Intelligent control method for complex terrain infrastructure line construction site environment risk

Abstract

The invention discloses an intelligent control method for complex terrain infrastructure line construction site environment risks, and belongs to the field of signal control and processing, and the method comprises the steps: obtaining multi-source environment data through a self-adaptive sensing network; when the data is abnormal or a blind area exists, triggering the unmanned aerial vehicle to obtain high-precision spatial data; fusing the multi-source data and combining the multi-source data with geological information to generate digital risk feature information; inputting the risk feature information into a dynamic risk assessment model to obtain a risk level signal; comparing the risk level signal with a risk threshold in real time, generating a dynamic avoidance control instruction based on a comparison result, and issuing the dynamic avoidance control instruction to the construction equipment in real time; and dynamically adjusting model control parameters by taking a control response result of the construction equipment as feedback data. According to the invention, through a closed loop feedback mechanism, intelligent control of environmental risks is realized, and construction safety and management efficiency are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of signal control and processing, and in particular to an intelligent control method for environmental risks at construction sites of infrastructure lines in complex terrain. Background Art

[0002] During the construction of infrastructure projects in complex terrain, real-time monitoring and intelligent control of construction site environmental risks are core challenges in ensuring project safety. Traditional methods rely primarily on manual inspections and periodic monitoring by fixed sensors, which have fundamental flaws in building effective control systems. The monitoring system is completely disconnected from the control of construction equipment, resulting in the inability to convert risk signals into equipment control instructions in real time; the static threshold alarm mechanism lacks the ability to dynamically evaluate the execution effect of control instructions, making it impossible to optimize control parameters based on the actual evolution of risks; more seriously, existing technologies are unable to generate adaptive control strategies based on the real-time status of the equipment, and control instructions are severely decoupled from equipment capabilities. These systemic flaws lead to frequent problems such as high miscontrol rates and delayed responses, and are unable to meet the real-time control needs of construction in complex terrain.

[0003] While the development of multi-source sensing technology has enabled modern monitoring systems to acquire richer environmental data, they still haven't resolved the core issue of a broken control loop. A single data source makes it difficult to support accurate control decisions, resulting in a significant mismatch between control instructions and risk profiles. The lack of a complete "monitoring-decision-control-feedback" closed-loop mechanism prevents adaptive optimization of control parameters. The separation of control instruction generation from equipment operating conditions makes it difficult to formulate dynamic avoidance strategies. These technical bottlenecks prevent existing systems from guaranteeing the timeliness of control instructions or the accuracy of risk control when faced with complex and dynamic terrain changes. Summary of the Invention

[0004] To address these issues, the present invention provides an intelligent control method for environmental risks at construction sites of infrastructure projects located in complex terrain. This method utilizes multidimensional information fusion and a dynamic decision-making mechanism to achieve automated and precise management of construction risks by comparing and selecting data from diverse sources. This method improves the timeliness and reliability of early warnings, enhances intelligent control of construction risks and emergency response capabilities, and provides technical support for the safe construction of infrastructure projects located in complex terrain.

[0005] The above objectives can be achieved through the following solutions: An intelligent control method for environmental risks at construction sites of infrastructure lines in complex terrain comprises: acquiring multi-source environmental data through an adaptive sensing network arranged at the construction site, wherein the multi-source environmental data includes construction activity parameters and environmental parameters; automatically triggering a drone to perform multimodal scanning to acquire high-precision spatial data when the multi-source environmental data exceeds an abnormality threshold dynamically set based on historical data or a monitoring blind spot occurs; fusing the multi-source environmental data with the high-precision spatial data, and extracting features related to potential environmental risks in combination with geological information to generate digital risk feature information; inputting the digital risk feature information into a dynamic risk assessment model for processing to obtain a risk level signal; comparing the risk level signal with a preset risk threshold in real time, generating dynamic avoidance control instructions based on the comparison result and the real-time construction status of the construction site, and issuing them to construction equipment in real time; using the control response results of the construction equipment as feedback data to dynamically adjust the control parameters of the risk assessment model to optimize the accuracy of the risk control instructions.

[0006] Optionally, the acquisition of multi-source environmental data through an adaptive perception network arranged at the construction site includes: the adaptive perception network not only collects environmental parameters, but also collects construction activity parameters at the same time, wherein the environmental parameters include geological deformation data and meteorological data, and the construction activity parameters include construction equipment vibration frequency, blasting vibration intensity and vehicle traffic load; performing spatiotemporal correlation analysis on the environmental parameters and construction activity parameters, identifying local environmental changes caused by specific construction behaviors, and generating multi-source environmental data.

[0007] Optionally, when the multi-source environmental data exceeds an abnormality threshold dynamically set based on historical data or a monitoring blind spot appears, it includes: dynamically adjusting the abnormality threshold based on historical data, current construction activity parameters and environmental parameters, and generating an abnormality triggering drone scanning when the change of the multi-source environmental data exceeds the abnormality threshold; using a spatial interpolation algorithm to perform modeling and prediction based on the sensor data in the adaptive perception network, when the difference between the actual monitoring data and the predicted data exceeds the set tolerance range or there is a trend of insufficient data coverage in a certain area, a monitoring blind spot triggering drone scanning is generated; when the drone receives any one of the abnormality trigger or blind spot trigger, it starts multimodal scanning to obtain high-precision spatial data of the abnormal area or blind spot.

[0008] Optionally, the generating of digital risk characteristic information includes: generating fused data based on the multi-source environmental data and high-precision spatial data; spatially aligning the fused data with the geological information, and identifying deviations between the fused data and the geological information, wherein the deviations include geological deformation, surface undulations, and groundwater anomalies; using the deviations as risk characteristics and combining them with construction activity parameters to generate digital risk characteristic information that can characterize the risk caused by construction behavior.

