Intelligent decision-making method and system for post-disaster emergency repair of road infrastructure network

Abstract

The invention relates to an intelligent decision-making method and system for post-disaster emergency repair of a road infrastructure network. The method comprises the following steps: acquiring an aerial image of a post-disaster road infrastructure by using an unmanned aerial vehicle; features of the road infrastructure damage area are automatically identified and interpreted, and a multi-dimensional damage feature vector is generated; obtaining a first-aid repair team coordinate and existing first-aid repair resource information, calculating a spatial geometric distance between each damage area and the first-aid repair team, generating a distance matrix, performing normalization processing, and integrating the normalized distance matrix and the multi-dimensional damage feature vector into a three-dimensional tensor; the three-dimensional tensor is mapped into an importance degree matrix in a dimensionality reduction mode, an emergency dispatching scheme of a first-aid repair team is formulated through an optimization algorithm, and the first-aid repair team is assigned; and when the first-aid repair of any damaged area is completed, the steps are repeated until the first-aid repair of all the areas is completed. Compared with the prior art, according to the dynamic planning of the emergency repair real-time process, the optimal matching of the emergency repair task and the resource is ensured, and the resource use efficiency is maximized.

Description

Technical Field

[0001] The present invention relates to the field of emergency repair, and in particular to an intelligent decision-making method and system for post-disaster emergency repair of highway infrastructure networks. Background Art

[0002] Highway networks are infrastructure that carry economic and social development and people's daily travel, and are widely distributed everywhere. In recent years, extreme weather events have occurred frequently, such as natural disasters like heavy rain, floods, landslides, and debris flows, which have frequently damaged highway infrastructure. In particular, problems such as road section collapses, landslides, and blockages caused by slope disasters directly affect traffic flow and endanger people's lives and property. Against this background, the repair and restoration of post-disaster highway infrastructure have become a key task in emergency repair management.

[0003] Traditional methods for post-disaster damage assessment of highways rely on manual or inefficient satellite remote sensing data analysis. The transit cycle speed is slow and it is unable to accurately and finely quantify and reflect the complexity of the disaster situation, resulting in a lag in decision-making during the post-disaster repair process. It often takes weeks of communication and negotiation to determine the construction operation plan, which is not conducive to restoring highway traffic safety and normal order. In addition, due to the mesh distribution of highway networks, after a disaster occurs, multiple regional road sections may be damaged simultaneously, and the damage degrees are different, showing a situation of multiple disaster points occurring simultaneously and multiple disaster types coupling. Therefore, how to scientifically and efficiently survey and evaluate the disaster situations of damaged roads at each point and reasonably dispatch limited repair teams has become a key issue in improving the post-disaster traffic recovery effect and minimizing disaster losses.

[0004] Chinese Patent Application CN117746625A discloses an emergency disposal system for highway emergencies, including a highway monitoring system, an information processing module, a command and dispatch module, an execution unit, a drone rapid response system, an alarm system, and a database module. The emergency disposal system for highway emergencies can quickly detect emergencies through the highway monitoring system and transmit them to the information processing module for analysis. The command and dispatch module conducts command and control based on the results analyzed by the information processing module to make the alarm system, the execution unit, and the drone rapid response system respond. Finally, the database module collects and collates the processing processes of the alarm system, the drone rapid response system, and the execution unit. This patent application proposes to use drones for the handling of emergencies, but only mentions this operation, and there is no description of actual emergency disposal planning and repair operations. There is still a large blank in the field of using drones for post-disaster disposal.

[0005] Based on this, developing an efficient post-disaster emergency repair decision-making system for highways, combining intelligent artificial intelligence technology to quickly evaluate the disaster situation and optimize the repair plan, has become an urgent issue to be solved. Summary of the Invention

[0006] The object of the present invention is to overcome the defects of the above-mentioned existing technologies and provide an intelligent decision-making method and system for post-disaster emergency repair of highway infrastructure networks, which can achieve rapid assessment of post-disaster highway facilities and optimal scheduling of repair resources.

[0007] The object of the present invention can be achieved by the following technical solutions:

[0008] An intelligent decision-making method for post-disaster emergency repair of highway infrastructure networks, the method comprising:

[0009] Step 1, using an unmanned aerial vehicle to collect aerial images of post-disaster highway infrastructure;

[0010] Step 2, based on the encoder-decoder model in deep learning, automatically identifying and interpreting the features of the damaged areas of highway infrastructure in the aerial images, and using the road damage investigation and interpretation model to generate multi-dimensional damage feature vectors;

[0011] Step 3, obtaining the coordinates of the repair teams and the information of existing repair resources, calculating the spatial geometric distances between each damaged area and the repair teams, generating a distance matrix and performing normalization processing;

[0012] Step 4, integrating the normalized distance matrix and the multi-dimensional damage feature vectors into a three-dimensional tensor;

[0013] Step 5, combining the features of different damaged areas of highway infrastructure, the distances between the repair teams, and the information of existing repair resources, reducing the dimension of the three-dimensional tensor to map it into an importance matrix, and using an optimization algorithm to formulate an emergency scheduling plan for the repair teams and assign the repair teams.

