Landslide disaster early warning method, equipment and storage medium based on machine learning

Abstract

The landslide disaster early warning method, device and storage medium based on machine learning provided by the embodiment of the present invention first obtain geological information and rainfall information of the target area; then input the geological information and rainfall information into the landslide disaster early warning model to obtain the landslide hazard result; the present invention adopts a boosting integrated algorithm, which integrates a homogeneous machine learning algorithm to perform parallel training of multiple learners of the same type, and for the key hyperparameters in the boosting integrated algorithm, a Bayesian optimization method based on a tree structure is used to calculate, and the hyperparameters are no longer searched in a large range by random search or network search, but by maximizing the expected value of probability, the effect of the previous iteration is evaluated in each iteration, and the selection interval of the next iteration is inferred, so as to achieve the effect of rapid convergence, thereby improving the prediction precision and accuracy. At the same time, the algorithm performance and generalization ability are stronger, thereby improving the accuracy of landslide disaster early warning.

Description

Technical Field

[0001] The present invention belongs to the technical field of disaster early warning, and in particular relates to a landslide disaster early warning method, device and storage medium based on machine learning. Background Art

[0002] Landslide disasters occur in groups, with many points and wide areas, causing serious damage. Regional landslide disaster warning is an important means of effective disaster prevention and mitigation. The warning model is the core of the successful implementation of the warning business and one of the important contents of landslide disaster prevention and mitigation. However, due to the complex landslide induction mechanism in the study area, insufficient investigation and monitoring big data and analysis methods, the traditional regional geological disaster warning model has problems such as limited warning accuracy and insufficient refinement.

[0003] In recent years, with the rapid development of big data and artificial intelligence technology, machine learning technology has gradually been widely used in geological disaster early warning and has achieved good results. However, the performance and generalization ability of the model still need to be improved, and the landslide disaster early warning is not accurate enough. Summary of the invention

[0004] In view of this, the present invention provides a landslide disaster warning method, device and storage medium based on machine learning, aiming to solve the problem that the performance and generalization ability of the existing landslide disaster warning model still need to be improved and the landslide disaster warning is not accurate enough.

[0005] A first aspect of an embodiment of the present invention provides a landslide disaster early warning method based on machine learning, comprising:

[0006] Obtain geological and rainfall information of the target area;

[0007] Input geological information and rainfall information into the landslide hazard early warning model to obtain landslide hazard results;

[0008] Among them, the landslide disaster warning model includes multiple learners trained in parallel; the landslide disaster warning model is trained according to the boosting ensemble algorithm; the hyperparameters in the boosting ensemble algorithm are calculated according to the Bayesian optimization method based on the tree structure.

[0009] A second aspect of an embodiment of the present invention provides a landslide disaster early warning device based on machine learning, comprising:

[0010] An acquisition module is used to obtain geological information and rainfall information of the target area;

[0011] The early warning module is used to input geological information and rainfall information into the landslide disaster early warning model to obtain the landslide hazard results;

[0012] Among them, the landslide disaster warning model includes multiple learners trained in parallel; the landslide disaster warning model is trained according to the boosting ensemble algorithm; the hyperparameters in the boosting ensemble algorithm are calculated according to the Bayesian optimization method based on the tree structure.

[0013] A third aspect of an embodiment of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the landslide disaster warning method based on machine learning as described in the first aspect above are implemented.

[0014] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the landslide disaster warning method based on machine learning as described in the first aspect above are implemented.

[0015] The landslide disaster early warning method, device and storage medium based on machine learning provided by the embodiment of the present invention first obtain geological information and rainfall information of the target area; then input the geological information and rainfall information into the landslide disaster early warning model to obtain the landslide hazard result; wherein the landslide disaster early warning model includes multiple learners trained in parallel; the landslide disaster early warning model is trained according to the boosting integrated algorithm; the hyperparameters in the boosting integrated algorithm are calculated according to the Bayesian optimization method based on the tree structure. The present invention adopts the boosting integrated algorithm, which integrates homogeneous machine learning algorithms, and is no longer a single learner training, but a parallel training of multiple learners of the same type, and the weak learner is trained into a strong learner. For the key hyperparameters in the boosting integrated algorithm, the Bayesian optimization method based on the tree structure is used for calculation, and the hyperparameters include the number of integrated weak learners and the model parameters of the learner itself. The hyperparameters are no longer searched in a large range by random search or network search, but by maximizing the expected value of probability, the effect of the previous iteration is evaluated in each iteration, and the selection interval of the next iteration is inferred, so as to achieve the effect of rapid convergence, thereby improving the prediction precision and accuracy. At the same time, the algorithm performance and generalization ability are stronger, thereby improving the accuracy of landslide disaster warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0017] Figure 1is a flow chart of an implementation of a landslide disaster early warning method based on machine learning provided by an embodiment of the present invention;

