A construction risk management method and system for pavement engineering
Through multimodal deep learning model and graph attention network, pavement diagram structure is built, section data is dynamically updated, construction risks are predicted, and the shortcomings of relying on experience and historical data in the existing technology are solved, and the risk of pavement engineering is efficient. Accurate management.
Patent Information
- Application Number
- CN202410733554.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-07
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-06-07
AI Technical Summary
In the existing technology, pavement engineering construction risk management relies on empirical judgment and historical data, and lacks forward-looking and dynamic adjustment capabilities, resulting in unpredictable sudden risks, delays in construction periods and increased costs.
Using multimodal deep learning model combined with a graph attention network, build a pavement diagram structure, obtain section data and dynamic updates, predict construction risks, including geological types, construction equipment data and weather forecast data, identify potential risks and provide specific risks Type and probability.
It improves the accuracy and forward-looking nature of construction risk prediction, realizes automated and intelligent risk management, reduces blindness in decision-making and resource waste, optimizes the allocation of construction resources, and ensures the reliable implementation of construction plans.
Smart Images

Figure CN118586709B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of intelligent analysis of engineering risks, and in particular to a construction risk management method and system for pavement engineering. Background Art
[0002] As the core link of infrastructure construction, the success of pavement engineering construction is directly related to the service life and safety of the road. In this process, construction risk management is particularly important.
[0003] At present, the risk management of pavement construction mainly relies on simple analysis of experience and historical data, such as analysis by experts based on the experience of historical accident data. However, the coverage of these data sources is limited and cannot reflect the dynamically changing risk factors during the construction process. In addition, at present, the construction party is generally organized to learn and publicize risks only after pavement problems or accidents occur, resulting in the inability to predict potential risks in advance at an early stage and a lack of foresight. When sudden risks occur, it often leads to delays in construction schedules and increased construction costs.
[0004] In response to the above problems, the industry has not yet proposed a better technical solution. Summary of the invention
[0005] The present application provides a construction risk management method, system, storage medium, computer program product and electronic device for pavement engineering, which are used to at least solve the problem in the current related technology that the construction risk management of pavement engineering can only be based on the manager's experience learning and post-management, and sudden construction risks cannot be effectively handled.
[0006] In a first aspect, an embodiment of the present application provides a construction risk management method for a pavement engineering project, comprising: obtaining a section data group of each section in a target pavement to be managed, the section data group comprising a section geological type, construction equipment data, a planned construction time and weather forecast data corresponding to the section; the section geological type comprises soil type information, geological structure information and soil layer thickness information; the construction equipment data comprises equipment type information, equipment status information and equipment usage frequency; the planned construction time comprises a construction progress plan and a construction phase time plan; the weather forecast data comprises temperature information, humidity information, precipitation information and wind force information corresponding to a preset time period in the future; based on the section data group corresponding to each section, a pavement graph structure is constructed; the graph structure comprises a plurality of nodes and edge connections, each of the nodes uniquely corresponds to a section, and the node features of the nodes are defined by the corresponding section data group; the edge connections are defined according to section association events between different nodes, and the section association events are used to indicate at least one of the following: sections with adjacent spatial positions, sections with overlapping planned construction times, and sections sharing the same construction equipment; based on a graph attention network The method comprises the steps of: processing the pavement map structure by a network to update the node features of each node in the pavement map structure; inputting the updated pavement map structure into a construction risk prediction model, so that the construction risk prediction model predicts the corresponding pavement construction risk prediction results for each node, and the pavement construction risk prediction results include the pavement construction risk type and risk probability; the construction risk prediction model adopts a multimodal deep learning model, which comprises an input layer, a convolution layer, an LSTM layer, an attention layer, an MLP layer and an output layer; the input layer is used to receive the updated target node feature vector of the corresponding target node; the convolution layer is used to extract the spatial features corresponding to the target node feature vector through a convolution operation to determine the corresponding first intermediate feature; the LSTM layer is used to capture the temporal dependency of the target node feature vector to determine the corresponding second intermediate feature; the attention layer is used to weightedly fuse the first intermediate feature and the second intermediate feature based on the attention weight to determine the corresponding attention comprehensive feature; the MLP layer is used to extract the high-order feature representation corresponding to the attention comprehensive feature; and the output layer is used to predict the construction risk prediction result corresponding to the high-order feature representation.
[0007] In a second aspect, an embodiment of the present application provides a construction risk management system for pavement engineering, comprising: a data acquisition unit, for acquiring a section data group of each section in the target pavement to be managed, the section data group comprising the section geological type, construction equipment data, planned construction time and weather forecast data corresponding to the section; the section geological type comprises soil type information, geological structure information and soil layer thickness information; the construction equipment data comprises equipment type information, equipment status information and equipment usage frequency; the planned construction time comprises a construction progress plan and a construction phase time plan; the weather forecast data comprises temperature information, humidity information, precipitation information and wind force information corresponding to a preset time period in the future; a graph structure construction unit, for constructing a pavement graph structure based on the section data group corresponding to each section; the graph structure comprises a plurality of nodes and edge connections, each of the nodes uniquely corresponds to a section, and the node features of the nodes are defined by the corresponding section data group; the edge connections are defined based on section association events between different nodes, and the section association events are used to indicate at least one of the following: sections with adjacent spatial positions, sections with overlapping planned construction times, and sections sharing the same construction equipment; a node feature update unit , used to process the pavement graph structure based on the graph attention network to update the node features of each node in the pavement graph structure; a risk prediction unit, used to input the updated pavement graph structure into the construction risk prediction model, so that the construction risk prediction model predicts the corresponding pavement construction risk prediction results for each node respectively, and the pavement construction risk prediction results include the pavement construction risk type and risk probability; the construction risk prediction model adopts a multimodal deep learning model, which includes an input layer, a convolution layer, an LSTM layer, an attention layer, an MLP layer and an output layer; the input layer is used to receive the updated target node feature vector of the corresponding target node; the convolution layer is used to extract the spatial features corresponding to the target node feature vector through a convolution operation to determine the corresponding first intermediate feature; the LSTM layer is used to capture the temporal dependency of the target node feature vector to determine the corresponding second intermediate feature; the attention layer is used to weightedly fuse the first intermediate feature and the second intermediate feature based on the attention weight to determine the corresponding attention comprehensive feature; the MLP layer is used to extract the high-order feature representation corresponding to the attention comprehensive feature; the output layer is used to predict the construction risk prediction result corresponding to the high-order feature representation.