[0014] Furthermore, the process of obtaining the features of the damaged areas of the highway infrastructure includes:

[0015] Based on the encoder-decoder model, performing layer-by-layer feature extraction of high-dimensional features on the damaged areas of highway infrastructure in the aerial images, pooling operations to extract features of different scales, restoring the high-dimensional features to an image and reconstructing the boundaries, shapes, and details of the damaged areas, automatically identifying and extracting the features of each damage type in the damaged areas of highway infrastructure in the aerial images; combining the features of the damage types, parsing the semantic information in the aerial images, and automatically quantifying and interpreting the types, ranges, and geometric characteristics of each damage in the damaged areas of highway infrastructure in the aerial images, and outputting them as the features of the damaged areas of the highway infrastructure.

[0016] Even further, the specific process of the road damage investigation and interpretation model generating the multi-dimensional damage feature vectors includes:

[0017] Obtain the characteristics of the damaged area of the highway infrastructure. According to the preset minimum division unit value, divide the damaged area of the highway infrastructure into units, and divide the damaged area of the highway infrastructure into multiple damaged units; respectively convert each damaged unit into a structured state vector, where the state vector contains the characteristics and their parameters of the damaged area of the highway infrastructure in the corresponding damaged unit; combine the state vectors, and classify the severity of the damage of the damaged unit according to the preset damage assessment criteria; combine the severity of the damage of the damaged unit with the characteristics of the damaged area of the highway infrastructure, convert it into a structured state vector, and output it as a multi-dimensional damage feature vector.

[0018] Furthermore, a polarization attention module and an efficient channel attention module are embedded in the network structure of the encoder-decoder model.

[0019] Furthermore, the polarization attention module performs convolution operations on the aerial image through convolution kernels of different sizes, obtains feature maps of different scales, and splices or weights and fuses the feature maps to form a feature representation containing multi-scale information, improving the recognition ability of the encoder-decoder model for aerial images under complex backgrounds; the efficient channel attention module calculates the kernel size of one-dimensional convolution according to the number of channels of the aerial image, and applies the one-dimensional convolution to aerial image recognition, improving the performance of the encoder-decoder model and reducing the computational complexity.

[0020] Furthermore, the training process of the encoder-decoder model includes:

[0021] Obtain the highway scene and its disaster damage feature annotation dataset;

[0022] Standardize and enhance the dataset;

[0023] Input the dataset into the encoder-decoder model for training;

[0024] Introduce the Dice Loss function in the training to optimize the overlap measure of the segmentation result and retain the cross-entropy loss and IoU loss;

[0025] Use the gradient backpropagation algorithm to continuously update the model weights until the encoder-decoder model is trained;

[0026] Evaluate the performance of the trained encoder-decoder model and further optimize the encoder-decoder model.

[0027] Furthermore, the optimized encoder-decoder model is loaded onto the edge computing platform of the drone, and the computational complexity of identifying the aerial image is reduced through model quantization and pruning techniques, and a lightweight inference framework is used to adapt to the device environment.

[0028] Further, the multi-dimensional damage feature vector includes the damage type, the parameters corresponding to the damage type, and the damage degree, where

[0029] the damage type includes subgrade damage, pavement damage, or bridge and tunnel damage;

[0030] the parameters corresponding to the damage type include the collapse area, crack length, width, the position of the center of the damaged section, and the specific value of the buried area of the highway by the landslide;

[0031] the damage degree includes slight, moderate, or severe.

[0032] Further, when the emergency repair of any damaged area is completed, repeat the above steps 1 - 5, re-formulate the emergency dispatch plan for the repair team, and conduct a new round of assignment until the emergency repair of all areas is completed.

[0033] An intelligent decision-making system for post-disaster emergency repair of highway infrastructure network, the system includes:

[0034] A road network facility disaster situation acquisition module, which uses unmanned aerial vehicles to collect aerial images of post-disaster highway infrastructure;

[0035] A highway facility disaster situation automatic interpretation module, which automatically identifies and interprets the characteristics of the damaged areas of highway infrastructure in the aerial images based on the encoder-decoder model in deep learning;

[0036] A facility damage condition vectorization module, which generates a multi-dimensional damage feature vector based on the road damage investigation and interpretation model and combines the characteristics of the damaged areas of the highway infrastructure;

[0037] A repair path space normalization module, which obtains the coordinates of the repair team and the existing repair resource information, calculates the spatial geometric distance between each damaged area and the repair team, generates a distance matrix and conducts normalization processing;

[0038] A state mapping module for feature fusion, which integrates the normalized distance matrix and the multi-dimensional damage feature vector into a three-dimensional tensor;

[0039] A highway facility intelligent repair decision-making module, which combines the characteristics of different damaged areas of highway infrastructure, the distances between repair teams, and the existing repair resource information, reduces the three-dimensional tensor to a importance matrix, and uses an optimization algorithm to formulate an emergency dispatch plan for the repair team and assign the repair team.

