Road collapse assessment method and device and computer readable storage medium

Abstract

The invention provides a road collapse assessment method and device and a computer readable storage medium. The method comprises the steps of obtaining first settlement data of a sample site and a road collapse label corresponding to the sample site; inputting the first settlement data into a preset prediction model to obtain a road collapse prediction probability of the sample site; wherein the preset prediction model comprises N first models and N second models which are stacked in sequence; determining a predicted road collapse label according to the road collapse prediction probability, and updating model parameters of a preset prediction model according to the predicted road collapse label and the road collapse label to obtain a target prediction model; and obtaining second settlement data of a to-be-predicted site, and inputting the second settlement data into the target prediction model to obtain a road collapse probability of the to-be-predicted site. According to the invention, the road collapse risk of the to-be-predicted place can be evaluated, and the accuracy of predicting the road collapse risk is improved.

Description

Technical Field

[0001] The present application relates to the field of geological disaster monitoring and risk assessment, and in particular to a road collapse assessment method, device and computer-readable storage medium. Background Art

[0002] In the detection of geological disasters, it is often necessary to collect local historical settlement information. Interferometric Synthetic Aperture Radar (InSAR) is usually used for large-scale, long-term, and high-precision earth observation. The basic principle of InSAR technology is to accurately measure the three-dimensional spatial position and small changes of a certain location on the surface by constructing the geometric relationship between satellite system parameters, posture parameters, orbital parameters, and radar echo phase, so as to obtain geographic data with extremely high spatial resolution and millimeter-level accuracy. The historical settlement information obtained based on InSAR data often has a lot of messy data and high precision, which makes it difficult to combine historical settlement information with long-term ground changes for settlement assessment.

[0003] Related technologies often use long short-term memory (LSTM) networks to predict settlement probability, but the LSTM network structure is complex and requires a lot of memory and computing resources when processing long-term data. At the same time, due to the excessive number of model parameters in the LSTM model, overfitting is prone to occur during the training process, resulting in reduced accuracy of the prediction results. With the acceleration of urbanization, the complexity of urban underground space has increased significantly, resulting in frequent road collapse incidents, which seriously threatens the safety of life and property of urban residents. Therefore, how to provide a more accurate road collapse assessment method is an urgent problem to be solved. Summary of the invention

[0004] In view of this, the present application provides a road collapse assessment method, device and computer-readable storage medium, which can implement road collapse risk assessment for a predicted location and improve the accuracy of predicting road collapse risk.

[0005] A first aspect of an embodiment of the present application provides a road collapse assessment method, comprising: Obtain first settlement data of a sample location, and a road collapse label corresponding to the sample location; input the first settlement data into a preset prediction model to obtain a road collapse prediction probability of the sample location; wherein the preset prediction model includes N first models and second models stacked in sequence, and obtaining the road collapse prediction probability of the sample location includes: inputting the first settlement data into the N first models to obtain a first output result, and inputting the first output result into the second model to obtain a road collapse prediction probability; determining a predicted road collapse label according to the road collapse prediction probability, and updating the model parameters of the preset prediction model according to the predicted road collapse label and the road collapse label to obtain a target prediction model; obtaining second settlement data of the location to be predicted, and inputting the second settlement data into the target prediction model to obtain the road collapse probability of the location to be predicted.

[0006] Compared with the related art, the embodiments of the present application have at least the following advantages: Compared with the LSTM neural network used in the prior art, the dependency of capturing long-term data is poor. The present application adopts a preset prediction model for training, which can effectively mine the long-term relationship between the first settlement data and the road collapse label, thereby increasing the accuracy of predicting the probability of road collapse at the predicted location. At the same time, the network structure of the first model is simple, easy to implement, and consumes low resources. By stacking N first models, the depth of the model can be increased. While maintaining a simple neural network structure, the characteristics of long sequence data can be better captured, and the accuracy of the prediction results can be improved. Through the second model, new features can be obtained from the output results of the N first models, thereby completing the output of the road collapse probability. On the other hand, the structure of the neural network architecture used in the present application is simple, the model hardware deployment is superior in friendliness, and it is easier to apply to the end side.

[0007] In a second aspect, an embodiment of the present application further provides a road collapse assessment device, comprising: An acquisition unit, used to acquire first settlement data of a sample location and a road collapse label corresponding to the sample location; A first training unit is used to input the first settlement data into a preset prediction model to obtain a road collapse prediction probability at a sample location; wherein the preset prediction model includes N first models and second models stacked in sequence, and obtaining the road collapse prediction probability at the sample location includes: inputting the first settlement data into the N first models to obtain a first output result, and inputting the first output result into the second model to obtain the road collapse prediction probability; The second training unit determines a predicted road collapse label according to the road collapse prediction probability, updates a model parameter of a preset prediction model according to the predicted road collapse label and the road collapse label, and obtains a target prediction model; The prediction unit obtains the second settlement data of the location to be predicted, inputs the second settlement data into the target prediction model, and obtains the road collapse probability of the location to be predicted.

[0008] In a third aspect, an embodiment of the present application further provides an electronic device, the electronic device comprising a processor and a memory, the memory being used to store instructions, and the processor being used to call the instructions in the memory so that the electronic device executes the road collapse assessment method as in the first aspect.

[0009] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the road collapse assessment method as in the first aspect.

[0010] The technical effects obtained by the above-mentioned second, third and fourth aspects are similar to the technical effects obtained by the corresponding technical means in the first aspect, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 A schematic diagram of an implementation environment of a road collapse assessment method provided in an embodiment of the present application.

