A processing method, device, and processing equipment for a flood vulnerability prediction model

Abstract

The present application provides a processing method, device, and processing equipment for a flood vulnerability prediction model, offering a novel model training architecture. By using a single-sample data augmentation method based on one-dimensional multi-feature structured data and coupling and applying a consistency regularization semi-supervised learning method, it reduces the instability of prediction results caused by data problems, can better complete model training, provides the flood vulnerability prediction accuracy of the model, and has good practical application value.

Description

Technical Field

[0001] This application relates to the field of flood susceptibility prediction, and particularly to a processing method, device, and processing equipment for a flood susceptibility prediction model. Background Art

[0002] Flood is one of the most common and destructive natural disasters worldwide. Therefore, accurately identifying and predicting areas vulnerable to rainstorm floods is of great significance for rainstorm flood prevention and reducing disaster losses.

[0003] Flood susceptibility refers to the likelihood of a flood disaster occurring in a specific area under local topographical, hydrological, and meteorological conditions. Flood susceptibility prediction is a means of predicting flood-prone areas, which can provide data support for flood response work, thereby enabling early response to floods or reducing the damage caused by floods.

[0004] However, the inventors of this application found that when predicting flood susceptibility through the currently widely used supervised learning method, due to the scarcity of data volume and one-sidedness of data records in the flood dataset, there will be a problem of uncertainty in the performance of the trained model.

[0005] Semi-Supervised Learning is a machine learning method that lies between supervised learning and unsupervised learning, and is mainly applied to situations where labeled data is scarce or sample labeling is difficult. In semi-supervised learning, the model can use a small amount of labeled data and a large amount of unlabeled data for learning. Compared with supervised learning, the semi-supervised learning method can make full use of the information of unlabeled data, thereby obtaining better generalization performance and higher prediction accuracy.

[0006] In semi-supervised learning methods, the utilization methods of unlabeled data can be divided into two paradigms: ProxyLabel and Consistency Regularization.

[0007] For the proxy label method, first, the proxy label method depends on the prediction results of the existing model. If the performance of the initial model is poor, the obtained proxy labels may also be inaccurate, thus affecting the training and performance of the model; second, the proxy label method performs poorly in dealing with noise and mislabeled data because it cannot distinguish between true labels and proxy labels.

[0008] For the consistency regularization method, which is a common semi-supervised learning regularization technique, aiming to improve the generalization ability of the model by constraining the consistency between different inputs or different model predictions. Consistency regularization usually generates new samples by slightly perturbing unlabeled samples and requires these new samples to be consistent with the original samples in the model output. This perturbation technique usually adopts data augmentation methods to enhance the performance of the model through data augmentation of a single sample and the consistency constraint with the original sample.

[0009] In the fields of images, natural language, frequency signals, etc., semi-supervised learning methods based on consistency regularization have achieved remarkable results. However, in the field of flood susceptibility prediction, flood inventory datasets usually exist in a one-dimensional multi-feature structured form. Currently, data augmentation techniques for such datasets mainly target multiple samples, lacking data augmentation methods for a single sample. Due to this limitation, semi-supervised learning methods with consistency regularization are difficult to be effectively applied in the field of flood susceptibility prediction. Summary of the Invention

[0010] This application provides a processing method, device, and processing equipment for a flood susceptibility prediction model, providing a novel model training architecture. By using a single-sample data augmentation method based on one-dimensional multi-feature structured data and coupling with a consistency regularization semi-supervised learning method, it reduces the instability of prediction results caused by data problems, can effectively complete model training, improve the flood susceptibility prediction accuracy of the model, and has good practical application value.

[0011] In a first aspect, this application provides a processing method for a flood susceptibility prediction model, and the method includes:

[0012] Obtain a flood inventory dataset, where the flood inventory dataset includes a first flood inventory dataset with labels and a second flood inventory dataset without labels. The label identifies whether there is a flood occurrence in the corresponding flood inventory data part, and the flood inventory dataset uses the quantitative values of preset flood impact factors as features;

[0013] Based on the k-means clustering method and the SMOTE method, perform single-sample data augmentation on the second flood inventory dataset in the flood inventory data to obtain an augmented dataset;

[0014] Based on the two datasets of the flood inventory dataset and the augmented dataset, combine the consistency regularization semi-supervised learning method to train a multi-layer perceptron model to obtain a flood susceptibility prediction model, where the flood susceptibility prediction model is used to predict the corresponding flood susceptibility based on the flood impact factor feature values input into the model.

[0015] In a second aspect, the present application provides a processing device for a flood vulnerability prediction model. The device includes:

[0016] An acquisition unit configured to acquire a flood inventory data set. The flood inventory data set includes a first flood inventory data set with labels and a second flood inventory data set without labels. The labels identify whether there is a flood occurrence in the corresponding part of the flood inventory data. The flood inventory data set uses the quantitative values of preset flood impact factors as features.

[0017] An enhancement unit configured to perform single-sample data enhancement on the second flood inventory data set in the flood inventory data based on the k-means clustering method and the SMOTE method to obtain an enhanced data set.