[0009] Optionally, the constructed dynamic risk assessment model includes: constructing a multi-dimensional feature embedding layer to map the heterogeneous features in the digitized risk feature information to a unified vector space to achieve semantic alignment of data of different dimensions; constructing a feature fusion intermediate layer to use a self-attention mechanism to perform weighted fusion on the feature vectors output by the multi-dimensional feature embedding layer; constructing a risk assessment function to evaluate the evolution trend of the risk based on the fused features and output the corresponding comprehensive risk assessment value; constructing a hierarchical decision output layer to generate a hierarchical risk level signal based on the comprehensive risk assessment value as the final output.

[0010] Optionally, inputting the digitized risk feature information into a dynamic risk assessment model for processing includes: dividing the digitized risk feature information into geometric features, time series features, and environmental correlation features, and inputting each of them into a multidimensional feature embedding layer in the dynamic risk assessment model; in a feature fusion middle layer of the dynamic risk assessment model, performing multi-layer feature fusion on the output of the multidimensional feature embedding layer to obtain a comprehensive risk assessment value; and generating a risk level signal based on the comprehensive risk assessment value, wherein the risk level signals include "no risk", "low risk", "medium risk", and "high risk".

[0011] Optionally, the automatic selection and output of the corresponding dynamic avoidance instruction signal includes: constructing an instruction library containing a variety of dynamic avoidance instructions, and the instructions in the instruction library are associated with different risk levels and construction equipment status; obtaining the precise location and construction status information of all construction equipment in the current construction area in real time; intelligently matching the risk level signal with the construction equipment location and construction status information, and automatically selecting a suitable dynamic avoidance instruction signal from the instruction library.

[0012] Optionally, the automatic selection of a suitable dynamic avoidance instruction signal from the instruction library includes: generating a text describing the current risk scenario based on the risk level signal and the location of the construction equipment; evaluating the maneuverability and operating capability under the current risk scenario based on the status information of the construction equipment, and generating an equipment capability evaluation result; and generating a dynamic avoidance instruction signal suitable for the current scenario using the risk scenario text and the equipment capability evaluation result.

[0013] Optionally, the real-time adjustment of the parameters of the risk assessment model includes: the dynamic risk assessment model generates a confidence parameter for evaluating the reliability of the risk level signal while generating a risk level signal; the execution result of the dynamic avoidance instruction signal is used as new feedback data, and compared with the risk level signal and the confidence parameter to evaluate the assessment accuracy of the risk assessment model; and the parameters of the risk assessment model are updated using the Bayesian inference method based on the comparison result, so that the risk assessment model can more accurately reflect the actual risk.

[0014] Based on the same inventive concept, the present invention also provides an intelligent control system for environmental risks at construction sites of infrastructure lines in complex terrain. The system includes: a data acquisition module: used to acquire multi-source environmental data through an adaptive perception network arranged at the construction site, and to acquire high-precision spatial data through multi-modal scanning by unmanned aerial vehicles; an information fusion module: used to fuse the multi-source environmental data with the high-precision spatial data, and extract features related to potential environmental risks in combination with geological information to generate digital risk feature information; a risk assessment module: used to input the digital risk feature information into a dynamic risk assessment model for processing to obtain a risk level signal; an instruction selection module: used to compare the risk level signal with a preset risk threshold in real time, and automatically select and output a corresponding dynamic avoidance instruction signal based on the comparison result, current construction status and equipment position; a model optimization module: used to use the execution result of the dynamic avoidance instruction signal as feedback data, adjust the parameters of the risk assessment model in real time, and optimize the accuracy of risk identification and decision-making.

[0015] Compared with the prior art, the present invention has the following advantages: 1. The present invention uses a dynamic collaborative mechanism between an adaptive sensing network and drones to build a multi-dimensional data acquisition system covering the entire construction area, deeply integrating geological information with real-time construction parameters to provide high-precision input for control command generation. This mechanism solves the mismatch between control commands and risk situations caused by a traditional single data source, significantly improving the accuracy of command generation. Risk feature analysis based on multi-source data fusion can accurately match the status of construction equipment with the risk level, and output dynamic control commands such as equipment deceleration, detour, or emergency shutdown. This data-driven decision-making mechanism significantly reduces the error control rate and enhances the proactive risk avoidance capability of construction equipment.

[0016] 2. This invention innovatively constructs a closed-loop control system: command generation – equipment execution – feedback optimization. It captures equipment response data in real time and uses it as feedback signals to feed into the risk assessment model. Dynamic parameter adjustments continuously optimize the command generation logic. This design possesses continuous learning capabilities, completely eliminating the response lag inherent in static thresholds. This closed-loop mechanism not only shortens risk response time but also enables real-time, precise control of construction equipment's risk avoidance behavior, significantly improving control stability and reliability under complex working conditions.

[0017] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 It is a flow chart of an intelligent control method for environmental risks at construction sites of infrastructure lines in complex terrain according to an embodiment of the present invention.

[0020] Figure 2 It is a line graph of geological deformation data and dynamic anomaly threshold values ​​according to an embodiment of the present invention.

[0021] Figure 3 3 is a spatial distribution diagram of geological deformation deviation according to an embodiment of the present invention.

[0022] Figure 4 It is a structural diagram of an intelligent control system for environmental risks at construction sites of infrastructure lines in complex terrain according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0024] Reference Figure 1 One embodiment of the present invention proposes an intelligent control method for environmental risks at construction sites of infrastructure projects located in complex terrain. This method utilizes multi-dimensional information fusion and a dynamic decision-making mechanism to improve the timeliness and reliability of early warnings, enhance intelligent control of construction risks and emergency response capabilities, and provide technical support for the safe construction of infrastructure projects in complex terrain.

[0025] The method of this embodiment specifically includes: Acquiring multi-source environmental data through an adaptive sensing network deployed at the construction site, wherein the multi-source environmental data includes construction activity parameters and environmental parameters; Specifically, the adaptive sensing network consists of various sensors deployed within the construction area, such as sensors for monitoring geological deformation, weather stations for collecting meteorological data, and sensors for recording the operating status of construction equipment. These sensors can continuously collect environmental parameters and construction activity parameters at the construction site in real time.

