Road water damage monitoring method, device, equipment and medium

Abstract

The invention provides a road water damage monitoring method, device and equipment and a medium, and the method comprises the steps: obtaining the road section information of a target road, the road section information comprises the road attribute of a road section, the historical water damage information of the road section, and the position information of the road section; determining a geographical influence area range corresponding to each road section of the target road according to the position information of the road sections; acquiring grid weather data of at least one weather grid covered by each geographical influence area range in the target time period; and utilizing a trained road water damage monitoring model to obtain water damage risk monitoring information of the target road according to the grid weather data of each geographical influence area range, the road attributes of the road sections and the historical water damage information of the road sections. Therefore, the accuracy of the road water damage monitoring result is improved.

Description

Technical Field

[0001] This application relates to the technical field of road disaster monitoring, and particularly to a method, device, equipment and medium for monitoring road water damage. Background Art

[0002] Road water damage refers to the phenomenon that roads along the line are damaged due to the action of water. There are many limitations in traditional road water damage monitoring technologies. For example, when using the threshold method, that is, static threshold alarm based on single-point cumulative rainfall, it is difficult to comprehensively reflect the superimposed influence of precipitation at multiple surrounding points in time and space; while using shallow machine learning models such as logistic regression or random forest for road water damage monitoring, it is difficult to effectively capture the spatio-temporal correlation characteristics of precipitation sequences over a long time span. Thus, the monitoring results are not accurate enough. Therefore, how to improve the accuracy of road water damage monitoring results has become an urgent task. Summary of the Invention

[0003] In view of this, this application provides a method, device, equipment and medium for monitoring road water damage, which can use a road water damage monitoring model to achieve accurate monitoring of road water damage and improve the accuracy of road water damage monitoring results.

[0004] In a first aspect, a method for monitoring road water damage is provided, including: obtaining section information of a target road, where the section information includes the road attributes of the section, the historical water damage information of the section, and the location information of the section; determining the geographical influence area range corresponding to each section of the target road according to the location information of the section; obtaining grid weather data of at least one weather grid covered by each geographical influence area range within a target time period; and obtaining water damage risk monitoring information of the target road by using a trained road water damage monitoring model according to the grid weather data of each geographical influence area range, the road attributes of the section, and the historical water damage information of the section.

[0005] In a second aspect, a device for monitoring road water damage includes: an information acquisition module for obtaining section information of a target road, where the section information includes the road attributes of the section, the historical water damage information of the section, and the location information of the section; a regional range determination module for determining the geographical influence area range corresponding to each section of the target road according to the location information of the section; a data acquisition module for obtaining grid weather data of at least one weather grid covered by each geographical influence area range within a target time period; and a monitoring information obtaining module for obtaining water damage risk monitoring information of the target road by using a trained road water damage monitoring model according to the grid weather data of each geographical influence area range, the road attributes of the section, and the historical water damage information of the section.

[0006] In a third aspect, an electronic device is provided, including a processor, a memory, and a program stored on the memory and capable of running on the processor. When the program is executed by the processor, the steps of any one of the road water damage monitoring provided in the embodiments of the present application are implemented.

[0007] In a fourth aspect, a computer-readable storage medium is provided. Instructions are stored on the computer-readable storage medium. When the instructions are executed by a processor, the steps of any one of the road water damage monitoring provided in the embodiments of the present application are implemented.

[0008] In summary, the road water damage monitoring method, device, electronic device, and medium provided in the present application have the following beneficial effects: According to the location information of the road section, the geographical influence area range corresponding to each road section of the target road is determined, which can focus on specific geographical areas related to road water damage, reduce the interference of data in irrelevant areas, and improve the pertinence and effectiveness of monitoring. By obtaining the grid weather data of at least one weather grid covered by each geographical influence area range within the target time period, more refined weather conditions of different regions can be obtained through the grid weather data. And compared with traditional rough weather data, the grid weather data can capture the climate differences and changes in local areas, providing more accurate meteorological input for water damage risk monitoring. Moreover, using the trained road water damage monitoring model, based on the grid weather data of each geographical influence area range, the road attributes of the road section, and the historical water damage information of the road section, the water damage risk monitoring information of the target road is obtained, enabling the road water damage monitoring model to perform water damage risk monitoring based on multi-source heterogeneous data, thereby quickly and accurately monitoring the water damage risk of the target road, significantly improving the timeliness and accuracy of road water damage risk monitoring, and enhancing the precision of road water damage monitoring results. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those skilled in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0010] Figure 1 A flowchart showing a road water damage monitoring method provided by an embodiment of the present application;

[0011] Figure 2 A structural diagram showing a road water damage monitoring model provided by an embodiment of the present application;

[0012] Figure 3 A structural diagram showing another road water damage monitoring model provided by an embodiment of the present application;

[0013] Figure 4 Shows a schematic structural diagram of a road water damage monitoring device provided by an embodiment of the present application;

[0014] Figure 5 Shows a schematic structural diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0015] In order to make the above and other features and advantages of the present application clearer, the present application will be further described below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for the purpose of explaining to those skilled in the art and are merely exemplary, not restrictive.