[0040] Compared with the prior art, the beneficial effects of the present invention include:

[0041] 1. The present invention utilizes drones to collect high - resolution aerial images of post - disaster highway infrastructure in real - time. Combining with the encoder - decoder model in the deep learning network, it can quickly and accurately identify the damage types of highway facilities and quantify the damage degree, and make post - disaster emergency repair decisions, avoiding the lag and blindness of traditional decisions, and significantly improving the scientificity and timeliness of emergency repair decisions. In the case where multiple road sections may be damaged simultaneously, with different degrees of damage, presenting a situation of concurrent multi - disaster points and coupled multi - disaster types, the present invention can establish and integrate a distance matrix and a multi - dimensional damage feature vector, combine spatial distance information with damage features, and then, combining the characteristics of damaged areas of different highway infrastructure, the distances between various emergency repair teams, and the existing emergency repair resource information, use an optimization algorithm to accurately mobilize resources, reasonably dispatch limited emergency repair teams, improve the post - disaster traffic recovery effect and resource utilization rate, minimize disaster losses, and also reduce the emergency repair cost;

[0042] 2. The present invention uses drones to collect aerial images in real - time. To avoid the problems of complex backgrounds and redundant features that may occur in the images collected by drones, a polarization attention module and an efficient channel attention module are embedded in the encoder - decoder model, enhancing the model's ability to identify damaged areas under complex backgrounds, improving the generalization ability and recognition accuracy of the interpretation model, making the disaster situation assessment more comprehensive and detailed, and making the formulation of the emergency repair team's emergency dispatch plan more reasonable;

[0043] 3. When training the encoder - decoder model, for the actual application scenario, the Dice Loss function is introduced for training, making the model pay more attention to the overlapping areas of the segmentation results, effectively improving the model's performance in dealing with unbalanced data sets. After training with a diverse highway scene data set, the model can finally have the generalization ability in different highway scenes, be able to accurately extract multi - scale semantic features of highway disaster - damaged areas, be more adaptable to damage detection on complex post - disaster highways, and ensure that the identification of damage types and quantification of damage degree can be accurate and error - free;

[0044] 4. In the present invention, the encoder - decoder model is loaded onto the edge computing platform of the drone. While acquiring the aerial images of post - disaster highway infrastructure, it can also process the images, quickly and accurately identify the damage types of highway facilities and quantify the damage degree, further improving the efficiency of formulating post - disaster emergency repair decisions;

[0045] 5. In the emergency repair decision - making of the present invention, it can dynamically plan according to the real - time progress of the emergency repair, ensure the best match between the emergency repair tasks and resources, maximize the resource utilization efficiency, avoid inefficient emergency repair resource allocation, reduce the repair time at the same time, and reduce the post - disaster recovery cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1This is the flowchart of the method of the present invention;

[0047] Figure 2 This is the block diagram of the system structure of the present invention. Specific embodiments

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] Embodiment 1

[0050] This embodiment aims to disclose an intelligent decision-making method for post-disaster emergency repair of highway infrastructure networks. The method steps are as Figure 1 shown, and the specific step process is as follows:

[0051] S1. Use an unmanned aerial vehicle to collect aerial images of post-disaster highway infrastructure.

[0052] S2. Based on the encoder-decoder model in deep learning, automatically identify and interpret the features of the damaged areas of highway infrastructure in the aerial images, and use the road damage investigation and interpretation model to generate multi-dimensional damage feature vectors.

[0053] The process of obtaining the features of the damaged areas of highway infrastructure includes:

[0054] Based on the encoder-decoder model, perform layer-by-layer feature extraction of high-dimensional features on the damaged areas of highway infrastructure in the aerial images, perform pooling operations to extract features of different scales, restore the high-dimensional features to the image, and reconstruct the boundaries, shapes, and details of the damaged areas, and automatically identify and extract the features of each damage type in the damaged areas of highway infrastructure in the aerial images;

[0055] Combine the features of the damage types, analyze the semantic information in the aerial images, automatically quantify and interpret the types, ranges, and geometric characteristics of each damage in the damaged areas of highway infrastructure in the aerial images, and output them as the features of the damaged areas of highway infrastructure.

[0056] Specifically, in the encoder-decoder model, the encoder part extracts features layer by layer from the aerial images of damaged sections through a multi-layer convolutional neural network, converting the local and global information in the images into high-level feature representations. These features can capture the structural and functional features of different regions in the images, including but not limited to damage types such as crack width, roadbed collapse, and buried area. At the same time, the encoder reduces the spatial dimension of the image through pooling operations, extracting features at different scales, so that the network can understand damaged regions of different sizes and shapes. Then, the decoder part gradually restores the high-dimensional features extracted by the encoder into an output image with spatial resolution through deconvolution layers, accurately reconstructing the boundaries, shapes, and details of the damaged regions to form a high-precision segmentation map of the damaged regions. The decoder can combine the global and local features provided by the encoder, effectively process the complex background and noise in the images, accurately identify the damaged regions, and eliminate misjudgments. Through this encoding-decoding process, various damaged regions in the images can be automatically identified and segmented, and the recognition results include not only the damage types but also specific damage features, such as key parameters like the collapsed area, crack length, and landslide buried area.

