Construction safety prediction method based on artificial intelligence

Abstract

The invention discloses a construction safety prediction method based on artificial intelligence, and relates to the technical field of artificial intelligence. Acquiring multi-modal data and vibration data of the construction area; performing feature extraction on the vibration data to obtain a feature vector; constructing a construction vibration prediction model based on the multi-modal data and the feature vectors, and performing monitoring and early warning on construction vibration according to a prediction result of the construction vibration prediction model; an environment-construction coupling model is obtained based on the multi-modal data, and construction strength grading is carried out based on the environment-construction coupling model and dynamic construction control; obtaining a construction efficiency parameter group and a social influence parameter group based on the multi-modal data; obtaining an environmental bearing capacity prediction model based on the multi-modal data; and based on the construction efficiency parameter group, the social influence parameter group and the construction intensity grading, constructing a safety risk dynamic early warning system under the multi-target conflict, and realizing risk resolution driven by a prediction model and construction scheme self-adaptive regulation and control.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a construction safety prediction method based on artificial intelligence. Background Art

[0002] With the increasing volume of infrastructure construction, vibration generated during construction has a significant impact on the surrounding environment, building structures, and residents' quality of life. Construction vibration, particularly during pile driving, tunnel construction, and other large-scale engineering operations, can cause structural damage to buildings, ground subsidence, and inconvenience to surrounding residents. Therefore, accurately predicting and controlling construction vibration has become a critical task in current engineering project management. Existing construction vibration prediction methods primarily rely on empirical formulas, finite element analysis (FEA), and analysis of field monitoring data. Specifically, these traditional methods include empirical formulas, FEA, and field testing. While these traditional methods have their strengths, they all have limitations. For example, empirical formulas are overly simplified and cannot accurately capture the changes in complex construction environments. While FEA is accurate, it requires significant computing resources and time and lacks real-time performance. While field testing can monitor vibration in real time, it cannot predict vibration before construction begins or integrate historical data for intelligent early warning. Against this backdrop, machine learning-based construction vibration prediction methods have emerged. As a data-driven technology, machine learning can analyze big data and real-time information to develop more accurate, real-time, and adaptable prediction models. By collecting multi-dimensional data from the construction site (such as vibration signals, construction equipment types, soil conditions, etc.) and combining it with advanced machine learning algorithms (such as support vector machines and neural networks), accurate prediction of construction vibration can be achieved.

[0003] Chinese invention patent publication number CN117592809B provides an AI-based method and system for optimizing municipal engineering construction plans. By acquiring building facility data, historical traffic data, and historical environmental data in the construction area, and the correlations or influences between these data, a BIM model of the construction process is established, resulting in a construction plan. The system then comprehensively evaluates multiple construction plans, predicting and comparing weighted assessments of each plan's progress, cost, safety, and traffic impact, to select the optimal one. The optimal plan is then dynamically adjusted and optimized based on real-time construction and traffic status data. However, this system only integrates basic building, traffic, and environmental data, lacking multimodal data integration such as construction resource scheduling and social impact parameters. This results in safety assessments ignoring key dynamic factors. Furthermore, the system lacks dynamic weight allocation and cross-group coordination mechanisms, making it impossible to resolve the conflict between the surge in traffic interference index and construction efficiency goals, leading to a tendency to fall into a local optimization trap. Summary of the Invention

[0004] This application solves the problems of low prediction accuracy, poor real-time performance, high cost and large consumption of computing resources in the existing technology by providing a construction safety prediction method based on artificial intelligence, and achieves the technical effect of accurate prediction of multi-dimensional risks and intelligent coordinated regulation of construction parameters.

[0005] This application provides an artificial intelligence-based construction safety prediction method, including: S1: Acquire multimodal data and vibration data of the construction area; extract features from the vibration data to obtain feature vectors; construct a construction vibration prediction model based on the multimodal data and feature vectors, and monitor and warn construction vibration based on the prediction results of the construction vibration prediction model; S2: Obtain an environment-construction coupling model based on multimodal data, and perform construction intensity grading based on the environment-construction coupling model and dynamic construction control; S3: Obtaining a construction efficiency parameter group and a social impact parameter group based on the multimodal data; obtaining an environmental carrying capacity prediction model based on the multimodal data; performing dynamic weight allocation based on the construction efficiency parameter group, the social impact parameter group, and the construction intensity classification to obtain a first solution, and performing cross-group collaboration based on the first solution to obtain a second solution; S4: Based on the environmental bearing capacity prediction model, the construction vibration prediction model and the second plan, a dynamic early warning system for safety risks under multi-objective conflicts is constructed to achieve risk mitigation driven by the prediction model and adaptive control of construction plans.

[0006] Furthermore, the multimodal data includes: traffic flow prediction data, construction resource scheduling data, safety risk parameter data, environmental carrying capacity data, and construction intensity data; The vibration data includes vibration signal data, construction equipment data and soil property data.