[0016] In the following description, many specific details are set forth to provide a thorough understanding of the present application. However, it is obvious to those skilled in the art that the present application does not need to be practiced with these specific details. In other cases, well-known steps or operations are not described in detail to avoid obscuring the present application.

[0017] On the one hand, an embodiment of the present application provides a road water damage monitoring method, which is applied to a road water damage monitoring device. Figure 1 Shows a schematic flow diagram of a road water damage monitoring method provided by an embodiment of the present application, as Figure 1 shown, the road water damage monitoring method may include the following steps.

[0018] Step S11, obtain the section information of the target road, where the section information includes the road attributes of the section, the historical water damage information of the section, and the location information of the section.

[0019] The target road involved in an embodiment of the present application may be any road on the map, and may include, but is not limited to, rural roads, town roads, county roads, national roads, and highways. The target road may include one or more sections.

[0020] The section information involved in an embodiment of the present application may be determined according to the information provided by multiple data platforms. For example, the historical water damage information of the section may be determined according to the water damage records provided by the data platform of the traffic maintenance department. The road attributes of the section may be determined according to the project archives provided by the data platform of the traffic construction department. The location information of the section may be determined according to the electronic map.

[0021] The road attributes of a section involved in an embodiment of the present application may include, but are not limited to, the roadbed material of the section, the slope protection level, the gradient, the altitude of the section, the distance between the section and the river, and the culvert density, etc. The historical flood damage information of the section may include, but is not limited to, the historical flood damage level, the historical flood damage frequency, and the weather data of the historical flood damage. The location information of the section may include, but is not limited to, the longitude and latitude information of the starting point of the section and the longitude and latitude information of the ending point of the section. In addition, the section information also includes the name and identifier of the section.

[0022] In an embodiment of the present application, the road flood damage monitoring device may obtain data related to the target road from multiple data platforms according to the road identifier of the target road, and preprocess the data of the obtained multiple data sources to obtain the section information of the target road. Among them, the road identifier may be the road name. The preprocessing includes cleaning data and integrating data.

[0023] Step S12, determine the geographical influence area range corresponding to each section of the target road according to the location information of the section.

[0024] The geographical influence area range where the section involved in an embodiment of the present application is located refers to the natural geographical area range that can affect the flood damage of the section.

[0025] In an embodiment of the present application, the road flood damage monitoring device may extract the starting point and ending point of each section from the location information of the section, match the starting point and ending point of each section with the geographical area data, and divide the geographical influence area range corresponding to each section according to the geographical influence area range division rule.

[0026] Among them, the geographical area division rule can be inferred based on the historical flood damage records and the geographical information around the road. For example, the geographical area division rule includes that the section is located between two mountains, and the area where these two mountains are located is determined as the geographical influence area range of the section, etc.

[0027] Step S13, obtain the grid weather data of at least one weather grid covered by each geographical influence area range within the target time period.

[0028] The target time period involved in an embodiment of the present application may be the current time period or the predicted time period. Among them, the current time period is a time period that traces back a preset duration forward with the current moment as the end point. The predicted time period is a time period that extends backward a preset duration with the current moment as the starting point. Optionally, the preset duration is at least 72 hours.

[0029] The grid weather data involved in an embodiment of this application refers to the weather forecast data obtained by dividing a geographical area into grid cells. The grid weather data may include, but is not limited to, basic weather elements and weather phenomena. Among them, the basic weather elements may include temperature, humidity, precipitation, wind direction, and wind speed, etc. The weather phenomena may be one of sunny, cloudy, overcast, fog, haze, rain, snow, thunderstorm, etc. A weather grid refers to a square grid with a fixed side length. Optionally, the fixed side length can be any integer from 1 to 10, and the unit is kilometers.

[0030] It should be noted that when the target time period is the current time period, the grid meteorological data of the current time period can be obtained to achieve real-time monitoring of road water damage. When the target time period is the predicted time period, the grid meteorological data of the predicted time period can be obtained to achieve the prediction of road water damage.