[0057] It should be noted that due to the complex background of UAV images, there are often a large amount of redundant information unrelated to landslide bodies in the feature space, which may lead to local information loss or noise interference. To solve this problem, a polarization attention module and an efficient channel attention module are embedded in the network structure of the encoder-decoder model.

[0058] The polarization attention module performs convolution operations on aerial images through convolutional kernels of different sizes, obtains feature maps at different scales, and splices or weights and fuses the feature maps to form a feature representation containing multi-scale information, enhancing the recognition ability of the encoder-decoder model for aerial images under complex backgrounds; the efficient channel attention module calculates the kernel size of one-dimensional convolution according to the number of channels of the aerial images and applies one-dimensional convolution to aerial image recognition, which can improve the performance of the encoder-decoder model and reduce the computational complexity.

[0059] The training process of the encoder-decoder model includes:

[0060] Obtain a dataset of highway scenes and their disaster damage feature annotations;

[0061] Normalize and enhance the dataset;

[0062] Input the dataset into the encoder-decoder model for training;

[0063] Introduce the Dice Loss function during training to optimize the overlap metric of the segmentation results and retain the cross-entropy loss and IoU loss;

[0064] The gradient backpropagation algorithm is used to continuously update the model weights until the encoder-decoder model is trained;

[0065] The performance of the trained encoder-decoder model is evaluated, and the encoder-decoder model is further optimized.

[0066] The specific optimization process includes: after training, the performance of the model on the validation set is evaluated. By calculating metrics such as accuracy, recall, and F1 value, it is ensured that the model has good generalization ability in different highway scenarios. If the model performance does not meet the expectations, the training parameters are adjusted, the training data is increased, or the model structure is improved, and then training and evaluation are carried out again until the model performance meets the requirements.

[0067] The optimized encoder-decoder model is loaded onto the edge computing platform of the drone. Through model quantization and pruning techniques, the computational complexity of identifying aerial images is reduced, and a lightweight inference framework is used to adapt to the device environment. The drone can then process the images while collecting highway images and accurately identify the damage types of highway facilities and quantify the damage degree.

[0068] In another embodiment, based on the encoder-decoder structure in deep learning, the features of the highway infrastructure damage area in the aerial image are extracted. Specifically, the semantic segmentation model is used to extract the features of the damage types in the image, and the weight model is used to analyze the semantic information of the damaged aerial image, automatically quantifying and interpreting the type, scope, and geometric characteristics of the damage, and outputting them as the features of the highway infrastructure damage area.

[0069] The specific process of the road damage investigation and interpretation model generating a multi-dimensional damage feature vector includes:

[0070] Obtain the features of the highway infrastructure damage area. According to the preset minimum division unit value, the highway infrastructure damage area is divided into units, and the highway infrastructure damage area is divided into multiple damage units; each damage unit is respectively transformed into a structured state vector, and the state vector contains the features and their parameters of the highway infrastructure damage area in the corresponding damage unit; combined with the state vector, the damage severity of the damage unit is graded according to the preset damage assessment criteria;

[0071] The damage severity of the damage unit is combined with the features of the highway infrastructure damage area, transformed into a structured state vector, and output as a multi-dimensional damage feature vector.

[0072] In this embodiment, the road damage investigation and interpretation model is trained with a large amount of highway engineering expert annotation data.

[0073] The multi-dimensional damage feature vector includes the damage type, the parameters corresponding to the damage type, and the damage degree, where,

[0074] The types of damage include subgrade damage, pavement damage or bridge-tunnel damage;

[0075] The parameters corresponding to the damage type include the collapse area, crack length, width, the position of the center of the damaged section, and the specific value of the buried area of the highway by the landslide;

[0076] The degree of damage includes slight, moderate or severe.

[0077] The state vector can be expressed as: S=(T1, T2, T3, P1, P2, P3, M, L, K, W, Y), where: T1, T2 and T3 represent the damage type codes, which are subgrade damage, pavement damage and bridge-tunnel damage respectively. When such damage is detected in the damaged unit, the value is 1, otherwise the value is 0; P1, P2, P3, M, L, K, W and Y respectively represent the attribute values of the subclasses under the major damage type category, that is, the parameters corresponding to the damage type, specifically the collapse area, crack length, width, the position of the center of the damaged section, and the specific value of the buried area of the highway by the landslide.

[0078] The rating of the damage degree is represented by G. This characterization index is divided into 3 levels. Among them, slight means that the road surface is buried and there are slight cracks, without structural damage, and vehicles can pass at low speed; moderate means that there are large road surface cracks and small-scale collapses, with local structural damage, and local passage requires traffic control; severe means that there are large-scale subgrade collapses and slope slides and other structural damages, and all types of damages are serious, and vehicles cannot pass. The basis for the division between the damage degree levels can be determined by the local emergency repair command leading group according to the implementation of the emergency plan. This level provides data support for the subsequent optimization of emergency repair decisions and resource allocation.