[0018] A training unit configured to train a multi-layer perceptron model based on the flood inventory data set and the enhanced data set in combination with the consistency regularization semi-supervised learning method to obtain a flood vulnerability prediction model. The flood vulnerability prediction model is used to predict the corresponding flood vulnerability based on the flood impact factor feature values input into the model.

[0019] In a third aspect, the present application provides a processing device including a processor and a memory. A computer program is stored in the memory. When the processor calls the computer program in the memory, it executes the method provided in the first aspect of the present application or any possible implementation manner of the first aspect of the present application.

[0020] In a fourth aspect, the present application provides a computer-readable storage medium storing multiple instructions suitable for being loaded by a processor to execute the method provided in the first aspect of the present application or any possible implementation manner of the first aspect of the present application.

[0021] From the above content, the following beneficial effects of the present application can be obtained:

[0022] Aiming at the flood vulnerability prediction target based on consistency regularization semi-supervised learning, the present application provides a novel model training architecture. By using a single-sample data enhancement method based on one-dimensional multi-feature structured data and coupling the application of the consistency regularization semi-supervised learning method, it reduces the instability of the prediction results caused by data problems, can better complete model training, improves the flood vulnerability prediction accuracy of the model, and has good practical application value. Description of the Drawings

[0023] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those skilled in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0024] Figure 1 It is a schematic flowchart of a processing method for the flood susceptibility prediction model of the present application;

[0025] Figure 2 It is a schematic architecture diagram of a consistency regularization semi-supervised learning method of the present application;

[0026] Figure 3 It is a schematic diagram of the result of the AUC value of the model of the present application changing with the k value;

[0027] Figure 4 It is a schematic structural diagram of a processing device for the flood susceptibility prediction model of the present application;

[0028] Figure 5 It is a schematic structural diagram of a processing device of the present application. Specific embodiments

[0029] The following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0030] Terms such as "first" and "second" in the specification, claims and drawings of the present application are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or modules does not necessarily limit to those clearly listed steps or modules, but may include other steps or modules not clearly listed or inherent to these processes, methods, products or devices. The naming or numbering of steps in the present application does not mean that the steps in the method flow must be executed in the time / logical order indicated by the naming or numbering. The already named or numbered process steps can change the execution order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0031] The division of modules in this application is a logical division. In actual implementation, there may be other division methods. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Additionally, the couplings, direct couplings, or communication connections shown or discussed among each other can be through some interfaces. The indirect couplings or communication connections between modules can be electrical or other similar forms, which are not limited in this application. Moreover, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed into multiple circuit modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this application.

[0032] Before introducing the processing method of the flood vulnerability prediction model provided by this application, the background content involved in this application will be introduced first.

[0033] The processing method, device, and computer-readable storage medium of the flood vulnerability prediction model provided by this application can be applied to a processing device, providing a novel model training architecture. By using a single-sample data augmentation method based on one-dimensional multi-feature structured data and coupling and applying a consistency regularization semi-supervised learning method, it reduces the instability of the prediction results caused by data problems, can complete model training with better effects, improves the flood vulnerability prediction accuracy of the model, and has good practical application value.

[0034] The execution subject of the processing method of the flood vulnerability prediction model mentioned in this application can be a processing device for the flood vulnerability prediction model, or different types of processing devices such as a server, a physical host, or a user equipment (UE) that integrates the processing device for the flood vulnerability prediction model. Among them, the processing device for the flood vulnerability prediction model can be implemented in a hardware or software manner. The UE can specifically be a terminal device such as a smart phone, a tablet computer, a notebook computer, a desktop computer, or a personal digital assistant (PDA). The processing device can be set up in the form of a device cluster.

[0035] It can be understood that the key point of the solution of this application lies in model training processing, and it is model training processing carried out on the basis of existing data. Therefore, the processing device that executes the processing method of the flood vulnerability prediction model of this application, or rather, the processing device that carries the corresponding application service of the processing method of the flood vulnerability prediction model of this application, only needs to meet the requirements in terms of data processing ability in actual situations, and its specific device type and device deployment form can be flexibly configured according to actual needs.

[0036] For the specific data collection work, the processing device needs to be configured with corresponding data collection functions (involving software and hardware configuration work).

[0037] Similarly, for the actual application after the trained model, the processing device also needs to be configured according to the specific application scenario.

[0038] As an example, in this case, the processing device may specifically include a first device part for training the model in the background and a second device part for applying the model on-site.

[0039] Next, the processing method of the flood susceptibility prediction model provided by this application will be introduced.

[0040] First, refer to Figure 1 , Figure 1 which shows a schematic flowchart of a processing method of the flood susceptibility prediction model provided by this application. The processing method of the flood susceptibility prediction model provided by this application may specifically include the following steps S101 to S103:

[0041] Step S101: Obtain a flood inventory dataset. Among them, the flood inventory dataset includes a first flood inventory dataset with labels and a second flood inventory dataset without labels. The label identifies whether there is a flood occurrence in the corresponding part of the flood inventory data. The flood inventory dataset uses the quantitative values of preset flood impact factors as features;

[0042] It can be understood that to train the flood susceptibility prediction model, training samples for training need to be configured. Here, the flood inventory dataset configured is the initial training sample obtained. Subsequently, further processing of this initial training sample is required to obtain a better-performing overall training sample.