[0031] In an embodiment of this application, the road water damage monitoring device matches the range of each geographical impact area with the weather grid system, and all the weather grids covered by the range of each geographical impact area can be obtained. Then, according to the grid numbers of the covered weather grids, the grid weather data of each covered weather grid in the target time period is obtained from the meteorological platform, that is, the grid weather data of the geographical impact area range of each section of the target road is obtained.

[0032] Step S14, using the trained road water damage monitoring model, based on the grid weather data of each geographical impact area range, the road attributes of the section, and the historical water damage information of the section, to obtain the water damage risk monitoring information of the target road.

[0033] The road water damage monitoring model involved in an embodiment of this application is constructed based on a large model with spatio-temporal enhancement, which can enhance the correlation between time features and space features, so as to obtain more accurate monitoring results of road water damage. The large model can adopt a Transformer model or other deep learning models. The trained road water damage monitoring model refers to the road water damage monitoring model trained through road water damage samples.

[0034] In addition, the input data of the road water damage monitoring model includes grid weather data for each geographical impact area, road attributes of the road section, and historical water damage information of the road section. The above input data is multi-source input data, that is, the formats of the input data are not unified. For example, the grid weather data exists in the form of a time series and can be expressed as [time period, longitude, latitude, hourly rainfall,...]. The historical water damage information of the road section can exist as structured data, and each row of data can be expressed as [road section number, time period, longitude, latitude, water damage level, water damage frequency,...]. The road attributes of the road section exist as structured data, and each row of data can be expressed as [road section number, subgrade material, slope protection level, slope, distance from the road section to the river, culvert density,...]. That is to say, the road water damage monitoring model has the ability to process multi-source data.

[0035] The water damage risk monitoring information involved in an embodiment of the present application can be the current monitoring information of the water damage risk or the predicted information of the water damage risk. Among them, the current monitoring information of the water damage risk can be used to indicate the water damage situation of the target road at the current moment. The predicted information of the water damage risk can be used to indicate the water damage situation of the target road at a certain future moment.

[0036] It should be noted that when the grid weather data in the target time period is the grid weather data in the current time period, the road water damage monitoring model outputs the current monitoring information of the water damage risk to achieve real-time monitoring of the road water damage. When the grid weather data in the target time period is the grid weather data in the predicted time period, the road water damage monitoring model outputs the predicted information of the water damage risk.

[0037] In an embodiment of the present application, the road water damage monitoring device uses the grid weather data for each geographical impact area, the road attributes of the road section, and the historical water damage information of the road section as input data, inputs them into the trained road water damage monitoring model, and through the processing of the road water damage monitoring model, obtains the water damage risk monitoring information of the target road.

[0038] It should be noted that if only a certain road section of the target road is monitored, the grid weather data for the geographical impact area corresponding to the road section is used as the input of the road water damage monitoring model.

[0039] In some of the above embodiments, according to the location information of the road sections, the geographical influence area range corresponding to each road section of the target road can be determined, which can focus on specific geographical areas related to road washouts, reduce the interference of data in irrelevant areas, and improve the pertinence and effectiveness of monitoring. Obtaining the grid weather data of at least one weather grid covered by each geographical influence area range within the target time period can obtain more refined weather conditions of different regions through the grid weather data. And compared with the traditional rough weather data, the grid weather data can capture the climate differences and changes in local areas, providing more accurate meteorological input for washout risk monitoring. Moreover, using the trained road washout monitoring model, based on the grid weather data of each geographical influence area range, the road attributes of the road sections, and the historical washout information of the road sections, the washout risk monitoring information of the target road is obtained, enabling the road washout monitoring model to monitor the washout risk according to multi-source heterogeneous data, thereby quickly and accurately monitoring the washout risk of the target road, significantly improving the timeliness and accuracy of road washout risk monitoring, and enhancing the precision of road washout monitoring results.

[0040] Figure 2 The following shows a schematic structural diagram of a road washout monitoring model provided by an embodiment of the present application, as Figure 2 shown, the road washout monitoring model 20 may include a multi-source data fusion layer 21, a spatio-temporal enhanced convolution layer 22, and a multi-task output layer 23.

[0041] The multi-source data fusion layer 21 is used to construct a spatio-temporal feature tensor for each road section based on multi-source input data, and map the spatio-temporal feature tensor into a spatio-temporal feature vector.

[0042] The multi-source data fusion layer 21 involved in an embodiment of the present application can integrate multi-source input data, break the barriers between different data sources, and unify the originally scattered information into a feature space. Mapping the spatio-temporal feature tensor into a spatio-temporal feature vector is a conversion from the original data to high-dimensional features. The spatio-temporal feature vector can more concisely and effectively represent the characteristics of the road section in different time and space dimensions, reduce the dimension and complexity of the data, while retaining the key information, facilitating the subsequent processing and analysis of the model.

