use by the spatiotemporal graph neural network prediction module. The OWL ontology is a formalized, machine-understandable "dictionary" and "rulebook" specifically created for the field of "construction management." It uses the W3C standard OWL language to explicitly define all concepts involved in construction sites, their properties, and the relationships between them.

[0033] The spatiotemporal graph constructed by the spatiotemporal graph neural network prediction module includes nodes and edges. The nodes include nodes representing process tasks, nodes representing material inventory, nodes representing personnel attendance, and nodes representing equipment status. The edges are used to represent process dependencies, resource allocation relationships, and cost relationships between nodes. The spatiotemporal graph neural network prediction module also inputs weather data, supply chain fluctuation index, and design change frequency as global features into the spatiotemporal correlation graph.

[0034] Specifically, the spatiotemporal graph neural network prediction module first constructs a spatiotemporal correlation graph based on the standardized data stream output by the streaming data fusion module. The node instantiation includes process task nodes identified by unique process numbers and associated with planned duration and actual progress percentage attributes; material inventory nodes identified by material codes and associated with current inventory quantity and purchase unit price attributes; personnel attendance nodes identified by employee numbers and associated with attendance status and job type attributes; and equipment status nodes identified by equipment numbers and associated with runtime and fault code attributes. The specific implementation of edge relationships is to connect nodes using weighted directed edges. Process dependency relationship edges represent the process constraints between processes by setting predecessor and successor relationships and logical delay weights. Resource allocation relationship edges connect personnel nodes or equipment nodes with the process task nodes they serve by configuring resource allocation coefficients. Cost association relationship edges quantify the transmission effect of material price fluctuations or equipment failures on the cost of process nodes by defining cost impact factors. Meanwhile, this module also uses precipitation and temperature from weather data, logistics delay rate from supply chain volatility index, and number of change orders from design change frequency as global feature vectors. After projection through a fully connected layer, these vectors are concatenated with the feature vectors of each node, enabling the graph neural network to synchronously perceive the potential impact of external environmental changes on all nodes in the graph during message transmission, thereby achieving joint and accurate prediction of project progress and cost risks.

[0035] The adaptive early warning module adopts a Bayesian online learning framework. The adaptive early warning module is configured to dynamically adjust the early warning thresholds for schedule deviations and cost overruns based on feedback from the false alarm rate and false negative rate of historical early warnings. The initial threshold of the early warning threshold is set to ±10% of the baseline value.

[0036] Specifically, the adaptive early warning module uses a Bayesian online learning framework to dynamically optimize the early warning threshold. The expression for the Bayesian online learning framework is as follows: in, This indicates that these are the model parameters at time t. Specifically, in this system, it refers to the dynamically adjusted warning threshold, which is the core objective that we need to estimate and optimize. This indicates that this is newly observed data evidence at time t. Specifically, it is a comparison between the warning results issued by the system at the current time and the actual progress or cost status, including historical records of false alarms and missed alarms; Let represent the posterior probability distribution at time t-1, which is the distribution considering all historical data up to time t-1. Then, we consider the threshold parameter. The latest understanding of uncertainty is that, at the start of the iteration, this distribution is initialized by the prior distribution; This represents the new posterior probability distribution calculated at time t, which represents the distribution after incorporating the latest evidence. Then, for the current threshold parameter The latest and most accurate understanding of this distribution serves as the direct basis for early warning decisions; This represents the likelihood function. It represents the likelihood function assuming the current threshold. In the case of "truth," we observe the current evidence. How likely is it? This indicates a direct proportionality; during initialization, the warning thresholds for schedule deviations and cost overruns are set to ±10% of the baseline plan as the prior probability distribution. After the time-space graph neural network prediction module outputs the schedule and cost prediction results for future time windows, this module compares the predicted values ​​with the actual schedule and cost data, generates warning judgment results, and records the number of historical false alarms and missed alarms. Based on the core principle of Bayes' theorem, this module uses the prior threshold distribution as a basis, and uses the continuously accumulated false alarm rate and missed alarm rate as observation evidence to calculate the posterior probability. The probability distribution parameters of the warning threshold are dynamically corrected through the Bayesian update formula, so that the threshold is adaptively adjusted in the direction that minimizes the overall loss function. The system forms a closed-loop optimization process by continuously monitoring the feedback of warning accuracy. When the predicted risk value exceeds the current dynamic threshold, a graded warning signal is immediately triggered and automatically pushed to the project management personnel terminal, thereby realizing the intelligent warning capability that continuously optimizes itself as the project progresses.