[0043] Among them, the acquisition and processing of the flood inventory dataset here are usually the acquisition and processing of existing data in actual situations. For example, receiving a manually entered flood inventory dataset, or extracting the flood inventory dataset from a local storage location, or downloading / receiving the flood inventory dataset from a certain device.

[0044] Of course, in some cases, real-time collection of the flood inventory dataset is not excluded. In this process, different link data processing work such as docking the data source of the original data may also be involved.

[0045] It should be understood that the model training architecture adopted in this application belongs to the consistency regularization semi-supervised learning method / architecture. In this regard, a flood inventory dataset that uses labels to mark whether there is a flood occurrence (for example, 1 indicates a flood and 0 indicates no flood) and a flood inventory dataset that does not use labels to mark whether there is a flood occurrence are configured. This situation belongs to the category of the prior art, so specific explanations will not be elaborated here.

[0046] Among them, the substantial content of the flood inventory dataset uses the quantitative values of preset flood impact factors as features. Therefore, it can also be understood by the feature set corresponding to the flood impact factors. Here, it corresponds to the working principle of the flood susceptibility prediction model to be trained. The flood susceptibility prediction model is used to predict the corresponding flood susceptibility based on the flood impact factor feature values input into the model.

[0047] It can be understood that a prediction process of flood susceptibility usually involves multiple and different flood impact factors (variables, and the specific feature conditions of the corresponding features are represented by different numerical values). The specific flood impact factors involved can be directly configured manually, directly searched by the machine through the search method, or optimized through search within the factor range configured manually (i.e., the initial factor set). Considering that how to configure the flood impact factors is not the focus of the solution of this application, specific explanations will not be elaborated here either.

[0048] Step S102: Based on the k-means clustering method and the SMOTE method, perform single-sample data augmentation on the second flood inventory dataset in the flood inventory data to obtain an augmented dataset;

[0049] It can be understood that in the field of flood susceptibility prediction, in actual situations, the flood inventory dataset usually exists in a one-dimensional multi-feature structured form. Currently, the data augmentation techniques for such datasets mainly target multiple samples and lack a data augmentation method for a single sample. Due to this limitation, the consistency regularization semi-supervised learning method is difficult to be effectively applied in the field of flood susceptibility.

[0050] In this regard, this application can perform further data processing on the basis of the flood inventory data obtained previously to obtain training samples adapted to the consistency regularization semi-supervised learning method.

[0051] Specifically, this application focuses on the second flood inventory dataset without tags (used to mark whether floods occur). Through the coupled application of the k-means clustering method and the SMOTE method, a single-sample data augmentation method is implemented to perform single-sample data augmentation on the second flood inventory dataset, so as to overcome the uncertainty problem of the model performance of the subsequently trained model due to the scarcity of the data volume and the one-sidedness of the data records in the flood dataset. The newly obtained flood inventory dataset can be recorded as the augmented dataset.

[0052] Regarding the k-means clustering method involved, it can also be called the k-means clustering algorithm (K-Means Clustering Algorithm, KMA). It is an existing clustering algorithm, and its main processing content can be understood as follows:

[0053] Pre-divide the data into K groups, then randomly select K objects as the initial clustering centers, and then calculate the distances between each object and each seed clustering center, and assign each object to the clustering center closest to it. The clustering center and the objects assigned to it represent a cluster. For each assigned sample, the clustering center of the cluster will be recalculated based on the existing objects in the cluster. This process will be repeated continuously until a certain termination condition is met. The termination condition can be that no (or the minimum number) of objects are reassigned to different clusters, no (or the minimum number) of clustering centers change anymore, and the sum of squared errors is locally minimized.

[0054] Regarding the SMOTE method involved, that is, the Synthetic Minority Oversampling Technique, it is an improved scheme based on the random oversampling algorithm. Since the random oversampling adopts the strategy of simply replicating samples to increase the minority class samples, this is likely to cause the problem of model overfitting, that is, the information learned by the model is too specific and not general enough. The basic idea of the SMOTE algorithm is to analyze the minority class samples and artificially synthesize new samples according to the minority class samples and add them to the dataset. Its main processing content can be understood as follows:

[0055] (1) For each sample x in the minority class, calculate its distances to all samples in the minority class sample set with the Euclidean distance as the standard to obtain its k nearest neighbors; (2) Set a sampling ratio according to the sample imbalance ratio to determine the sampling magnification N. For each minority class sample x, randomly select several samples from its k nearest neighbors. Suppose the selected nearest neighbor is o; (3) For each randomly selected nearest neighbor o, construct a new sample with the original sample according to the formula o(new) = o + rand(0,1) * (x - o).

[0056] Step S103: Based on the two datasets of the flood inventory dataset and the enhanced dataset, combine the consistency regularization semi-supervised learning method to train a multi-layer perceptron model to obtain a flood susceptibility prediction model, where the flood susceptibility prediction model is used to predict the corresponding flood susceptibility based on the flood impact factor eigenvalue input into the model.

