Machine learning-based construction engineering construction progress early warning method and system

Abstract

The application relates to the technical field of building engineering management, and discloses a building engineering construction progress early warning method and system based on machine learning, which collects multi-dimensional original data from a construction site sensor, a BIM model, an IoT device and a historical database; extracts features related to the construction progress from the multi-dimensional original data, combines and transforms the extracted features to obtain composite features; inputs the composite features into a progress prediction model, predicts the construction progress in a future period of time through the progress prediction model, and outputs a prediction result; determines an initial early warning threshold value by analyzing historical construction data; in the construction process, dynamically adjusts the early warning threshold value according to real-time monitoring data and the prediction result; when the construction progress deviates from the predicted progress and exceeds the dynamically adjusted early warning threshold value, early warning is triggered; and the application improves adaptability to complex construction environments and prediction accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of construction engineering management, and particularly relates to a construction engineering construction progress early warning method and system based on machine learning. BACKGROUND

[0002] In the construction engineering construction process, effective management of construction progress is crucial, and timely and accurate grasp of construction progress is of great significance for reasonable arrangement of resources and ensuring timely delivery of the project. However, the current traditional construction engineering construction progress early warning method mostly relies on manual experience or simple linear analysis model, and it is difficult to comprehensively consider the complex and changeable factors in the construction process, such as weather changes, material supply delays, labor shortages, equipment failures, etc. These methods often have the problems of low early warning accuracy and inability to dynamically adapt to changes in the construction environment, resulting in the inability to timely and effectively warn and adjust the construction progress. SUMMARY

[0003] The present application aims to solve the above problems, and designs a construction engineering construction progress early warning method and system based on machine learning.

[0004] The first aspect of the present application provides a construction engineering construction progress early warning method based on machine learning, which comprises the following steps:

[0005] Collecting multi-dimensional raw data from construction site sensors, BIM models, IoT devices and historical databases;

[0006] Extracting features related to construction progress from the multi-dimensional raw data, combining and transforming the extracted features to obtain composite features;

[0007] Inputting the composite features into a progress prediction model, predicting the construction progress in a future period of time through the progress prediction model, and outputting the prediction result;

[0008] Determining an initial early warning threshold by analyzing historical construction data, and dynamically adjusting the early warning threshold according to real-time monitoring data and prediction results during the construction process;

[0009] Triggering an early warning when the construction progress deviates from the predicted progress and exceeds the dynamically adjusted early warning threshold.

[0010] Optionally, in the first implementation manner of the first aspect of the present application, the extracting features related to construction progress from the multi-dimensional raw data, and combining and transforming the extracted features to obtain composite features comprises:

[0011] Inputting the multi-dimensional raw data in the form of time series into an LSTM network, capturing the trend changes and periodic regularities of construction progress in long time series through a gating mechanism, and outputting a feature vector representing the progress trend and period.

[0012] The construction tasks of the construction project are represented in a graph structure, the graph structure is learned by using a graph neural network, complex dependency relationships between the tasks are mined through a node feature updating and a message passing mechanism, and a feature matrix representing a task correlation degree is output;

[0013] The HOG algorithm is used to extract geometric features of the engineering components in the BIM model.

[0014] Optionally, in a second implementation manner of the first aspect of the present application, the HOG algorithm is used to extract geometric features of the engineering components in the BIM model, including:

[0015] Based on the geometric attributes and semantic information of the components, the three-dimensional geometric information of the components is converted into a parameterized representation;

[0016] The texture features and material properties of the component surface are extracted, the gradient changes of the component surface in different directions are calculated, and a feature histogram is formed;

[0017] The components are divided into multiple local regions, features are extracted and fused respectively, and local geometric properties of the components are captured

[0018] The continuous gradient features are discretized and converted into digital feature vectors, and the geometric feature information of the components is obtained.