[0043] The spatio-temporal enhanced convolution layer 22 is used to obtain a time-enhanced feature representation and a space-enhanced feature representation corresponding to the spatio-temporal feature vector according to the time attention weight and the space attention weight, and fuse the time-enhanced feature representation and the space-enhanced feature representation to generate a spatio-temporal feature representation of each road section.

[0044] In an embodiment of the present application, the spatio-temporal enhanced convolution layer 22 introduces the time attention weight and the space attention weight, enabling the road washout monitoring model 20 to dynamically adjust the feature representation according to the importance of different time and space positions.

[0045] The temporal attention weight focuses on the influence degree of different time points on the road water damage risk, while the spatial attention weight focuses on the feature contributions of different spatial positions. Through the above two mechanisms, the model can more accurately capture the key information in spatio-temporal features and enhance the learning and representation ability of important features.

[0046] Moreover, the temporal enhanced feature representation and the spatial enhanced feature representation are fused, enabling the model to comprehensively and accurately understand the characteristics of road segments under different spatio-temporal backgrounds, providing richer feature support for subsequent multi-task outputs.

[0047] The multi-task output layer 23 is used to output the results of multiple road water damage-related tasks for each road segment according to the spatio-temporal feature representation of each road segment, so as to obtain the water damage risk monitoring information of the target road.

[0048] The multi-task output layer 23 involved in an embodiment of the present application can achieve task collaborative processing, output multiple tasks, realize information sharing and collaborative processing between different tasks, and improve the generalization ability and efficiency of the model. And the number and tasks of the output tasks can be adjusted according to actual needs.

[0049] The road water damage-related tasks involved in an embodiment of the present application at least include classification tasks, regression tasks, and interpretive tasks. Among them, the result of the classification task is the probability of water damage occurrence, the result of the regression task is the severity level of water damage, and the interpretive task is the key disaster-causing grid and disaster-causing period.

[0050] The water damage risk monitoring information of the target road involved in an embodiment of the present application may include: the probability of water damage occurrence for each road segment, the severity level of water damage for each road segment, and the key disaster-causing grid and disaster-causing period for each road segment.

[0051] The key disaster-causing grid and disaster-causing period involved in an embodiment of the present application refer to the key weather grid and key period that trigger water damage to the road segment.

[0052] In an embodiment of the present application, the multi-task output layer 23 outputs the results of multiple tasks for each road segment according to the spatio-temporal feature representation of each road segment, and combines the results of all road segments to obtain the water damage risk monitoring information of the target road.

[0053] In the above embodiments, through the processing of the multi-source data fusion layer and the spatio-temporal enhancement convolution layer, the road washout model can efficiently process multi-source input data, quickly and more accurately extract the spatio-temporal features of each road section, capture the key influencing factors of road washout risk, and through the multi-task output layer, the road washout risk can be evaluated from different perspectives. Therefore, through the collaborative work of the multi-source data fusion layer, the spatio-temporal enhancement convolution layer, and the multi-task output layer, the comprehensive, accurate, and efficient monitoring of road washout risk is achieved.

[0054] In some embodiments, the spatio-temporal feature tensor of each road section may include a time dimension, a space dimension, and a feature dimension.

[0055] In one embodiment of the present application, the time dimension involved can be a preset duration, where each hour represents a time step.

[0056] In one embodiment of the present application, the space dimension is an n×m weather grid. n and m can be determined according to the number of weather grids covered by the geographical influence area range. For example, if the number of weather grids is 8, the space dimension is 4×2.

[0057] In one embodiment of the present application, the feature dimension at least includes the cumulative rainfall intensity index, rainfall intensity, terrain vulnerability coefficient, culvert density, historical washout frequency, and elevation difference between the road section and the meteorological grid.

[0058] Among them, the cumulative rainfall intensity index can dynamically reflect the time decay effect of precipitation by weighted accumulation of historical precipitation data. The time decay effect means that the precipitation in the time period farther from the last moment of the target time period has a smaller impact on road washout. The cumulative rainfall intensity index CRII can be calculated according to the following formula.

[0059]

[0060] ω(t)=e -0.1(t-T)

[0061] Among them, R(t) represents the precipitation in the t time periods before the last time period in the target time period, and one hour is one time period. T represents the number of time periods covered by the target time period.

[0062] The terrain vulnerability coefficient TVI can be calculated according to the following formula.

[0063] TVI=λ s S+λ k K+λ e E (2)

[0064] Among them, λ s 、λ k and λ eIt represents the weight parameter. S represents the slope factor, and its physical meaning is that the greater the slope, the higher the risk of runoff scouring. S is not less than 0 and not greater than 1.