[0008] In a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the steps of the construction risk management method for pavement engineering of any embodiment of the present application.
[0009] In a fourth aspect, an embodiment of the present application provides a storage medium on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of the construction risk management method for pavement engineering of any embodiment of the present application are implemented.
[0010] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the construction risk management method for pavement engineering of any embodiment of the present application.
[0011] The construction risk management method of a pavement engineering provided in this application can produce at least the following technical effects:
[0012] (1) By acquiring detailed road segment data sets, including geological types, construction equipment data, planned construction time, and weather forecast data, the actual conditions of the construction site can be fully and multi-dimensionally reflected. By constructing a road map structure and using a graph attention network to process the road segment data set, the association information between different road segments can be fully utilized and dynamically updated, improving the accuracy and comprehensiveness of the data. Furthermore, a multimodal deep learning model is adopted, combining the convolutional layer to extract spatial features, the LSTM layer to capture temporal dependencies, the attention layer to weightedly fuse features, and the MLP layer to extract high-order feature representations, achieving deeper feature extraction and fusion, thereby improving the accuracy of construction risk prediction.
[0013] (2) The prediction model predicts the risk type and risk probability of pavement construction in different sections for each node, and can identify potential risks in advance. Using dynamic data such as weather forecast data and construction schedule, the model can update and predict future risks in real time, making construction management more forward-looking. Therefore, the system's early warning of risks helps the construction party to take effective preventive measures or adjust plans before the risk occurs, reduce the impact of risks, and avoid construction delays and cost increases.
[0014] (3) Through the multimodal deep learning model, multi-dimensional information is integrated to conduct construction risk assessment, realizing a comprehensive analysis of risk factors in complex construction environments. The construction risk prediction results can provide specific risk types and risk probabilities, provide construction parties with clear risk response strategies, and reduce the blindness and uncertainty of decision-making. Therefore, through the system to achieve automated and intelligent risk management, the efficiency of construction management is effectively improved, enabling construction parties to quickly respond to and deal with sudden risks, and raising the overall level of construction management.
[0015] (4) Based on the construction equipment data and planned construction time, the model can predict the risk of equipment use in different road sections, guide the construction party to reasonably arrange equipment use and scheduling, and avoid resource waste and equipment damage. The road section association events are established based on factors such as spatial proximity, time overlap, and equipment sharing, and are used to define the edges in the graph structure, so that the model can comprehensively consider multiple factors for risk prediction and optimize the allocation of construction resources.
[0016] This technical solution effectively solves the shortcomings of traditional pavement engineering construction risk management, which relies on experience-based judgment and lacks foresight and dynamic adjustment capabilities. It significantly improves the scientificity, accuracy and timeliness of construction risk management, thereby ensuring the reliable implementation of pavement engineering construction plans. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to 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 application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0018] Figure 1 A flowchart showing an example of a construction risk management method for a pavement project according to an embodiment of the present application;
[0019] Figure 2 A schematic diagram of structural connection of an example of a construction risk prediction model according to an embodiment of the present application is shown;
[0020] Figure 3 A structural block diagram of an example of a construction risk management system for a pavement project according to an embodiment of the present application is shown;
[0021] Figure 4 It is a schematic structural diagram of an embodiment of an electronic device of the present application. DETAILED DESCRIPTION
[0022] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0023] In the technical solution of this application, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved shall comply with the provisions of relevant laws and regulations and shall not violate public order and good morals.
[0024] Figure 1 A flowchart showing an example of a construction risk management method for a pavement project according to an embodiment of the present application is shown.
[0025] Regarding the executor of the method of the embodiment of the present application, it can be any controller or processor with computing or processing capabilities. By combining a multimodal deep learning model and a graph attention network, it not only improves the accuracy and real-time performance of construction risk prediction, but also realizes multi-dimensional and comprehensive risk assessment, which has significant technical effects and practical application value.
[0026] In some examples, it may be integrated and configured in an electronic device or terminal by means of software, hardware, or a combination of software and hardware, and the type of the terminal or electronic device may be diverse, such as a mobile phone, a tablet computer, or a desktop computer, etc.
[0027] like Figure 1 As shown, in step S110, a road section data group of each road section in the target road surface to be managed is obtained.
[0028] Here, the road section data group includes the road section geological type, construction equipment data, planned construction time and weather forecast data corresponding to the road section. The road section geological type includes soil type information, geological structure information and soil layer thickness information. The construction equipment data includes equipment type information, equipment status information and equipment use frequency. The planned construction time includes the construction progress plan and the construction stage time plan. The weather forecast data includes temperature information, humidity information, precipitation information and wind information corresponding to the future preset time period.