[0019] Optionally, in a third implementation manner of the first aspect of the present application, the composite features are input into the progress prediction model, the progress prediction model is used to predict the construction progress in a future period of time, and a prediction result is output, including:

[0020] The composite features are input into the progress prediction model, the composite features enter the XGBoost branch, the composite features are divided and evaluated layer by layer through the XGBoost branch, and a first construction progress prediction result is obtained;

[0021] The composite features also enter the random forest branch, each decision tree independently analyzes and judges the composite features when processing, and through node splitting and decision-making, a respective construction progress prediction result is obtained, the prediction results of all the decision trees are summarized, a voting method is used, and a second construction progress prediction result is obtained;

[0022] The first construction progress prediction result and the second construction progress prediction result are weighted and fused by using a weight determined through cross-validation, and a final prediction result is obtained.

[0023] Optionally, in a fourth implementation manner of the first aspect of the present application, the composite features are divided and evaluated layer by layer through the XGBoost branch, and a first construction progress prediction result is obtained, including:

[0024] According to the potential importance of the characteristics to the construction progress prediction, the composite characteristics are divided into different subsets to form initial branches under the root node;

[0025] For each subset after the division of the root node, the characteristics that can distinguish the data samples are selected again from the remaining composite characteristics to divide the next layer of nodes;

[0026] When the characteristics have been used for division, leaf nodes are formed, and the data samples in each leaf node correspond to a construction progress prediction result;

[0027] The prediction results of all leaf nodes are summarized to obtain a first construction progress prediction result.

[0028] Optionally, in a fifth implementation manner of the first aspect of the present application, the initial warning threshold is determined by analyzing historical construction data, and the warning threshold is dynamically adjusted according to real-time monitoring data and prediction results during the construction process, including:

[0029] An initial construction progress state estimation value is set for the Kalman filter algorithm, including the mean and variance of the current construction progress;

[0030] Based on the construction progress state estimation value at the previous time, the Kalman filter is used to calculate the construction progress prediction state at the current time;

[0031] The real-time collected construction progress data is compared with the prediction result, and the estimation of the construction progress state is automatically adjusted according to the difference Kalman filter algorithm to dynamically adjust the warning threshold.

[0032] Optionally, in a sixth implementation manner of the first aspect of the present application, the initial warning threshold is determined by analyzing historical construction data, and the warning threshold is dynamically adjusted according to real-time monitoring data and prediction results during the construction process, and further includes:

[0033] The real-time state of the construction progress is taken as the input of the deep Q network, and based on the current state, the deep Q network selects an action from the action space to determine a threshold adjustment strategy;

[0034] The advantages and disadvantages of each threshold adjustment action are evaluated through a reward feedback mechanism, and the deep Q network continuously optimizes the threshold adjustment strategy according to the reward feedback obtained by each action.

[0035] The second aspect of the present application provides a building engineering construction progress warning system based on machine learning, which includes:

[0036] The acquisition module is used for acquiring multi-dimensional original data from construction site sensors, BIM models, IoT devices and historical databases;

[0037] An extraction module is configured to extract features related to construction progress from multi-dimensional raw data, combine and transform the extracted features to obtain composite features;

[0038] A prediction module is configured to input the composite features into a progress prediction model, predict the construction progress in a future period of time through the progress prediction model, and output a prediction result;

[0039] An adjustment module is configured to determine an initial early warning threshold by analyzing historical construction data, and dynamically adjust the early warning threshold according to real-time monitoring data and the prediction result in the construction process.

[0040] An early warning module is configured to trigger early warning when the construction progress deviates from the predicted progress and exceeds the dynamically adjusted early warning threshold.

[0041] The third aspect of the present application provides a machine learning-based construction progress early warning device, which comprises a memory and at least one processor, and the memory stores instructions; the at least one processor invokes the instructions in the memory to enable the machine learning-based construction progress early warning device to perform the steps of the machine learning-based construction progress early warning method according to any one of the above.

[0042] The fourth aspect of the present application provides a computer readable storage medium, which stores instructions, and the instructions are executed by a processor to implement the steps of the machine learning-based construction progress early warning method according to any one of the above.