[0026] When the multi-source environmental data exceeds an abnormal threshold dynamically set based on historical data or a monitoring blind spot appears, the drone is automatically triggered to perform multimodal scanning to obtain high-precision spatial data; Specifically, the adaptive perception network analyzes multi-source environmental data collected. If abnormal data fluctuations or interruptions in data flow are detected in a certain area, it is determined to be a risk or a monitoring blind spot. At this point, the data processing center immediately sends a command to the drone, dispatching it to perform multimodal scanning of the designated abnormal area or blind spot to obtain high-precision spatial data.

[0027] fusing the multi-source environmental data with high-precision spatial data, and extracting features related to potential environmental risks in combination with geological information to generate digital risk feature information; Specifically, multi-source environmental data collected by the adaptive sensing network is fused with high-precision spatial data acquired by drones and aligned with existing geological information. The fused data is used to extract key features associated with potential risks. These features are then quantified and encoded into digital risk signature information.

[0028] Inputting the digital risk characteristic information into the constructed dynamic risk assessment model for processing to obtain a risk level signal; Specifically, digital risk signature information is fed into a dynamic risk assessment model. The model comprehensively analyzes and evaluates these signatures, ultimately generating a risk level signal that reflects the severity of the risk at the current construction site.

[0029] Comparing the risk level signal with a preset risk threshold in real time, generating dynamic avoidance control instructions based on the comparison result and the real-time construction status of the construction site, and issuing the instructions to the construction equipment in real time; Specifically, the risk level signal output by the dynamic risk assessment model is compared in real time with multiple preset risk thresholds. If the risk level signal exceeds a certain threshold, the most appropriate dynamic avoidance instruction is automatically selected and issued from a set of instructions based on the current location and working status of the construction equipment.

[0030] The control response results of the construction equipment are used as feedback data to dynamically adjust the control parameters of the risk assessment model and optimize the accuracy of risk control instructions.

[0031] Specifically, the system continuously monitors changes in construction site environmental data after instructions are issued. Based on whether risks are effectively controlled after instructions are executed, feedback data is generated and used to adjust the internal parameters of the dynamic risk assessment model in real time. Through this feedback learning mechanism, the model's recognition and decision-making capabilities are continuously improved to adapt to the ever-changing construction environment and risk patterns.

[0032] Adaptive perception networks and drones are used to collaboratively acquire multi-source environmental data, and combined with dynamic risk assessment models and feedback loops for processing, to achieve intelligent risk control, precise assessment, and dynamic decision-making, significantly improving construction site safety and management efficiency.

[0033] Optionally, acquiring multi-source environmental data through an adaptive sensing network deployed at the construction site includes: The adaptive sensing network not only collects environmental parameters, but also construction activity parameters. The environmental parameters include geological deformation data and meteorological data, and the construction activity parameters include construction equipment vibration frequency, blasting vibration intensity, and vehicle traffic load. Specifically, the adaptive sensing network deploys geological deformation sensors, inclinometers, meteorological monitoring stations, and vibration sensors. The geological deformation sensors and inclinometers collect geological deformation data. The meteorological monitoring stations collect meteorological data. Vibration sensors are deployed near construction equipment and roadbeds to collect equipment vibration frequency, blasting vibration intensity, and vehicle traffic load. These heterogeneous sensors form a collaborative sensing network that synchronously transmits environmental data and construction activity data.

[0034] For example, at a slope construction site, geological deformation sensors continuously record minute slope displacements. Meanwhile, vibration sensors on nearby roads record vibration data from passing heavy trucks. A weather monitoring station records the day's rainfall and wind speed.

[0035] A spatiotemporal correlation analysis is performed on the environmental parameters and construction activity parameters to identify local environmental changes caused by specific construction behaviors and generate multi-source environmental data.

[0036] Specifically, the spatiotemporal correlation analysis of environmental parameters and construction activity parameters is to identify local environmental changes caused by specific construction activities. This analysis can be achieved through a spatiotemporal correlation model. The model can combine different types of data, such as geological deformation data and construction activity data Align them in time and space. By analyzing their correlation within a specific area and time period, the degree of impact of construction activities on environmental parameters can be calculated. For example, the correlation coefficient can be calculated using a formula to determine the weight of the impact of construction activity data on geological deformation data, thereby distinguishing between normal construction impacts and potential risks. The spatiotemporal correlation model can be expressed as: , in, Represents the signal of local environmental change, Represents geological deformation data, represents construction activity data, represents the spatial correlation kernel function, Represents the time-correlated kernel function. In this way, local environmental changes caused by construction activity data can be separated from geological deformation data.

[0037] For example, after a blasting operation, a vibration sensor recorded blasting vibration data with a vibration intensity of 0.8. Spatiotemporal correlation model analysis revealed that within the following two hours, a geological deformation sensor 20 meters from the blasting site recorded a vertical displacement of 0.5 mm. The model confirmed a strong spatiotemporal correlation between this displacement and the blasting vibration data. This correlation analysis result was generated as part of the multi-source environmental data.

[0038] Optionally, when the multi-source environmental data exceeds an abnormal threshold dynamically set based on historical data or a monitoring blind spot occurs, the method includes: Dynamically adjust the anomaly threshold based on historical data, current construction activity parameters, and environmental parameters. When changes in multi-source environmental data exceed the anomaly threshold, an anomaly is generated that triggers a drone scan. Specifically, the dynamic adjustment of the anomaly threshold can be achieved through a statistical model based on a sliding time window. The model continuously analyzes historical data and current construction activity parameters to obtain the fluctuation range of multi-source environmental data under normal conditions. When the new data collected exceeds this range, the model will calculate its deviation value. If the deviation value exceeds a preset statistical significance level, a signal is generated to trigger the drone scan. Figure 2 The figure below shows how multi-source environmental data changes over time. Geological deformation data is updated in real time, while anomaly thresholds are dynamically adjusted based on historical data and construction activity parameters. When the data curve exceeds the dynamic anomaly threshold, an anomaly trigger signal is generated to initiate drone scanning. The formula for dynamically adjusting the threshold can be expressed as: , in, represents the dynamic anomaly threshold, and Represents the mean and standard deviation of historical data, is a time-dependent weighting coefficient, is the weighted coefficient of the construction activity parameter. In this way, the abnormal threshold can be adapted to the changes of the construction site in real time, thereby improving the accuracy of the early warning.