[0065] K represents the soil erosion resistance factor, and its physical meaning is the erosion resistance level mapped according to the soil type. For example, when the soil is clay, K = 0.2; when the soil is silt, K = 0.7; when the soil is sand, K = 0.9.

[0066] H represents the hydrological exposure factor, and its physical meaning is that the closer to the river channel, the greater the risk. H is not less than 0 and not greater than 1.

[0067] E represents the engineering protection factor, and its physical meaning is that the more perfect the protection, the lower the vulnerability. E is not less than 0 and not greater than 1.

[0068] Figure 3 It shows the structural schematic diagram of another road water damage monitoring model provided by an embodiment of the present application. Figure 3 On the Figure 2 basis, it shows the specific structure of the multi-source data fusion layer 21 and the specific structure of the spatio-temporal enhanced convolution layer 22.

[0069] As Figure 3 shown, the multi-source data fusion layer 21 includes a spatio-temporal tensor construction layer 211, an embedding layer 212, and a position encoding layer 213.

[0070] It should be noted that the embedding layer 212 and the position encoding layer 213 can be implemented using existing embedding layer structures and position encoding layer structures. The spatio-temporal tensor construction layer 211 can be implemented using a deep neural network structure.

[0071] The spatio-temporal tensor construction layer 211 is used to construct the spatio-temporal feature tensor of each section according to the multi-source input data.

[0072] The embedding layer 212 is used to map the spatio-temporal feature tensor into a spatio-temporal feature vector.

[0073] In an embodiment of the present application, the embedding layer 212 is used to map the spatio-temporal feature tensor of each section into a spatio-temporal feature vector respectively. Compared with the spatio-temporal feature tensor, the elements in the moment feature vector are all represented by 0 and 1.

[0074] The position encoding layer 213 is used to perform position encoding on the spatio-temporal feature vector to obtain the encoded spatio-temporal feature vector.

[0075] In an embodiment of the present application, the position encoding layer 213 is used to perform position encoding on the spatio-temporal feature vector of each section respectively to obtain the encoded spatio-temporal feature vector of each section.

[0076] It should be noted that, compared with the spatio-temporal feature vector of the embedding layer 212, position signals are added to the encoded spatio-temporal feature vector, where the position signals refer to the signals describing the relative or absolute positions of elements in the sequence.

[0077] As Figure 3 shown, the spatio-temporal enhancement convolutional layer 22 includes a temporal attention head 221, a spatial attention head 222, a gated fusion mechanism layer 223, a first residual connection normalization layer 224, a feed-forward neural network layer 225, and a second residual connection normalization layer 226.

[0078] The temporal attention head 221 is used to perform weighted summation on the spatio-temporal feature vector according to the temporal attention weights to obtain a temporally enhanced feature representation.

[0079] The temporal attention weights involved in the embodiments of the present application reflect the importance of different time steps. The temporally enhanced feature representation refers to the feature representation after temporally enhancing the spatio-temporal feature vector. For example, x j is the feature vector of the j-th time step in the spatio-temporal feature vector, and h j is the weighted feature of the j-th time step in the temporally enhanced feature representation.

[0080] The spatial attention head 222 is used to perform weighted summation on the spatio-temporal feature vector according to the spatial attention weights to obtain a spatially enhanced feature representation.

[0081] It should be noted that the temporal attention head 221 can capture the evolution law in the precipitation process, and the spatial attention head 222 can identify the key weather grids affecting the weather.

[0082] The spatial attention weights involved in the embodiments of the present application represent the importance of different spatial nodes. The spatially enhanced feature representation refers to the feature representation after spatially enhancing the spatio-temporal feature vector. For example, x j is the feature vector of the j-th spatial node (i.e., weather grid) in the spatio-temporal feature vector, and L j is the weighted feature of the j-th spatial node in the spatially enhanced feature representation.

[0083] The gated fusion mechanism layer 223 is used to generate fusion weights using a gated mechanism, and perform weighted fusion on the temporally enhanced feature representation and the spatially enhanced feature representation to obtain an initial spatio-temporal feature representation.

[0084] The gated fusion mechanism layer 223 involved in an embodiment of the present application can adaptively adjust the fusion weights according to the input data, control which features in the temporally enhanced feature representation and the spatially enhanced feature representation are retained or suppressed, so as to perform selective fusion.

[0085] The first residual normalization layer 224 is used to perform residual connection and normalization on the initial spatio-temporal feature representation to obtain a first intermediate spatio-temporal feature representation.