[0043] In the technical solution provided by the present application, multi-dimensional raw data is collected from construction site sensors, BIM models, IoT devices and historical databases; features related to construction progress are extracted from the multi-dimensional raw data, and the extracted features are combined and transformed to obtain composite features; the composite features are input into a progress prediction model, the construction progress in a future period of time is predicted through the progress prediction model, and a prediction result is output; an initial early warning threshold is determined by analyzing historical construction data, and the early warning threshold is dynamically adjusted according to real-time monitoring data and the prediction result in the construction process; early warning is triggered when the construction progress deviates from the predicted progress and exceeds the dynamically adjusted early warning threshold; the present application fully integrates various information in the construction process, can comprehensively reflect the actual construction situation, provides rich data support for accurate construction progress early warning, improves the adaptability and prediction accuracy in complex construction environments, automatically adjusts the early warning standard according to the dynamic changes in the construction process, avoids the problem of inaccurate early warning caused by fixed threshold, and improves the timeliness and reliability of early warning. BRIEF DESCRIPTION OF DRAWINGS

[0044] Various other advantages and benefits will become apparent to those of ordinary skill in the art, upon reading the following detailed description of the preferred embodiment. The accompanying drawings are included to provide a description of preferred embodiments, and are not meant to limit the present application.

[0045] Figure 1 A flow chart of the machine learning-based construction progress early warning method provided by the embodiment of the present application is provided.

[0046] Figure 2 A structural schematic diagram of the machine learning-based construction progress early warning system provided by the embodiment of the present application is provided.

[0047] Figure 3 A structural schematic diagram of the machine learning-based construction progress early warning device provided by the embodiment of the present application is provided. DETAILED DESCRIPTION

[0048] The terms "first", "second", "third", "fourth" and the like in the description and claims of the present application and in the above DETAILED DESCRIPTION, if any, are used for distinguishing between similar objects talking about the application and do not necessarily have to appear in the application in this order or in sequence. It is to be understood that the data used in these descriptions replacing other data, as appropriate, so that the embodiments described herein can be carried out in other than the order discussed herein without departing from the scope of the application. Additionally, the terms "comprise" or "have" and variations thereof in the description and claims of the present application are intended to cover both the positive and negative cases, for example, a process, method, article, manufacture, or device that comprises a list of steps or units is not necessarily limited to those steps or units clearly recited, but can include other steps or units that are not expressly listed or inherent to such process, method, article, manufacture, or device.

[0049] For the sake of understanding, the specific flow of the embodiment of the present application is described below, please refer to Figure 1 The flow chart of the machine learning-based construction progress early warning method provided by the embodiment of the present application, the method specifically includes the following steps:

[0050] Step 101, collecting multi-dimensional original data from construction site sensors, BIM models, IoT devices and historical databases;

[0051] In this embodiment, the sensor network deployed at the construction site, such as temperature and humidity sensors, displacement sensors, RFID devices, the construction site sensor collects environmental parameters, equipment operation data, personnel working status in real time, environmental parameters such as temperature and humidity, wind speed, rainfall, equipment operation data such as equipment start and stop time, working time, fault frequency, personnel working status such as personnel attendance, working position, operation behavior; the BIM model extracts the geometric size of the engineering component, the construction process requirement, the spatial position relationship, and the progress plan information; the IoT device obtains the material access data; the historical database collects the construction progress, resource input, risk event and other data of the same type of engineering in the past, and in the collection process, standardized data interface and protocol are used to ensure that data from different sources can be accurately transmitted and stored.

[0052] Step 102, extracting features related to construction progress from multi-dimensional raw data, combining and transforming the extracted features to obtain composite features;

[0053] In this embodiment, the time series form of multi-dimensional raw data is input into the LSTM network, the trend change and periodicity of the construction progress in the long time sequence are captured through the gating mechanism, and the feature vector representing the progress trend and period is output; the construction task of the building engineering is represented in a graph structure, the graph structure is learned by using the graph neural network, the complex dependency relationship between tasks is mined through the node feature update and message passing mechanism, and the feature matrix representing the task correlation degree is output; the HOG algorithm is used to extract the geometric features of the engineering components in the BIM model.

[0054] In this embodiment, the components are identified based on the geometric attributes and semantic information of the components, the three-dimensional geometric information of the components is converted into parameterized representation; the texture features and material properties of the component surface are extracted, the gradient change of the component surface in different directions is calculated, and a feature histogram is formed; the component is divided into multiple local regions, the features are extracted and fused respectively, the local geometric properties of the component are captured, the continuous gradient features are discretized, and the digital feature vector is converted, and the geometric feature information of the component is obtained.