[0057] It can be understood that after obtaining the new training samples of the enhanced dataset through the previous step S102, it can form an overall new training sample with the original flood inventory dataset as the training sample.

[0058] In this case, the model training logic of the consistency regularization semi-supervised learning method can be followed to perform specific model training processing on the initial model specifically selected as the multi-layer perceptron model, so as to obtain a flood susceptibility prediction model that has completed training and can be put into actual use.

[0059] Among them, the multi-layer perceptron model (Multilayer Perceptron, MLP) is a basic and widely used artificial neural network model. Its structure consists of multiple layers, including three parts: an input layer, at least one hidden layer, and an output layer. With its powerful non-linear mapping ability and flexible structure design, MLP demonstrates excellent performance in multiple fields such as classification, regression, and pattern recognition.

[0060] For the model training process, the following main processing contents may be involved:

[0061] In each round of model training, a set of training samples is input into the model, enabling the model to perform corresponding flood susceptibility predictions to achieve forward propagation. Then, based on the flood susceptibility prediction results output by the model, the loss function is calculated, and the model parameters are optimized according to the calculation results of the loss function to achieve backward propagation. In this way, when the model training requirements such as training duration, training times, or prediction accuracy are met, the model can be completed. At this time, the model is the flood susceptibility prediction model that can be put into actual use.

[0062] From Figure 1 As can be seen from the illustrated embodiments, for the flood susceptibility prediction target based on the consistency regularization semi-supervised learning, this application provides a novel model training architecture. By using a single-sample data augmentation method based on one-dimensional multi-feature structured data and coupling the application of the consistency regularization semi-supervised learning method, the instability of the prediction results caused by data problems is reduced, and the model training can be better completed, providing the flood susceptibility prediction accuracy of the model, which has good practical application value.

[0063] Next, continue with the above Figure 1Elaborate in detail on the steps of the illustrated embodiment and its possible embodiments in practical applications.

[0064] As an exemplary embodiment, step S102 performs data augmentation on the second flood inventory dataset in the flood inventory data based on the k-means clustering method and the SMOTE method to obtain an augmented dataset, which specifically may include:

[0065] Cluster the second flood inventory dataset in the flood inventory data through the k-means clustering method to obtain multiple clustering categories;

[0066] For the multi-sample clustering categories, use the SMOTE method to generate new samples among the multiple samples included in the multi-sample clustering categories, and at the same time do not process the samples included in the single-sample clustering categories, so as to achieve data augmentation for the flood inventory data and obtain an augmented dataset.

[0067] It can be understood that during the clustering process, while determining the clustering categories, samples with similarity based on the cluster centers will also be determined accordingly. These samples are the samples included in the clustering categories. Generally speaking, the clustering process is to divide the original dataset into multiple clustering categories, each category representing a specific data pattern or feature combination, and the samples within each group have similar characteristics under a certain similarity metric.

[0068] For the clustering results, there are two cases. The first case is the multi-sample clustering category (the clustering category contains multiple samples), and the second case is the single-sample clustering category (the clustering category contains only one sample). And this application aims at the multi-sample clustering categories among them and performs data augmentation operations.

[0069] Specifically, for the multi-sample clustering categories, use the SMOTE method to generate new samples among the multiple samples included in the clustering categories. These new samples inherit the feature distribution of the original samples, increasing the diversity and sample size of the dataset, which helps to improve the model training effect; for the single-sample clustering category, the data augmentation result is defined as the sample itself. In this case, since there are not enough samples to apply the SMOTE method, the original sample is retained as the output of data augmentation, so as to ensure the robustness of the data augmentation process.

[0070] Furthermore, the following set of pseudo-code examples can be used to help understanding:

[0071]

[0072] For the above pseudo-code, there is:

[0073] The first line defines the method inputs, including: dataset D (containing both labeled and unlabeled data), the number of clusters K (i.e., how many clusters the dataset is expected to be divided into), and the parameter settings of the SMOTE algorithm;

[0074] The second line defines the method output, i.e., the dataset after data augmentation, called the augmented dataset D_augmented;

[0075] Lines 3 to 5 are the first step of the method, clustering analysis. The K-means algorithm is used to cluster the dataset D to obtain K clusters. The purpose of this step is to group the samples based on their similarity. For each cluster i, all the samples in that cluster are extracted to form a sample set, denoted as sample set Si. In this way, each cluster corresponds to a sample set that contains all the samples in that cluster.

[0076] The second step of the method is data augmentation. First, an empty dataset, the augmented dataset D_augmented, is initialized to store all the new samples generated after data augmentation. This part is in line 6.

[0077] Lines 7 - 12 first iterate through the sample set Si in each cluster i, and then perform data augmentation operations on each sample set. If the number of samples in the sample set Si is greater than 1, it means there are enough samples for SMOTE data augmentation. In this case, the SMOTE method is applied to the sample set Si to generate a new sample set, the new sample set Ni, and the new sample set Ni generated by the SMOTE method is added to the augmented dataset D_augmented to expand the dataset (lines 8 - 10). If there is only one sample in the sample set Si, then the single sample in the sample set Si is directly added to the augmented dataset D_augmented. In the case of only one sample, the SMOTE method cannot be applied, so the sample is directly retained (lines 11 - 12).