[0029] It should be understood that various methods can be used to implement the collection of data for various road sections, and no restrictions are imposed for the time being. In some embodiments, soil type information, geological structure information and soil thickness information are obtained by using geological survey equipment (such as geological radar, drilling equipment) through on-site exploration, satellite remote sensing and other methods. In addition, the equipment type, equipment status (such as normal, faulty, under maintenance) and equipment usage frequency (such as daily usage time, cumulative number of times used) are obtained from the Internet of Things sensors and equipment management systems of the construction equipment. In addition, from the construction management system, the construction schedule (such as the start and end time of each stage) and the construction stage time plan (such as the time arrangement of specific tasks) are determined based on the input information. In addition, the temperature, humidity, precipitation and wind force information for the preset time period in the future can also be obtained from the meteorological service provider to ensure real-time updating of the data.
[0030] Therefore, multi-dimensional data collection ensures a comprehensive understanding of the road construction scenarios to be carried out, which helps to accurately predict risks. In addition, risk prediction analysis through real-time acquisition and update of data improves the timeliness and accuracy of data and avoids prediction errors caused by data lag.
[0031] In step S120, a road map structure is constructed based on the road segment data group corresponding to each road segment.
[0032] Here, the graph structure contains multiple nodes and edge connections, each node uniquely corresponds to a road section, and the node characteristics of the node are defined by the corresponding road section data group. The edge connection is defined based on the road section association events between different nodes, and the road section association events are used to indicate at least one of the following: road sections with adjacent spatial locations, road sections with overlapping planned construction times, and road sections that share the same construction equipment.
[0033] In some embodiments, each road segment is regarded as a node, and node features are defined based on the acquired road segment data set, and the node features include geological type features, equipment usage features, time features, and weather features. In addition, the distance between adjacent road segments is calculated based on geographic coordinates, and an edge connection is created if the distance is less than a set threshold; based on the planned construction time, an edge connection is created if the construction time of two road segments overlaps; and an edge connection is created if two road segments share the same construction equipment.
[0034] Therefore, the nodes and edges in the graph structure constructed by the above scheme can effectively reflect the spatial, temporal and resource associations between different road sections, and improve the model's ability to simulate actual construction conditions. In addition, through the diversity of edge connections (space, time, equipment), the model can adapt to different construction environments and conditions. In addition, the design based on the graph structure is easy to expand and adjust, and more nodes or edges can be added according to actual construction needs or plan adjustments.
[0035] In step S130, the road graph structure is processed based on the graph attention network to update the node features of each node in the road graph structure.
[0036] Graph Attention Network (GAT) is a neural network model that applies attention mechanism to graph data, aiming to enhance the performance of graph neural network by introducing attention mechanism. The core idea of Graph Attention Network is to aggregate the information of neighbor nodes through attention mechanism. Specifically, GAT introduces node-to-node adaptive weights, which are learned and dynamically adjusted to better aggregate the information of neighbor nodes.
[0037] Here, GAT is used to process the road map structure, and the attention mechanism is used to dynamically update and optimize the node features corresponding to each road section in the graph structure, effectively capturing the important correlation between nodes. Through multi-level and multi-angle feature fusion, the accuracy of feature updating is improved, and the model's adaptability to complex construction environments is enhanced.
[0038] In step S140, the updated pavement map structure is input into the construction risk prediction model, so that the construction risk prediction model predicts the corresponding pavement construction risk prediction result for each node, and the pavement construction risk prediction result includes the pavement construction risk type and risk probability.
[0039] Here, the pavement map structure is processed through the construction risk prediction model, and the deep learning model technology is used to process and predict pavement construction risks in real time, providing timely and accurate decision support for construction management.
[0040] Figure 2 A schematic diagram of the structural connection of an example of a construction risk prediction model according to an embodiment of the present application is shown.
[0041] like Figure 2 As shown, the construction risk prediction model 200 adopts a multimodal deep learning model, which includes an input layer 210, a convolutional layer 220, an LSTM layer 230, an attention layer 240, an MLP layer 250 and an output layer 260.
[0042] The input layer 210 is used to receive the updated target node feature vector corresponding to the target node.
[0043] In some embodiments, the node feature vectors updated by the graph attention network are received, each node feature vector integrating multiple information such as geological type, construction equipment data, planned construction time, and weather forecast data. In addition, the input data is standardized through the input layer, such as normalization of feature values, to meet the requirements of the deep learning model.
[0044] The multimodal data (geology, equipment, time, weather) are effectively integrated through the input layer to ensure the integrity of the input features, improve the convergence speed and stability of the model, and avoid training problems caused by features of different scales.
[0045] The convolution layer 220 is used to extract the spatial features corresponding to the target node feature vector through a convolution operation to determine the corresponding first intermediate features.
[0046] In some embodiments, a one-dimensional convolution operation is used to extract local spatial features in a node feature vector, and the size and number of convolution kernels are adjusted according to the feature dimension and actual needs. In addition, multi-layer convolution operations gradually extract high-level spatial features to enhance the richness and abstraction ability of feature representation.
[0047] Therefore, the local pattern and structural information in the node features can be effectively extracted through the convolution operation, improving the quality of feature representation. In addition, by gradually extracting higher-level features through multi-layer convolution stacking, complex feature relationships can be effectively captured.
[0048] The LSTM layer 230 is used to capture the temporal dependency of the target node feature vector to determine the corresponding second intermediate feature.
[0049] In some implementations, the LSTM layer is used to capture the temporal dependencies in the node feature vectors, which can memorize and process the long-range dependencies in the sequence data and adapt to dynamically changing construction data. In addition, by stacking multiple layers of LSTM, the ability to capture temporal features can be significantly improved, and the model's understanding of temporal information can be enhanced.
[0050] Therefore, the LSTM layer effectively captures the time dependency of node features, identifies dynamically changing risk factors during the construction process, and can adapt to the processing requirements of time series data such as construction schedules, thereby improving the model's ability to predict dynamic risks.