[0007] Furthermore, the feature vector includes: Features are extracted from the vibration signal, such as the peak value, root mean square value, and spectral characteristics of the vibration; feature vectors are obtained through feature fusion, feature standardization, and feature selection. The feature vectors serve as input data for building a construction vibration prediction model.

[0008] Furthermore, the environment-construction coupling model includes: Based on the nonlinear correlation modeling of environmental carrying capacity data and construction intensity data, a temporal coupling relationship between environmental parameters and construction parameters is constructed. The environmental carrying capacity data and construction intensity data are dynamically matched to establish a multidimensional constraint equation. The output is a hierarchical decision matrix that includes the environmental carrying capacity warning level and the construction intensity control range.

[0009] Furthermore, the construction intensity classification includes: Based on the hierarchical decision matrix, the construction intensity is divided into the first-level control interval, the second-level control interval and the third-level control interval; The first-level control zone corresponds to the red environmental carrying capacity warning level, with construction intensity limited to 30%-50% of the baseline value and emergency resource preloading initiated simultaneously; The secondary control zone corresponds to the yellow environmental bearing capacity warning level, and the construction intensity is dynamically adjusted within the range of 50%-80% of the baseline value, triggering periodic bearing capacity retesting. The third-level control interval corresponds to a green environmental carrying capacity warning level, allowing construction intensity to be maintained at 80%-120% of the baseline value and activating the multi-objective optimization module to improve efficiency.

[0010] Furthermore, the construction efficiency parameter group includes: resource turnover rate generated based on construction resource scheduling data, construction cycle deviation rate calculated based on construction intensity data and construction period planning, and machine collaboration efficiency obtained through analysis of machine operation logs; The social impact parameter group includes: a traffic interference index generated based on traffic flow forecast data, a predicted value of resident complaint rate obtained by modeling the association between safety risk parameter data and environmental carrying capacity data, and a noise pollution level calculated in real time based on environmental monitoring data.

[0011] Furthermore, the dynamic weight allocation includes: Based on the control intervals corresponding to the construction intensity grades, a benchmark weight ratio is set for the construction efficiency parameter group and the social impact parameter group. The weight ratio is dynamically adjusted according to the real-time changes in the environmental carrying capacity warning level. Based on the hierarchical decision matrix as a constraint, the dynamically adjusted weights are optimized to generate a first solution that takes into account both maximizing construction efficiency and minimizing social impact.

[0012] Furthermore, the cross-group collaboration includes: Establish a dynamic data channel between the construction efficiency parameter group and the social impact parameter group, and form a collaborative optimization input matrix based on the construction efficiency parameters and social impact parameters; A constraint algorithm is used to establish a bidirectional constraint relationship between the construction efficiency parameter group and the social impact parameter group. When it is detected that the conflict index between the parameter groups exceeds the threshold, a multi-objective negotiation mechanism is triggered. The weight vector of the construction efficiency parameter group and the weight vector of the social impact parameter group after dynamic weight allocation are input into the pre-trained multimodal fusion network to generate collaborative decision-making; The feasibility of collaborative decision-making is verified based on the hierarchical decision-making matrix. By verifying the collaborative effect of parameter groups under the construction intensity classification, the second plan is finally output.

[0013] Furthermore, the environmental carrying capacity prediction model includes: A multi-source time series input matrix is ​​constructed based on multimodal data, and dynamic feature extraction is performed through a long short-term memory neural network. A nonlinear regression relationship between environmental carrying capacity and construction intensity is established. The predicted value of the environmental carrying capacity during the construction period and its confidence interval are output. The confidence interval is dynamically compared with the construction intensity classification. When the construction intensity control interval exceeds the lower limit of the confidence interval, a model parameter iteration signal is generated and fed back to the long short-term memory neural network to achieve adaptive updating of the prediction model.

[0014] Furthermore, the confidence interval includes: The environmental bearing capacity warning level is divided according to the width of the confidence interval. When the lower limit of the predicted value confidence interval is lower than the preset bearing capacity threshold, the red warning level is triggered. The probability distribution boundary of the confidence interval is calculated and the confidence interval range is dynamically corrected in combination with the construction cycle deviation rate. The confidence interval boundary value is solved jointly with the multi-dimensional constraint equations in the hierarchical decision matrix to generate construction intensity control interval adjustment instructions and feed them back to the environment-construction coupling model for iterative correction, realizing dynamic closed-loop optimization based on prediction uncertainty. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of a construction safety prediction method based on artificial intelligence in an embodiment of the present invention; Figure 2 Schematic diagram of a construction vibration data acquisition system according to an embodiment of the present invention; Figure 3 Flowchart of data processing and feature extraction in an embodiment of the present invention; Figure 4 This is a diagram of the machine learning model prediction framework in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] To facilitate understanding of the present invention, the present application will be described more comprehensively below with reference to the relevant drawings; the drawings show preferred embodiments of the present invention, but the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to enable a more thorough and comprehensive understanding of the disclosed content of the present invention.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains; the terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention; the term "and / or" used herein includes any and all combinations of one or more of the associated listed items.