[0055] Step 103, inputting the composite features into the progress prediction model, predicting the construction progress in a future period of time through the progress prediction model, and outputting the prediction result;

[0056] In this embodiment, the composite features are input into the progress prediction model. The composite features enter the XGBoost branch, and are divided and evaluated layer by layer by the XGBoost branch to obtain a first construction progress prediction result. The composite features also enter the random forest branch, and each decision tree independently analyzes and judges the composite features when processing. Through node splitting and decision-making, a respective construction progress prediction result is obtained. The prediction results of all decision trees are summarized, and a second construction progress prediction result is obtained by voting. The first construction progress prediction result and the second construction progress prediction result are weighted and fused by the weight determined by cross-validation to obtain a final prediction result.

[0057] In this embodiment, the constructed composite features are format-converted and standardized to meet the input requirements of the progress prediction model. The data is checked for missing values and abnormal values. Missing values are handled by filling with adjacent values or replacing with statistical quantities. Abnormal values are corrected or removed according to business logic to ensure the data quality of the input model. The composite feature data first enters the XGBoost branch. The model divides and evaluates the input features layer by layer based on the internally trained tree structure. XGBoost allocates data samples to different child nodes according to the established division rules, starting from the root node, and gradually determines the class or prediction value interval to which the sample belongs. In this process, the model considers the contribution of each feature to the construction progress prediction, quickly processes large-scale composite feature data, and obtains a preliminary construction progress prediction result.

[0058] The same set of composite feature data is input into the random forest branch. The random forest is composed of multiple decision trees, each of which is trained based on a subset obtained by sampling with replacement from the original data and a random selection of part of the features. When processing the input data, each decision tree independently analyzes and judges the composite features, and gives a respective construction progress prediction result through node splitting and decision rules. Finally, the random forest summarizes the prediction results of all decision trees and obtains the construction progress prediction result of the random forest branch by voting or averaging.

[0059] The prediction results output by the XGBoost branch and the random forest branch are weighted and fused according to the weights determined in advance through cross-validation; XGBoost is good at processing large-scale data and complex features, and random forest has good robustness to outliers and noise; through weight fusion, the advantages of both are fully utilized, so that the prediction result after fusion can more accurately reflect the future construction progress; the prediction result after fusion will be calibrated in combination with the actual situation of the construction project; special circumstances that may occur during construction, such as important holidays and sudden policy adjustments, are considered, and the prediction result is appropriately modified, and finally a detailed construction progress prediction result in a future period of time is output, including the estimated completion time of each sub-project, the overall engineering progress percentage, the time node prediction of key nodes and other information.

[0060] In this embodiment, the composite features are divided into different subsets according to their potential importance to construction progress prediction, forming initial branches under the root node; for each subset after root node division, the features that can further distinguish data samples are selected from the remaining composite features to divide the next layer of nodes; when all features have been used for division, leaf nodes are formed, and the data samples in each leaf node correspond to a construction progress prediction result; the prediction results of all leaf nodes are summarized to obtain the first construction progress prediction result;

[0061] In this embodiment, the composite feature data constructed through preprocessing and feature engineering is arranged in a format recognizable by the XGBoost branch and loaded into the data processing module of the branch; the integrity and consistency of the data are ensured, and invalid data that may affect the prediction accuracy are removed; the branch selects the most discriminative features as the basis for root node division according to the potential importance of the features to construction progress prediction from the composite feature set; for example, features such as material supply delay days and key equipment failure rate with high correlation to construction progress are preferentially selected, and all data samples are divided into different subsets according to the value range of the feature, forming initial branches under the root node;

[0062] For each subset after root node division, the branch continues to select the features that can further distinguish data samples from the remaining composite features to divide the next layer of nodes; this process is repeated continuously, and as the number of node layers increases, the data subsets are divided more and more finely; at each division, the branch considers the influence of different feature combinations on construction progress prediction to ensure that the data samples in each sub-node after division have higher similarity in construction progress-related attributes;