[0078] Finally, the augmented dataset D_augmented after data augmentation is returned (line 13).

[0079] For the model training architecture of the consistency regularization semi-supervised learning method, as can be seen from the previous content, this application has made certain modifications to the one-dimensional multi-feature structured data involved in using it for the application scenario of flood susceptibility prediction and introduced data augmentation operations.

[0080] Briefly, the data augmentation of the original consistency regularization semi-supervised learning method uses the RandAugment and back-translation methods for data augmentation of image data and text data respectively.

[0081] In this application, a single-sample data augmentation method is adopted to augment the flood vulnerability prediction dataset. During the data augmentation process, for a given single unlabeled sample input x, the augmented data input x' is generated through the single-sample data augmentation method.

[0082] In addition, to more vividly understand the configuration work of this application corresponding to the model training architecture of the consistency regularization semi-supervised learning method, reference can also be made to Figure 2 a schematic diagram of an architecture of the consistency regularization semi-supervised learning method of this application shown in Figure 2 From which it can be seen that this application has further optimization settings.

[0083] Specifically, as an exemplary embodiment, for the consistency regularization semi-supervised learning method involved in this application, the following configuration content can be available:

[0084] Use binary cross-entropy loss to replace the original loss term;

[0085] Perform model predictions on the second flood inventory dataset and the augmented dataset respectively to obtain the output probability distributions in two aspects, and calculate the distance between the output probability distributions in the two aspects through KL divergence as the unsupervised loss term;

[0086] The loss after model prediction on the first flood inventory dataset is used as the supervised loss term;

[0087] The overall loss is the sum of the supervised loss term and λ times the unsupervised loss term.

[0088] Specifically, for labeled data, traditional supervised losses of the consistency regularization semi-supervised learning method, such as cross-entropy loss, are used to ensure the performance of the model on labeled data. And this application takes into account that since the data used is binary classification data, binary cross-entropy loss is used to replace the original loss term.

[0089] In addition, for the unlabeled data generated through data augmentation, the consistency loss of the model output between the original unlabeled sample and the augmented sample is calculated to encourage the model to maintain stable predictions for input changes. That is, calculate the output of both on the same model, and calculate the distance between their output probability distributions through the Kullback-Leibler divergence as the unsupervised loss term

[0090] On the other hand, for labeled samples, there is a normal supervised loss term;

[0091] In this case, the overall loss term is the sum of the supervised loss term and λ times the unsupervised consistency loss term.

[0092] During the training process, the model is trained by iteratively optimizing the overall loss function. In each iteration, the supervised loss on a batch of labeled data and the consistency loss corresponding to a batch of unlabeled data generated by data augmentation are calculated, and then the model parameters are updated to minimize these losses.

[0093] It can be seen that the embodiments herein show how to achieve better model training effects, and the present application provides an optimized and novel architecture setting for the consistency regularization semi-supervised learning method.

[0094] In addition, for the initial model involved in the present application and the flood susceptibility prediction model obtained after training the initial model, that is, the multi-layer perceptron model, as an exemplary embodiment, it can specifically be a fully connected layer artificial neural network configured with 1 input layer, 2 hidden layers, and 1 output layer. The number of nodes in each layer is 13, 128, 128, and 2. The activation function of the middle layer of the network is the ReLU activation function, and the activation function between the hidden layer and the output layer is the Sigmoid activation function.

[0095] It can be understood that specific implementation solutions are provided herein for the model architecture involved in the present application and its specific parameters / structures.

[0096] At the same time, for the flood susceptibility prediction model of the present application, it can also involve model setting work such as hyperparameters. These works can obviously be adjusted according to actual needs. Of course, it also includes some routine operations. It can be understood that the present application does not expand too much on the routine operations that can be adopted.

[0097] Taking the tuning process of hyperparameters as an example, as an exemplary embodiment, the configuration method of the flood susceptibility prediction model of the present application can further include:

[0098] Carry out hyperparameter tuning for the clustering category k value and / or the weight λ of the unsupervised loss term involved in the consistency regularization semi-supervised learning method.

[0099] For the clustering category k value, it can be understood that it is a parameter involved in the clustering process of the K-means clustering method, and the present application can take it as a hyperparameter to carry out hyperparameter tuning to promote better model training / prediction effects.

[0100] Among them, in terms of the specific hyperparameter k value, it can help to select the best data augmentation data in the single-sample augmentation process.

[0101] Similarly, in the embodiments herein, the present application also regards the weight λ of the unsupervised loss term involved above as another hyperparameter to carry out hyperparameter tuning to promote better model training / prediction effects.

[0102] Among them, in specific aspects, the hyperparameter λ can help to better carry out the backpropagation process of the model and refer to the parameter adjustment of the model with higher efficiency / accuracy.