[0012] Figure 2 A flowchart of the steps of a road collapse assessment method provided in one embodiment of the present application.

[0013] Figure 3 A flowchart of the steps of a road collapse assessment method provided in yet another embodiment of the present application.

[0014] Figure 4 A schematic diagram of the structure of a preset prediction model provided in one embodiment of the present application.

[0015] Figure 5 A schematic diagram of the structure of a first model provided in one embodiment of the present application.

[0016] Figure 6 A schematic diagram of the structure of a second model provided in one embodiment of the present application.

[0017] Figure 7 A functional module diagram of a road collapse assessment device provided in one embodiment of the present application.

[0018] Figure 8 A schematic diagram of the structure of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application is described in detail below in conjunction with the accompanying drawings and specific implementation methods. It should be noted that the implementation methods of the present application and the features in the implementation methods can be combined with each other without conflict.

[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present application. The described implementations are only part of the implementations of the present application, rather than all the implementations.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0022] It should be further noted that, in this article, the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0023] In this application, "at least one" means one or more, and "more" means two or more than two. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and drawings of this application are used to distinguish similar objects, rather than to describe a specific order or sequence.

[0024] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0025] To facilitate understanding, some illustrations of concepts related to the embodiments of the present application are given by way of example for reference.

[0026] 1. Interferometric synthetic aperture radar interferometry technology Synthetic Aperture Radar Interferometry, InSAR is a combination of synthetic aperture radar (SAR) and interferometry technology. Its basic principle is to accurately measure the three-dimensional spatial position of a point on the surface and its tiny changes by constructing the geometric relationship between satellite system parameters, attitude parameters, orbit parameters and radar echo phase.

[0027] 2. Long Short-Term Memory Networks LSTM, sLSTM and mLSTM Long Short-Term Memory (LSTM) is a time-recurrent neural network that is specially designed to solve the long-term dependency problem of general RNN (recurrent neural network).

[0028] LSTM introduces three important gates: Input Gate, Forget Gate, and Output Gate. The Input Gate has a tanh activation function and a memory blocking effect, which determines how much new information is added. The Forget Gate determines whether to discard information from the previous output. Output Gate: The Output Gate is responsible for determining the information output at the current moment.

[0029] After a long period of development, LSTM has multiple variants, such as sLSTM (with scalar memory and update) and mLSTM (with matrix memory and covariance update rule). Multiple variants are integrated into the residual block to form the xLSTM architecture.

[0030] sLSTM: By introducing exponential activation functions into the input and forget gates, as well as the normalizer state to balance the gated activation values.

[0031] mLSTM: uses matrices to store and retrieve information, improves storage capacity, and uses covariance matrices to update memory cells.

[0032] xLSTM Architecture The xLSTM architecture is built by stacking residual blocks and using pre-layer normalization (preLayerNorm) to form a residual backbone.

[0033] 3. Normalization layer The normalization layer has two functions: preventing gradient explosion and gradient vanishing. Normalization is also called standardization. In the process of back propagation of the neural network, the calculation of the gradient of each layer must use the data (feature map) of this layer multiplied by the gradient passed from the previous layer to obtain the gradient of this layer. The data of this layer is normalized and limited to a mean of zero, so there will be no gradient explosion when multiplying. In the process of back propagation of the neural network, the calculation of the gradient of each layer must be multiplied by the gradient passed from the previous layer. If the gradient value of each layer is relatively small, then the smaller the gradient is during the back propagation process, the gradient will vanish. Therefore, normalization can prevent gradient vanishing.

[0034] 4. Block Diagonal Linear Layer The Block Diagonal Linear Layer is a special type of linear layer whose weight matrix consists of multiple independent blocks arranged on the main diagonal to form a block diagonal matrix. In the Block Diagonal Linear Layer, each block only interacts with the corresponding part of the input, thereby simulating multiple independent sets of linear transformations.

[0035] 5. Swish activation function The swish activation function is a nonlinear activation function, and its mathematical expression is: in, is the sigmoid function.

[0036] The swish activation function can increase the generalization ability of the model while improving the expressiveness of the model.

[0037] 6. Causal Convolutional Layer The ‌Causal Convolution Layer‌ is a convolution layer specifically designed for processing time series data. It ensures that the output of each event step depends only on the current and past inputs, and not on future inputs. This feature enables the causal convolution layer to maintain temporal consistency and prediction accuracy when processing time series data.

[0038] 7. Projection layer The projection layer is used to map data to a new feature space. It maps data from the original feature space to the new feature space by multiplying the weight matrix. The projection layer is divided into an upper projection layer and a lower projection layer. The upper projection layer maps the input data to a higher-dimensional feature space to help the model capture more complex features. The lower projection layer maps the data to a lower-dimensional feature space and is often used for data dimensionality reduction to reduce the complexity of the data while retaining important information as much as possible.

[0039] 8. Single-dimensional convolutional layer A single-dimensional convolution layer, also called a one-dimensional convolution layer, is an operation in a convolutional neural network (CNN) that is mainly used to process one-dimensional data sequences, such as time series data, audio signals, text data, etc. Unlike two-dimensional convolution layers used for image processing, one-dimensional convolution layers only consider one dimension of the data, usually the time dimension or a single dimension in the sequence.