[0063] When a preset division termination condition is reached, such as the number of samples in a node being less than a certain threshold, or all features have been used for division, a leaf node is formed; the data samples in each leaf node correspond to a construction progress prediction result, which is usually based on the historical construction progress data statistics of the samples in the leaf node, such as calculating the average construction progress completion time, progress deviation mean, etc. of the samples in the leaf node, as the construction progress prediction value corresponding to the leaf node;

[0064] The prediction results of all leaf nodes are summarized through an integration mechanism within the branch; for a new input composite feature sample, the branch will determine its belonging leaf node according to its division path in each node, and output the prediction result of the leaf node as the first construction progress prediction result of the sample; in the integration process, the branch will consider the weight or credibility of different branch paths, and optimize and adjust the final prediction result to improve the accuracy and reliability of the prediction.

[0065] Step 104, determine the initial warning threshold by analyzing historical construction data, and dynamically adjust the warning threshold according to real-time monitoring data and prediction results during construction;

[0066] In this embodiment, the initial construction progress state estimation value is set for the Kalman filter algorithm, including the mean and variance of the current construction progress; based on the construction progress state estimation value at the last time, the current construction progress prediction state is calculated using the Kalman filter; the real-time collected construction progress data is compared with the prediction result, and the estimation of the construction progress state is automatically adjusted according to the difference Kalman filter algorithm to dynamically adjust the warning threshold.

[0067] In this embodiment, various sensors and data collection equipment are deployed at the construction site to obtain real-time construction progress data, such as daily completed engineering quantity, time nodes of started and completed sub-projects, resource input quantity, etc.; these data are collected and transmitted to the data processing center at fixed time intervals to form continuous construction progress time series data; the initial construction progress state estimation value is set for the Kalman filter algorithm, including the mean of the current construction progress, i.e. the current estimated construction progress completion condition, and the variance reflecting the uncertainty degree of the estimation value, which can be determined based on the construction progress plan, previous similar engineering experience or preliminary data at the beginning of the project;

[0068] Based on the construction progress state estimation value at the last time, the current construction progress prediction state is calculated using the prediction mechanism of the Kalman filter; in this process, the algorithm will consider the trend of the construction progress changing with time, such as the expected progress under the normal construction speed, and also consider the potential influence of possible interference factors, such as weather, equipment failure, etc. on the construction progress, so as to obtain a preliminary current time construction progress prediction value and the corresponding prediction variance;

[0069] When the real-time collected construction progress data arrives, the actual data is compared with the predicted value obtained in the prediction stage; according to the difference between the two, the Kalman filtering algorithm automatically adjusts the estimation of the construction progress state; if the actual progress is faster than the predicted progress, the algorithm will correspondingly increase the estimation of the current construction progress mean value and reduce the estimation uncertainty; on the contrary, if the actual progress lags behind, the algorithm will lower the mean value estimation and possibly increase the variance to reflect greater uncertainty; with the continuous arrival of new real-time construction progress data, the above prediction and update process will continue to circulate; with each iteration, the Kalman filtering algorithm will further optimize the estimation of the construction progress state according to the latest data, so that the real-time processing result of the construction progress can closely follow the changes of the actual construction situation, providing accurate and timely data support for the subsequent dynamic adjustment of the early warning threshold.

[0070] In this embodiment, the real-time state of the construction progress is taken as the input of the deep Q network, and based on the current state, the deep Q network selects an action from the action space to determine a threshold adjustment strategy; the pros and cons of each threshold adjustment action are evaluated through the reward feedback mechanism, and the deep Q network continuously optimizes the threshold adjustment strategy according to the reward feedback obtained by each action.