[0039] For example, over the past month, the average geological deformation data for a certain area was 0.1 mm, with a standard deviation of 0.05 mm. Today, a heavy excavator begins operating in the area. Based on the excavator's operating parameters, the model dynamically adjusts the threshold to the average value plus an additional factor to account for vibration. When the sensor detects that the geological deformation data reaches 0.3 mm, exceeding the new dynamic threshold, an anomaly trigger signal is generated.

[0040] Based on the sensor data in the adaptive perception network, spatial interpolation algorithms are used to build models and predict data. When the difference between the actual monitoring data and the predicted data exceeds the set tolerance range or there is a trend of insufficient data coverage in a certain area, a monitoring blind spot is generated, triggering a drone scan; Specifically, spatial interpolation algorithms can be implemented using methods such as Kriging interpolation or inverse distance weighted interpolation. Data points from all sensors in the adaptive sensing network are used to construct a two- or three-dimensional spatial model representing the environmental parameters of the construction site. This model can predict environmental data for locations where sensors are not deployed or where sensors fail. When the predicted data for a certain area differs significantly from the actual monitored data, or when the model's prediction confidence in certain areas falls below a preset value, it is determined that there is a trend of insufficient data coverage. These situations will generate a signal that triggers drone scanning of the monitoring blind spot.

[0041] For example, in one area of ​​a slope, sensor data is sparse. The data processing center uses the Kriging interpolation algorithm to model the deformation data in this area and predict its deformation value. The model finds that at one predicted point, the predicted deformation value differs by more than 0.2 mm from the data at the nearest actual monitoring point, indicating an excessively large prediction error. Simultaneously, the model finds that the prediction confidence level for another area is less than 90%, also indicating a trend of insufficient data coverage. Both of these situations generate a monitoring blind spot trigger signal.

[0042] When the UAV receives any of the abnormal triggers or blind spot triggers, it starts multimodal scanning to obtain high-precision spatial data of the abnormal area or blind spot.

[0043] Specifically, upon receiving an anomaly trigger signal or blind spot trigger signal, the drone will immediately transition from standby mode to flight mode. The drone formation will automatically plan a flight path based on the coordinates of the anomaly area or blind spot specified in the signal. During flight, the drones utilize onboard equipment such as lidar, high-resolution visible light cameras, or multispectral cameras to perform multimodal scanning of the target area. Lidar can acquire high-precision three-dimensional point cloud data; visible light cameras can capture image information such as surface cracks and vegetation changes; and multispectral cameras can capture data such as geological composition and hydrological anomalies.

[0044] For example, when a drone receives an anomaly trigger signal containing the GPS coordinates of the anomaly area, the drone formation automatically plans a flight to 100 meters above the target area based on these coordinates. The drones then activate their onboard lidar to scan the area, acquiring high-precision 3D point cloud data to pinpoint surface cracks and subsidence areas. Simultaneously, high-resolution cameras capture surface images to reveal details of cracks and vegetation decline.

[0045] Optionally, generating digital risk characteristic information includes: generating fused data based on the multi-source environmental data and the high-precision spatial data; Specifically, multi-source environmental data and high-precision spatial data are spatiotemporally registered. This registration process utilizes a fusion algorithm based on timestamps and geographic coordinates to ensure precise alignment of data points from different sources in both time and space. For example, high-precision 3D point cloud data acquired by drones can be fused with time-series deformation data acquired by surface deformation sensors to form a dynamic 3D dataset with both temporal and spatial geometry information. The generation of fused data can be achieved using a weighted fusion function: , in, represents the generated fusion data, Represents multi-source environmental data, Represents high-precision spatial data, and It is a weighting factor set according to data type and quality.

[0046] For example, a surface deformation sensor with coordinates (100, 200, 50) records a displacement of 0.5 mm at 10:30. Simultaneously, a drone scans the area at 10:30, acquiring a high-precision 3D point cloud containing these coordinates. The data processing center performs spatiotemporal registration on these two data points to generate a single 3D data set containing the point's displacement information.

[0047] spatially aligning the fused data with the geological information, and identifying deviations between the fused data and the geological information, wherein the deviations include geological deformation, surface relief, and subsurface hydrological anomalies; Specifically, geological information includes preset geological maps, hydrological models, and soil parameters. This information is spatially aligned with the fused data. After alignment, the data processing center uses a deviation identification algorithm to compare the actual monitoring values ​​in the fused data with the theoretical predicted values ​​in the geological information. The deviation identification algorithm can calculate the difference between the actual surface undulation and the theoretical elevation in the geological model, or the difference between the groundwater level predicted by the hydrological model and the pore water pressure actually monitored by the sensor. Figure 3The figure shows the deviation size and spatial distribution between the actual monitoring value and the geological model prediction value. In the figure, the darker the area, the greater the deviation, representing the potential risk area. The deviation identification formula can be expressed as: , in, represents the identified deviation, represents the actual monitoring value in the fused data, Represents the theoretical predicted value in geological information.

[0048] For example, a geological model predicts a surface elevation of 150 meters in a certain area. The high-precision point cloud in the fused data shows the actual elevation is 150.3 meters. The deviation identification algorithm calculates a 0.3-meter deviation and flags it as a surface anomaly. Meanwhile, a hydrological model predicts a groundwater level of 10 meters, but sensors measure a groundwater level of 12 meters, a 2-meter deviation that is flagged as a groundwater anomaly.