[0018] Figure 2 It is a schematic diagram of the training process of the landslide disaster early warning model provided by an embodiment of the present invention;

[0019] Figure 3 is a schematic diagram of boosting integration of various learners provided in an embodiment of the present invention;

[0020] Figure 4 is a structural schematic diagram of a landslide disaster early warning device based on machine learning provided by an embodiment of the present invention;

[0021] Figure 5 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0023] Figure 1 FIG. 1 is a flow chart of the implementation of the landslide disaster early warning method based on machine learning provided by an embodiment of the present invention. Figure 1 As shown, in some embodiments, a landslide disaster early warning method based on machine learning includes:

[0024] S210, obtaining geological information and rainfall information of the target area;

[0025] S220, inputting geological information and rainfall information into a landslide disaster early warning model to obtain a landslide hazard result;

[0026] Among them, the landslide disaster warning model includes multiple learners trained in parallel; the landslide disaster warning model is trained according to the boosting ensemble algorithm; the hyperparameters in the boosting ensemble algorithm are calculated according to the Bayesian optimization method based on the tree structure.

[0027] In an embodiment of the present invention, the learner includes but is not limited to at least one of the following: K-nearest neighbor, support vector machine, multilayer perceptron neural network and logistic regression. For different types of machine learning algorithms, each is integrated separately, and the gradient boosting machine GBM (Gradient Boosting Machine) model in the boosting ensemble learning algorithm is used, that is, multiple weak learners output are combined into a strong learner to improve the prediction accuracy. The Bayesian optimization method is used to automatically adjust the hyperparameters in the boosting ensemble learning algorithm. Compared with network search and random search, it does not require manpower, and when trying the next set of hyperparameters, it will refer to the previous evaluation effect, so the search time is saved. The present invention uses the Hyperopt library in Python for parameter adjustment, which uses the Tree Parzen Estimator (TPE). Taking the random forest algorithm (RandomForest, RF) as an example, the number of random forest learners in the boosting ensemble, the minimum number of samples required for each random forest learner node split, and the maximum depth are all hyperparameters of the model. The present invention can complete the training of the model through the following embodiments.

[0028] Figure 2 FIG. 1 is a schematic diagram of the training process of the landslide disaster warning model provided by an embodiment of the present invention. Figure 2 As shown, in some embodiments, the method further includes:

[0029] Step 1, construct a landslide disaster machine learning sample dataset;

[0030] Step 2, using a tree-structured Bayesian optimization method to determine the hyperparameters of multiple learners in a pre-set boosting ensemble learning algorithm;

[0031] Step 3, classify and calculate the landslide disaster machine learning sample data set according to multiple learners, and evaluate the classification results of each learner to obtain the evaluation results of each learner;

[0032] Step 4: According to the Stacking integration strategy and the evaluation results of each learner, multiple learners are integrated to obtain a landslide disaster warning model.

[0033] In some embodiments, step 1 includes: selecting positive samples and negative samples, wherein the positive samples are landslide points and the negative samples are non-landslide points; obtaining multiple preset disaster factor features, and using the disaster factor features as feature parameters, and using the positive samples and negative samples as output feature parameters to construct an initial sample data set; performing attribute extraction and cleaning on the initial sample data set to obtain a landslide disaster machine learning sample data set.

[0034] In the embodiment of the present invention, the positive sample is the landslide point, and the negative sample is the non-landslide point. The negative sample is determined by the spatiotemporal sampling method, that is, random sampling of the space outside the buffer zone of 3 times the size of the warning grid unit centered on the positive sample. In order to reduce the imbalance between the positive and negative sample data and the spatial autocorrelation, the sampling ratio is 1:2, the spatial position of the negative sample is determined, and then the time attribute of the negative sample is determined at random time within the flood season. A county where landslide disasters often occur is selected as the research object, and 1926 positive and negative samples of landslide disasters in 8 years are determined, including 613 landslide points (positive samples) and 1313 non-landslide points (negative samples).