[0086] The feed-forward neural network 225 is used to perform non-linear mapping on the first intermediate spatio-temporal feature representation to obtain a second intermediate spatio-temporal feature representation.

[0087] The second residual normalization layer 226 is used to perform residual connection and normalization on the second intermediate spatio-temporal feature representation to obtain a spatio-temporal feature representation.

[0088] It should be noted that the time attention head 221, the spatial attention head 222, the gated fusion mechanism layer 223, the first residual connection normalization layer 224, the feed-forward neural network layer 225, and the second residual connection normalization layer 226 can all be implemented using existing structures.

[0089] In addition, Figure 3 shows multiple tasks (including Task 1 to Task 3) simultaneously processed by the multi-task output layer 23. The multi-task output layer 23 can implement multiple tasks using multiple independent convolutional networks respectively.

[0090] In some embodiments, the model training of the road water damage monitoring model can be performed using training samples. The training samples can include positive samples and negative samples. Among them, the positive samples can include data from several hours before the occurrence of the road. The negative samples can randomly select data from any period when the road has not suffered water damage. The data can include grid weather data, road attributes, and historical water damage information. And both the positive and negative samples also include labels. The labels annotate the water damage level and the probability of water damage occurrence.

[0091] In some embodiments, the positive samples can include a first road water damage sample and a second road water damage training sample. The first road water damage sample is a water damage sample caused by a single-point heavy rain, and the second road water damage sample is a road water damage sample caused by continuous rainfall.

[0092] In an embodiment of the present application, the road water damage monitoring method further includes a training step of the road water damage monitoring model. The training step specifically includes: training the initial road water damage monitoring model using the first road water damage sample to obtain a road water damage monitoring model after the first stage of training; freezing the spatial attention head in the road water damage monitoring model after the first stage of training; training the frozen road water damage monitoring model using the second road water damage sample to obtain a road water damage monitoring model after the second stage of training.

[0093] The training step of the road water damage monitoring model involved in an embodiment of the present application belongs to staged training, including the first stage of training and the second stage of training, and combines the attention freezing mechanism to gradually optimize the model.

[0094] The training objective of the first stage is that the road flood damage monitoring model needs to learn how to identify the flood damage features caused by single-point heavy rain. In the first-stage training, the spatial attention head in the road flood damage monitoring model trained by the first road flood damage samples learns to focus on the key weather grids of each section.

[0095] Freezing the spatial attention head after the first-stage training can retain the recognition ability of the key weather grids learned in the first stage, while avoiding being covered by the specific features of continuous rainfall samples in the second-stage training.

[0096] The training objective of the second stage is that the road flood damage monitoring model needs to learn how to identify the flood damage features caused by continuous rainfall. In the second-stage training, other parts of the road flood damage monitoring model are further optimized. At the same time, since the spatial attention head is frozen, the model still retains the recognition ability of the key weather grids.

[0097] In the above embodiments, through staged training and the attention freezing mechanism, the key area recognition ability is retained by freezing the spatial attention head, and at the same time, the second-stage training is used to optimize other parts of the model, so as to improve the accuracy of the road flood damage monitoring model, enhance the generalization ability of the model, so that the model has better generalization ability when processing different types of flood damage samples, and can adapt to the road flood damage monitoring tasks in different environments.

[0098] In some embodiments, in order to train the number of samples, the sequence length of the training samples can be randomly scaled to simulate different precipitation rhythms. And randomly mask some non-key weather grids to force the model to focus on the core area.

[0099] In some embodiments, the road flood damage monitoring method may further include generating an alarm message when the flood damage risk monitoring information of the target road meets the alarm condition. Among them, the alarm condition may include that the flood damage occurrence probability of the section reaches a preset probability. The preset probability may be above 40%. The alarm message may include the section name, the flood damage occurrence probability, and the flood damage level.

[0100] In some embodiments, the road flood damage monitoring method may further include determining corresponding treatment suggestions according to the flood damage risk monitoring information of the target road.

[0101] Another aspect of the embodiments of the present application provides a road flood damage monitoring device. Figure 4 The structural schematic diagram of a road flood damage monitoring device provided by an embodiment of the present application is shown, as Figure 4 shown, the road flood damage monitoring device 40 may include the following several modules.

[0102] An information acquisition module 41 is configured to acquire section information of a target road, where the section information includes road attributes of the section, historical flood damage information of the section, and location information of the section.

[0103] A regional range determination module 42 is configured to determine the geographical influence area range corresponding to each section of the target road according to the location information of the section.

[0104] A data acquisition module 43 is configured to acquire grid weather data of at least one weather grid covered by each geographical influence area range within a target time period.