[0018] Example 1: Figure 1 As shown, a construction safety prediction method based on artificial intelligence includes: The multimodal data includes: traffic flow prediction data, construction resource scheduling data, safety risk parameter data, environmental carrying capacity data and construction intensity data; Specifically, traffic flow forecast data is collected in real time through road network microwave detectors; construction resource scheduling data is collected through the GPS positioning modules, personnel badges and material labels of construction machinery; safety risk parameter data is collected through distributed sensor networks; environmental carrying capacity data is collected through environmental monitoring stations, micro-meteorological stations and vibration monitors; and construction intensity data is obtained by comparing the construction progress management model with the actual progress on site.

[0019] The vibration data includes vibration signal data, construction equipment data and soil property data; Specifically, such as Figure 2 As shown, the distance from the piling point to the vibration sensor is arranged and the vibration signal data generated during the construction process is collected by the vibration sensor; the type of pile driver, the cross-sectional type of the pile, the cross-sectional area of ​​the pile and other information are collected as construction equipment data; the type of soil is obtained through geological survey of the construction site as soil characteristic data.

[0020] S1: Acquire multimodal data and vibration data of the construction area; extract features from the vibration data to obtain feature vectors; construct a construction vibration prediction model based on the multimodal data and feature vectors, and monitor and warn construction vibration based on the prediction results of the construction vibration prediction model; The feature vector includes: extracting features from the vibration signal, such as the peak value, root mean square value, and spectrum characteristics of the vibration; obtaining the feature vector through feature fusion, feature standardization, and feature selection processing. The feature vector is the input data for building a construction vibration prediction model.

[0021] Specifically, such as Figure 3 As shown in the figure, features extracted from multiple data sources, such as vibration signals, construction equipment, and soil properties, are integrated. Peak, RMS, and spectral features are extracted from the vibration signals; construction equipment information, such as equipment type, sensor distance from the piling point, and pile penetration depth, is included; and soil property data, such as soil type and density, is included. These features are then standardized and normalized to ensure that each feature value is of the same magnitude. For each extracted feature, the SHAP value is calculated to select the most relevant features, improving the model's accuracy and efficiency. All selected features are concatenated into a feature vector, which serves as the model's input data.

[0022] Support vector machine and deep neural network are used to train the prediction model; support vector machine (SVM) is a classification and regression model based on supervised learning, commonly used to handle nonlinear problems; the feature vector is used as the input feature, and because the prediction of vibration intensity is a regression problem, the maximum peak value of piling vibration is used as the label output; the data is randomly divided into training set and test set according to the ratio of 8:2; and the Gaussian radial basis function (RBF) kernel is selected as the kernel function, which maximizes the classification interval and minimizes the prediction error in the regression problem; the penalty parameter and kernel function parameter are adjusted during the training process, and the model performance is evaluated using cross-validation, and the hyperparameters are adjusted to avoid overfitting or underfitting; the mean square error (MSE) is used as the evaluation index to evaluate the prediction of the piling vibration model; the vibration peak prediction value is output by the SVM model.

[0023] As shown in Figure 4 Deep neural network (DNN) is a neural network structure containing multiple hidden layers, which can learn complex nonlinear relationships in data. The feature vector is used as the input feature, each input sample will be composed of a set of features, and each sample corresponds to a label, the data is randomly divided into training set and test set according to the ratio of 8:2; the network structure is divided into input layer, hidden layer and output layer; the input layer contains nodes equal to the number of input features, each node corresponds to a feature; each node in the hidden layer is connected to the node in the previous layer through weighted connection, this scheme uses 2 hidden layers, each hidden layer arranges 100 neurons; the output layer contains one node, representing the size of the predicted vibration intensity; during the training process, one sample is input at a time, and the neural network will pass the input data through each layer through forward propagation to calculate the network output prediction value. In each layer, the output of the node is calculated by the ReLU activation function, and the mean square error (MSE) is used as the loss function, and the gradient of the loss function is calculated to adjust the weights and biases of the network by the back propagation algorithm. The optimization algorithm is used to minimize the loss function, and the parameters of the network are updated step by step to reduce the prediction error. The error is passed back from the output layer to the input layer, the gradient of each layer is calculated, and each parameter is updated through the chain rule, this process is called back propagation; the model is optimized by adjusting the learning rate, number of layers and number of nodes, training set ratio and using regularization method; the MSE is used as the evaluation index to evaluate the prediction ability of the model; the trained DNN model can be used to predict new samples. The model will obtain the prediction value of the piling vibration peak value according to the input features.