[0035] In the embodiment of the present invention, the disaster factor characteristics include at least one of the following: slope aspect, slope, elevation, landform type, stratum lithology, number of historical disaster points, distance from faults, distance from rivers, average annual rainfall, distance from houses, distance from roads, rainfall on the day, and daily rainfall in the previous 1 to 15 days. That is, a total of 27 influencing factors are involved in model construction.

[0036] After determining the characteristics of the disaster-causing factors, the study area was gridded with 200m×200m units, and the above 27 influencing factors were matched as input feature parameters, and whether the landslide occurred (the positive sample of the landslide is 1, and the negative sample of the non-landslide is 0) was used as the output feature parameter. A landslide disaster warning sample set for a certain county area (that is, the initial sample set mentioned above) was constructed, with 1926 samples, and then the sample set was divided into a training set and a test set at a ratio of 4:1. The input features and parameters of the sample data set are as follows:

[0037] Table 1 Sample dataset input features and parameters

[0038]

[0039] In the embodiment of the present invention, the initial sample data set is subjected to attribute extraction and cleaning, specifically as follows:

[0040] (1) The Pearson correlation coefficient method was used to calculate the correlation coefficients between slope, aspect, elevation, landform type, and stratigraphic lithology variables. The coefficients were all close to 0, indicating that there was little correlation between these variables, indicating that the influence between the variables had almost no effect on the accuracy of machine learning.

[0041] (2) Random negative sample points, import the DEM elevation data of a county into the software at the same time, then import the training set and test point data, and finally perform spatial overlay analysis on the positive and negative samples and the geological environment feature library to extract the geological environment characteristic attributes of the positive and negative samples; based on the time attributes of the positive and negative samples, query and extract the rainfall factor characteristic attributes of the positive and negative samples, thereby constructing the entire sample data set.

[0042] (3) Clean the sample data set. If there are few null values ​​in the original data set, they can be deleted directly. If the proportion of null values ​​is large, they can generally be replaced by the results calculated by the cubic spline interpolation method. They cannot be deleted directly. This type of data cleaning refers to interpolating or removing missing values ​​from the data and identifying and processing outliers (noise values); normalizing the sample eigenvalues, that is, dividing each eigenvalue by the sum of the absolute values ​​of each eigenvalue. After the transformation of the sample matrix, the sum of the absolute values ​​of the eigenvalues ​​of each sample is 1; scaling the sample eigenvalues, that is, setting the minimum and maximum values ​​of each column in the sample matrix to the same interval, unifying the range of the eigenvalues ​​of each column. In general, the eigenvalues ​​are scaled to the interval [0, 1].

[0043] Figure 3 is a schematic diagram of boosting integration of various learners provided in an embodiment of the present invention. Figure 3 As shown, in some embodiments, step 2 includes: using a tree-structured Bayesian optimization method to optimize the objective function of the gradient boosting machine model in a pre-set boosting ensemble learning algorithm to obtain a minimum loss function; wherein the gradient boosting machine model includes multiple learners, the input of the objective function is a set of hyperparameters, and the output is an optimal hyperparameter sequence.

[0044] In the embodiment of the present invention, taking the random forest algorithm as an example:

[0045] (1) Python's Hyperopt library can perform gradient boosting machine (GBM) calculations while treating the objective function as a black box, considering only its input and output. The input is a set of hyperparameters, namely the number of random forest ensembles, the minimum number of samples required for node splitting in each random forest, and the maximum depth; the output is the value to be minimized, namely the cross-validation loss, which is set to 1-ROC AUC (the AUC value of the ROC curve). The performance of the hyperparameters on the validation set is evaluated, but the data set is not divided into a fixed validation set and training set. Instead, a 5-fold cross-validation is used.

[0046] (2) Domain space. In Bayesian optimization, a hyperparameter is not selected sequentially (network) or randomly, but according to the probability distribution of each hyperparameter. Therefore, the range can be set larger and the Bayesian algorithm can be used for reasoning. The number of ensembles and the minimum number of samples required for node splitting in each random forest are selected using discrete uniform distribution, and the maximum depth is selected using logarithmic uniform distribution.