[0049] The deviation is used as a risk feature and combined with construction activity parameters to generate digital risk feature information that can characterize the risk caused by the construction behavior.

[0050] Specifically, identified deviations are treated as risk characteristics and associated with construction activity parameters. The data processing center uses a risk attribution model to analyze the causal relationship between deviations and specific construction behaviors. The attribution model attributes deviations to specific construction behaviors based on factors such as temporal sequence, spatial distance, and physical impact intensity, thereby generating digital risk characteristic information that characterizes the risk caused by the construction behavior. The formula for generating risk characteristic information can be expressed as: , in, Represents digital risk characteristic information, Represents the identified deviation. Function It represents the risk attribution model, which takes deviation and construction activity parameters as input and outputs the final risk characteristic information.

[0051] For example, after a blasting operation, sensors detected a surface crack that widened by 2 millimeters. Risk attribution model analysis revealed a strong correlation between the crack's appearance and the intensity of the blasting vibration. The model then generated a risk signature containing parameters related to the crack and the causative construction behavior.

[0052] Optionally, the constructed dynamic risk assessment model includes: Construct a multi-dimensional feature embedding layer to map the heterogeneous features in the digitized risk feature information into a unified vector space, achieving semantic alignment of data of different dimensions; Specifically, the multidimensional feature embedding layer receives digitized risk feature information. These heterogeneous features include the geometric parameters of surface cracks, temporal variations in geological deformation, and environmental data related to construction activities. The embedding layer uses various linear or nonlinear functions to map these heterogeneous features into a common high-dimensional vector space. For example, a multilayer perceptron can be used to convert unstructured text descriptions into vectors, or a convolutional neural network can be used to extract image features. Ultimately, all features are vectorized, allowing them to be compared and integrated in a unified vector space.

[0053] For example, a time series feature of geological deformation data is mapped into a 128-dimensional vector. An environmental feature associated with blasting vibration is also mapped into a 128-dimensional vector. These two vectors are in the same vector space and can be subsequently fused.

[0054] Construct a feature fusion intermediate layer and use the self-attention mechanism to perform weighted fusion on the feature vectors output by the multi-dimensional feature embedding layer; Specifically, the feature fusion intermediate layer receives multiple feature vectors output by the multi-dimensional feature embedding layer. This intermediate layer uses the self-attention mechanism to perform weighted fusion of these vectors. The self-attention mechanism can calculate the importance of each feature vector to all other feature vectors and assign different weights to them. In this way, the model can automatically identify the most important risk features in the current scenario, such as the rapid expansion of surface cracks is more important than slight vehicle vibration. The fusion process can be expressed by the formula, where the weighted fusion output vector The input feature vector set can be Calculation yields: , in, represent the query matrix, key matrix and value matrix respectively, Represents the dimension of the feature vector. This formula is used to calculate the contribution of each feature vector to the overall risk assessment and generate a fused comprehensive feature vector.

[0055] For example, the multidimensional feature embedding layer outputs three feature vectors, representing geological deformation, groundwater anomalies, and construction equipment vibration. The feature fusion intermediate layer applies a self-attention mechanism and finds that the geological deformation vector and the groundwater anomaly vector have the strongest correlation. Therefore, these two vectors are given a higher weight during fusion, while the construction equipment vibration vector is given a lower weight.

[0056] Build a risk assessment function to evaluate the evolution trend of risks based on the fused features and output the corresponding comprehensive risk assessment value; Specifically, the risk assessment function receives the comprehensive feature vector generated by the feature fusion intermediate layer. Based on a time series analysis model, this function models the historical data of the comprehensive feature vector and assesses the current risk evolution trend. Based on this evolution trend and the current risk level, the model outputs a comprehensive risk assessment value, a continuous value that represents the severity of the risk.

[0057] For example, the model receives a composite feature vector whose historical data indicates that the rate of geological deformation is accelerating at a rate of 0.1 mm / hour. Based on this trend, the risk assessment function predicts that the geological deformation will reach a critical value of 0.5 mm within the next four hours and outputs a high composite risk assessment value of 0.85.

[0058] Construct a hierarchical decision output layer to generate a graded risk level signal based on the comprehensive risk assessment value as the final output.

[0059] Specifically, the hierarchical decision output layer receives the comprehensive risk assessment value generated by the risk assessment function. This output layer pre-sets multiple numerical intervals, each corresponding to a risk level signal. The output layer maps the comprehensive risk assessment value to the corresponding interval and generates the corresponding hierarchical risk level signal.

[0060] For example, the pre-set rules for the hierarchical decision output layer are: comprehensive risk assessment values ​​between 0.0 and 0.2 are "no risk," 0.2-0.5 are "low risk," 0.5-0.8 are "medium risk," and 0.8-1.0 are "high risk." If the model generates a comprehensive risk assessment value of 0.85, the output layer maps it to the "high risk" range and generates a "high risk" risk level signal.

[0061] Optionally, inputting the digitized risk feature information into the constructed dynamic risk assessment model for processing includes: The digital risk feature information is divided into geometric features, temporal features and environmental correlation features, and inputted into the multidimensional feature embedding layer of the dynamic risk assessment model respectively; Specifically, before being input into the dynamic risk assessment model, digitized risk feature information is classified according to its intrinsic properties. Parameters such as the length, width, and depth of surface cracks are classified as geometric features. Data such as the changing trends of geological deformation rates and blasting vibration intensity are classified as time series features. Data on the correlation between construction activities and environmental changes are classified as environmental correlation features. These classified heterogeneous features are then fed into the multidimensional feature embedding layer of the dynamic risk assessment model for independent preprocessing and vectorization.