[0051] The attention layer 240 is used to weightedly fuse the first intermediate feature and the second intermediate feature based on the attention weight to determine the corresponding attention comprehensive feature.
[0052] In some embodiments, the features extracted by the convolution layer and the LSTM layer are weighted and fused through the attention mechanism, and the attention weights of different features are calculated to highlight important features and suppress minor features.
[0053] Therefore, the key features are highlighted through the attention mechanism, the importance of different features is comprehensively considered, the feature weights are dynamically adjusted, the information with the most predictive value is extracted, the effectiveness of feature fusion is improved, and the predictive performance of the model is enhanced.
[0054] The MLP layer 250 is used to extract the high-level feature representation corresponding to the comprehensive attention features.
[0055] Here, a multi-layer perceptron (MLP) is used to extract high-order features from comprehensive features, including multiple fully connected layers and activation functions (such as ReLU) for nonlinear transformation and feature combination. Thus, the MLP layer extracts high-order features through nonlinear transformation, captures complex feature relationships, and enhances the expressiveness of the model.
[0056] The output layer 260 is used to predict the construction risk prediction result corresponding to the high-order feature representation.
[0057] In some implementations, the output layer may adopt a fully connected layer and output the final construction risk prediction result according to the guidance of the activation function and the loss function.
[0058] Through the collaborative work of various parts of the above multimodal deep learning model, the model can integrate information from multiple data sources, capture complex spatial and temporal features, dynamically adjust feature weights, and ultimately output high-precision risk prediction results, achieving a comprehensive, accurate, and dynamic prediction of pavement construction risks. This improves the accuracy and foresight of risk prediction, provides strong technical support for construction management, and effectively reduces construction delays and cost increases caused by construction risks.
[0059] It should be noted that, in the present embodiment, the convolution layer 220 and the LSTM layer 230 adopt a parallel structure, which processes the original input features in parallel and can extract spatial and temporal features at the same time. Compared with the serial sequential structure, this parallel processing method enables the model to capture the complex relationships in the data more comprehensively. In addition, through the parallel structure, the features extracted by the convolution layer 220 and the LSTM layer 230 can be comprehensively fused in the attention layer 240, so that the model can better combine spatial and temporal features, and has a stronger feature fusion effect than the serial structure. In addition, by processing multiple features simultaneously through the convolution layer 220 and the LSTM layer 230, compared with the serial sequential structure, the model has higher robustness to noise and outliers in the data, which can effectively reduce the probability of false positives of risks.
[0060] In the practice of this application, combined with the characteristics of the construction risk prediction scenario of pavement engineering, we also made a series of enhanced designs for the graph attention network and construction risk prediction model to improve their advantages in this scenario. More details will be expanded in conjunction with other examples below.
[0061] In some examples of the embodiments of the present application, when constructing the road map structure, in order to further improve the expressive power of the model, the concept of edge connection strength can be introduced, which can be defined according to the node association degree, so as to more accurately reflect the relationship between different nodes. The edge connection strength can intuitively indicate the strength of the association between two nodes, and it can be comprehensively determined based on multiple factors, such as spatial location proximity, construction time overlap, and construction equipment sharing.
[0062] More specifically, each edge connection in the road graph structure is also provided with a corresponding edge connection strength, which is determined in the following manner:
[0063] S ij =[S spatial ,S time ,S equipment ], Formula (1)
[0064]
[0065] In the formula, S ijrepresents the edge connection strength between node i and node j; S equipment Indicates the degree of device sharing, S time represents the overlap of planned time, S spatial represents the spatial proximity of the road segment, α, β and γ represent the weight coefficients of the corresponding single relationship strength respectively; d ij Represents the geographical distance between the road sections corresponding to node i and node j; Overlap(T i ,T j ) represents the overlapped construction time between the sections corresponding to nodes i and j, T i and T j They represent the planned construction time intervals of the road sections corresponding to node i and node j respectively; Sh(i,j) represents the number of construction equipment shared by the road sections corresponding to node i and node j, and Tot(i) and Tot(j) represent the total number of construction equipment in the road sections corresponding to node i and node j respectively.
[0066] By calculating the spatial proximity, construction time overlap, and equipment sharing, and combining these factors to determine the edge connection strength, the relationship between different road sections can be more accurately reflected, so that the graph structure is not just a simple connection of nodes, but can more accurately express the complex relationship between nodes. In addition, through the above-mentioned further improved graph structure design, when the subsequent GAT updates the node features, it can more accurately consider the actual correlation between different nodes, thereby improving the accuracy of feature updates.
[0067] In some examples of the embodiments of the present application, when GAT updates the nodes in the graph structure, the feature vector h of each node may be updated. i and h j Perform linear transformation to obtain new feature representation. In addition, the trainable linear transformation matrices W1, W2, and W3 are used to process features of different dimensions. Then, for each dimension, the attention score e between node i and each neighbor node j is calculated. ij , the attention score combines the linear transformation result of the node feature and the weighted result of the correlation S. Then, the attention weight is obtained by performing Softmax normalization on the attention score of each dimension. Then, according to the weight coefficient β of different dimensions, the final attention weight α is comprehensively calculated ij Finally, using the integrated attention weight α ij , perform weighted summation of the features of neighboring nodes and update the feature h of node i i′ .
[0068] More specifically, the graph attention network is used to update the node features of each node in the road graph structure in the following way:
[0069] For each information dimension in the edge connection strength, the corresponding attention weight is calculated separately:
[0070]
[0071]
[0072] In the formula, and Respectively represent the attention scores between node i and node j for spatial proximity, construction time overlap, and equipment sharing degree; a spatial 、a time 、a equipment They represent trainable weight vectors respectively; W1, W2, W3 represent trainable linear transformation matrices respectively; h i and h j Represents the feature vectors of node i and node j respectively; || represents the feature concatenation operation; b spatial , b time and b equipment They represent the influence coefficient of the correlation degree respectively; and Respectively represent the attention weights between node i and node j in terms of spatial proximity, construction time overlap, and equipment sharing degree; Represents the set of neighbor nodes of node i.