[0079] S3. Obtain the coordinates of the emergency repair team and the existing emergency repair resource information, calculate the spatial geometric distance between each damaged area and the emergency repair team, generate a distance matrix and perform normalization processing.

[0080] For example, assuming there are m damaged points and n emergency repair team stations, the system will calculate the Euclidean distance between each damaged point and each emergency repair team station to form an m×n distance matrix D, where D ij represents the distance between the i-th damaged point and the j-th emergency repair team station, which can be calculated by the Euclidean distance or other geographical distance measurement methods: D ij =sqrt[(x i -x j ) 2 +(y i -y j ) 2. Then, the system normalizes the distance matrix to ensure that the distance data is within a certain numerical range, preventing large spatial distance data from interfering with the calculation of the subsequent importance matrix and facilitating comparative analysis.

[0081] S4. Integrate the normalized distance matrix and the multi-dimensional damage feature vector into a three-dimensional tensor.

[0082] The normalized distance matrix and the damage status vector are integrated into a three-dimensional tensor X with the shape of m×n×d. Here, m represents the number of damage points, n represents the number of repair teams, and d represents the feature dimensions of each damage point, including spatial distance, damage degree, repair resources, and traffic importance. This fusion method combines spatial distance information with damage features, providing comprehensive data support for subsequent intelligent decision-making. Each combination between a damage point and each repair team corresponds to a d-dimensional feature vector, containing the attributes of the damage point and the distance information with a specific repair team.

[0083] S5. Combine the characteristics of different highway infrastructure damage areas, the distances between repair teams, and the existing repair resource information, reduce the dimension of the three-dimensional tensor to map it into an importance matrix, and use an optimization algorithm to formulate an emergency dispatch plan for repair teams and assign repair teams.

[0084] Step S5 combines the characteristics of different highway infrastructure damage areas, the distances between repair teams, and the existing repair resource information, reduces the dimension of the three-dimensional tensor X to map it into an importance matrix C, and determines which repair team should be assigned to each damage point according to the minimum cost (or maximum benefit) in the matrix, that is, applies the Hungarian algorithm to solve the optimal matching.

[0085] According to the importance matrix C, apply the Hungarian algorithm to solve the optimal matching. The core of the Hungarian algorithm is to find a set of independent zero elements in the matrix, so that the rows and columns where these zero elements are located are different, and their sum (or cost) reaches the minimum (or maximum benefit). In this embodiment, it is to find the most suitable repair team for each damage point to optimize the overall repair benefit.

[0086] The dimensionality reduction of the fused three-dimensional tensor X is adopted to simplify the calculation and optimize the model, reducing the computational complexity and memory occupancy to ensure efficient operation. The weight w in the importance matrix C iThe determination is made using the analytic hierarchy process (AHP). The benefit is that decision makers can determine the benefit objectives of emergency repairs based on actual conditions, such as the minimum emergency repair cost, the shortest time, or the most repaired sections, with sufficient decision-making flexibility. The Hungarian algorithm has low complexity and is very efficient for small and medium-sized problems. It can quickly obtain the optimal solution and ensure the best match between emergency repair tasks and resources, thereby maximizing the efficiency of resource use, reducing repair time, optimizing scheduling sequence, and improving emergency repair benefits during post-disaster recovery.

[0087] S6: When the emergency repair of any damaged area is completed, repeat the above steps, re-formulate the emergency dispatch plan for the repair team, and conduct a new round of assignments until all areas are repaired.

[0088] When one or several damaged road sections are repaired, all elements of the importance matrix row corresponding to the damaged point become positive infinity or negative infinity, and the latest location of the repair team, the repair force replenishment, and the location of the newly appeared damaged road section are updated. The above steps are repeated to generate a new importance matrix C, and a new round of assignments is continued. In this way, the disaster repair information is updated in a timely and dynamic manner, and the repair efficiency is maximized.

[0089] Below, assuming that a practical application is carried out, a specific application example of the decision-making process in step S5 of the method proposed in this embodiment is given.

[0090] The image acquisition and image processing in the early stage went smoothly. The importance matrix of this emergency repair decision is as follows: C ij =w1D ij +βw2E ij +w3G i +W4V i , where w i is the weight, D ij is the normalized distance between the damage point and the repair team. The closer it is to 1, the closer it is to the repair team (dimensionless value); E ij G is the estimated time (in hours) required to repair damaged point i by the repair team j based on the damage status vector, repair equipment and materials, etc. β is the scaling factor of the repair time. To avoid the extreme value effect of a large repair time on the importance calculation, the inverse of the minimum value can be used to repair the road sections that are easy to repair first; i The severity of the damage to the repair point is divided into slight, moderate and severe damage, usually represented by 1, 3 and 5 respectively; V i Indicates the traffic importance of the road section where the damage point is located. Scoring is based on road type, with expressways, trunk roads and rural roads being given scores of 5, 3 and 1 respectively.