[0105] A monitoring information obtaining module 44 is configured to use a trained road flood damage monitoring model to obtain flood damage risk monitoring information of the target road according to the grid weather data of each geographical influence area range, the road attributes of the section, and the historical flood damage information of the section.

[0106] In the above embodiments, determining the geographical influence area range corresponding to each section of the target road according to the location information of the section can focus on specific geographical areas related to road flood damage, reduce the interference of data in irrelevant areas, and improve the pertinence and effectiveness of monitoring. Acquiring the grid weather data of at least one weather grid covered by each geographical influence area range within the target time period can obtain more refined weather conditions of different regions through the grid weather data. And compared with traditional rough weather data, the grid weather data can capture the climate differences and changes in local areas, providing more accurate meteorological input for flood damage risk monitoring. Moreover, using the trained road flood damage monitoring model to obtain the flood damage risk monitoring information of the target road according to the grid weather data of each geographical influence area range, the road attributes of the section, and the historical flood damage information of the section enables the road flood damage monitoring model to monitor the flood damage risk according to multi-source heterogeneous data, thereby quickly and accurately monitoring the flood damage risk of the target road, significantly improving the timeliness and accuracy of road flood damage risk monitoring, and improving the accuracy of road flood damage monitoring results.

[0107] In some embodiments, the road flood damage monitoring device 40 further includes a training module.

[0108] The training module is specifically configured to train an initial road flood damage monitoring model with a first road flood damage sample to obtain a road flood damage monitoring model after the first-stage training, where the first road flood damage sample is a road flood damage sample caused by a single-point heavy rain; freeze the spatial attention head in the road flood damage monitoring model after the first-stage training; and train the frozen road flood damage monitoring model with a second road flood damage sample to obtain a road flood damage monitoring model after the second-stage training, where the second road flood damage sample is a road flood damage sample caused by continuous rainfall.

[0109] It should be understood that the specific features, operations, and details described above regarding the method of the present application can also be similarly applied to the device of the present application, or vice versa. Additionally, each step of the method of the present application described above can be executed by the corresponding components or units of the device or system of the present application.

[0110] It should be understood that each module / unit of the device of the present application can be implemented in whole or in part by software, hardware, firmware, or a combination thereof. Each module / unit can be embedded in the processor of the electronic device in the form of hardware or firmware or independent of the processor, or stored in the memory of the electronic device in the form of software for the processor to call to execute the operations of each module / unit. Each module / unit can be implemented as an independent component or module, or two or more modules / units can be implemented as a single component or module.

[0111] In another aspect of the present application, an electronic device is provided. Figure 5 The structural schematic diagram of an electronic device provided according to an embodiment of the present application is shown, as Figure 5 shown, the electronic device 50 includes a processor 51, a memory 52, and a program stored on the memory and capable of running on the processor. When the program is executed by the processor, it implements the steps of the road water damage monitoring method provided in any of the above embodiments.

[0112] In an embodiment, the electronic device 50 may include a processor, a memory, a network interface, a communication interface, etc. connected through a system bus. The processor of the electronic device 50 can be used to provide necessary computing, processing, and / or control capabilities. The memory of the electronic device 50 can include a non-volatile storage medium and an internal memory. The non-volatile storage medium can store an operating system, a computer program, etc. The internal memory can provide an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface and the communication interface of the electronic device 50 can be used to connect and communicate with external devices through a network.

[0113] In another aspect of the present application, a computer-readable storage medium is provided. Instructions are stored on the computer-readable storage medium. When the instructions are executed by the processor, they implement the steps of the road water damage monitoring method provided in any of the above embodiments.

[0114] Those skilled in the art can understand that the method steps of this application can be completed by a computer program instructing relevant hardware such as electronic devices or processors. The computer program can be stored in a non-transitory computer-readable storage medium. When the computer program is executed, the steps of this application are caused to be executed. Depending on the situation, any reference to a memory, storage, or other medium in this article may include non-volatile or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0115] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification as long as such a combination does not exist in contradiction.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, rather than to limit them; although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for monitoring road water damage, characterized in that, Including: Obtaining section information of a target road, where the section information includes the road attributes of the section, the historical flood damage information of the section, and the location information of the section; Determining the geographical influence area range corresponding to each section of the target road according to the location information of the section; Obtaining grid weather data of at least one weather grid covered by each geographical influence area range within a target time period; Using a trained road flood damage monitoring model, based on the grid weather data of each geographical influence area range, the road attributes of the section, and the historical flood damage information of the section, to obtain the flood damage risk monitoring information of the target road.