[0040] In the detection of geological disasters, it is often necessary to collect accurate historical settlement information of the local area. Interferometric Synthetic Aperture Radar (InSAR) is usually used for large-scale, long-term, and high-precision earth observation. The basic principle of InSAR technology is to accurately measure the three-dimensional spatial position and slight changes of a certain location on the surface by constructing the geometric relationship between satellite system parameters, posture parameters, orbital parameters, and radar echo phase, so as to obtain extremely high spatial resolution and millimeter-level accuracy. The historical settlement information obtained based on InSAR data often has a lot of messy data and high precision, which makes it difficult to combine historical settlement information with long-term ground changes for settlement assessment. Related technologies often use Long Short-Term Memory (LSTM) networks to predict settlement probability, but the LSTM network structure is complex. When processing long-term data, it requires a lot of memory and computing resources. At the same time, due to the excessive number of model parameters of the LSTM model, overfitting is prone to occur during the training process, resulting in a decrease in the accuracy of the prediction results.

[0041] With the acceleration of urbanization, the complexity of urban underground space has increased significantly, resulting in frequent road collapse incidents, which seriously threatens the safety of life and property of urban residents. Therefore, how to provide a more accurate road collapse assessment method is an urgent problem to be solved.

[0042] In view of the above, the embodiments of the present application provide a road collapse assessment method, device and computer-readable storage medium, which can implement road collapse risk assessment for a predicted location and improve the accuracy of predicting road collapse risk.

[0043] Please refer to Figure 1 , Figure 1 A schematic diagram of an implementation environment of a road collapse assessment method provided in an embodiment of the present application.

[0044] Those skilled in the art will understand that Figure 1The schematic diagram of the implementation environment of the road collapse assessment method shown is merely an example. The implementation environment described in the embodiment of the present application is to more clearly illustrate the technical solution of the embodiment of the present application, and does not constitute a limitation on the technical solution provided in the embodiment of the present application. A person of ordinary skill in the art can know that with the emergence of new business scenarios, the technical solution provided in the embodiment of the present application is also applicable to similar technical problems.

[0045] exist Figure 1 In the embodiment of the present application, the implementation environment of the road collapse assessment method provided by the embodiment of the present application may include a terminal 110 and a server 120. The server 120 may be integrated with the road collapse assessment device provided by the embodiment of the present application. Among them, the terminal 110 may be, but is not limited to, other user devices (not shown in the figure) that establish a communication connection with the device 120. The terminal 110 and the server 120 are connected via a network.

[0046] Optionally, in this embodiment, the terminal may be a terminal device configured with a target client, and may include but is not limited to at least one of the following: a mobile phone (such as an Android phone, an iOS phone, etc.), a laptop, a tablet computer, a PDA, a MID (Mobile Internet Devices), a PAD, a desktop computer, a smart TV, etc. The above network may include but is not limited to: a wired network, a wireless network, wherein the wired network includes: a local area network, a metropolitan area network and a wide area network, and the wireless network includes: Bluetooth, WIFI and other networks that implement wireless communication. The above server may be a single server, or a server cluster consisting of multiple servers, or a cloud server. The above is only an example, and this embodiment does not make any limitation to this.

[0047] The server 120 can execute the road collapse assessment method provided in the embodiment of the present application to implement the road collapse risk assessment of the location to be predicted in the actual measurement scenario, and improve the accuracy of predicting the road collapse risk.

[0048] Please refer to Figure 2 , Figure 2 is a flowchart of the steps of the road collapse assessment method provided by an embodiment of the present application. The order of the steps in the flowchart can be changed according to different requirements, and some steps can be omitted. The road collapse assessment method can be applied to Figure 1 The server 120 is shown, but is not limited to this, and the embodiments of the present application are not limited to this.

[0049] The specific process of this embodiment is as follows Figure 2 As shown, the following steps are included: Step 201: Acquire first settlement data of a sample location and a road collapse label corresponding to the sample location.

[0050] This embodiment collects settlement data of sample cities, analyzes and mines the historical settlement data of the city, obtains a series of first settlement data for each sample location, determines the settlement results of the first settlement data, and labels the first settlement data with corresponding settlement labels. For example, several accident locations of sample cities are collected, each accident location is used as a sample location, the settlement data of a sample location for 12 months is used as the first settlement data, and the road collapse result of the 13th month is used as the corresponding road collapse label.

[0051] In one embodiment, obtaining first settlement data of a sample location and a road collapse label corresponding to the sample location includes: obtaining first historical settlement data of the sample location over a historical period; dividing the first historical settlement data into monitoring data and verification data; converting the monitoring data into first settlement data in the form of a time series; and generating a road collapse label for the sample location based on the verification data.

[0052] In this embodiment, the first historical settlement data of the sample location in the historical period is divided into monitoring data and verification data. The verification data is used to generate a road collapse label to represent the prediction result of the monitoring data. The above road collapse label is used to represent the road collapse road condition and the non-road collapse road condition. Different road collapse labels can be represented by different character strings. For example, for the settlement data of the sample location for 13 consecutive months, the settlement data for 13 months is divided into the detection values ​​of the first 12 months and the verification data of the 13th month. The settlement data of the previous 12 months are stored in the form of time series. For example, the time series corresponding to sample location 1 is [settlement value 1, settlement value 2, settlement value 3, settlement value 4, settlement value 5, settlement value 6, settlement value 7, settlement value 8, settlement value 9, settlement value 10, settlement value 11, settlement value 12]. The road collapse condition corresponding to the settlement value of the 13th month is determined. According to the road collapse condition, the accident location 1 is labeled with the corresponding road collapse label. For example, the collapsed road surface condition is recorded as a road collapse condition, and the non-road collapse condition is recorded as a non-road collapse. Among them, the road collapse condition can be represented by 1, and the non-road collapse condition can be represented by 0.