[0091] The process of obtaining each parameter of the importance matrix is as follows:

[0092] First, the AHP method is used to determine the weights w in the importance matrix. i .

[0093] As a subjective assignment evaluation method, the AHP method decomposes complex decision-making problems into elements at different levels by constructing a hierarchical structure model and uses pairwise weight assignment to evaluate the importance of these elements. In this embodiment, the relative importance of each factor is evaluated through scoring and discussion by senior experts in the field of highway emergency.

[0094] The basic steps for evaluating the relative importance of each factor include the following aspects:

[0095] 1. Construct a hierarchical structure model: Determine the goal layer, criterion layer (factor layer), and solution layer.

[0096] 2. Construct a judgment matrix: Score the relative importance between each factor through expert judgment.

[0097] 3. Calculate the weights: Calculate the weight of each factor through methods such as consistency test and normalization.

[0098] 4. Comprehensive evaluation: Apply the obtained weights to subsequent decision-making.

[0099] In this embodiment, the goal layer is to determine the importance value of each element in the importance matrix, and the criterion layer consists of four factors: D ij , G i , E ij , V i . The solution layer represents the weights of each factor. According to the requirements of the analytic hierarchy process, experts will score the importance between each pair of criteria.

[0100] The scoring criteria are as follows: 1 indicates that two criteria are equally important; 3 indicates that one criterion is slightly more important than the other; 5 indicates that one criterion is significantly more important than the other; 7 indicates that one criterion is very much more important than the other; 9 indicates that one criterion is extremely more important than the other; 2, 4, 6, and 8 are the median values between the above pairwise judgments. The specific scoring results are shown in Table 1.

[0101] Table 1 Index

[0102] Comparison criterion <![CDATA[D ij > <![CDATA[E ij > <![CDATA[G i > <![CDATA[V i > <![CDATA[D ij > 1 1 / 3 1 / 3 1 / 5 <![CDATA[E ij > 3 1 1 1 / 3 <![CDATA[G i > 3 1 1 1 / 2 <![CDATA[V i > 5 3 2 1

[0103] The scoring results are shown in Table 2.

[0104] Table 2 AHP Hierarchical Analysis Results

[0105]

[0106] D ij , Gi , E ij , V i The importance of the four factors in the decision-making system is quantitatively reflected by the magnitudes of the eigenvectors and weight values in Table 2.

[0107] V i The eigenvector of the factor is 1.933, and the corresponding weight value is as high as 48.313%, indicating that the traffic importance of the section where the damage point is located occupies an absolute dominant position in the evaluation system.

[0108] G i The eigenvector of the factor is 0.914, and the weight value is 22.854%, which is an important factor in the evaluation system.

[0109] E ij The eigenvector of the factor is 0.832, and the weight value is 20.805%. In contrast, the weight value of the D ij factor is relatively low, at 8.028%, and the eigenvector is 0.321, indicating that the importance of this factor in decision-making is relatively low.

[0110] After rounding each weight coefficient, it becomes the weight value w in the importance matrix i , so the final weight coefficients are w1 = 8, w2 = 21, w3 = 23, w4 = 48.

[0111] Suppose there are currently m = 3 damage points and n = 2 repair teams.

[0112] Normalized distances between damage points and repair teams:

[0113]

[0114] Repair duration (assuming the unit is hours):

[0115]

[0116] Severity of damage points: G i = [5 3 1]

[0117] Traffic importance: V i = [5 3 1]

[0118] The scaling factor β is 1 / 8, and the final importance matrix is calculated as:

[0119]

[0120] Apply the Hungarian algorithm to solve the optimal matching and select the plan with the maximum benefit.

[0121] Conclusion: Sort each row in the matrix and select the plan with the greatest benefit. For damage point 1, assign repair team 2 (expected to be repaired in 15 hours). For disaster-damaged point 2, arrange team 1 for repair (expected to be repaired in 10 hours). For the remaining and newly added disaster-damaged points, reassign according to the new importance matrix.

[0122] The specific assignment and solution process using the Hungarian algorithm is as follows:

[0123] Since the Hungarian algorithm requires a square matrix, it needs to be extended to a 3×3 matrix. Add a virtual repair team (designated as team 3), and set its cost to a maximum value (indicating infeasibility):

[0124]

[0125] Row transformation: Find the minimum value in each row of the cost matrix, and subtract the minimum value from each element in the row to make at least one 0 appear in each row. The minimum value in the first row is 390.5, the minimum value in the second row is 241.65, and the minimum value in the third row is 97.6.

[0126] After row transformation, the new matrix is obtained:

[0127]

[0128] Column transformation: Find the minimum value in each column of the transformed matrix, and subtract the minimum value from each element in the column. The minimum value in the first column is 0, and the minimum value in the second column is 7.65.