[0103] As an example here, taking the flood vulnerability dataset of a certain city as an example, the number of labeled data samples in the dataset is 532, and the number of unlabeled data samples is 5000. After adjusting the weight of the unsupervised loss term, when λ > 1, the model diverges. After debugging, the final value of λ is 0.5, and the value of k ranges from 20 to 3000. The model for each k value is trained 10 times, and its model AUC value is recorded. Figure 3 A schematic diagram showing a result of the change of the model AUC value of the present application with the k value is shown. In Figure 3 Among them, the box plot in the vertical axis direction is the model AUC value trained 10 times under the corresponding K value. The average value of the model AUC value under each K value is also shown in the figure. The line connecting the average values is the change curve. Finally, the box plot with the subscript x is the AUC value of the supervised learning MLP model.

[0104] Among them, the AUC value (Area Under Curve), which is defined as the area enclosed by the ROC curve and the coordinate axes, is a commonly used index in model evaluation metrics.

[0105] In specific operations, the selection of the k value gives priority to the average accuracy of the AUC value, and secondly, pays attention to the stability of the model. Based on this, as an exemplary embodiment, the present application can specifically select 240 as the optimal value of the k value.

[0106] After testing with the test set, the AUC value of the model is 0.918, and the accuracy of the MLP trained by supervised learning is 0.886. Compared with supervised learning, the performance of the model trained by this method has been greatly improved.

[0107] After the model training is completed, obviously, it can be put into actual use for flood vulnerability prediction processing in a specific / specified area.

[0108] Regarding this, as an exemplary embodiment, the processing method of the flood vulnerability prediction model of the present application may further include:

[0109] Obtain the target flood impact factor eigenvalue of the target area;

[0110] Input the target flood impact factor eigenvalue into the flood vulnerability prediction model, so that the flood vulnerability prediction model predicts the corresponding flood vulnerability;

[0111] Extract the flood vulnerability prediction result of the target area output by the flood vulnerability prediction model.

[0112] It can be understood that in practice, the target area is usually determined according to the currently initiated flood susceptibility prediction task. The initiation method of the flood susceptibility prediction task is flexible. It can be a manually initiated task, a task forwarded by other devices, or a task autonomously initiated according to a pre-configured task initiation strategy. All of these are possible.

[0113] After determining the target area, the corresponding flood impact factor characteristic values of it, that is, the target flood impact factor characteristic values, which can also be referred to as a feature set, can be obtained. This data corresponds to the features involved in the training samples mentioned above.

[0114] In this way, after inputting the target flood impact factor characteristic values into the configured flood susceptibility prediction model, the flood susceptibility prediction results output after the model performs flood susceptibility prediction processing can be extracted.

[0115] At this time, the flood susceptibility prediction results can also be stored, displayed, forwarded, further processed, or a prompt for completing flood susceptibility can be output. Obviously, the subsequent data application link for the flood susceptibility prediction results can also be flexibly configured and adjusted according to the pre-configured application requirements and real-time application requirements. The present application does not make specific limitations.

[0116] The above is an introduction to the processing method of the flood susceptibility prediction model provided by the present application. To facilitate better implementation of the processing method of the flood susceptibility prediction model provided by the present application, the present application also provides a processing device for the flood susceptibility prediction model from the perspective of functional modules.

[0117] Refer to Figure 4 , Figure 4 which is a schematic structural diagram of a processing device for the flood susceptibility prediction model of the present application. In the present application, the processing device 400 for the flood susceptibility prediction model specifically may include the following structure:

[0118] An acquisition unit 401, configured to acquire a flood inventory data set, where the flood inventory data set includes a first flood inventory data set with labels and a second flood inventory data set without labels. The label identifies whether there is a flood occurrence in the corresponding part of the flood inventory data. The flood inventory data set uses the quantitative values of preset flood impact factors as features;

[0119] An enhancement unit 402, configured to perform single-sample data enhancement on the second flood inventory data set in the flood inventory data based on the k-means clustering method and the SMOTE method to obtain an enhanced data set;

[0120] A training unit 403, which is used to train a multi-layer perceptron model based on two datasets, namely a flood inventory dataset and an enhanced dataset, by combining a consistency regularization semi-supervised learning method to obtain a flood susceptibility prediction model. The flood susceptibility prediction model is used to predict the corresponding flood susceptibility based on the flood impact factor eigenvalue input into the model.

[0121] In an exemplary embodiment, the enhancement unit 402 is specifically configured to:

[0122] Cluster the second flood inventory dataset in the flood inventory data through the k-means clustering method to obtain multiple clustering categories;

[0123] For the multi-sample clustering categories, the SMOTE method is used to generate new samples among the multiple samples included in the multi-sample clustering categories, while not processing the samples included in the single-sample clustering categories, so as to achieve data enhancement for the flood inventory data and obtain an enhanced dataset.

[0124] In another exemplary embodiment, for the consistency regularization semi-supervised learning method, the following configuration content is available:

[0125] Use binary cross-entropy loss to replace the original loss term;

[0126] The second flood inventory dataset and the enhanced dataset are respectively used for model prediction to obtain the output probability distributions in two aspects, and the KL divergence is used to calculate the distance between the output probability distributions in the two aspects as the unsupervised loss term;

[0127] The loss after model prediction of the first flood inventory dataset is used as the supervised loss term;

[0128] The total loss is the sum of the supervised loss term and λ times the unsupervised loss term.