[0053] In one embodiment, obtaining first historical settlement data of a sample location over a historical period includes: obtaining second historical settlement data of a sample city over a historical period; clustering the second historical settlement data based on a spatiotemporal clustering method to determine the accident location of the sample city; generating the sample location and the second historical settlement data based on the accident location to generate the first historical settlement data.

[0054] In this embodiment, by analyzing the second historical settlement data of the sample city, all accident sites are obtained, that is, the sample sites are determined. Each sample site is sampled at a continuous time interval to obtain the first historical settlement data of each location. For example, the InSAR time series data and satellite orbit data of the satellite are obtained, and the InSAR time series data and orbit data of the satellite are processed to obtain a high-precision settlement map of the city. According to the high-precision settlement map of the city, the second historical settlement data of the city is obtained. The second historical settlement data is processed to obtain the first historical settlement data corresponding to the sample site.

[0055] Optionally, after processing the InSAR time series data and orbital data of the satellite, a high-precision sedimentation map of the city is obtained, including: generating an interference map according to the InSAR time series data and orbital data of the satellite, and clustering the interference map pairs based on the time-space threshold method. The time-space threshold method screens the interference map pairs according to the event interval and the spatial baseline (i.e., the change in the satellite position) to ensure the quality of the interference map. The interference map pairs are corrected using digital elevation model (Digital Elevation Model, DEM) data, and the corrected interference map pairs are filtered to reduce noise and errors. The corrected interference map pairs are phase-unwrapped to obtain continuous phase information, and the phase map after unwrapping is subjected to secondary difference processing, and atmospheric errors are eliminated to generate a high-precision sedimentation map. The high-precision sedimentation map is used to represent the sedimentation situation of the surface.

[0056] In one embodiment, after obtaining the historical settlement data of the sample location over a historical period, the method further includes: obtaining an evaluation data set. Obtaining the evaluation data set includes: randomly sampling non-accident locations from the second historical settlement data of the sample city; obtaining third historical settlement data of the non-accident location, dividing the third historical settlement data into third settlement data and verification data of the third settlement data, and generating a road collapse label corresponding to the non-accident location based on the verification data of the third settlement data.

[0057] In this embodiment, the time series length of the third sedimentation data is the same as the time series length of the first sedimentation data. The evaluation data set is used to test the training degree of the prediction model.

[0058] Step 202, input the first settlement data into a preset prediction model to obtain the road collapse prediction probability of the sample location; wherein the preset prediction model includes N first models and second models stacked in sequence, and obtaining the road collapse prediction probability of the sample location includes: inputting the first settlement data into the N first models to obtain a first output result, and inputting the first output result into the second model to obtain the road collapse prediction probability.

[0059] In this embodiment, the road collapse prediction probability is output by the model, which is used to represent the output probability of the road collapse label. For example, the output road collapse prediction probability is 0.8, and 0.8 means that the probability of outputting the road collapse road condition label is 0.8. The preset prediction model is an xLSTM network structure. The first model uses mLSTM to update the node state, and the second model uses sLSTM to update the node state. N mLSTM blocks are stacked with sLSTM blocks to form a complete xLSTM structure. By stacking N first models, the depth of the model can be increased, the characteristics of long sequence data can be better captured, and the accuracy of the prediction results can be improved. Through the second model, the new features obtained from the aforementioned process can be better used to complete the output of the road collapse probability.

[0060] Combination Figure 3 and Figure 4 Step 202 is described. Figure 3 A flowchart of a road collapse assessment method provided in another embodiment of the present application is shown below: Figure 3 Specifically, it is a flowchart of the steps of obtaining the road collapse prediction probability method. Figure 4 A schematic diagram of the structure of a preset prediction model provided in one embodiment of the present application.

[0061] like Figure 3 As shown, the first settlement data is input into N first models to obtain a first output result, and the first output result is input into a second model to obtain a road collapse prediction probability, including: Step 301, input the first settlement data into the first first model, and input the output result of the first first model into the second first model, repeat the above process of inputting the output result of the previous first model into the next first model, until the last first model among the N first models outputs the first output result.

[0062] like Figure 4 As shown, when the first settlement data is input into the first first model, the first feature is obtained, and the first feature is input into the second first model to obtain the second feature; the above process of inputting the output result of the previous first model into the next first model is repeated, that is, for the ith first model, the i-1th feature of the output of the i-1th model is received, and the ith feature is output, and the ith first model also inputs the ith feature into the i+1th feature, traversing N first models until the Nth first model, and outputting the first output result according to the N-1th feature.

[0063] Step 302: Input the first output result into the second model to obtain the predicted probability of road collapse at the sample location.

[0064] The second model outputs a road collapse prediction probability according to the first output result. A road collapse prediction label can be determined according to the road collapse prediction probability.