[0073] The attention weights of each information dimension are combined to determine the comprehensive attention weight:
[0074]
[0075] In the formula, α ij represents the comprehensive attention weight between node i and node j; β spatial , β time and β equipment Represents the learnable weight coefficients of each information dimension.
[0076] Update the node features of each node according to the comprehensive attention weight:
[0077]
[0078] In the formula, h i′ represents the updated node feature of node i, σ represents the activation function, and W represents the trainable weight matrix.
[0079] Through the above operation mode, GAT can more accurately capture the multi-dimensional correlation between nodes, improve the accuracy of feature representation and the accuracy of construction risk prediction. By linearly transforming the feature vector of each node and combining the multi-dimensional correlation, the features of the node can be more accurately extracted and represented. The multi-dimensional feature representation is used to capture information such as spatial position, construction time and equipment sharing, which improves the expressiveness of node features. In addition, by calculating and synthesizing the multi-dimensional attention scores, the attention weights between nodes can be more accurately calculated, which can dynamically adjust the influence of different dimensions, so that the model can more comprehensively consider different factors when making risk predictions and improve the accuracy of predictions. In addition, GAT, which adopts multi-dimensional correlations, can maintain high robustness and generalization capabilities in complex construction scenarios, and can accurately identify and predict potential construction risks under the combined action of multiple factors, thereby improving the effectiveness of construction management.
[0080] Here, the enhanced design details of each local module in the construction risk prediction model will be expanded.
[0081] In some examples of the embodiments of the present application, the convolution layer is a multi-scale convolution layer, and the LSTM layer is a bidirectional LSTM layer.
[0082] More specifically, by using convolution kernels of different sizes (such as 3×3, 5×5, and 7×7), spatial features of different scales can be extracted. This multi-scale convolution can capture spatial patterns and local features of different sizes, making the model more expressive of spatial relationships in road segment data. For example, in pavement construction, the spatial features of different road segments may vary greatly, and these differences can be effectively extracted through multi-scale convolution layers, enabling the model to better identify potential construction risks. For example, for geographically adjacent road segments, smaller convolution kernels can capture subtle spatial features, while for road segments with larger spans, larger convolution kernels can extract more extensive spatial features.
[0083] In addition, through the bidirectional LSTM layer, the forward and backward time dependencies can be captured simultaneously, and richer time features can be extracted, so that the model can more comprehensively understand the time dependencies in the construction progress data, and enhance the processing ability of time series data. For example, in pavement engineering construction, the timing of construction progress and plans is crucial for risk prediction. The bidirectional LSTM layer can capture the time dependencies and changing trends in the construction process, helping to identify potential construction delays and resource scheduling problems. For example, for sections with overlapping construction times, the bidirectional LSTM layer can more accurately predict the risks caused by time conflicts.
[0084] Therefore, through the combination of multi-scale convolutional layers and bidirectional LSTM layers, the model can extract spatial and temporal features at the same time. The model can more comprehensively analyze and process multi-dimensional data, improve its adaptability to complex construction environments, more accurately predict construction risks, reduce false alarms and omissions, and enhance the robustness and reliability of the model in practical applications. For example, for sections of road with shared construction equipment, the model can comprehensively consider the spatial location, construction time, and equipment usage to provide a more accurate risk assessment.
[0085] In some examples of the embodiments of the present application, the attention layer adopts a multi-head attention layer with dynamic weight adjustment and adaptive feature fusion mechanism, which is used to weightedly fuse the first intermediate feature and the second intermediate feature in the following manner:
[0086] Calculate the query vector Q based on the first intermediate feature:
[0087] Q=W Q ·Fused(h)+b Q , Formula (13)
[0088] Where W Q and b Q They represent the weight matrix and bias term of the query vector respectively, and Fused(h) represents the first intermediate feature.
[0089] Calculate the key vector K based on the second intermediate feature:
[0090] K=W K ·h t +b K , Formula (14)
[0091] Where W K and b K They represent the weight matrix and bias term of the key vector respectively, and K represents the second intermediate feature.
[0092] Calculate the value vector V based on the second intermediate feature:
[0093] V=W V ·h t +b V , Formula (15)
[0094] Where W V and b V denote the weight matrix and bias term of the key vector respectively.
[0095] Calculate the dynamic weight of each attention head and combine the dynamic weights for multi-head attention calculation:
[0096] β i =softmax(W βConcat(Q,K,V)), Formula (15)
[0097]
[0098] In the formula, α i represents the attention weight of the i-th attention head; β i represents the dynamic weight of the i-th attention head; W β The weight matrix represents the dynamic weight, Concat(Q,K,V) represents the concatenation of Q, K and V; head i represents the output feature of the i-th attention head, They represent the query vector, key vector and value vector of the i-th attention head respectively; m represents the total number of attention heads.
[0099] The outputs of all attention heads are adaptively fused and linearly transformed to determine the corresponding comprehensive attention features:
[0100]
[0101] F A =σ(W A ·F e +b A ), Formula (18)
[0102] Where W A and b A denote the weight matrix and bias term of the linear transformation layer, m denotes the total number of attention heads, and F A Represents the comprehensive characteristics of attention.