[0047] (3) Hyperparameter optimization algorithm. TPE (Tree Parzen Estimator) is a tree-structured Bayesian optimization algorithm used to solve the global optimization problem of black-box functions. The TPE algorithm can adaptively adjust the size of the parameter search space and find the global optimal solution in the fewest possible iterations.

[0048] TPE is defined using two density functions: :

[0049] (1)

[0050] in, x is a sequence of hyperparameters, y * represents the threshold value, let , represents a certain quantile of the TPE algorithm, used to divide l ( x )and g ( x ), the range is between (0,1). l ( x ) is the observation space To establish, the loss corresponding to the observation space Less than y * , using the remaining observations to build g ( x ). TPE selects expected improvement (EI) as the acquisition function, which is used to determine where to collect the next sample point. , so the Bayesian formula is used for conversion:

[0051] = (2)

[0052] The above formula shows that in order to maximize EI, it is hoped that x Down l ( x ) has a high probability g ( x ) probability is low. When the algorithm returns the one with the maximum EI, select x * :

[0053] (3)

[0054] in, x * This is the optimal hyperparameter sequence required.

[0055] The hyperparameters and search steps of the boosting integration of each algorithm selected by the present invention and the final determined hyperparameter values ​​are shown in the table below.

[0056] Table 2 Hyperparameters and search steps of each algorithm

[0057]

[0058] In some embodiments, step 3 includes: using a preset evaluation index to evaluate the classification result of each learner to obtain an evaluation result of each learner.

[0059] In this embodiment of the present invention, after the boosting integrated model calculation in step 2 above, the optimal hyperparameter sequence of each algorithm and the minimum loss function (objective function) value, i.e., 1-ROC AUC, can be obtained. The early warning model of each integrated algorithm can be evaluated based on this result.

[0060] 1. Random Forest (RF) integration

[0061] This algorithm combines bagging, decision trees, and random attribute selection. The algorithm process is random sample selection: n samples are randomly selected from the sample set using bagging. The random feature selection is to randomly select n attributes (n < m) from m attributes, determine the root node in the selected attributes, and build a decision tree. Repeat k times in a row, and the k trees built form a random forest. Finally, the average combined prediction results are used to obtain the final prediction result using the mean or voting.

[0062] The calculated cross-validation accuracy of the early warning model is 0.953 at most. Based on this, when the threshold is set to 0.5, the total early warning accuracy is the highest, which is 0.961. At the same time, the corresponding landslide recall rate is 0.915, the non-landslide recall rate is 0.984, the accuracy rate (ACC) is 0.953, and the model evaluation index AUC value is 0.982.

[0063] 2. K-nearest Neighbor (KNN) integration

[0064] When performing classification warning, the majority voting method is selected, that is, the K samples in the training set that are closest to the predicted sample features are predicted as the category with the largest number of samples. The core process requires calculating the distance between the predicted sample and the training sample, and the Euclidean distance is generally used to determine the similarity. The category label Y can be divided into positive samples (landslide points, recorded as 1) and negative samples (no landslide points, recorded as 0). Classification function , Euclidean distance It can be expressed as:

[0065] (4)

[0066] (5)

[0067] in: , , …, ; , , …, For two different sets of sample data; is the number of characteristic attributes.

[0068] Before classification, the classification algorithm should be understood as system parameters. , distance function , Category Collection , classification function , discriminant function , training samples , samples to be predicted .in This represents the number of nearest neighboring data points selected. Represents the weight value of each dimension of the sample point vector when calculating the distance. Represents the weight of each sample point in the nearest neighbor set of a sample point. Discriminant function , which is used to determine which category the sample point belongs to. ,express Belongs to category , On the contrary. Training sample set The category is known and the sample set to be predicted Category unknown.

[0069] It is calculated that when the threshold is set to 0.25, the total precision is the highest, which is 0.949. At the same time, the corresponding landslide recall rate is 0.922, the non-landslide recall rate is 0.963, the accuracy rate (ACC) is 0.938, and the model evaluation index AUC value is 0.972.