[0037] A method for intelligent monitoring and early warning of project progress and cost includes the following steps: S1. Collect streaming data through edge devices deployed at the construction site; S2. Perform time alignment and semantic fusion on the collected multi-source asynchronous data; S3. Based on the fused data, construct a spatiotemporal relationship diagram of process-resource-cost; S4. Use a spatiotemporal graph neural network model to jointly predict schedule deviations and cost overrun risks; S5. Apply an adaptive threshold mechanism to dynamically trigger early warnings based on prediction results and feedback learning.

[0038] Specifically, S1: Streaming data is collected through edge devices deployed at the construction site. This step utilizes the edge computing data acquisition module described in Example 1, integrating devices such as RFID readers, UWB positioning base stations, and smart cameras through an edge smart gateway supporting LoRa or NB-IoT communication protocols to collect multi-source data streams from the construction site in real time. RFID readers collect material tracking data, UWB positioning base stations collect personnel and equipment location information, and smart cameras identify the progress status of construction procedures through a built-in YOLOv8 model. The edge smart gateway ensures data integrity through a breakpoint resume mechanism in the event of network fluctuations. S2: Time alignment and semantic fusion are performed on the collected multi-source asynchronous data. This step utilizes the streaming data fusion module described in Example 1, employing the Apache Flink streaming engine for data processing. During time alignment, a dynamic time warping algorithm is used to process data streams with sampling frequency differences within a factor of 10, mapping asynchronous data streams such as the 10Hz sampling frequency of UWB data, the 1Hz sampling frequency of RFID data, and the 0.2Hz sampling frequency of video recognition results onto a unified time base. The semantic fusion process utilizes a knowledge graph based on the OWL ontology to perform semantic disambiguation on the multi-source data, establishing a unified terminology system to ensure semantic consistency of the same concepts in different data sources. S3: Based on the fused data, construct a spatiotemporal relationship graph of processes, resources, and costs. This step constructs a spatiotemporal relationship graph containing nodes and edges. Nodes include nodes representing process tasks, material inventory, personnel attendance, and equipment status. Edges are used to represent process dependencies, resource allocation relationships, and cost relationships between nodes. Process dependency edges reflect the logical sequence of construction processes, resource allocation edges describe the configuration relationships between personnel, equipment, materials, and specific processes, and cost relationship edges quantify the impact of changes in each factor on project costs. Simultaneously, weather data, supply chain volatility index, and design change frequency are input as global features into the spatiotemporal relationship graph, enabling the graph network to comprehensively consider the combined effects of internal construction factors and external environmental changes. S4: Jointly predict schedule deviations and cost overrun risks using a spatiotemporal graph neural network model. This step uses the message passing mechanism of the spatiotemporal graph neural network to propagate information between nodes and learns the dynamic evolution of the project by combining time series features. During the joint prediction process, a graph attention mechanism is used to adaptively learn the impact weights of weather changes and supply chain fluctuations on different process nodes. Through dynamic adjustment of attention weights, the model can identify the sensitivity of each process node under different external conditions, thereby improving the accuracy of the prediction. The graph attention mechanism automatically learns and assigns different attention weights based on the correlation between weather changes and process delays, and the correlation between supply chain fluctuations and material cost changes in historical data, enabling the model to focus on key process nodes that are significantly affected by external factors. S5: Apply an adaptive threshold mechanism to dynamically trigger early warnings based on prediction results and feedback learning. This step uses the adaptive early warning module described in Example 1, employing a Bayesian online learning framework to dynamically adjust the early warning thresholds for schedule deviations and cost overruns based on feedback from historical false alarm and missed alarm rates. The initial setting of the early warning threshold is ±10% of the baseline value. As early warning accuracy data accumulates during system operation, the Bayesian learning algorithm automatically optimizes the early warning threshold based on the statistical distribution of false alarm and missed alarm rates. When the prediction results indicate that the schedule is lagging or the risk of cost overruns exceeds the dynamic threshold, the system immediately sends an early warning signal to project managers and provides risk source location and suggested measures.