[0103] In an embodiment of the present application, a dynamic weight adjustment mechanism is integrated to dynamically adjust the weights of each attention head according to different input features, so that each attention head can focus on different feature dimensions and the relationship between features, thereby achieving more accurate feature fusion. An adaptive feature fusion mechanism is integrated, and the features extracted by different attention heads can flexibly adjust the fusion method according to the changes in the input data, which can better capture the interactive relationship between different features, improve the feature expression ability of the model, and enable the model to more accurately understand and process multi-dimensional features in complex construction environments. Therefore, based on dynamic adjustment and adaptive fusion, the model can flexibly adjust the feature fusion method according to different input features. Through more accurate feature fusion and expression, the robustness and adaptability of the model in handling different construction scenarios are improved, false alarms and missed alarms are reduced, and the decision-making support capabilities of construction management are improved.
[0104] It should be noted that designing the MLP layer between the attention layer and the output layer can significantly improve the model's feature extraction and expression capabilities, thereby improving the accuracy and robustness of construction risk prediction. Specifically, the MLP layer can further process the comprehensive features of attention, extract high-order feature representations, and through multiple fully connected layers, the MLP layer can capture more complex and deep feature information and improve the richness of feature representation. In addition, through multi-layer fully connected networks, the MLP layer can enhance the generalization ability of the model and reduce overfitting.
[0105] In some examples of the embodiments of the present application, a comprehensive loss function consisting of classification loss, weighted loss and time series consistency loss is selected for the design of the loss function of the construction risk prediction model. Classification loss is used to measure the difference between the model prediction and the actual category, and is the basis of all classification tasks, which can ensure the basic accuracy of the overall prediction. Weighted loss can enhance the model's prediction ability for important categories by assigning different weights to different categories, especially high-risk categories. For example, in construction risk management, high-risk categories (such as geological disasters and major safety accidents) often require special attention. Weighted loss can ensure the model's sensitivity to these categories and plays an important role. In the construction process, temporal continuity and stability are very important, and temporal consistency loss ensures the model's prediction consistency in the time dimension, which can significantly improve the model's performance on time series data, and is a factor that cannot be ignored in actual construction scenarios.
[0106] More specifically, the loss function L of the construction risk prediction model is:
[0107] L=λ CE L CE +λ Weighted L Weighted +λ Temporal L Temporal , Formula (19)
[0108]
[0109] Where, L CE represents the classification loss term, L Weighted represents the weighted loss term, L Temporal represents the temporal consistency loss term; λ CE ,λ Weighted ,λ Temporal Respectively represent the weight coefficients of each loss item; N represents the number of samples, C represents the total number of risk categories, and y i,c represents the actual label of sample i in category c, It represents the predicted probability of sample i in category c; w cis the risk weight of category c, which is pre-set, and the risk weight of the high-risk category is greater than the risk weight of the low-risk category; T is the length of the time series, Represents the prediction result at time t.
[0110] Through the comprehensive loss function provided by the embodiment of the present application, the multi-dimensional requirements in construction risk management are fully considered, including classification accuracy, attention to high-risk categories, and continuity in the time dimension. As a result, the model can provide comprehensive and accurate risk prediction in a variety of construction scenarios, enhancing the adaptability and practicality of the model. In addition, by introducing temporal consistency loss, the model can handle the time dependence in the construction process, ensure the continuity and stability of the prediction results in time, thereby improving the robustness of the model, reducing the prediction instability caused by time series fluctuations, and enabling the model to maintain efficient and reliable prediction performance over a long period of time. In addition, the introduction of weighted loss enables the model to pay more attention to high-risk categories, thereby improving the ability to identify key risk events, ensuring that construction management personnel can obtain early warning information of high-risk events in a timely manner, take measures in advance to control risks, and improve the safety and management efficiency of the construction process. Through the embodiments of the present application, classification losses ensure basic classification accuracy, weighted losses increase attention to high-risk events, and time consistency losses improve the prediction stability of the time dimension, so that the model can adapt to complex and changeable construction environments, provide reliable risk predictions under a variety of construction conditions, adapt to different geological types, construction equipment conditions and weather conditions, and improve the application value of the model in actual construction.
[0111] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of actions combined, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.
[0112] Figure 3 A structural block diagram of an example of a construction risk management system for a pavement project according to an embodiment of the present application is shown.
[0113] like Figure 3 As shown, the construction risk management system 300 for pavement engineering includes a data acquisition unit 310 , a graph structure construction unit 320 , a node feature update unit 330 and a risk prediction unit 340 .
[0114] The data acquisition unit 310 is used to obtain the section data group of each section in the target road surface to be managed, and the section data group includes the section geological type, construction equipment data, planned construction time and weather forecast data corresponding to the section; the section geological type includes soil type information, geological structure information and soil layer thickness information; the construction equipment data includes equipment type information, equipment status information and equipment usage frequency; the planned construction time includes construction progress plan and construction stage time plan; the weather forecast data includes temperature information, humidity information, precipitation information and wind information corresponding to a preset time period in the future.
[0115] The graph structure construction unit 320 is used to construct a road surface graph structure based on the road segment data group corresponding to each road segment; the graph structure includes multiple nodes and edge connections, each of the nodes uniquely corresponds to a road segment, and the node characteristics of the node are defined by the corresponding road segment data group; the edge connection is defined based on the road segment association events between different nodes, and the road segment association events are used to indicate at least one of the following: road segments with adjacent spatial positions, road segments with overlapping planned construction times, and road segments sharing the same construction equipment.
[0116] The node feature updating unit 330 is used to process the road map structure based on the graph attention network to update the node features of each node in the road map structure.
[0117] The risk prediction unit 340 is used to input the updated pavement map structure into the construction risk prediction model, so that the construction risk prediction model predicts the corresponding pavement construction risk prediction result for each node, and the pavement construction risk prediction result includes the pavement construction risk type and risk probability.