[0070] 3. Support Vector Machine (SVM) integration

[0071] The purpose of this algorithm is to find the optimal hyperplane to maximize the classification interval, that is, to find a line so that the point closest to the line can be the farthest (find a line so that the line closest to the river bank can be the farthest). The optimization objective formula is:

[0072] (6)

[0073] Seek The minimum value of is as follows:

[0074] (7)

[0075] Finally, the Lagrange multiplier method is used to find the minimum value:

[0076] (8)

[0077] The final optimization is:

[0078] (9)

[0079] The final decision boundary is composed of sample points where 𝛼 is not zero (i.e., boundary points). Only when non-boundary points 𝛼 are equal to zero (support vectors) will the boundary be affected.

[0080] When low-dimensional linearly inseparable needs to be mapped to high-dimensional linearly separable, a kernel function needs to be introduced , so that the vector can be n dimensional space for linear analysis. This article cites the radial basis kernel function: rbf , the distribution probability of the original sample features is increased by the Gaussian distribution function, so that the linear inseparable problem in the low-dimensional space is converted into a linear separable problem in the high-dimensional space.

[0081] (10)

[0082] in, is the hyperplane normal vector, The value to increase the margin so that the hyperplane does not pass through the origin. is the squared Euclidean distance between two eigenvectors; 𝜎 is a free parameter.

[0083] It is calculated that when the threshold is set to 0.5, the total precision is the highest, which is 0.958. At the same time, the corresponding landslide recall rate is 0.907, the non-landslide recall rate is 0.983, the accuracy rate (ACC) is 0.947, and the model evaluation index AUC value is 0.982.

[0084] 4. Multilayer Perceptron Neural Network Algorithm (MLP) Integration

[0085] The model consists of an input layer, a hidden layer, and an output layer. The output of the hidden layer is as follows:

[0086] (10)

[0087] in, is the input data of the input layer; is the weight (also called connection coefficient); is the bias term; is the activation function, usually a sigmoid function or a tanh function.

[0088] The hidden layer to the output layer can be regarded as a multi-category logistic regression (softmax regression), and the output formula is as follows:

[0089] (11)

[0090] in, is the output of the hidden layer.

[0091] It is calculated that when the threshold is set to 0.5, the total precision is the highest, which is 0.957. At the same time, the corresponding landslide recall rate is 0.924, the non-landslide recall rate is 0.973, the accuracy rate (ACC) is 0.949, and the model evaluation index AUC value is 0.981.

[0092] 5. Logistic Regression (LR) integration

[0093] The algorithm estimates parameters through the maximum likelihood estimation method, where the Logistic function (also called Sigmoid function) is in the form of

[0094] (12)

[0095] For landslide disaster warning, factors such as geological environment and rainfall that affect whether a landslide occurs can be used as independent variables, and whether a landslide occurs or not can be used as a categorical dependent variable. is the probability of landslide occurrence, ranging from [0,1], is the probability that a landslide will not occur. / is the probability ratio of landslide occurrence to non-occurrence, and its natural logarithm is taken :

[0096] (13)

[0097] (14)

[0098] in, is the probability of landslide occurrence; , , …, is the independent variable (influencing factor); , , …, is the regression coefficient of the logistic regression algorithm.

[0099] It is calculated that when the threshold is set to 0.5, the total precision is the highest, which is 0.952. At the same time, the corresponding landslide recall rate is 0.911, the non-landslide recall rate is 0.972, the accuracy rate (ACC) is 0.946, and the model evaluation index AUC value is 0.980.

[0100] In summary, the random forest ensemble model performed best, with the highest accuracy (ACC=0.953), the best model generalization ability (AUC=0.982), and no overfitting.

[0101] In some embodiments, multiple learners are integrated according to the Stacking integration strategy and the evaluation results of each learner to obtain a landslide hazard warning model, including: taking any learner as a meta-model and other learners as base models to obtain multiple Stacking integration models; wherein; the characteristics of the meta-model are obtained by weighted averaging the prediction results of each base model according to the test accuracy of each base model; and according to the evaluation results, determining the landslide hazard warning model from the multiple Stacking integration models.

[0102] In the embodiment of the present invention, the Stacking strategy is selected to fuse the five final integrated algorithms in step 3. Stacking is an integrated learning method, which is usually used to improve the performance of machine learning models. That is, the evaluation results of multiple basic models are used as input, and then stacked using a metamodel, and then these evaluation results are integrated to obtain a new integrated model.