[0039] In step S2, the time alignment of multi-source asynchronous data specifically involves using a dynamic time warping algorithm to process data streams with sampling frequency differences within 10 times.

[0040] Specifically, in the time alignment process of step S2, for asynchronous data streams with sampling frequencies within 10 times, such as UWB positioning data (10Hz), RFID material data (1Hz), and video recognition results (0.2Hz), a dynamic time warping algorithm is used for processing. The core principle of this algorithm is to calculate the minimum curved path between two time series of different lengths using dynamic programming technology to find the optimal nonlinear alignment mapping relationship, thereby overcoming the information distortion problem caused by the fixed sampling rate resampling method. In specific implementation, firstly, an independent temporal data stream window is opened for each data source in the Apache Flink stream processing engine. Then, the dynamic time warping algorithm calculates the cumulative distance matrix between the UWB high-frequency sequence points and the RFID or video recognition low-frequency sequence points, and finds the path that minimizes the overall alignment cost through backtracking. Finally, the high-frequency data points are adaptively matched to the timestamps of the low-frequency sequences, achieving accurate time axis alignment without changing the original data form, thus laying an accurate time reference for subsequent semantic fusion and joint prediction.

[0041] In step S4, the joint prediction uses a graph attention mechanism to adaptively learn the weights of the impact of weather changes and supply chain fluctuations on different process nodes.

[0042] Specifically, for each process node in the spatiotemporal correlation graph, a query vector, a key vector, and a value vector are generated by calculating the correlation score between it and global feature vectors such as weather data and supply chain fluctuation index. The query vector originates from the feature representation of the target process node, while the key vector and value vector are obtained by linear transformation of the global feature vector. By scaling dot product attention calculations, the model assigns a dynamic weight to each global feature, which reflects the potential impact of specific external conditions on the process node. This information aggregation method based on attention weights enables the model to differentiate and integrate external environmental information during message transmission, rather than processing it equally. For example, when facing rainstorms, the model automatically increases the attention weight of weather features on outdoor pouring process nodes while reducing their impact coefficient on indoor installation processes. This enables refined modeling of external disturbance factors and improves the accuracy and interpretability of joint prediction of schedule deviations and cost overrun risks.

[0043] A computer-readable storage medium having a computer program stored thereon, characterized in that, when executed by a processor, the program implements the steps of an engineering progress and cost early warning method based on a spatiotemporal graph neural network.

[0044] Specifically, the program first calls the edge device interface module to drive RFID readers, UWB positioning base stations, and smart cameras deployed at the construction site to collect streaming data via LoRa or NB-IoT communication protocols, and enables a breakpoint resume mechanism to ensure data integrity. Then, the program starts the streaming data processing module to call the Apache Flink engine's DataStream API to apply a dynamic time warping algorithm to align the collected multi-source asynchronous data streams, and loads a predefined OWL ontology knowledge graph file to perform semantic disambiguation and unification of the data. Next, the program's graph construction module instantiates process task nodes, material inventory nodes, personnel attendance nodes, and equipment status nodes based on the semantically fused data. It then constructs a spatiotemporal relational graph topology based on preset process dependencies, resource allocation, and cost association rules, while injecting global features such as externally acquired weather data into the graph structure. The program's core prediction module then loads a pre-trained spatiotemporal graph neural network model. This model integrates a graph attention mechanism to achieve weighted aggregation of message passing between nodes and performs joint prediction of execution progress and cost based on historical sequence data. Finally, the program runs an adaptive early warning module that uses a Bayesian online learning algorithm to dynamically update the early warning threshold based on the prediction results and real-time feedback. When the threshold is triggered, an early warning signal is automatically generated and sent to the management terminal through a message queue interface, thus completely reproducing the entire process of the early warning method.