[0118] The construction risk prediction model adopts a multimodal deep learning model, which includes a cascaded input layer, a convolutional layer, an LSTM layer, an attention layer, an MLP layer and an output layer;
[0119] The input layer is used to receive the updated target node feature vector of the corresponding target node;
[0120] The convolution layer is used to extract the spatial features corresponding to the target node feature vector through a convolution operation to determine the corresponding first intermediate features;
[0121] The LSTM layer is used to capture the temporal dependency of the target node feature vector to determine the corresponding second intermediate feature;
[0122] The attention layer is used to weightedly fuse the first intermediate feature and the second intermediate feature based on the attention weight to determine the corresponding attention comprehensive feature;
[0123] The MLP layer is used to extract the high-order feature representation corresponding to the comprehensive attention feature;
[0124] The output layer is used to predict the construction risk prediction result corresponding to the high-order feature representation.
[0125] In some embodiments, an embodiment of the present application provides a non-volatile computer-readable storage medium, in which one or more programs including execution instructions are stored. The execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute the steps of any of the above-mentioned pavement engineering construction risk management methods of the present application.
[0126] In some embodiments, the embodiments of the present application also provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the steps of any one of the above-mentioned pavement engineering construction risk management methods.
[0127] In some embodiments, the embodiments of the present application also provide an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the construction risk management method for pavement engineering.
[0128] Figure 4 is a schematic diagram of the hardware structure of an electronic device for executing a construction risk management method for a pavement project provided by another embodiment of the present application, such as Figure 4 As shown, the device includes:
[0129] One or more processors 410 and memory 420, Figure 4 A processor 410 is taken as an example.
[0130] The device for executing the construction risk management method for pavement engineering may further include: an input device 430 and an output device 440 .
[0131] The processor 410, the memory 420, the input device 430 and the output device 440 may be connected via a bus or other means. Figure 4 The example of connecting through bus is taken in the following.
[0132] The memory 420, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the construction risk management method for pavement engineering in the embodiment of the present application. The processor 410 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 420, that is, the construction risk management method for pavement engineering in the above method embodiment is realized.
[0133] The memory 420 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 420 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 420 may optionally include a memory remotely arranged relative to the processor 410, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0134] The input device 430 may receive input digital or character information and generate signals related to user settings and function control of the electronic device. The output device 440 may include a display device such as a display screen.
[0135] The one or more modules are stored in the memory 420, and when executed by the one or more processors 410, the construction risk management method for pavement engineering in any of the above method embodiments is executed.
[0136] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.
[0137] The electronic devices of the embodiments of the present application exist in various forms, including but not limited to:
[0138] (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.
[0139] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, etc.
[0140] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.
[0141] (4) Other onboard electronic devices with data interaction functions, such as on-board devices installed in vehicles.
[0142] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0143] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A construction risk management method for pavement engineering, comprising: Obtain a section data group for each section in the target road surface to be managed, wherein the section data group includes the section geological type, construction equipment data, planned construction time and weather forecast data corresponding to the section; the section geological type includes soil type information, geological structure information and soil layer thickness information; the construction equipment data includes equipment type information, equipment status information and equipment use frequency; the planned construction time includes a construction progress plan and a construction stage time plan; the weather forecast data includes temperature information, humidity information, precipitation information and wind force information corresponding to a preset time period in the future; Based on the road segment data group corresponding to each road segment, a road surface graph structure is constructed; the graph structure includes a plurality of nodes and edge connections, each of the nodes uniquely corresponds to a road segment, and the node characteristics of the nodes are defined by the corresponding road segment data group; The edge connection is defined according to the road segment association events between different nodes, wherein the road segment association events are used to indicate at least one of the following: road segments with adjacent spatial positions, road segments with overlapping planned construction times, and road segments sharing the same construction equipment; Processing the road graph structure based on a graph attention network to update node features of each node in the road graph structure; Inputting the updated pavement graph structure into the construction risk prediction model, so that the construction risk prediction model predicts the corresponding pavement construction risk prediction result for each node, wherein the pavement construction risk prediction result includes the pavement construction risk type and risk probability; The construction risk prediction model adopts a multimodal deep learning model, which includes an input layer, a convolutional layer, an LSTM layer, an attention layer, an MLP layer and an output layer; The input layer is used to receive the updated target node feature vector of the corresponding target node; The convolution layer is used to extract the spatial features corresponding to the target node feature vector through a convolution operation to determine the corresponding first intermediate features; The LSTM layer is used to capture the temporal dependency of the target node feature vector to determine the corresponding second intermediate feature; The attention layer is used to weightedly fuse the first intermediate feature and the second intermediate feature based on the attention weight to determine the corresponding attention comprehensive feature; The MLP layer is used to extract the high-order feature representation corresponding to the comprehensive attention feature; The output layer is used to predict the construction risk prediction result corresponding to the high-order feature representation.
2. The method according to claim 1, wherein: Each edge connection in the road map structure is also provided with a corresponding edge connection strength, which is determined in the following manner: S ij =[S spatial ,S time ,S equipment ] In the formula, S ij represents the edge connection strength between node i and node j; S equipment Indicates the degree of device sharing, S time represents the overlap of planned time, S spatial represents the spatial proximity of the road segment, α, β and γ represent the weight coefficients of the corresponding single relationship strength; d ij Represents the geographical distance between the road sections corresponding to node i and node j; Overlap(T i , T j ) represents the overlapped construction time between the sections corresponding to nodes i and j, T i and T j They represent the planned construction time intervals of the road sections corresponding to node i and node j respectively; Sh(i, j) represents the number of construction equipment shared by the road sections corresponding to node i and node j, and Tot(i) and Tot(j) represent the total number of construction equipment in the road sections corresponding to node i and node j respectively.