[0103] The present invention uses 4 of the above 5 algorithms as base models and the remaining one as a meta-model, thus obtaining 5 Stacking fusion models. The fusion strategies are as follows:

[0104] Table 3 Stacking fusion strategy

[0105]

[0106] The present invention optimizes the Stacking algorithm. The traditional Stacking algorithm uses a five-fold cross-validation method to divide the data set into five equal parts. Five base models are formed by five data subsets. When the first layer predicts the test set, the average value of the prediction results of the five base models is taken as the feature of the second layer. There may be a situation where the data is unevenly divided, resulting in poor prediction results. Therefore, the present invention performs a weighted average of the prediction results according to the test accuracy of the base model, and obtains the result as the feature of the second layer.

[0107] Calculation shows that among the five Stacking fusion strategies, the fusion algorithm with random forest ensemble as meta-model performs the best. Its accuracy, recall rate and precision are the best among the five ensemble algorithms, and the values ​​of various evaluation indicators are only lower than the random forest ensemble in the boosting ensemble algorithm. At the same time, the evaluation indicators of the five models under the Stacking fusion strategy with various boosting ensemble algorithms as meta-models are higher than various machine learning boosting ensemble algorithms.

[0108] Beneficial effects of the present invention:

[0109] The present invention takes into account 27 influencing factors that may cause landslide disasters, and based on this, proposes a method for constructing a machine learning sample data set as well as sample feature extraction and cleaning methods, providing comprehensive and scientific basic data for landslide disaster early warning.

[0110] In view of the shortcoming of low prediction accuracy of various machine learning algorithms, the present invention proposes to adopt a boosting ensemble algorithm, which integrates homogeneous machine learning algorithms. Instead of training a single learner, multiple learners of the same type are trained in parallel to train weak learners into strong learners.

[0111] The present invention adopts a tree-structured Bayesian optimization method to calculate the key hyperparameters in the boosting ensemble algorithm. The hyperparameters include the number of weak learners and the model parameters of the learners themselves. The hyperparameters are no longer searched in a large range by random search or network search, but by maximizing the expected value of probability. In each iteration, the effect of the previous iteration is evaluated and the selection interval of the next iteration is inferred, thereby achieving a rapid convergence effect and improving the prediction precision and accuracy. At the same time, the algorithm performance and generalization ability are stronger.

[0112] The five prediction methods and results of the present invention show that the random forest ensemble model performs best, followed by the multilayer perceptron neural network ensemble model, the support vector machine ensemble model and the logistic regression ensemble model, and the nearest neighbor ensemble model is not ideal in comparison. This provides a reference for the selection of machine learning algorithms by relevant researchers in the future.

[0113] The present invention proposes a Stacking ensemble fusion strategy, and the results show that the ensemble algorithm with random forest as the meta-model and the other four models as the basic model performs best; among them, the input feature selection of the second layer of the Stacking algorithm is optimized.

[0114] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0115] Figure 4 Schematic diagram of the structure of the landslide disaster warning device based on machine learning provided by an embodiment of the present invention. Figure 4 As shown, in some embodiments, the landslide disaster early warning device 4 based on machine learning includes:

[0116] An acquisition module 410 is used to acquire geological information and rainfall information of a target area;

[0117] The early warning module 420 is used to input geological information and rainfall information into the landslide disaster early warning model to obtain landslide hazard results;

[0118] Among them, the landslide disaster warning model includes multiple learners trained in parallel; the landslide disaster warning model is trained according to the boosting ensemble algorithm; the hyperparameters in the boosting ensemble algorithm are calculated according to the Bayesian optimization method based on the tree structure.

[0119] Optionally, the landslide disaster warning device 4 based on machine learning also includes: a training module for constructing a landslide disaster machine learning sample data set; using a tree-structured Bayesian optimization method to determine the hyperparameters of multiple learners in a pre-set boosting ensemble learning algorithm; classifying and calculating the landslide disaster machine learning sample data set according to multiple learners, and evaluating the classification results of each learner to obtain an evaluation result of each learner; integrating multiple learners according to the Stacking integration strategy and the evaluation result of each learner to obtain a landslide disaster warning model.