[0065] Figure 5 This is a schematic diagram of the structure of the first model provided in one embodiment of the present application. Figure 5 As shown, the first model includes: a first linear layer 510, a first projection layer 520, a first module 530, a second module 540 and a second projection layer 550, the first projection layer 520 includes two upper projection layers, and the first module 530 includes: an upper projection layer, a causal convolution layer, a swish activation layer, three block diagonal linear layers ( Figure 5 The nine squares in the figure are used to represent the block diagonal layer, wherein the block diagonal of the nine squares is filled with black, and one nine square corresponds to one block diagonal layer), a one-dimensional convolution layer, and a group normalization layer; the second module 540 includes: an upper projection layer and a swish activation layer; wherein: the first linear layer 510 is used to extract features from the input data; the first projection layer 520 is used to output two projection data to the output result of the first linear layer 510, so as to obtain a first sub-feature and a second sub-feature; the first module 530 is used to obtain a third sub-feature according to the first sub-feature; the second module 540 is used to obtain a fourth sub-feature according to the second sub-feature; the second projection layer 550 is used to output features corresponding to the first model according to the third sub-feature and the fourth sub-feature.

[0066] Further, if Figure 5 As shown, the input data is first processed by the first linear layer 510, and then passed through the first projection layer 520 to generate two projection outputs: one of the projection outputs is passed to the first module 530, and the other projection output is passed to the second module 540. For the first module 530, after receiving the data, the data is input into the causal convolution layer and the swish layer to obtain four output data, one of which is input into the single-dimensional convolution layer, and the other three are respectively input into a block diagonal layer, and the query (q), key (k) and value components (v) are generated through the block diagonal projection matrix. Each block diagonal linear layer consists of four diagonal blocks (i.e., "heads"). Each diagonal block has an independent unit state. In the first model, each diagonal block is updated using an mLSTM block. The query (q), key (k) and value component (v) are used for further processing by the mLSTM block, causing changes in the state unit. The unit state update formula of mLSTM is formula (1), as shown: Formula (1); in, represents the matrix cell state at the current time step, represents the value vector of the current time step, A key vector representing the current time step.

[0067] The hidden state update formula is formula (2): Formula (2); in, represents the hidden state at the current time step, represents the output gate, represents the query vector at the current time step, Indicates the normalized state.

[0068] Figure 6 This is a schematic diagram of the structure of the second model provided in one embodiment of the present application. Figure 6 As shown, the second model includes: a second linear layer 610, a third module 620, a fourth module 630, a group normalization layer 640 and a third projection layer 650; the third module 620 includes: a causal convolution layer, a swish activation layer and two block diagonal linear layers (block diagonal linear layers are used Figure 6 The fourth module 630 includes: two block diagonal linear layers; the third projection layer 650 includes two upper projection layers and one lower projection layer, wherein; the second linear layer 610 is used to extract features from the input data to obtain the fifth sub-feature and the sixth sub-feature; the third module 620 is used to obtain the seventh sub-feature according to the fifth sub-feature; the fourth module 630 is used to obtain the eighth sub-feature according to the sixth sub-feature; the group normalization layer 640 is used to process the seventh sub-feature and the eighth sub-feature, and input the processing results of the group normalization layer 640 into the third projection layer 650.

[0069] Further in Figure 6 In the figure, the first output result x(t) outputs two copies of data after passing through the second linear layer 610, one copy is processed according to the third module 620, and the other copy is processed by the fourth module 630. The output data of the third module 620 and the fourth module 630 are input into the group normalization layer 640. After the group normalization layer 640 inputs the data into two upper projection layers, the results of the two upper projection layers are input into the lower projection layer to obtain the output result of the settlement probability value.

[0070] Among them, for the third module 620, the input data is processed by the causal convolution layer and the swish activation function, and then the data is output through two block diagonal linear layers. Each block diagonal linear layer consists of four diagonal blocks (i.e., "heads"). Each diagonal block has an independent unit state, and in the second model, each diagonal block uses sLSTM to update the unit state. For the fourth module 630, the input data is output through two block diagonal linear layers.

[0071] The cell state update formula of sLSTM is formula (3), including: Formula (3); in, represents the cell state at the current time step, represents the forget gate, represents the input gate, Represents the input at the current time step.

[0072] For each unit's input gate The calculation formula is shown in formula (4), the forget gate of each unit The calculation formula is shown in formula (5), including: Formula (4); Formula (5); in, and Represent the weight matrices of the input gate and the forget gate respectively, and represents the recurrent weight matrix of the input gate and the forget gate, and represents the bias term of the input gate and the forget gate, represents the hidden state at the previous time step.

[0073] Step 203, determining a predicted road collapse label according to the road collapse prediction probability, updating model parameters of a preset prediction model according to the predicted road collapse label and the road collapse label, and obtaining a target prediction model.

[0074] The output result of the second model needs to be passed through the Sigmoid function to obtain an output probability between 0 and 1. In this embodiment, the probability of the road collapse label output by the preset prediction model is determined to be a road collapse for the probability exceeding the prediction probability threshold. For example, 1 is used to represent road collapse and 0 is used to represent non-road collapse, and the output probability value represents the probability of the road collapse. If the probability threshold is 0.5, the output result exceeding 0.5 is regarded as a road collapse, and the output result with a probability not exceeding 0.5 is regarded as a non-road collapse.

[0075] In one embodiment, model parameters of a preset prediction model are updated according to predicted road collapse labels and road collapse labels to obtain a target prediction model, including: updating parameters of the preset prediction model according to the difference between the predicted road collapse labels and the road collapse labels at the sample locations to obtain an updated model; continuously updating parameters of the updated model, calculating the loss function of the updated model, and when the loss function converges, determining that the updated model is the target prediction model; and / or obtaining an evaluation data set, inputting the evaluation data set into the updated model to obtain an evaluation result, calculating the accuracy of the evaluation result, and when the accuracy no longer increases, determining that the updated model is the target prediction model.