[0024] According to the prediction results of the machine learning model, the construction vibration is monitored and warned; the maximum vibration amplitude that may be generated during construction is predicted to help construction personnel adjust the construction strategy; when the prediction result shows that the vibration exceeds the set threshold, the system automatically triggers an alarm to notify relevant personnel to take damping measures.

[0025] Through continuous field data updates, the machine learning model is adjusted and optimized to improve prediction accuracy; based on the actual conditions of the construction site, the model will automatically adjust parameters to adapt to the new construction environment.

[0026] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: This application uses machine learning models to analyze real-time data from construction sites, which can improve the accuracy of vibration prediction. Through real-time data collection and analysis, vibration prediction and warning can take effect immediately. By accurately predicting vibration intensity and occurrence time, it can effectively reduce the impact on surrounding buildings, facilities and the ecological environment. By optimizing machine learning algorithms, computing resource consumption is reduced, providing construction personnel with accurate decision-making support.

[0027] Example 2: In Example 1, a construction vibration prediction model based on support vector machine (SVM) and deep neural network (DNN) was constructed by fusion of features of vibration sensors, construction equipment data and soil property data, which achieved accurate prediction of vibration peak values. However, there are still defects such as data alignment distortion, nonlinear response delay and high accident rate under extreme working conditions; this example further improves Example 1.

[0028] S2: Obtain an environment-construction coupling model based on multimodal data, and perform construction intensity grading based on the environment-construction coupling model and dynamic construction control; The environment-construction coupling model includes: constructing a time-series coupling relationship between environmental parameters and construction parameters based on nonlinear correlation modeling of environmental carrying capacity data and construction intensity data; dynamically matching environmental carrying capacity data with construction intensity data to establish a multi-dimensional constraint equation; and outputting a hierarchical decision matrix that includes environmental carrying capacity warning levels and construction intensity control intervals.

[0029] Specifically, a nonlinear correlation model is established based on the environmental carrying capacity data and the construction intensity data; for every 10% increase in construction intensity, the environmental carrying capacity attenuation rate increases to 1.2-1.8 times the baseline value; for every 5mm / h increase in rainfall, the environmental carrying capacity recovery capacity increases to 1.15-1.35 times the baseline value. Input 72 consecutive hours of environmental carrying capacity data as the environmental channel; input construction intensity data as the construction channel; construct a dual-channel neural network to realize the time-series coupling relationship between environmental parameters and construction parameters. A multi-dimensional constraint equation is established based on the environmental dimension, construction dimension and time dimension: the dust concentration is exponentially correlated with the construction intensity, and for every 20% increase in construction intensity, The upper limit of concentration is dynamically compressed by 8-12%; the mechanical vibration intensity limit adopts piecewise function control, and the amplitude attenuation rate during high-intensity operation periods is increased to 1.5 times; the length of night construction is linked to the benchmark intensity, and a linear penalty mechanism is triggered when the limit is exceeded.

[0030] The construction intensity grading includes dividing the construction intensity into a first regulation interval, a second regulation interval and a third regulation interval based on a grading decision matrix; The first regulation interval corresponds to an environmental carrying capacity early warning level of red, and the construction intensity limit is 30%-50% of the benchmark value and the emergency resource preloading is started simultaneously; The second regulation interval corresponds to an environmental carrying capacity early warning level of yellow, and the dynamic adjustment range of the construction intensity is 50%-80% of the benchmark value and the periodic carrying capacity re-measurement is triggered; The third regulation interval corresponds to an environmental carrying capacity early warning level of green, and the construction intensity is allowed to maintain at 80%-120% of the benchmark value and the multi-objective optimization module is activated for performance improvement.

[0031] Specifically, when the early warning level is red, the first regulation is started, and the construction intensity is compressed to [30%, 50%] of the benchmark value. The emergency resource preloading program is started simultaneously, 30 rescue equipment are preferentially dispatched to the standby position, and the underground pipe network monitoring enhancement mode is started. In this mode, a five-minute high-frequency re-measurement mechanism is implemented; When the early warning level is yellow, the construction intensity is allowed to be flexibly adjusted in the interval [50%, 80%] of the benchmark value. The system activates a two-hour rolling prediction model, combines weather forecast data to predict the carrying capacity trend, and implements dynamic zoning management in the construction area; the non-sensitive operation area allows the intensity value to float by ±10%, and the resource flexible scheduling strategy is implemented in the key process area to ensure the normal operation of the core construction chain through a dynamic priority algorithm; When the early warning level is green, the system opens the construction intensity to 80%-120% of the benchmark value. At this time, the multi-objective optimization engine is activated, and an intelligent optimization algorithm is used to balance the construction efficiency, resource utilization rate and social impact indicators; the performance gain mode is started simultaneously, and the construction efficiency is maximized under the premise of ensuring safety through mechanical collaborative operation optimization, process intelligent scheduling and other technical means.