[0129] In another exemplary embodiment, the multi-layer perceptron model is a fully connected layer artificial neural network configured with 1 input layer, 3 hidden layers and 1 output layer. The number of nodes in each layer is 13, 128, 128 and 2. The activation function of the middle layer of the network is the ReLU activation function, and the activation function between the hidden layer and the output layer is the Sigmoid activation function.

[0130] In another exemplary embodiment, during the model training process, the device further includes a parameter tuning unit 404, which is used to:

[0131] Perform hyperparameter tuning for the clustering category k value and / or the weight λ of the unsupervised loss term involved in the consistency regularization semi-supervised learning method.

[0132] In yet another exemplary embodiment, after the hyperparameter tuning process, the value of the clustering category k is 240 and λ is 0.5.

[0133] In yet another exemplary embodiment, the apparatus further includes an application unit 405 for:

[0134] Obtain the target flood impact factor eigenvalue of the target area;

[0135] Input the target flood impact factor eigenvalue into the flood susceptibility prediction model, so that the flood susceptibility prediction model predicts the corresponding flood susceptibility based on it;

[0136] Extract the flood susceptibility prediction result of the target area output by the flood susceptibility prediction model.

[0137] The present application also provides a processing device from the perspective of the hardware structure. Refer to Figure 5 , Figure 5 FIG. shows a schematic structural diagram of the processing device of the present application. Specifically, the processing device of the present application may include a processor 501, a memory 502, and an input / output device 503. When the processor 501 executes the computer program stored in the memory 502, it implements the steps of the processing method of the flood susceptibility prediction model in the corresponding Figure 1 embodiment; or, when the processor 501 executes the computer program stored in the memory 502, it implements the functions of each unit in the corresponding Figure 4 embodiment. The memory 502 is used to store the computer program required for the processor 501 to execute the processing method of the flood susceptibility prediction model in the above Figure 1 corresponding embodiment.

[0138] Exemplarily, the computer program can be divided into one or more modules / units. One or more modules / units are stored in the memory 502 and executed by the processor 501 to complete the present application. One or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the computer device.

[0139] The processing device may include, but is not limited to, the processor 501, the memory 502, and the input / output device 503. Those skilled in the art can understand that the schematic diagram is only an example of the processing device and does not constitute a limitation on the processing device. It may include more or fewer components than shown, or combine certain components, or different components. For example, the processing device may further include a network access device, a bus, etc. The processor 501, the memory 502, the input / output device 503, etc. are connected through the bus.

[0140] The processor 501 can be a Central Processing Unit (CPU), or it can also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the processing device and connects various parts of the entire device using various interfaces and lines.

[0141] The memory 502 can be used to store computer programs and / or modules. The processor 501 realizes various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 502, and by calling the data stored in the memory 502. The memory 502 mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function, etc.; the data storage area can store data created according to the use of the processing device, etc. In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disks, memory, plug-in hard disks, Smart Media Cards (SMCs), Secure Digital (SD) cards, Flash Cards, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage devices.

[0142] When the processor 501 is used to execute the computer program stored in the memory 502, the following functions can be specifically realized:

[0143] Obtain a flood inventory dataset, where the flood inventory dataset includes a first flood inventory dataset with labels and a second flood inventory dataset without labels. The label identifies whether there is a flood occurrence in the corresponding part of the flood inventory data. The flood inventory dataset uses the quantitative value of a preset flood impact factor as a feature;

[0144] Based on the k-means clustering method and the SMOTE method, perform single-sample data augmentation on the second flood inventory dataset in the flood inventory data to obtain an augmented dataset;

[0145] Based on two types of datasets, namely the flood inventory dataset and the enhanced dataset, a multi-layer perceptron model is trained by combining the consistency regularization semi-supervised learning method to obtain a flood susceptibility prediction model. The flood susceptibility prediction model is used to predict the corresponding flood susceptibility based on the characteristic values of flood impact factors input into the model.

[0146] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the processing device, processing equipment and their corresponding units of the flood susceptibility prediction model described above can refer to Figure 1 the description of the processing method of the flood susceptibility prediction model in the corresponding embodiment, and will not be elaborated here specifically.

[0147] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions, or by controlling relevant hardware through instructions. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0148] Therefore, this application provides a computer-readable storage medium, which stores multiple instructions that can be loaded by a processor to execute the steps of the processing method of the flood susceptibility prediction model in this application, such as Figure 1 the description of the processing method of the flood susceptibility prediction model in the corresponding embodiment, and the specific operation can refer to Figure 1 the description of the processing method of the flood susceptibility prediction model in the corresponding embodiment, and will not be elaborated here.

[0149] Among them, the computer-readable storage medium may include: Read Only Memory (ROM), Random Access Memory (RAM), magnetic disk or optical disk, etc.