[0071] In this embodiment, in the construction progress management scenario, the state and action required by the deep Q network are first defined; the state space covers real-time construction progress data, such as the deviation between the current actual progress and the planned progress, the progress change trend, the resource input situation, external environmental factors and other information; the action space is various possible threshold adjustment operations, such as increasing the delay warning threshold by 10%, decreasing it by 5%, or adjusting the resource shortage warning threshold in stages, etc., to build a basic framework for the operation of DQN;

[0072] The real-time state of the construction progress is continuously perceived, and these information is converted into a state vector that can be understood by DQN; based on the current state, DQN selects an action from the action space, i.e. determines a threshold adjustment strategy; for example, when it is detected that the weather is bad recently, the construction efficiency is reduced, and the actual progress starts to show a lagging trend, DQN may choose to moderately lower the delay warning threshold, so that the system can issue warnings more timely; then, the system adjusts the warning threshold according to the selected strategy and observes the effect after the adjustment;

[0073] A reward feedback mechanism is established to evaluate the pros and cons of each threshold adjustment action. If the warning system can timely and accurately issue a warning after adjusting the threshold, and the relevant personnel take measures based on the warning to effectively avoid delays in the construction period or waste of resources and other problems, a positive reward is given. Conversely, if there is a false alarm, a missed alarm, or the adjustment fails to have a positive effect on construction progress management, a negative reward is given. For example, if the threshold adjustment successfully avoids a stoppage due to delayed material supply, the system will give a higher positive reward. If frequent false alarms disrupt the work order, a negative reward is given.

[0074] DQN continuously optimizes its threshold adjustment strategy based on the reward feedback obtained from each action. In the context of continuous progress in the construction process and the emergence of new states, DQN continuously selects actions, performs threshold adjustments, and obtains rewards, and gradually learns a better strategy based on these experiences. Over time and with experience accumulation, DQN gradually masters how to adjust the warning threshold to maximize the accuracy and effectiveness of construction progress warning in different construction scenarios, achieving continuous optimization of the threshold adjustment strategy to better adapt to complex and changing construction environments.

[0075] Step 105, when the construction progress deviates from the predicted progress and exceeds the dynamically adjusted warning threshold, triggering a warning.

[0076] In this embodiment, when the actual construction progress of the construction site deviates from the predicted progress output by the progress prediction model, and the degree of deviation exceeds the dynamically adjusted warning threshold optimized by the Kalman filter and the deep Q network, the warning mechanism will be triggered immediately. At this time, a preliminary analysis of the deviation cause will be automatically performed, and real-time collected construction data such as personnel attendance, equipment operation status, material supply delay records, and weather changes will be combined to determine whether the progress deviation is caused by resource shortage, equipment failure, adverse weather, or other factors. Subsequently, detailed warning information containing the specific value of the deviation, the potential impact range, historical similar situation handling experience, and other content will be generated according to the severity and type of the warning, and will be accurately pushed to project managers, construction team leaders, supervisory personnel, and other relevant responsible persons through multiple channels such as SMS, email, and construction management platform pop-up windows, ensuring that all parties can quickly grasp the abnormal construction progress situation and take effective corrective measures in a timely manner to ensure the smooth progress of the project.

[0077] Please refer to Figure 2 The building engineering construction progress warning system based on machine learning provided by the embodiment of the present application has the structure as shown in the figure.

[0078] The acquisition module is used to acquire multi-dimensional raw data from construction site sensors, BIM models, IoT devices, and historical databases.

[0079] The extraction module is configured to extract features related to the construction progress from multi-dimensional raw data, combine and transform the extracted features to obtain composite features.

[0080] The prediction module is configured to input the composite features into a progress prediction model, predict the construction progress in a future period of time through the progress prediction model, and output a prediction result.

[0081] The adjustment module is configured to determine an initial early warning threshold by analyzing historical construction data, and dynamically adjust the early warning threshold according to real-time monitoring data and the prediction result in the construction process.

[0082] The early warning module is configured to trigger early warning when the construction progress deviates from the predicted progress and exceeds the dynamically adjusted early warning threshold.

[0083] Figure 3 The structure of the building construction progress early warning device based on machine learning provided by the embodiment of the present application, the building construction progress early warning device based on machine learning 300 can be different due to configuration or performance, and can include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, one or more storage application programs 333 or data 332 storage media 330 (for example, one or more mass storage devices). Among them, the memory 320 and the storage medium 330 can be temporary storage or persistent storage. The program stored in the storage medium 330 can include one or more modules (not shown in the figure), each module can include a series of instruction operations in the building construction progress early warning device based on machine learning 300. Further, the processor 310 can be configured to communicate with the storage medium 330, execute a series of instruction operations in the storage medium 330 on the building construction progress early warning device based on machine learning 300, so as to realize the method provided by the above embodiment.