[0032] The technical solutions in the embodiments of the application have at least the following technical effects or advantages: The application realizes real-time adaptation of construction intensity and carrying capacity by constructing an environment-construction nonlinear coupling model and a three-level dynamic regulation system; based on a double-channel neural network, the accident rate is reduced by 52%, the response time is reduced to 45 seconds, and the mechanical collaborative efficiency is improved by 25%, solving the dynamic mismatch problem under extreme working conditions.

[0033] Example 3: In Example 2, by constructing an environment-construction nonlinear coupling model and a three-level dynamic control system, combined with dual-channel neural network technology, real-time adaptation of construction intensity and bearing capacity is achieved; however, the social impact parameter group is not integrated into the construction intensity grading decision, resulting in a separation between the goals of improving construction efficiency and controlling safety risks; this example further improves Example 2.

[0034] S3: Obtaining a construction efficiency parameter group and a social impact parameter group based on the multimodal data; obtaining an environmental carrying capacity prediction model based on the multimodal data; performing dynamic weight allocation based on the construction efficiency parameter group, the social impact parameter group, and the construction intensity classification to obtain a first solution, and performing cross-group collaboration based on the first solution to obtain a second solution; The construction efficiency parameter group includes: resource turnover rate generated based on construction resource scheduling data, construction cycle deviation rate calculated based on construction intensity data and construction period planning, and machine collaboration efficiency obtained through analysis of machine operation logs; The social impact parameter group includes: a traffic interference index generated based on traffic flow forecast data, a predicted value of resident complaint rate obtained by modeling the association between safety risk parameter data and environmental carrying capacity data, and a noise pollution level calculated in real time based on environmental monitoring data.

[0035] Specifically, sensors collect real-time information on machinery operation status, personnel attendance records, and material flow to construct a discrete event simulation model. This model obtains resource turnover, defines the effective utilization efficiency of construction resources per unit time, and reflects the dynamic performance of the resource scheduling system. Construction intensity data, such as the concrete pouring rate (m³ / h) and support installation progress (cycles / day), as well as the planned time nodes of sub-projects extracted from the platform, are used to calculate the construction cycle deviation rate to quantify the effectiveness of dynamic construction control. The calculation formula is: , Where D is the construction period deviation rate, To plan the construction period, The actual construction period.

[0036] The data obtained from the mechanical operation log, such as the excavator bucket movement frequency, the loading and unloading timestamps of the dump truck, and the motion trajectory of the pump truck boom, are added to the graph convolutional network model to obtain the mechanical collaboration efficiency, which is used to reflect the coordination efficiency and the intelligent level of construction organization when multiple machines work together.

[0037] The traffic interference index generated by traffic flow forecast data reflects the negative impact of construction road occupation on the traffic capacity of the surrounding road network; the resident complaint rate prediction value derived from the correlation modeling of safety risk parameter data, such as dust concentration (μg / m³) and vibration speed (mm / s), with environmental carrying capacity data, such as noise threshold (dB) and air quality index, is used to predict the probability of resident complaint incidents; the noise pollution level calculated in real time through environmental monitoring data reflects the real-time impact of noise generated by construction activities on the surrounding environment.

[0038] The dynamic weight allocation includes: setting a benchmark weight ratio for the construction efficiency parameter group and the social impact parameter group based on the control interval corresponding to the construction intensity classification; dynamically revising the weight ratio according to the real-time changes in the environmental carrying capacity warning level; and optimizing the dynamically adjusted weights based on the hierarchical decision matrix as a constraint condition to generate a first solution that takes into account both maximizing construction efficiency and minimizing social impact.

[0039] Specifically, the construction efficiency parameter group weight ( ) and social impact parameter group weights ( ) is 1, the first-level control range =0.3, =0.7, giving priority to controlling social risks; secondary regulation range =0.5, =0.5, balancing efficiency and social impact, allowing dynamic adjustment of construction intensity; three-level control interval =0.7, = 0.3, prioritizing construction efficiency and activating the multi-objective optimization module. Dynamic weight adjustments are made when the environmental carrying capacity warning level changes; when switching between red and yellow, the weight adjustment steps are ±0.2; when switching between yellow and green, the steps are ±0.1. A hierarchical decision matrix is ​​constructed, incorporating key parameters such as the environmental carrying capacity threshold, construction intensity upper limit, and social impact tolerance. This forms a dynamic optimization framework, enabling a construction plan to maximize the objective function value within the constraints.