[0150] Since the instructions stored in the computer-readable storage medium can execute the steps of the processing method of the flood susceptibility prediction model in this application, such as Figure 1 the corresponding embodiment, the beneficial effects that can be achieved by the processing method of the flood susceptibility prediction model in this application can be realized. For details, please refer to the previous description and will not be elaborated here. Figure 1

[0151] ​The above has introduced in detail the processing method, device, processing equipment and computer-readable storage medium of the flood vulnerability prediction model provided by this application. Specific examples are used in this article to elaborate on the principle and implementation manner of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those skilled in the art, according to the idea of this application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A method for processing a flood susceptibility prediction model, characterized in that: The method comprises: Acquire a flood inventory dataset, wherein the flood inventory dataset includes a first flood inventory dataset carrying a tag and a second flood inventory dataset not carrying the tag, the tag identifies whether a flood occurs in a corresponding flood inventory data portion, and the flood inventory dataset uses a quantitative value of a preset flood impact factor as a feature; Based on the k-means clustering method and the SMOTE method, single sample data enhancement is performed on the second flood inventory data set in the flood inventory data to obtain an enhanced data set; Based on the two datasets, the flood inventory dataset and the enhanced dataset, a multilayer perceptron model is trained in combination with a consistency regularized semi-supervised learning method to obtain a flood susceptibility prediction model, wherein the flood susceptibility prediction model is used to predict the corresponding flood susceptibility based on the characteristic values ​​of the flood influencing factors input into the model; The method of performing data enhancement on the second flood inventory data set in the flood inventory data based on the k-means clustering method and the SMOTE method to obtain an enhanced data set includes: Clustering the second flood inventory data set in the flood inventory data by using the k-means clustering method to obtain a plurality of cluster categories; For a multi-sample clustering category, the SMOTE method is used to generate new samples between multiple samples included in the multi-sample clustering category, and the samples included in the single-sample clustering category are not processed, so as to achieve data enhancement of the flood inventory data and obtain the enhanced data set; For the consistency regularized semi-supervised learning method, the following configuration contents are included: Use binary cross entropy loss to replace the original loss term; The second flood inventory dataset and the enhanced dataset are respectively subjected to model prediction to obtain output probability distributions in two aspects, and the distance between the output probability distributions in the two aspects is calculated by KL divergence as an unsupervised loss term; The loss after the model prediction of the first flood inventory dataset is used as the supervision loss item; The overall loss is the sum of the supervised loss term and λ times the unsupervised loss term.

2. The method according to claim 1, characterized in that The multilayer perceptron model is a fully connected artificial neural network configured with 1 input layer, 2 hidden layers and 1 output layer. The number of nodes in each layer is 13, 128, 128 and 2, the activation function of the middle layer of the network is the ReLU activation function, and the activation function between the hidden layer and the output layer is the Sigmoid activation function.

3. The method according to claim 1, characterized in that During the model training process, the method further includes: A hyperparameter adjustment process is performed for the clustering category k value and / or the weight λ of the unsupervised loss term involved in the consistency regularized semi-supervised learning method.

4. The method according to claim 3, characterized in that After the hyperparameter adjustment process, the cluster category k value is 240 and λ is 0.

5.

5. The method according to claim 1, characterized in that The method further comprises: Obtain the characteristic value of the target flood impact factor in the target area; Inputting the target flood impact factor characteristic value into the flood susceptibility prediction model, so that the flood susceptibility prediction model is based on the corresponding flood susceptibility predicted; The flood susceptibility prediction result of the target area output by the flood susceptibility prediction model is extracted.

6. A processing device for a flood susceptibility prediction model, characterized in that: The device comprises: an acquisition unit, configured to acquire a flood inventory data set, wherein the flood inventory data set includes a first flood inventory data set carrying a tag and a second flood inventory data set not carrying the tag, the tag identifies whether a flood occurs in a corresponding flood inventory data portion, and the flood inventory data set uses a quantitative value of a preset flood impact factor as a feature; an enhancement unit, configured to perform single sample data enhancement on the second flood inventory data set in the flood inventory data based on a k-means clustering method and a SMOTE method to obtain an enhanced data set; A training unit is used to train a multilayer perceptron model based on the flood inventory dataset and the enhanced dataset in combination with a consistency regularized semi-supervised learning method to obtain a flood susceptibility prediction model, wherein the flood susceptibility prediction model is used to predict the corresponding flood susceptibility based on the characteristic values ​​of the flood influencing factors input into the model; The enhancement unit is specifically used for: Clustering the second flood inventory data set in the flood inventory data by using the k-means clustering method to obtain a plurality of cluster categories; For a multi-sample clustering category, the SMOTE method is used to generate new samples between multiple samples included in the multi-sample clustering category, and the samples included in the single-sample clustering category are not processed, so as to achieve data enhancement of the flood inventory data and obtain the enhanced data set; For the consistency regularized semi-supervised learning method, the following configuration contents are included: Use binary cross entropy loss to replace the original loss term; The second flood inventory dataset and the enhanced dataset are respectively subjected to model prediction to obtain output probability distributions in two aspects, and the distance between the output probability distributions in the two aspects is calculated by KL divergence as an unsupervised loss term; The loss after the model prediction of the first flood inventory dataset is used as the supervision loss item; The overall loss is the sum of the supervised loss term and λ times the unsupervised loss term.

7. A processing device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and the processor executes the method according to any one of claims 1 to 5 when calling the computer program in the memory.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the method according to any one of claims 1 to 5.

Images