[0062] For example, the digital risk feature information for a specific risk event includes a crack length of 1.5 meters, a width of 0.5 millimeters, a deformation rate of 0.2 millimeters per hour over the past two hours, and a correlation with heavy truck traffic. The crack length and width are classified as geometric features, the deformation rate as a temporal feature, and the correlation with heavy truck traffic as an environmental feature.

[0063] In the feature fusion middle layer of the dynamic risk assessment model, multi-layer feature fusion is performed on the output of the multi-dimensional feature embedding layer to obtain a comprehensive risk assessment value; Specifically, the fusion middle layer performs multi-layer feature fusion on these feature vectors through multiple fusion units to generate a comprehensive risk assessment value. Different weights are assigned to different features based on their contribution to risk, ensuring that the fused comprehensive assessment value more accurately reflects the actual risk situation.

[0064] For example, the multidimensional feature embedding layer outputs three vectors: geometric features, temporal features, and environmental features. The feature fusion layer assigns the highest weight to the temporal feature vector, as it reflects the dynamic development trend of the risk; the second highest weight to the geometric feature vector, as it reflects the physical form of the risk; and the lower weight to the environmental feature vector, as it reflects potential triggers rather than direct risks.

[0065] A risk level signal is generated according to the comprehensive risk assessment value, wherein the risk level signal includes "no risk", "low risk", "medium risk" and "high risk".

[0066] Specifically, the hierarchical decision output layer generates a hierarchical risk level signal based on the comprehensive risk assessment value. The comprehensive risk assessment value is a continuous numerical value representing the severity of the risk. The hierarchical decision output layer presets multiple numerical intervals, each corresponding to a risk level. The hierarchical decision output layer maps the comprehensive risk assessment value to the corresponding interval and outputs the corresponding risk level signal.

[0067] Optionally, the automatically selecting and outputting a corresponding dynamic avoidance instruction signal includes: Build an instruction library containing a variety of dynamic avoidance instructions, where the instructions in the instruction library are associated with different risk levels and construction equipment status; Specifically, the instruction library is a structured database that stores a variety of dynamic avoidance instructions. Each instruction in the instruction library corresponds to one or more risk levels and construction equipment statuses. Instructions in the instruction library can include equipment deceleration, equipment detour, equipment shutdown, personnel evacuation, and remote alarm commands. Each instruction includes specific execution parameters.

[0068] For example, the instruction library stores a command with the following conditions: risk level is "medium risk," construction equipment status is "operating," and equipment type is "excavator." The command specifies that the equipment should slow down to 5 kilometers per hour and send a warning message.

[0069] Obtain the precise location and construction status information of all construction equipment in the current construction area in real time; Specifically, construction equipment is equipped with navigation and IoT sensors. Navigation is used to obtain the equipment's precise geographic location in real time. IoT sensors are used to obtain the equipment's construction status information. This equipment location and status information is transmitted in real time via wireless networks.

[0070] For example, the navigation coordinate information of an excavator is (120.15, 30.25), and the status information transmitted by its IoT sensor shows that the engine speed of the excavator is 1500 revolutions per minute and the working arm is performing excavation operations.

[0071] The risk level signal is intelligently matched with the construction equipment location and construction status information, and an appropriate dynamic avoidance instruction signal is automatically selected from the instruction library.

[0072] Specifically, the risk level signal is intelligently matched with real-time information on the location and status of construction equipment. This matching process is implemented through a decision-making matching algorithm. This algorithm takes the risk level, the precise location of the equipment, and its operating status as input and searches the instruction library to find all candidate instructions that meet the requirements. If multiple instructions meet the requirements, the matching algorithm sorts and selects them based on priority rules, ultimately automatically selecting the most appropriate dynamic avoidance instruction signal.

[0073] For example, a "high risk" signal is received, and a heavy truck is detected traveling at 30 kilometers per hour in a high-risk area. The matching algorithm uses the "high risk" signal, the truck's location, and its "high speed" status as inputs and searches the instruction library. The algorithm finds two matching instructions: one for "slow down to 10 kilometers per hour" and the other for "immediately stop work." Based on the priority rules, "immediately stop work" is assigned the highest priority and automatically selected for output.

[0074] Optionally, the automatically selecting an appropriate dynamic avoidance instruction signal from the instruction library includes: Generate a text describing the current risk scenario based on the risk level signal and the location of the construction equipment; Specifically, the risk level signal and device location information are converted into natural language or structured data to generate text describing the current risk scenario. This text can describe the risk type, location, and severity in detail.

[0075] For example, the dynamic risk assessment model generates a "medium risk" risk level signal. Meanwhile, the coordinates of a road roller are (120.18, 30.28). This information is integrated into a risk scenario text: "The medium risk area is near the road roller, please pay attention."

[0076] Assessing the maneuverability and operating capability of the construction equipment under the current risk scenario based on the status information of the construction equipment, and generating an equipment capability assessment result; Specifically, the equipment's maneuverability and operational capabilities are assessed based on its status information. This information includes engine speed, boom angle, hydraulic system pressure, and more. For example, the assessment is made as to whether the equipment's current speed supports rapid evacuation or whether the boom angle is suitable for emergency support operations. These assessment results are quantified as equipment capability assessments.

[0077] For example, consider the status of an excavator, which displays information indicating it is currently digging with its boom at full extension. Based on this information, the data processing center assesses that the excavator currently lacks rapid maneuverability and generates a capability assessment result: "Poor maneuverability, strong operational capability."

[0078] The risk scenario text and the equipment capability assessment results are used to generate a dynamic avoidance instruction signal suitable for the current scenario.

[0079] Specifically, the risk scenario text and device capability assessment results are used to generate dynamic avoidance command signals tailored to the current scenario. This command signal generation process is implemented through a decision-making model. This model takes the risk scenario and device capabilities as input and matches them against a command library to generate the optimal avoidance command signal. The command signal can include specific action parameters.