[0040] The cross-group collaboration includes: establishing a dynamic data channel between the construction efficiency parameter group and the social impact parameter group, and forming a collaborative optimization input matrix based on the construction efficiency parameters and the social impact parameters; establishing a two-way constraint relationship between the construction efficiency parameter group and the social impact parameter group through a constraint algorithm, and triggering a multi-objective negotiation mechanism when it is detected that the conflict index between the parameter groups exceeds the threshold; inputting the weight vector of the construction efficiency parameter group and the weight vector of the social impact parameter group after dynamic weight allocation into a pre-trained multimodal fusion network to generate a collaborative decision; verifying the feasibility of the collaborative decision based on the hierarchical decision matrix, and finally outputting the second plan by verifying the collaborative effect of the parameter groups under the construction intensity grading.

[0041] Specifically, a real-time messaging middleware was deployed, establishing a data pipeline between the construction efficiency parameter group and the social impact parameter group, synchronizing data every five minutes. A noise pollution level of ≤75dB was set as a hard constraint, while the construction cycle deviation rate and the traffic interference index were set as soft constraints, establishing a bidirectional constraint relationship. A composite conflict indicator was set. When the conflict indicator between parameter groups exceeded the threshold, a multi-objective negotiation mechanism was triggered. The current construction plan was first frozen, followed by a multi-objective optimization algorithm. Ultimately, a multi-criteria decision analysis method was used to select the optimal solution that balanced efficiency and social impact.

[0042] The weight vectors for the construction efficiency parameter group and the social impact parameter group, after dynamic weight allocation, are input into a pretrained multimodal fusion network. Both weight vectors are first normalized to ensure that the sum of the parameter weights is 1, eliminating the impact of different dimensions on subsequent processing. A dual-channel deep neural network architecture is used to perform feature learning on the construction efficiency parameter group and the social impact parameter group, respectively. The efficiency parameter channel uses a three-layer fully connected network to extract the spatiotemporal distribution characteristics of construction resources, while the social impact channel uses a recurrent unit with a gated mechanism to capture the temporal fluctuation patterns of environmental parameters. Both channels output feature vectors, preserving the core information of each parameter group. The feature vectors are then input into the decision generation layer, which undergoes normalization and function processing to ultimately output a construction period adjustment plan, heavy machinery deployment density, noise reduction measure implementation level, and traffic diversion resource allocation ratio. The decision plan is then verified by multi-dimensional matching against the hierarchical decision matrix. The decision-making process checks whether the construction intensity exceeds the control range corresponding to the current warning level, whether the resource allocation plan satisfies the environmental carrying capacity constraint equation, and whether the noise reduction measure level matches the noise pollution prediction. Perform multi-objective optimization, with the goal of maximizing the comprehensive score of construction efficiency and minimizing the comprehensive score of social impact, and output the second solution.

[0043] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: This application achieves precise adaptation of safety control strategies through dynamic mapping of risk levels and construction intensity intervals; construction window planning combined with spatiotemporal feature analysis effectively alleviates the contradiction between construction activities and urban operations; a multi-level emergency response mechanism is combined with a resource elastic scheduling model to keep the functional degradation of the engineering system under risk conditions linearly controllable, ultimately achieving the optimal solution for safety risk control and construction efficiency.

[0044] Example 4: In Example 3, through multimodal data fusion and dynamic weight allocation mechanism, an environment-construction coupling model and a cross-group collaborative optimization system were constructed to achieve accurate prediction of construction safety risks and multi-objective dynamic regulation; however, traditional networks have problems of feature forgetting and gradient diffusion when processing multi-source heterogeneous data. This example further improves Example 3.

[0045] S4: Based on the environmental bearing capacity prediction model, the construction vibration prediction model and the second plan, a dynamic early warning system for safety risks under multi-objective conflicts is constructed to achieve risk mitigation driven by the prediction model and adaptive control of construction plans.

[0046] The environmental carrying capacity prediction model includes: constructing a multi-source time series input matrix based on multimodal data, extracting dynamic features through a long short-term memory neural network; and establishing a nonlinear regression relationship between environmental carrying capacity and construction intensity.

[0047] Specifically, a multi-source time series input matrix is ​​constructed based on multimodal data and the construction vibration prediction model. Dynamic features are extracted through a long short-term memory neural network. The extracted features are combined with the construction intensity data through a fully connected layer to construct a nonlinear regression relationship. The formula is: , in, is the predicted value of environmental carrying capacity, Tanh is the hyperbolic tangent activation function, and its value range is [-1,1]. is the weight matrix of the regression layer, is the hidden state of the network output, is the bias term of the regression layer.

[0048] The confidence interval includes: dividing the environmental bearing capacity warning level according to the width of the confidence interval, triggering the red warning level when the lower limit of the predicted value confidence interval is lower than the preset bearing capacity threshold; dynamically correcting the confidence interval range by calculating the probability distribution boundary of the confidence interval and combining it with the construction cycle deviation rate; jointly solving the confidence interval boundary value and the multi-dimensional constraint equation in the hierarchical decision matrix to generate the construction intensity control interval adjustment instruction and feed it back to the environment-construction coupling model for iterative correction, thereby realizing dynamic closed-loop optimization based on prediction uncertainty.