[0076] In this embodiment, there are two conditions for judging whether the training is completed and generating the target prediction model. One of the conditions is that when the obtained loss function converges, it is judged that the training of the target prediction model is completed at this time; the other condition is that the updated model is evaluated with an evaluation data set so that the accuracy of the evaluation result no longer increases. If at least one of the two conditions is met, the training is considered to be completed and the target prediction model is generated. For example, 80% of the data set in the first sedimentation data is used as a training sample, and the remaining 20% ​​of the data set is used as an evaluation data set to evaluate the model performance. The total number of training iterations is set to 50, and the initial learning rate is 0.001. A weighted binary cross entropy loss function is used to deal with the problem of sample imbalance. Its formula is shown in formula (6): Formula (6); in, is the total number of samples, It is The true labels of samples, It is The predicted probability of a sample, and are the weight coefficients of positive and negative samples respectively. If the loss function of the model does not decrease for 50 iterations, or the accuracy on the validation set does not improve, the model is considered to have reached convergence and training is stopped.

[0077] In one embodiment, after generating the target prediction model, it is also necessary to verify the accuracy of the target prediction model and the reliability of the target prediction model, including: obtaining a verification data set, obtaining the output result of the target prediction model for the verification data set, and the output result includes: a consistency result and an inconsistency result; the consistency result indicates that the predicted road collapse label of the output result is the same as the road collapse label of the verification data set, and the inconsistency result indicates that the predicted road collapse label of the output result is different from the road collapse label of the verification data set. According to the output result, the reliability parameters of the target prediction model are calculated, and the reliability parameters of the reliability include at least the accuracy, the recall rate and the F1 score; according to the reliability parameters, the accuracy of the target prediction model is evaluated.

[0078] This embodiment uses the evaluation index calculation method based on the confusion matrix to calculate the accuracy, recall rate and F1 score, wherein the calculation formula (7) of the accuracy, the calculation formula (8) of the recall rate and the calculation formula (9) of the F1 score include: Calculation formula (7); Calculation formula (8); Calculation formula (9); Among them, TP is the number of samples whose label is positive and the model predicts positive results. TN is the number of samples whose label is negative and the model predicts negative results. FP is the number of samples whose label is negative but the model predicts positive results. FN is the number of samples whose label is positive but the model predicts negative results.

[0079] In an actual application scenario, the target prediction model trained by the present application achieved an accuracy of 0.886, a recall rate of 0.857, and an F1 score of 0.857, indicating that the present invention can more accurately extract the time series characteristics of InSAR data, thereby completing the assessment of road collapse risks.

[0080] Step 204, obtaining second settlement data of the location to be predicted, inputting the second settlement data into the target prediction model, and obtaining the road collapse probability of the location to be predicted.

[0081] It should be noted that the time series length of the second settlement data is the same as that of the first settlement data. For example, the first settlement data is the data for the 13th month predicted based on the historical data for the 12th month, and the second settlement data is also the data for the 13th month predicted based on the historical data for the 12th month. The output road collapse probability of the predicted location is the probability value of the output label 1 (i.e., the probability of outputting the road collapse label).

[0082] Compared with the related art, the embodiments of the present application have at least the following advantages: Compared with the LSTM neural network used in the prior art, the dependency of capturing long-term data is poor. The present application adopts a preset prediction model for training, which can effectively mine the long-term relationship between the first settlement data and the road collapse label, thereby increasing the accuracy of predicting the probability of road collapse at the predicted location. At the same time, the network structure of the first model is simple, easy to implement, and consumes low resources. By stacking N first models, the depth of the model can be increased. While maintaining a simple neural network structure, the characteristics of long sequence data can be better captured, and the accuracy of the prediction results can be improved. Through the second model, new features can be obtained from the output results of the N first models, thereby completing the output of the road collapse probability. On the other hand, the structure of the neural network architecture used in the present application is simple, the model hardware deployment is superior in friendliness, and it is easier to apply to the end side.

[0083] Based on the same idea as the road collapse assessment method in the above embodiment, the present application also provides a road collapse assessment device, which can be used to perform the above road collapse assessment method. For ease of explanation, the structural diagram of the road collapse assessment device embodiment only shows the parts related to the embodiment of the present application. It can be understood by those skilled in the art that the illustrated structure does not constitute a limitation on the device, and may include more or fewer components than shown in the diagram, or combine certain components, or arrange the components differently.

[0084] like Figure 7 As shown, the road collapse assessment device 70 includes an acquisition unit 710, a first training unit 720, a second training unit 730 and a prediction unit 740. In some embodiments, the above modules may be programmable software instructions stored in a memory and can be called and executed by a processor. It is understood that in other embodiments, the above modules may also be program instructions or firmware solidified in the processor.

[0085] The acquisition unit 710 is used to acquire first settlement data of a sample location and a road collapse label corresponding to the sample location.

[0086] The first training unit 720 is used to input the first settlement data into a preset prediction model to obtain the road collapse prediction probability of the sample location; wherein the preset prediction model includes N first models and second models stacked in sequence, and obtaining the road collapse prediction probability of the sample location includes: inputting the first settlement data into the N first models to obtain a first output result, and inputting the first output result into the second model to obtain the road collapse prediction probability; The second training unit 730 determines a predicted road collapse label according to the road collapse prediction probability, updates a model parameter of a preset prediction model according to the predicted road collapse label and the road collapse label, and obtains a target prediction model; The prediction unit 740 obtains the second settlement data of the location to be predicted, inputs the second settlement data into the target prediction model, and obtains the road collapse probability of the location to be predicted.