[0045] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Identical components are represented by the same reference numerals. Therefore, all equivalent changes made to the structure, shape, and principle of this application should be covered within the scope of protection of this application.

Claims

1. An intelligent monitoring and early warning system for project progress and cost, characterized in that, include: Edge computing data acquisition module: Deployed at the construction site, it is used to collect progress, resource and environmental data through various IoT sensing devices; Streaming data fusion module: connected to the edge computing data acquisition module, used for time alignment and semantic unification of multi-source asynchronous data; Spatiotemporal graph neural network prediction module: connected to the streaming data fusion module, used to construct a spatiotemporal correlation graph based on the fused data and perform joint prediction; Adaptive early warning module: connected to the spatiotemporal graph neural network prediction module, used to dynamically trigger early warning signals based on the prediction results.

2. The intelligent monitoring and early warning system for project progress and cost according to claim 1, characterized in that: The edge computing data acquisition module includes an edge smart gateway that supports LoRa or NB-IoT communication protocols. The edge smart gateway integrates an RFID reader, a UWB positioning base station, and a smart camera. The RFID reader is used to track the status of materials, the UWB positioning base station is used to obtain the location information of personnel and equipment, and the smart camera is equipped with a YOLOv8 model to automatically identify the progress of construction procedures.

3. The intelligent monitoring and early warning system for project progress and cost according to claim 1, characterized in that: The streaming data fusion module uses the Apache Flink streaming engine. It employs a dynamic time warping algorithm to time-align asynchronous data streams with different sampling frequencies, where the difference in sampling frequencies is no greater than 10 times. Then, it uses a knowledge graph based on the OWL ontology to perform semantic disambiguation on the multi-source data.

4. The intelligent monitoring and early warning system for project progress and cost according to claim 1, characterized in that: The spatiotemporal graph constructed by the spatiotemporal graph neural network prediction module includes nodes and edges. The nodes include nodes representing process tasks, nodes representing material inventory, nodes representing personnel attendance, and nodes representing equipment status. The edges are used to represent process dependencies, resource allocation relationships, and cost relationships between nodes.

5. The intelligent monitoring and early warning system for project progress and cost according to claim 4, characterized in that: The spatiotemporal graph neural network prediction module also inputs weather data, supply chain fluctuation index, and design change frequency as global features into the spatiotemporal correlation graph.

6. The intelligent monitoring and early warning system for project progress and cost according to claim 1, characterized in that: The adaptive early warning module adopts a Bayesian online learning framework. The adaptive early warning module is configured to dynamically adjust the early warning thresholds for schedule deviations and cost overruns based on feedback from the false alarm rate and false negative rate of historical early warnings. The initial threshold of the early warning threshold is set to ±10% of the baseline value.

7. A method for intelligent monitoring and early warning of project progress and cost, applicable to the intelligent monitoring and early warning system for project progress and cost as described in any one of claims 1-6, characterized in that, Includes the following steps: S1. Collect streaming data through edge devices deployed at the construction site; S2. Perform time alignment and semantic fusion on the collected multi-source asynchronous data; S3. Based on the fused data, construct a spatiotemporal relationship diagram of process-resource-cost; S4. Use a spatiotemporal graph neural network model to jointly predict schedule deviations and cost overrun risks; S5. Apply an adaptive threshold mechanism to dynamically trigger early warnings based on prediction results and feedback learning.

8. The intelligent monitoring and early warning method for project progress and cost according to claim 7, characterized in that: In step S2, the time alignment of multi-source asynchronous data specifically involves using a dynamic time warping algorithm to process data streams with sampling frequency differences within 10 times.

9. The intelligent monitoring and early warning method for project progress and cost according to claim 7, characterized in that: In step S4, the joint prediction uses a graph attention mechanism to adaptively learn the weights of the impact of weather changes and supply chain fluctuations on different process nodes.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the intelligent monitoring and early warning method for engineering progress and cost as described in any one of claims 7 to 9.

Images