[0129] After column transformation, the matrix is obtained:

[0130]

[0131] Cover all zero elements: Cover all zeros in the matrix with the fewest lines. Column 1 and row 3 can cover all zeros, and 2 lines are required. The number of lines required to cover the zero elements is less than the dimension 3 of the matrix, so the matrix needs to be adjusted.

[0132] Adjust the matrix: Find the smallest uncovered element 2.625, subtract 2.625 from all uncovered elements, and add 2.625 to the intersection element (row 3 and column 2).

[0133] Updated matrix:

[0134]

[0135] Re - covering zero elements: Cover all zeros with 3 lines (row 1, row 3, column 1). The number of lines is equal to the matrix dimension. Find the optimal match. Damage point 1 is assigned to Team 2 (corresponding to 400.775 in the original matrix, reduced to 0), damage point 2 is assigned to Team 1 (corresponding to 241.65, reduced to 0), and damage point 3 is assigned to Team 2 (corresponding to 105.25, reduced to 0).

[0136] After Team 2 is assigned to damage point 1, its position is updated. The remaining damage point 3 needs to recalculate the path, generate a new cost matrix, and repeat the above steps.

[0137] By fully considering the spatial distribution relationship between post - disaster highway damage and emergency repair resources, it is possible to reasonably plan and allocate emergency repair paths and tasks, and fully improve the efficiency and effect of post - disaster emergency repair and restoration work.

[0138] Embodiment 2

[0139] An intelligent decision - making system for post - disaster emergency repair of highway infrastructure network. The system is as Figure 2 shown, including:

[0140] The road network facility disaster situation collection module 310 uses unmanned aerial vehicles to collect aerial images of post - disaster highway infrastructure;

[0141] The highway facility disaster situation automatic interpretation module 320 automatically identifies and interprets the characteristics of damaged areas of highway infrastructure in aerial images based on the encoder - decoder model in deep learning;

[0142] The facility damage status vectorization module 330 generates multi - dimensional damage feature vectors based on the road damage investigation and interpretation model, combined with the characteristics of damaged areas of highway infrastructure;

[0143] The emergency repair path spatial normalization module 340 obtains the coordinates of emergency repair teams and existing emergency repair resource information, calculates the spatial geometric distance between each damaged area and the emergency repair teams, generates a distance matrix, and performs normalization processing;

[0144] The state mapping module 350 for feature fusion integrates the normalized distance matrix and multi - dimensional damage feature vectors into a three - dimensional tensor;

[0145] The intelligent emergency repair decision - making module 360 for highway facilities combines the characteristics of different damaged areas of highway infrastructure, the distances between emergency repair teams, and existing emergency repair resource information, reduces the three - dimensional tensor to a importance matrix through dimensionality reduction mapping, and uses an optimization algorithm to formulate an emergency dispatch plan for emergency repair teams and assign emergency repair teams.

[0146] The specific details of the above modules can be understood by referring to the relevant descriptions and effects in Embodiment 1.

[0147] Embodiment 3

[0148] Based on Embodiment 1, this embodiment provides an electronic device, including: one or more processors and a memory. One or more programs are stored in the memory, and the one or more programs include instructions for executing the intelligent decision-making method for post-disaster emergency repair of highway infrastructure network as described above.

[0149] At the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include other hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the above-mentioned intelligent decision-making method for post-disaster emergency repair of highway infrastructure network. Of course, in addition to the software implementation method, the present invention does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but may also be hardware or logic devices.

[0150] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of, for example, read-only memory (ROM) or flash memory (flash RAM). The memory is an example of a computer-readable medium.

[0151] Computer-readable media include permanent and non-permanent, removable and non-removable media and can store information by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0152] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. An intelligent decision-making method for post-disaster emergency repair of highway infrastructure network, characterized in that The method includes the following steps: Step 1: Use an unmanned aerial vehicle to collect aerial images of post-disaster highway infrastructure; Step 2: Based on the encoder-decoder model in deep learning, automatically identify and interpret the features of the damaged areas of highway infrastructure in the aerial images, and use the road damage survey and interpretation model to generate multi-dimensional damage feature vectors; Step 3: Obtain the coordinates of the emergency repair teams and the information of existing emergency repair resources, calculate the spatial geometric distances between each damaged area and the emergency repair teams, generate a distance matrix and perform normalization processing; Step 4: Integrate the normalized distance matrix and the multi-dimensional damage feature vectors into a three-dimensional tensor; Step 5: Combine the features of different damaged areas of highway infrastructure, the distances between emergency repair teams, and the information of existing emergency repair resources, map the three-dimensional tensor to an importance matrix by dimensionality reduction, and use an optimization algorithm to formulate an emergency dispatch plan for emergency repair teams and assign emergency repair teams.