[0080] For example, the risk scenario describes a high-risk area near a road roller, and the equipment capability assessment is poor maneuverability. Based on this information, the decision model selects and generates a command signal from the command library: "Immediately stop the road roller and evacuate 50 meters to a safe area behind it." This command signal is sent to the road roller.

[0081] Optionally, the real-time adjustment of risk assessment model parameters includes: The dynamic risk assessment model generates a confidence parameter for assessing the reliability of the risk level signal while generating the risk level signal; Specifically, the confidence parameter is a quantitative indicator that reflects the degree of certainty of the dynamic risk assessment model's generated risk level signal. The confidence parameter is generated based on the quality and diversity of the model's input features, as well as their match with historical training data. For example, when the input feature information closely matches high-risk patterns in the training data, the model generates a high-confidence risk level signal. The confidence parameter can be a value between 0 and 1, with larger values ​​indicating more reliable model predictions.

[0082] For example, a dynamic risk assessment model receives a set of risk signatures, including high-precision spatial data showing significant surface cracks. Because these signatures are strongly associated with high-risk events in the model's training data, the model generates a "high risk" rating and outputs a confidence parameter of 0.98, indicating a high degree of confidence in this judgment.

[0083] The execution result of the dynamic avoidance instruction signal is used as new feedback data, and compared with the risk level signal and the confidence parameter to evaluate the assessment accuracy of the risk assessment model; Specifically, the feedback data is compared with the risk level signal and confidence parameters previously generated by the model. This comparison process assesses whether the instruction effectively mitigated the risk. If the instruction successfully mitigated the risk and the model's prediction confidence is high, the model's prediction is considered accurate. If the risk still worsens after the instruction is executed and the model confidence is high, it indicates that the model has a high-confidence error.

[0084] For example, the dynamic risk assessment model generated a "high risk" signal with a confidence level of 0.95 and issued a stop-work order. Feedback data showed that geological deformation data stopped deteriorating within two hours of the order being executed. Comparing this result with the high-confidence prediction confirmed that the model's assessment was accurate.

[0085] Based on the comparison results, the Bayesian inference method is used to update the parameters of the risk assessment model so that the risk assessment model can more accurately reflect the actual risk.

[0086] Specifically, when a high-confidence error is detected in the model during comparison, the risk assessment model parameters are updated using Bayesian inference. This method treats the model's existing parameters as prior knowledge and high-confidence error feedback data as new evidence. According to the Bayesian formula, this new evidence is used to update the model's prior knowledge, resulting in a posterior probability distribution, which is then used to correct the model parameters. Corrections focus on key parameters that led to high-confidence errors to prevent similar errors from recurring. In this way, the risk assessment model continuously learns and optimizes, more accurately reflecting the actual risk situation.

[0087] For example, a model might have predicted a certain area as "high risk" with a confidence level of 0.95, but feedback data indicated the risk hadn't actually occurred. Using Bayesian inference, this high-confidence error serves as new evidence, leading to a significant adjustment of the model's feature weights related to that event. This prevents the model from making the same high-confidence misjudgment the next time it encounters a similar situation.

[0088] Based on the same inventive concept, Figure 4 As shown, the present invention also provides an intelligent control system for environmental risks at construction sites of infrastructure lines in complex terrain, the system comprising: Data acquisition module: used to acquire multi-source environmental data through an adaptive sensing network deployed at the construction site, and to acquire high-precision spatial data through multimodal scanning by drones; Information fusion module: used to fuse the multi-source environmental data with high-precision spatial data, and extract features related to potential environmental risks in combination with geological information to generate digital risk feature information; Risk assessment module: used to input the digital risk feature information into the dynamic risk assessment model for processing to obtain a risk level signal; Instruction selection module: used to compare the risk level signal with the preset risk threshold in real time, and automatically select and output the corresponding dynamic avoidance instruction signal based on the comparison result, current construction status and equipment location; Model optimization module: used to use the execution results of the dynamic avoidance instruction signal as feedback data, adjust the parameters of the risk assessment model in real time, and optimize the accuracy of risk identification and decision-making.

[0089] It should be noted that the electrical connections between the above-mentioned units do not necessarily mean direct connections of lines. Indirect connections are applicable to the embodiments of the present invention as long as the purpose of the present invention is achieved. The above description is only an exemplary embodiment of the present invention and is not intended to limit the scope of the present invention.

[0090] That is, any equivalent changes and modifications made according to the teachings of the present invention are still within the scope of the present invention. Those skilled in the art will readily conceive of other embodiments of the present invention after considering the disclosure of the specification and practical truths. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary technical means in the art not described herein.

Claims

1. An intelligent control method for environmental risks at construction sites of infrastructure lines in complex terrain, characterized by: The method comprises: Acquiring multi-source environmental data through an adaptive sensing network deployed at the construction site, wherein the multi-source environmental data includes construction activity parameters and environmental parameters; When the multi-source environmental data exceeds an abnormal threshold dynamically set based on historical data or a monitoring blind spot appears, the drone is automatically triggered to perform multimodal scanning to obtain high-precision spatial data; fusing the multi-source environmental data with high-precision spatial data, and extracting features related to potential environmental risks in combination with geological information to generate digital risk feature information; Inputting the digital risk characteristic information into the constructed dynamic risk assessment model for processing to obtain a risk level signal; Comparing the risk level signal with a preset risk threshold in real time, generating dynamic avoidance control instructions based on the comparison result and the real-time construction status of the construction site, and issuing them to the construction equipment in real time; The control response results of the construction equipment are used as feedback data to dynamically adjust the control parameters of the risk assessment model and optimize the accuracy of risk control instructions.

2. The intelligent control method for environmental risks at construction sites of infrastructure lines in complex terrain according to claim 1 is characterized in that: The method of acquiring multi-source environmental data through an adaptive sensing network deployed at the construction site includes: The adaptive sensing network not only collects environmental parameters, but also construction activity parameters. The environmental parameters include geological deformation data and meteorological data, and the construction activity parameters include construction equipment vibration frequency, blasting vibration intensity, and vehicle traffic load. A spatiotemporal correlation analysis is performed on the environmental parameters and construction activity parameters to identify local environmental changes caused by specific construction behaviors and generate multi-source environmental data.