3. The method according to claim 2, wherein: The graph attention network is used to update the node features of each node in the road graph structure in the following way: For each information dimension in the edge connection strength, the corresponding attention weight is calculated separately: In the formula, and Respectively represent the attention scores between node i and node j for spatial proximity, construction time overlap, and equipment sharing degree; a spatial 、a time 、a equipment They represent trainable weight vectors respectively; W1, W2, W3 represent trainable linear transformation matrices respectively; h i and h j Represent the feature vectors of node i and node j respectively; || represents the feature concatenation operation; b spatial , b time and b equipment They represent the influence coefficient of the correlation degree respectively; and Respectively represent the attention weights between node i and node j in terms of spatial proximity, construction time overlap, and equipment sharing degree; Represents the set of neighbor nodes of node i; The attention weights of each information dimension are combined to determine the comprehensive attention weight: In the formula, α ij Represents the comprehensive attention weight between node i and node j; β spatial , β time and β equipment Represents the learnable weight coefficients of each information dimension; Update the node features of each node according to the comprehensive attention weight: In the formula, h i’ represents the updated node feature of node i, σ represents the activation function, and W represents the trainable weight matrix.
4. The method according to claim 1, wherein: The convolution layer is a multi-scale convolution layer, and the LSTM layer is a bidirectional LSTM layer.
5. The method according to claim 4, wherein: The attention layer adopts a multi-head attention layer with dynamic weight adjustment and adaptive feature fusion mechanism, which is used to weightedly fuse the first intermediate feature and the second intermediate feature in the following manner: Calculate the query vector Q based on the first intermediate feature: Q=W Q ·Fused(h)+b Q Where W Q and b Q They represent the weight matrix and bias term of the query vector respectively, and Fused(h) represents the first intermediate feature; Calculate the key vector K based on the second intermediate feature: K=W K ·h t +b K Where W K and b K They represent the weight matrix and bias term of the key vector respectively, and K represents the second intermediate feature; Calculate the value vector V based on the second intermediate feature: V=W V ·h t +b V Where W V and b V Represent the weight matrix and bias term of the key vector respectively; Calculate the dynamic weight of each attention head and combine the dynamic weights for multi-head attention calculation: β i =softmax(W β ·Concat(Q,K,V)) In the formula, α i represents the attention weight of the i-th attention head; β i represents the dynamic weight of the i-th attention head; W β The weight matrix represents the dynamic weight, Concat(Q, K, V) represents the concatenation of Q, K and V; head i represents the output feature of the i-th attention head, They represent the query vector, key vector and value vector of the i-th attention head respectively; m represents the total number of attention heads; The outputs of all attention heads are adaptively fused and linearly transformed to determine the corresponding comprehensive attention features: F A =σ(W A ·F e +b A ) Where W A and b A denote the weight matrix and bias term of the linear transformation layer, m denotes the total number of attention heads, and F A Represents the comprehensive characteristics of attention.
6. The method according to claim 5, wherein: The loss function L of the construction risk prediction model is: L=λ CE L CE +λ Weighted L Weighted +λ Temporal L Temporal Where, L CE represents the classification loss term, L Weighted represents the weighted loss term, L Temporal represents the temporal consistency loss term; λ CE ,λ Weighted ,λ Temporal Respectively represent the weight coefficients of each loss item; N represents the number of samples, C represents the total number of risk categories, and y i,c represents the actual label of sample i in category c, represents the predicted probability of sample i in category c; w c is the risk weight of category c, which is pre-set and the risk weight of the high-risk category is greater than the risk weight of the low-risk category; T is the length of the time series, Represents the prediction result at time t.
7. A construction risk management system for a pavement project, comprising: A data acquisition unit is used to acquire a road section data group of each road section in the target road surface to be managed, wherein the road section data group includes the road section geological type, construction equipment data, planned construction time and weather forecast data corresponding to the road section; the road section geological type includes soil type information, geological structure information and soil layer thickness information; the construction equipment data includes equipment type information, equipment status information and equipment use frequency; the planned construction time includes a construction progress plan and a construction stage time plan; the weather forecast data includes temperature information, humidity information, precipitation information and wind force information corresponding to a preset time period in the future; A graph structure construction unit, for constructing a road graph structure based on the road segment data group corresponding to each road segment; the graph structure comprises a plurality of nodes and edge connections, each of the nodes uniquely corresponds to a road segment, and the node feature of the node is defined by the corresponding road segment data group; The edge connection is defined according to the road segment association events between different nodes, wherein the road segment association events are used to indicate at least one of the following: road segments with adjacent spatial positions, road segments with overlapping planned construction times, and road segments sharing the same construction equipment; A node feature updating unit, configured to process the road map structure based on a graph attention network to update the node features of each node in the road map structure; A risk prediction unit, used for inputting the updated pavement map structure into the construction risk prediction model, so that the construction risk prediction model predicts the corresponding pavement construction risk prediction result for each node, wherein the pavement construction risk prediction result includes the pavement construction risk type and risk probability; The construction risk prediction model adopts a multimodal deep learning model, which includes a cascaded input layer, a convolutional layer, an LSTM layer, an attention layer, an MLP layer and an output layer; The input layer is used to receive the updated target node feature vector of the corresponding target node; The convolution layer is used to extract the spatial features corresponding to the target node feature vector through a convolution operation to determine the corresponding first intermediate features; The LSTM layer is used to capture the temporal dependency of the target node feature vector to determine the corresponding second intermediate feature; The attention layer is used to weightedly fuse the first intermediate feature and the second intermediate feature based on the attention weight to determine the corresponding attention comprehensive feature; The MLP layer is used to extract the high-order feature representation corresponding to the comprehensive attention feature; The output layer is used to predict the construction risk prediction result corresponding to the high-order feature representation.