[0084] The building construction progress early warning device based on machine learning 300 can also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating devices 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art can understand that, Figure 3 The structure of the building construction progress early warning device based on machine learning shown does not constitute a limitation on the computer device provided by the present application, and can include more or fewer components than shown, or combine certain components, or different component arrangements.

[0085] The application further provides a computer readable storage medium, which can be a nonvolatile computer readable storage medium or a volatile computer readable storage medium, and instructions are stored in the computer readable storage medium, and the instructions make a computer execute steps of the machine learning based construction progress early warning method provided by each of the embodiments when the instructions are run on the computer.

[0086] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the device or apparatus or unit described above can refer to the corresponding process in the foregoing method embodiments, and will not be described here.

[0087] The integrated unit, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each of the embodiments of the application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0088] The basic principles, main features and advantages of the application are shown and described above. Those skilled in the art should understand that the application is not limited by the foregoing embodiments, and the foregoing embodiments and descriptions in the specification are only preferred examples of the application and are not intended to limit the application. Without departing from the spirit and scope of the application, various changes and improvements can be made to the application, and these changes and improvements all fall within the scope of the claimed application. The scope of protection of the application is defined by the appended claims and their equivalents.

Claims

1. A machine learning-based construction progress warning method for a construction project, characterized by, The method comprises the following steps: Collecting multi-dimensional raw data from construction site sensors, BIM models, IoT devices, and historical databases; Extracting features related to construction progress from the multi-dimensional raw data, combining and transforming the extracted features to obtain composite features; Inputting the composite features into a progress prediction model, predicting the construction progress in a future period of time through the progress prediction model, and outputting the prediction results; By analyzing historical construction data, an initial warning threshold is determined, and in the construction process, the warning threshold is dynamically adjusted according to real-time monitoring data and prediction results; When the construction progress deviates from the predicted progress and exceeds the dynamically adjusted warning threshold, a warning is triggered; The method comprises the following steps: The composite features are input into the progress prediction model, and the composite features enter the XGBoost branch, which performs layer-by-layer division and evaluation on the composite features to obtain the first construction progress prediction result; The composite features also enter the random forest branch, and each decision tree independently analyzes and judges the composite features during processing, obtains its own construction progress prediction result through node splitting and decision-making, and then the prediction results of all decision trees are summarized and a second construction progress prediction result is obtained through voting; The first construction progress prediction result and the second construction progress prediction result are weighted and fused according to the weight determined through cross-validation to obtain the final prediction result; The method comprises the following steps: According to the potential importance of the features to the construction progress prediction, the composite features are divided into different subsets to form initial branches under the root node; For each subset after the root node division, the features that can distinguish the data samples are selected again from the remaining composite features to divide the next layer of nodes; When the features have been used for division, leaf nodes are formed, and the data samples in each leaf node correspond to a construction progress prediction result; The prediction results of all leaf nodes are summarized to obtain the first construction progress prediction result; The method comprises the following steps: The multi-dimensional raw data in the form of time series is input into the LSTM network, and the trend changes and periodic rules of the construction progress in long time series are captured through the gating mechanism to output a feature vector representing the progress trend and period; The construction tasks of the building project are represented in a graph structure, and the graph structure is learned using a graph neural network to mine the complex dependency relationships between tasks through node feature updating and message passing mechanisms, and output a feature matrix representing the degree of task association; The HOG algorithm is used to extract geometric features of the engineering components in the BIM model; The method comprises the following steps: Based on the geometric attributes and semantic information of the components, the three-dimensional geometric information of the components is converted into a parameterized representation; extracting texture features and material properties of the component surface, calculating gradient changes of the component surface in different directions, and forming a feature histogram; dividing the component into multiple local regions, respectively extracting features and fusing them to capture local geometric properties of the component; discretizing continuous gradient features and converting them into a digital feature vector to obtain geometric feature information of the component.