3. The intelligent control method for environmental risks at construction sites of infrastructure lines in complex terrain according to claim 1 is characterized in that: When the multi-source environmental data exceeds an abnormal threshold dynamically set based on historical data or a monitoring blind spot appears, it includes: Dynamically adjust the anomaly threshold based on historical data, current construction activity parameters, and environmental parameters. When changes in multi-source environmental data exceed the anomaly threshold, an anomaly is generated that triggers a drone scan. Based on the sensor data in the adaptive perception network, spatial interpolation algorithms are used to build models and predict data. When the difference between the actual monitoring data and the predicted data exceeds the set tolerance range or there is a trend of insufficient data coverage in a certain area, a monitoring blind spot is generated, triggering a drone scan; When the UAV receives any of the abnormal triggers or blind spot triggers, it starts multimodal scanning to obtain high-precision spatial data of the abnormal area or blind spot.

4. The intelligent control method for environmental risks at construction sites of infrastructure lines in complex terrain according to claim 1 is characterized in that: Generating digital risk characteristic information includes: generating fused data based on the multi-source environmental data and the high-precision spatial data; spatially aligning the fused data with the geological information, and identifying deviations between the fused data and the geological information, wherein the deviations include geological deformation, surface relief, and subsurface hydrological anomalies; The deviation is used as a risk feature and combined with construction activity parameters to generate digital risk feature information that can characterize the risk caused by the construction behavior.

5. The intelligent control method for environmental risks at construction sites of infrastructure lines in complex terrain according to claim 1 is characterized in that: The constructed dynamic risk assessment model includes: Construct a multi-dimensional feature embedding layer to map the heterogeneous features in the digitized risk feature information into a unified vector space, achieving semantic alignment of data of different dimensions; Construct a feature fusion intermediate layer and use the self-attention mechanism to perform weighted fusion on the feature vectors output by the multi-dimensional feature embedding layer; Build a risk assessment function to evaluate the evolution trend of risks based on the fused features and output the corresponding comprehensive risk assessment value; Construct a hierarchical decision output layer to generate a graded risk level signal based on the comprehensive risk assessment value as the final output.

6. The intelligent control method for environmental risks at construction sites of infrastructure lines in complex terrain according to claim 5 is characterized in that: Inputting the digital risk characteristic information into the constructed dynamic risk assessment model for processing includes: The digital risk feature information is divided into geometric features, temporal features and environmental correlation features, and inputted into the multi-dimensional feature embedding layer of the dynamic risk assessment model respectively; In the feature fusion middle layer of the dynamic risk assessment model, multi-layer feature fusion is performed on the output of the multi-dimensional feature embedding layer to obtain a comprehensive risk assessment value; A risk level signal is generated according to the comprehensive risk assessment value, wherein the risk level signal includes "no risk", "low risk", "medium risk" and "high risk".

7. The intelligent control method for environmental risks at construction sites of infrastructure lines in complex terrain according to claim 1 is characterized in that: The generating of dynamic avoidance control instructions based on the comparison result and the real-time construction status of the construction site and issuing the instructions to the construction equipment in real time includes: Build an instruction library containing a variety of dynamic avoidance instructions, where the instructions in the instruction library are associated with different risk levels and construction equipment status; Obtain the precise location and construction status information of all construction equipment in the current construction area in real time; The risk level signal is intelligently matched with the construction equipment location and construction status information, and an appropriate dynamic avoidance instruction signal is automatically selected from the instruction library.

8. The intelligent control method for environmental risks at construction sites of infrastructure lines in complex terrain according to claim 7 is characterized in that: The automatically selecting an appropriate dynamic avoidance instruction signal from the instruction library includes: Generate a text describing the current risk scenario based on the risk level signal and the location of the construction equipment; Based on the status information of the construction equipment, assess the maneuverability and operation capability under the current risk scenario and generate an equipment capability assessment result; The risk scenario text and the equipment capability assessment results are used to generate a dynamic avoidance instruction signal suitable for the current scenario.

9. The intelligent control method for environmental risks at construction sites of infrastructure lines in complex terrain according to claim 1 is characterized in that: The control parameters of the dynamic adjustment risk assessment model include: The dynamic risk assessment model generates a confidence parameter for assessing the reliability of the risk level signal while generating the risk level signal; The execution result of the dynamic avoidance instruction signal is used as new feedback data, and compared with the risk level signal and the confidence parameter to evaluate the assessment accuracy of the risk assessment model; Based on the comparison results, the Bayesian inference method is used to update the parameters of the risk assessment model so that the risk assessment model can more accurately reflect the actual risk.

10. An intelligent control system for environmental risks at construction sites of infrastructure lines in complex terrain, characterized by: The system is applied to the intelligent control method for environmental risks at construction sites of infrastructure lines in complex terrain as described in any one of claims 1 to 9, and the system comprises: Data acquisition module: used to acquire multi-source environmental data through an adaptive sensing network deployed at the construction site, and to acquire high-precision spatial data through multimodal scanning by drones; Information fusion module: used to fuse the multi-source environmental data with high-precision spatial data, and extract features related to potential environmental risks in combination with geological information to generate digital risk feature information; Risk assessment module: used to input the digital risk feature information into the dynamic risk assessment model for processing to obtain a risk level signal; Instruction selection module: used to compare the risk level signal with the preset risk threshold in real time, generate dynamic avoidance control instructions based on the comparison result and the real-time construction status of the construction site, and issue them to the construction equipment in real time; Model optimization module: used to use the control response results of construction equipment as feedback data, dynamically adjust the control parameters of the risk assessment model, and optimize the accuracy of risk control instructions.

Images