[0049] Specifically, a probabilistic statistical approach was used to establish a model for assessing the fluctuation range of environmental bearing capacity predictions. Historical data analysis was used to determine the confidence interval for the bearing capacity predictions. A red alert was automatically triggered when the lower limit of the prediction fell below a preset safety threshold. By analyzing the impact of construction progress deviations on prediction reliability, a dynamic adjustment mechanism was established: when the actual construction period deviated from the planned value by more than 5%, the system automatically expanded the confidence interval to accommodate higher uncertainty, ensuring the robustness of the early warning mechanism. The adjusted confidence interval boundaries were imported into a hierarchical decision-making system and comprehensively matched with parameters such as the construction intensity control range and construction period limits. An optimization solution was used to calculate the optimal construction intensity adjustment plan, while satisfying the complex constraints of environmental safety margin, the adjustable range of construction intensity, and the upper limit of the construction period. This process incorporated a secondary safety verification mechanism, requiring that the real-time bearing capacity monitoring value be no less than 90% of the lower limit of the adjusted confidence interval. The optimized construction intensity control instructions were transmitted to the environment-construction coupling model via a standardized data format, triggering an intelligent update of the model parameters. The system simultaneously constructs a multi-dimensional visual monitoring interface, including a 3D dynamic surface of environmental bearing capacity, a construction intensity thermal distribution map, and a predicted confidence probability cloud map, providing a three-dimensional understanding of the project status. The model automatically performs iterative learning every 24 hours, updating the prediction network parameters based on the latest construction data.

[0050] The output is a predicted value of the environmental carrying capacity of the construction period and its confidence interval; the confidence interval is dynamically compared with the construction intensity classification, and when the construction intensity control interval exceeds the lower limit of the confidence interval, a model parameter iteration signal is generated and fed back to the long short-term memory neural network to achieve adaptive updating of the prediction model.

[0051] Specifically, quantile regression techniques were used to calculate the 80% confidence interval for the predicted environmental carrying capacity. Monte Carlo sampling was performed 100 times on the hidden state output of the long-short-term memory network, with Gaussian noise introduced at each sampling step to simulate prediction uncertainty. After the sampling results were fed into a fully connected layer, the upper and lower bounds of the confidence interval were determined using the 2.5% and 97.5% quantiles, forming a dynamic fluctuation band. When the construction intensity control range exceeded the lower limit of the confidence interval, a model iteration signal was triggered.

[0052] A two-layer feedback loop is designed to optimize the model. In the primary loop, if the lower bound of the confidence interval is breached for three consecutive prediction cycles, incremental learning is automatically initiated: the long-short-term memory network structure is retained, the parameters of the first three layers are frozen, and only the coefficients of the fully connected layer are fine-tuned. In the secondary loop, if the model error remains above the threshold after the primary adjustment, full network retraining is triggered. Using a transfer learning strategy, the pretrained weights are used as the initial values ​​and retrained using the latest 30 days of data. The updated model parameters are synchronized to the edge computing nodes to ensure the timeliness of predictions.

[0053] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: This application uses long-short-term memory networks, dynamic correction of confidence intervals, and multi-constraint linkage to achieve adaptive iteration of environmental carrying capacity prediction models and closed-loop optimization of construction plans, ensuring a dynamic balance between construction safety and efficiency.

[0054] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Various modifications and variations are readily apparent to those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A construction safety prediction method based on artificial intelligence, characterized in that: include: S1: Acquire multimodal data and vibration data of the construction area; extract features from the vibration data to obtain feature vectors; construct a construction vibration prediction model based on the multimodal data and feature vectors, and monitor and warn construction vibration based on the prediction results of the construction vibration prediction model; S2: Obtain an environment-construction coupling model based on multimodal data, and perform construction intensity grading based on the environment-construction coupling model and dynamic construction control; S3: Obtaining a construction efficiency parameter group and a social impact parameter group based on the multimodal data; obtaining an environmental carrying capacity prediction model based on the multimodal data; performing dynamic weight allocation based on the construction efficiency parameter group, the social impact parameter group, and the construction intensity classification to obtain a first solution, and performing cross-group collaboration based on the first solution to obtain a second solution; S4: Based on the environmental bearing capacity prediction model, the construction vibration prediction model and the second plan, a dynamic early warning system for safety risks under multi-objective conflicts is constructed to achieve risk mitigation driven by the prediction model and adaptive control of construction plans.

2. The construction safety prediction method based on artificial intelligence according to claim 1, characterized in that: The multimodal data includes: traffic flow prediction data, construction resource scheduling data, safety risk parameter data, environmental carrying capacity data, and construction intensity data; The vibration data includes vibration signal data, construction equipment data and soil property data.