2. The machine learning-based construction progress warning method of claim 1, wherein The initial warning threshold is determined by analyzing historical construction data, and the warning threshold is dynamically adjusted according to real-time monitoring data and prediction results during the construction process, including: setting an initial construction progress state estimation value for the Kalman filter algorithm, including the mean and variance of the current construction progress; based on the construction progress state estimation value at the last time, the current construction progress prediction state is calculated using Kalman filtering; comparing the real-time collected construction progress data with the prediction result, and automatically adjusting the estimation of the construction progress state according to the difference Kalman filter algorithm to dynamically adjust the warning threshold.

3. The machine learning based construction project progress early warning method of claim 1, wherein, The initial warning threshold is determined by analyzing historical construction data, and the warning threshold is dynamically adjusted according to real-time monitoring data and prediction results during the construction process, and further comprising: taking the real-time state of the construction progress as the input of the deep Q network, based on the current state, the deep Q network selects an action from the action space to determine a threshold adjustment strategy; evaluate the pros and cons of each threshold adjustment action through the reward feedback mechanism, and the deep Q network continuously optimizes the threshold adjustment strategy according to the reward feedback obtained by each action.

4. A machine learning based construction project progress early warning system, characterized in that, The system comprises: a collection module for collecting multi-dimensional raw data from construction site sensors, BIM models, IoT devices and historical databases; an extraction module for extracting features related to construction progress from multi-dimensional raw data, combining and transforming the extracted features to obtain composite features: inputting time series form multi-dimensional raw data into LSTM network, capturing trend changes and periodicity of construction progress in long time series through gating mechanism, outputting feature vector representing progress trend and period; representing the construction tasks of building engineering in graph structure, learning the graph structure using graph neural network, mining the complex dependency relationship between tasks through node feature update and message passing mechanism, outputting feature matrix representing the degree of task association; using HOG algorithm to extract geometric features of engineering components in BIM model; based on the geometric attributes and semantic information of the components, the three-dimensional geometric information of the components is converted into parameterized representation; extracting texture features and material properties of the component surface, calculating gradient changes of the component surface in different directions, and forming a feature histogram; dividing the component into multiple local regions, respectively extracting features and fusing them to capture local geometric properties of the component; discretizing continuous gradient features and converting them into a digital feature vector to obtain geometric feature information of the component. The prediction module is configured to input the composite features into a progress prediction model, and predict the construction progress in a future period of time through the progress prediction model, and output a prediction result. The composite features are input into the progress prediction model, and the composite features enter an XGBoost branch. The XGBoost branch is configured to perform layer-by-layer division and evaluation on the composite features, and obtain a first construction progress prediction result. The composite features also enter a random forest branch. When the random forest branch is processing, each decision tree independently analyzes and judges the composite features, and obtains a respective construction progress prediction result through node splitting and decision making. The prediction results of all the decision trees are summarized, and a second construction progress prediction result is obtained by using a voting method. The first construction progress prediction result and the second construction progress prediction result are weighted and fused by using a weight determined through cross validation, and a final prediction result is obtained. According to the potential importance of the features to the construction progress prediction, the composite features are divided into different subsets, and initial branches under a root node are formed. For each subset after the root node is divided, features capable of distinguishing data samples are selected from the remaining composite features, and the next layer of nodes is divided. When the features have been used for division, leaf nodes are formed, and the data samples in each leaf node correspond to a construction progress prediction result. The prediction results of all the leaf nodes are summarized, and the first construction progress prediction result is obtained. The adjustment module is configured to determine an initial early warning threshold value by analyzing historical construction data, and dynamically adjust the early warning threshold value according to real-time monitoring data and the prediction result in the construction process. The early warning module is configured to trigger early warning when the construction progress deviates from the predicted progress and exceeds the dynamically adjusted early warning threshold value. 5.A machine learning-based construction progress warning device for construction engineering, characterized by The machine learning-based construction progress early warning device includes a memory and at least one processor. The memory stores instructions. The at least one processor invokes the instructions in the memory, so that the machine learning-based construction progress early warning device performs each step of the machine learning-based construction progress early warning method according to any one of claims 1-3.

6. A computer-readable storage medium having stored thereon instructions, the computer-readable storage medium comprising: The instructions are executed by the processor to implement each step of the machine learning-based construction progress early warning method according to any one of claims 1-3.

Images