[0087] In one embodiment, obtaining first settlement data of a sample location and a road collapse label corresponding to the sample location includes: obtaining first historical settlement data of the sample location over a historical period; dividing the first historical settlement data into monitoring data and verification data; converting the monitoring data into first settlement data in the form of a time series; and generating a road collapse label for the sample location based on the verification data.

[0088] In one embodiment, obtaining first historical settlement data of a sample location over a historical period includes: obtaining second historical settlement data of a sample city over a historical period; clustering the second historical settlement data based on a spatiotemporal clustering method to determine the accident location of the sample city; generating the sample location and the second historical settlement data based on the accident location to generate the first historical settlement data.

[0089] In one embodiment, the first settlement data is input into N first models to obtain a first output result, and the first output result is input into a second model to obtain a predicted probability of road collapse, including: inputting the first settlement data into the first first model, and inputting the output result of the first first model into the second first model, repeating the above process of inputting the output result of the previous first model into the next first model, until the last first model among the N first models outputs the first output result; inputting the first output result into the second model to obtain the predicted probability of road collapse at the sample location.

[0090] In one embodiment, the first model includes: a first linear layer, a first projection layer, a first module, a second module and a second projection layer, the first projection layer includes two upper projection layers, the first module includes: an upper projection layer, a causal convolution layer, a swish activation layer, three block diagonal linear layers, a single-dimensional convolution layer and a group normalization layer; the second module includes: an upper projection layer and a swish activation layer; wherein: the first linear layer is used to extract features from input data; the first projection layer is used to perform two projection data outputs on the output result of the first linear layer to obtain a first sub-feature and a second sub-feature; the first module is used to obtain a third sub-feature based on the first sub-feature; the second module is used to obtain a fourth sub-feature based on the second sub-feature; the second projection layer is used to output a first output result based on the third sub-feature and the fourth sub-feature.

[0091] In one embodiment, the second model includes: a second linear layer, a third module, a fourth module, a group normalization layer and a third projection layer; the third module includes: a causal convolution layer, a swish activation layer and two block diagonal linear layers; the fourth module includes: two block diagonal linear layers; the third projection layer includes two upper projection layers and one lower projection layer, wherein: the second linear layer is used to extract features from the first output result to obtain a fifth sub-feature and a sixth sub-feature; the third module is used to obtain a seventh sub-feature based on the fifth sub-feature; the fourth module is used to obtain an eighth sub-feature based on the sixth sub-feature; the group normalization layer is used to process the seventh sub-feature and the eighth sub-feature, and input the processing results of the group normalization layer into the third projection layer.

[0092] In one embodiment, model parameters of a preset prediction model are updated according to predicted road collapse labels and road collapse labels to obtain a target prediction model, including: updating parameters of the preset prediction model according to the difference between the predicted road collapse labels and the road collapse labels of the sample locations to obtain an updated model; continuously updating parameters of the updated model, calculating the loss function of the updated model, and when the loss function value converges, determining that the updated model is the target prediction model; and / or; obtaining an evaluation data set, inputting the evaluation data set into the updated model to obtain an evaluation result, calculating the accuracy of the evaluation result, and when the accuracy no longer increases, determining that the updated model is the target prediction model.

[0093] Compared with the related art, the embodiments of the present application have at least the following advantages: Compared with the LSTM neural network used in the prior art, the dependency of capturing long-term data is poor. The present application adopts a preset prediction model for training, which can effectively mine the long-term relationship between the first settlement data and the road collapse label, thereby increasing the accuracy of predicting the probability of road collapse at the predicted location. At the same time, the network structure of the first model is simple, easy to implement, and consumes low resources. By stacking N first models, the depth of the model can be increased. While maintaining a simple neural network structure, the characteristics of long sequence data can be better captured, and the accuracy of the prediction results can be improved. Through the second model, new features can be obtained from the output results of the N first models, thereby completing the output of the road collapse probability. On the other hand, the structure of the neural network architecture used in the present application is simple, the model hardware deployment is superior in friendliness, and it is easier to apply to the end side.

[0094] Please refer to Figure 8 , Figure 8 This is a schematic diagram of an embodiment of an electronic device of the present application.

[0095] The electronic device 100 includes a memory 20, a processor 30, and a computer program 40 stored in the memory 20 and executable on the processor 30. When the processor 30 executes the computer program 40, the steps in the above-mentioned road collapse assessment method embodiment are implemented, for example Figure 2 Steps 201 to 204 shown, or, Figure 3 Steps 301 to 302 are shown.

[0096] Exemplarily, the computer program 40 may also be divided into one or more modules / units, which are stored in the memory 20 and executed by the processor 30. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program 40 in the electronic device 100. For example, it may be divided into the acquisition unit 710, the first training unit 720, the second training unit 730, and the prediction unit 740 shown.

[0097] Those skilled in the art will understand that the schematic diagram is merely an example of the electronic device 100 and does not constitute a limitation of the electronic device 100. The electronic device 100 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device 100 may also include input and output devices, network access devices, buses, etc.

[0098] The processor 30 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor, a single-chip microcomputer, or the processor 30 may also be any conventional processor, etc.

[0099] The memory 20 can be used to store the computer program 40 and / or the module / unit. The processor 30 realizes various functions of the electronic device 100 by running or executing the computer program and / or the module / unit stored in the memory 20 and calling the data stored in the memory 20. The memory 20 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data (such as audio data) created according to the use of the electronic device 100, etc. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices.