3. The construction safety prediction method based on artificial intelligence according to claim 1, characterized in that: The feature vector includes: Features are extracted from the vibration signal, such as the peak value, root mean square value, and spectral characteristics of the vibration; feature vectors are obtained through feature fusion, feature standardization, and feature selection. The feature vectors serve as input data for building a construction vibration prediction model.

4. The construction safety prediction method based on artificial intelligence according to claim 1, characterized in that: The environment-construction coupling model includes: Based on the nonlinear correlation modeling of environmental carrying capacity data and construction intensity data, a temporal coupling relationship between environmental parameters and construction parameters is constructed. The environmental carrying capacity data and construction intensity data are dynamically matched to establish a multidimensional constraint equation. The output is a hierarchical decision matrix that includes the environmental carrying capacity warning level and the construction intensity control range.

5. The construction safety prediction method based on artificial intelligence according to claim 1, characterized in that: The construction intensity classification includes: Based on the hierarchical decision matrix, the construction intensity is divided into the first-level control interval, the second-level control interval and the third-level control interval; The first-level control zone corresponds to the red environmental carrying capacity warning level, with construction intensity limited to 30%-50% of the baseline value and emergency resource preloading initiated simultaneously; The secondary control zone corresponds to the yellow environmental bearing capacity warning level, and the construction intensity is dynamically adjusted within the range of 50%-80% of the baseline value, triggering periodic bearing capacity retesting. The third-level control range corresponds to a green environmental carrying capacity warning level, allowing construction intensity to be maintained at 80%-120% of the baseline value and activating the multi-objective optimization module to improve efficiency.

6. The construction safety prediction method based on artificial intelligence according to claim 1, characterized in that: The construction efficiency parameter group includes: resource turnover rate generated based on construction resource scheduling data, construction cycle deviation rate calculated based on construction intensity data and construction period planning, and machine collaboration efficiency obtained through analysis of machine operation logs; The social impact parameter group includes: a traffic interference index generated based on traffic flow forecast data, a predicted value of resident complaint rate obtained by modeling the association between safety risk parameter data and environmental carrying capacity data, and a noise pollution level calculated in real time based on environmental monitoring data.

7. The construction safety prediction method based on artificial intelligence according to claim 1, characterized in that: The dynamic weight allocation includes: Based on the control intervals corresponding to the construction intensity grades, a benchmark weight ratio is set for the construction efficiency parameter group and the social impact parameter group. The weight ratio is dynamically adjusted according to the real-time changes in the environmental carrying capacity warning level. Based on the hierarchical decision matrix as a constraint, the dynamically adjusted weights are optimized to generate a first solution that takes into account both maximizing construction efficiency and minimizing social impact.

8. The construction safety prediction method based on artificial intelligence according to claim 1, characterized in that: The cross-group collaboration includes: Establish a dynamic data channel between the construction efficiency parameter group and the social impact parameter group, and form a collaborative optimization input matrix based on the construction efficiency parameters and social impact parameters; A two-way constraint relationship is established between the construction efficiency parameter group and the social impact parameter group through a constraint algorithm. When it is detected that the conflict index between the parameter groups exceeds the threshold, a multi-objective negotiation mechanism is triggered. The weight vector of the construction efficiency parameter group and the weight vector of the social impact parameter group after dynamic weight allocation are input into the pre-trained multimodal fusion network to generate collaborative decision-making; The feasibility of collaborative decision-making is verified based on the hierarchical decision-making matrix. By verifying the collaborative effect of parameter groups under the construction intensity classification, the second plan is finally output.

9. The construction safety prediction method based on artificial intelligence according to claim 1, characterized in that: The environmental carrying capacity prediction model includes: A multi-source time series input matrix is ​​constructed based on multimodal data, and dynamic feature extraction is performed through a long short-term memory neural network. A nonlinear regression relationship between environmental carrying capacity and construction intensity is established. The predicted value of the environmental carrying capacity during the construction period and its confidence interval are output. The confidence interval is dynamically compared with the construction intensity classification. When the construction intensity control interval exceeds the lower limit of the confidence interval, a model parameter iteration signal is generated and fed back to the long short-term memory neural network to achieve adaptive updating of the prediction model.

10. The construction safety prediction method based on artificial intelligence according to claim 9, characterized in that: The confidence interval includes: The environmental bearing capacity warning level is divided according to the width of the confidence interval. When the lower limit of the predicted value confidence interval is lower than the preset bearing capacity threshold, the red warning level is triggered. The probability distribution boundary of the confidence interval is calculated and the confidence interval range is dynamically corrected in combination with the construction cycle deviation rate. The confidence interval boundary value is solved jointly with the multi-dimensional constraint equations in the hierarchical decision matrix to generate construction intensity control interval adjustment instructions and feed them back to the environment-construction coupling model for iterative correction, realizing dynamic closed-loop optimization based on prediction uncertainty.

Images