2. The method according to claim 1, wherein The road flood damage monitoring model includes a multi-source data fusion layer, a spatio-temporal enhancement convolutional layer, and a multi-task output layer; The multi-source data fusion layer is used to construct a spatio-temporal feature tensor for each section based on multi-source input data, and map the spatio-temporal feature tensor into a spatio-temporal feature vector. The multi-source input data includes grid weather data, road attributes, and historical flood damage information; The spatio-temporal enhancement convolutional layer is used to obtain a time-enhanced feature representation and a space-enhanced feature representation corresponding to the spatio-temporal feature vector according to the time attention weight and the space attention weight, and fuse the time-enhanced feature representation and the space-enhanced feature representation to generate a spatio-temporal feature representation for each section; The multi-task output layer is used to respectively output the results of multiple road flood damage-related tasks for each section according to the spatio-temporal feature representation of each section, to obtain the flood damage risk monitoring information of the target road.

3. The method according to claim 1 or 2, characterized in that, The flood damage risk monitoring information of the target road includes: the flood damage occurrence probability of each section, the flood damage severity level of each section, and the key disaster-causing grids and disaster-causing time periods of each section.

4. The method according to claim 2, wherein The multi-source data fusion layer includes a spatio-temporal tensor construction layer, an embedding layer, and a position encoding layer; The spatio-temporal tensor construction layer is used to construct a spatio-temporal feature tensor for each section according to multi-source input data; The embedding layer is used to map the spatio-temporal feature tensor into a spatio-temporal feature vector; The position encoding layer is used to perform position encoding on the spatio-temporal feature vector to obtain an encoded spatio-temporal feature vector.

5. The method according to claim 2 or 4, characterized in that, The spatio-temporal feature tensor includes a time dimension, a space dimension, and a feature dimension. The feature dimension at least includes an accumulated rainfall intensity index, rainfall intensity, terrain vulnerability coefficient, culvert density, historical flood damage frequency, and the altitude difference between the section and the meteorological grid.

6. The method according to claim 2 or 4, characterized in that The spatio-temporal enhancement convolutional layer includes a time attention head, a space attention head, a gated fusion mechanism layer, a first residual connection normalization layer, a feed-forward neural network layer, and a second residual connection normalization layer; The time attention head is used to perform weighted summation on the spatio-temporal feature vector according to the time attention weight to obtain a time-enhanced feature representation; The space attention head is used to perform weighted summation on the spatio-temporal feature vector according to the space attention weight to obtain a space-enhanced feature representation; The gated fusion mechanism layer is used to generate a fusion weight using a gated mechanism, and perform weighted fusion on the time-enhanced feature representation and the space-enhanced feature representation to obtain an initial spatio-temporal feature representation; The first residual normalization layer is used to perform residual connection and normalization on the initial spatio-temporal feature representation to obtain a first intermediate spatio-temporal feature representation; The feed-forward neural network is used to perform non-linear mapping on the first intermediate spatio-temporal feature representation to obtain a second intermediate spatio-temporal feature representation; The second residual normalization layer is used to perform residual connection and normalization on the second intermediate spatio-temporal feature representation to obtain a spatio-temporal feature representation.

7. The method according to claim 6, characterized in that, It further includes a training step of the road water damage monitoring model, and the training step includes: Training an initial road water damage monitoring model with a first road water damage sample to obtain a road water damage monitoring model after the first stage of training, where the first road water damage sample is a road water damage sample caused by a single-point heavy rain; Freezing the spatial attention head in the road water damage monitoring model after the first stage of training; Training the frozen road water damage monitoring model with a second road water damage sample to obtain a road water damage monitoring model after the second stage of training, where the second road water damage sample is a road water damage sample caused by continuous rainfall.

8. A road water damage monitoring device, characterized in that, It includes: An information acquisition module, configured to acquire section information of a target road, where the section information includes the road attributes of the section, the historical water damage information of the section, and the location information of the section; A regional range determination module, configured to determine the geographical influence area range corresponding to each section of the target road according to the location information of the section; A data acquisition module, configured to acquire grid weather data of at least one weather grid covered by each geographical influence area range within a target time period; A monitoring information obtaining module, configured to use the trained road water damage monitoring model to obtain water damage risk monitoring information of the target road according to the grid weather data of each geographical influence area range, the road attributes of the section, and the historical water damage information of the section.

9. An electronic device, characterized in that, It includes a processor, a memory, and a program stored on the memory and executable on the processor. When the program is executed by the processor, it implements the steps of the road water damage monitoring method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, Instructions are stored on the computer-readable storage medium, and when the instructions are executed by the processor, they implement the steps of the road water damage monitoring method according to any one of claims 1-7.

Images