2. The intelligent decision-making method for post-disaster emergency repair of a highway infrastructure network according to claim 1, wherein The process of obtaining the features of the damaged areas of the highway infrastructure includes: Based on the encoder-decoder model, perform layer-by-layer feature extraction of high-dimensional features on the damaged areas of highway infrastructure in the aerial images, perform pooling operations to extract features of different scales, restore the high-dimensional features to images, and reconstruct the boundaries, shapes, and details of the damaged areas, and automatically identify and extract the features of each damage type in the damaged areas of highway infrastructure in the aerial images; combine the features of the damage types, analyze the semantic information in the aerial images, and automatically quantify and interpret the types, ranges, and geometric characteristics of each damage in the damaged areas of highway infrastructure in the aerial images, and output them as the features of the damaged areas of the highway infrastructure.

3. The intelligent decision-making method for post-disaster emergency repair of a highway infrastructure network according to claim 2, wherein The specific process of the road damage survey and interpretation model generating the multi-dimensional damage feature vectors includes: Obtain the features of the damaged areas of highway infrastructure, divide the damaged areas of highway infrastructure into units according to the preset minimum division unit value, and divide the damaged areas of highway infrastructure into multiple damaged units; convert each damaged unit into a structured state vector, where the state vector contains the features and their parameters of the damaged areas of highway infrastructure in the corresponding damaged unit; combine the state vectors, and classify the severity of the damage of the damaged units according to the preset damage assessment criteria; combine the severity of the damage of the damaged units with the features of the damaged areas of highway infrastructure, convert them into a structured state vector, and output it as a multi-dimensional damage feature vector.

4. The intelligent decision-making method for post-disaster emergency repair of a highway infrastructure network according to claim 1, wherein The polarization attention module and the efficient channel attention module are embedded in the network structure of the encoder-decoder model.

5. The intelligent decision-making method for post-disaster emergency repair of a highway infrastructure network according to claim 4, wherein The polarization attention module performs convolution operations on the aerial images through convolution kernels of different sizes, obtains feature maps of different scales, and splices or weights and fuses the feature maps to form a feature representation containing multi-scale information, improving the recognition ability of the encoder-decoder model for aerial images under complex backgrounds; the efficient channel attention module calculates the kernel size of one-dimensional convolution according to the number of channels of the aerial images, and applies the one-dimensional convolution to aerial image recognition, improving the performance of the encoder-decoder model and reducing the computational complexity.

6. The intelligent decision-making method for post-disaster emergency repair of a highway infrastructure network according to claim 1, characterized in that, The training process of the encoder-decoder model includes: Obtain a highway scene and its disaster damage feature annotation dataset; Normalize and enhance the dataset; Input the dataset into the encoder-decoder model for training; Introduce the Dice Loss function in the training to optimize the overlap metric of the segmentation result and retain the cross-entropy loss and IoU loss; Use the gradient backpropagation algorithm to continuously update the model weights until the encoder-decoder model is trained; Evaluate the performance of the trained encoder-decoder model and further optimize the encoder-decoder model.

7. An intelligent decision-making method for post-disaster emergency repair of highway infrastructure network according to claim 6, characterized in that The optimized encoder-decoder model is loaded onto the edge computing platform of the unmanned aerial vehicle. The computing complexity of identifying the aerial image is reduced through model quantization and pruning techniques, and a lightweight inference framework is used to adapt to the device environment.

8. An intelligent decision-making method for post-disaster emergency repair of highway infrastructure network according to claim 1, characterized in that, The multi-dimensional damage feature vector includes damage type, parameters corresponding to the damage type, and damage degree, where The damage type includes subgrade damage, pavement damage, or bridge and tunnel damage; The parameters corresponding to the damage type include the specific values of the collapse area, crack length, width, the position of the center of the damaged section, and the buried area of the landslide on the highway; The damage degree includes minor, moderate, or severe.

9. The intelligent decision-making method for post-disaster emergency repair of highway infrastructure network according to claim 1, characterized in that When any damaged area is repaired, repeat steps 1 - 5, re-formulate the emergency dispatch plan for the repair team, and conduct a new round of assignment until all areas are repaired.

10. An intelligent decision-making system for post-disaster emergency repair of highway infrastructure network, characterized in that, The system includes: A road network facility disaster situation acquisition module that uses unmanned aerial vehicles to acquire aerial images of post-disaster highway infrastructure; A highway facility disaster situation automatic interpretation module that automatically identifies and interprets the features of damaged areas of highway infrastructure in the aerial image based on the encoder-decoder model in deep learning; A facility damage condition vectorization module that generates a multi-dimensional damage feature vector based on the road damage investigation and interpretation model and combines the features of the damaged areas of the highway infrastructure; A repair path space normalization module that obtains the coordinates of the repair team and the existing repair resource information, calculates the spatial geometric distance between each damaged area and the repair team, generates a distance matrix, and performs normalization processing; A state mapping module for feature fusion that integrates the normalized distance matrix and the multi-dimensional damage feature vector into a three-dimensional tensor; A highway facility intelligent repair decision-making module that combines the features of different damaged areas of highway infrastructure, the distances between repair teams, and the existing repair resource information, reduces the three-dimensional tensor to a importance matrix through dimensionality reduction mapping, and uses an optimization algorithm to formulate an emergency dispatch plan for the repair team and assign the repair team.

Images