[0100] If the module / unit integrated in the electronic device 100 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

Claims

1. A road collapse assessment method, characterized in that: include: Acquire first settlement data of a sample location and a road collapse label corresponding to the sample location; Inputting the first settlement data into a preset prediction model to obtain the predicted probability of road collapse at the sample location; wherein the preset prediction model includes N first models and second models stacked in sequence, and obtaining the predicted probability of road collapse at the sample location includes: inputting the first settlement data into the N first models to obtain a first output result, and inputting the first output result into the second model to obtain the predicted probability of road collapse; Determining a predicted road collapse label according to the road collapse prediction probability, and updating a model parameter of the preset prediction model according to the predicted road collapse label and the road collapse label to obtain a target prediction model; The second settlement data of the location to be predicted is obtained, and the second settlement data is input into the target prediction model to obtain the road collapse probability of the location to be predicted.

2. The road collapse assessment method according to claim 1, characterized in that: The step of obtaining first settlement data of a sample location and a road collapse label corresponding to the sample location includes: Acquire the first historical settlement data of the sample location in the historical period; dividing the first historical settlement data into monitoring data and verification data; Converting the monitoring data into the first settlement data in the form of a time series; A road collapse label of the sample location is generated according to the verification data.

3. The road collapse assessment method according to claim 2, characterized in that: The obtaining of the first historical settlement data of the sample location in the historical period includes: Obtain the second historical settlement data of the sample city during the historical period; Based on the spatiotemporal clustering method, clustering the second historical settlement data to determine the accident location of the sample city; A sample location is generated according to the accident location and the second historical settlement data, and the first historical settlement data is generated.

4. The road collapse assessment method according to claim 1, characterized in that: The step of inputting the first settlement data into the N first models to obtain a first output result, and inputting the first output result into the second model to obtain the road collapse prediction probability comprises: Input the first settlement data into a first first model, and input the output result of the first first model into a second first model, and repeat the process of inputting the output result of the previous first model into the next first model, until the last first model among the N first models outputs the first output result; The first output result is input into the second model to obtain the predicted probability of road collapse at the sample location.

5. The road collapse assessment method according to claim 1, characterized in that: The first model includes: a first linear layer, a first projection layer, a first module, a second module, and a second projection layer, the first projection layer includes two upper projection layers, the first module includes: the upper projection layer, a causal convolution layer, a swish activation layer, three block diagonal linear layers, a single-dimensional convolution layer, and a group normalization layer; the second module includes: the upper projection layer and the swish activation layer; wherein: The first linear layer is used to extract features from input data; The first projection layer is used to perform two projection data outputs on the output result of the first linear layer to obtain a first sub-feature and a second sub-feature; The first module is used to obtain a third sub-feature according to the first sub-feature; The second module is used to obtain a fourth sub-feature according to the second sub-feature; The second projection layer is used to output the first output result according to the third sub-feature and the fourth sub-feature.

6. The road collapse assessment method according to claim 5, characterized in that: The second model includes: a second linear layer, a third module, a fourth module, the group normalization layer and a third projection layer; the third module includes: the causal convolution layer, the swish activation layer and two block diagonal linear layers; the fourth module includes: two block diagonal linear layers; the third projection layer includes two upper projection layers and one lower projection layer, wherein; The second linear layer is used to extract features from the first output result to obtain fifth sub-features and sixth sub-features; The third module is used to obtain a seventh sub-feature according to the fifth sub-feature; The fourth module is used to obtain an eighth sub-feature according to the sixth sub-feature; The group normalization layer is used to process the seventh sub-feature and the eighth sub-feature, and input the processing result of the group normalization layer into the third projection layer.

7. The road collapse assessment method according to claim 1, characterized in that: The step of updating the model parameters of the preset prediction model according to the predicted road collapse label and the road collapse label to obtain a target prediction model includes: According to the difference between the predicted road collapse label of the sample location and the road collapse label, updating the parameters of the preset prediction model to obtain an updated model; Continuously updating the parameters of the updated model, calculating the loss function of the updated model, and when the loss function converges, determining that the updated model is a target prediction model; and / or; Obtain an evaluation data set, input the evaluation data set into the updated model, obtain an evaluation result, calculate the accuracy of the evaluation result, and when the accuracy no longer increases, determine that the updated model is the target prediction model.

8. A road collapse assessment device, characterized in that: include: An acquisition unit, used to acquire first settlement data of a sample location and a road collapse label corresponding to the sample location; A first training unit is configured to input the first settlement data into a preset prediction model to obtain a road collapse prediction probability at the sample location; wherein the preset prediction model includes N first models and second models stacked in sequence, and obtaining the road collapse prediction probability at the sample location includes: inputting the first settlement data into the N first models to obtain a first output result, and inputting the first output result into the second model to obtain the road collapse prediction probability; A second training unit determines a predicted road collapse label according to the road collapse prediction probability, updates a model parameter of the preset prediction model according to the predicted road collapse label and the road collapse label, and obtains a target prediction model; The prediction unit obtains second settlement data of the location to be predicted, inputs the second settlement data into the target prediction model, and obtains the road collapse probability of the location to be predicted.

9. An electronic device, comprising a processor and a memory, characterized in that: The memory is used to store instructions, and the processor is used to call the instructions in the memory, so that the electronic device executes the road collapse assessment method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed on an electronic device, the electronic device executes the road collapse assessment method according to any one of claims 1 to 7.

Images