[0120] Optionally, a training module is used to select positive samples and negative samples, where the positive samples are landslide points and the negative samples are non-landslide points; obtain multiple preset disaster factor features, and use the disaster factor features as feature parameters, and use the positive samples and negative samples as output feature parameters to construct an initial sample data set; perform attribute extraction and cleaning on the initial sample data set to obtain a landslide disaster machine learning sample data set.

[0121] Optionally, a training module is used to optimize the objective function of the gradient boosting machine model in the pre-set boosting ensemble learning algorithm using a tree-structured Bayesian optimization method to obtain a minimum loss function; wherein the gradient boosting machine model includes multiple learners, the input of the objective function is a set of hyperparameters, and the output is an optimal hyperparameter sequence.

[0122] Optionally, a training module is used to evaluate the classification results of each learner using a preset evaluation index to obtain an evaluation result of each learner.

[0123] Optionally, a training module is used to use any learner as a meta-model and other learners as base models to obtain multiple Stacking integrated models; wherein; the characteristics of the meta-model are obtained by weighted averaging the prediction results of each base model according to the test accuracy of each base model; based on the evaluation results, a landslide hazard warning model is determined from multiple Stacking integrated models.

[0124] Optionally, the learner includes but is not limited to at least one of the following: K-nearest neighbor, support vector machine, multilayer perceptron neural network, and logistic regression.

[0125] Optionally, the disaster factor characteristics include at least one of the following: slope aspect, slope gradient, elevation, landform type, stratum lithology, number of historical disaster points, distance from faults, distance from rivers, average annual rainfall, distance from houses, distance from roads, rainfall on the day, and daily rainfall in the previous 1 to 15 days.

[0126] The landslide disaster warning device based on machine learning provided in this embodiment can be used to execute the above method embodiments. Its implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0127] Figure 4 Schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. Figure 4 As shown, an electronic device 4 provided by an embodiment of the present invention includes: a processor 40, a memory 41, and a computer program 42 stored in the memory 41 and executable on the processor 40. When the processor 40 executes the computer program 42, the steps in the above-mentioned output voltage calibration method based on LLC are implemented, for example Figure 2 Alternatively, when the processor 40 executes the computer program 42, the functions of each module / unit in the above-mentioned system embodiments are realized, for example Figure 3 The functions of each module are shown.

[0128] Exemplarily, the computer program 42 may be divided into one or more modules / units, one or more modules / units are stored in the memory 41, and are executed by the processor 40 to implement the present invention. One or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program 42 in the electronic device 4.

[0129] The electronic device 4 may be a mobile phone, MCU, ECU, industrial computer, etc., which is not limited here. The electronic device 4 may include, but is not limited to, a processor 40 and a memory 41. Those skilled in the art can understand that Figure 4It is only an example of the electronic device 4 and does not constitute a limitation of the electronic device 4. It 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 may also include input and output devices, network access devices, buses, etc.

[0130] The processor 40 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 or any conventional processor, etc.

[0131] The memory 41 may be an internal storage unit of the electronic device 4, such as a hard disk or memory of the electronic device 4. The memory 41 may also be an external storage device of the electronic device 4, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 4. Further, the memory 41 may also include both an internal storage unit and an external storage device of the electronic device 4. The memory 41 is used to store computer programs and other programs and data required by the electronic device. The memory 41 may also be used to temporarily store data that has been output or is to be output.

[0132] An embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps in the above-mentioned LLC-based output voltage calibration method embodiment are implemented.

[0133] The computer readable storage medium stores a computer program 42, which includes program instructions. When the program instructions are executed by the processor 40, all or part of the processes in the above-mentioned embodiment method are implemented. The computer program 42 can also be used to instruct related hardware to complete. The computer program 42 can be stored in a computer readable storage medium. When the computer program 42 is executed by the processor 40, the steps of each of the above-mentioned method embodiments can be implemented. Among them, the computer program 42 includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer readable medium can include: any entity or device capable of carrying 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.

[0134] The computer-readable storage medium may be an internal storage unit of the electronic device of any of the aforementioned embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of the electronic device and an external storage device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0135] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0136] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0137] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0138] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0139] In the embodiments provided by the present invention, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0140] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0141] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0142] If the integrated module / unit 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 invention 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, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. 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. Computer-readable media can include: any entity or device that can carry 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), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0143] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A landslide disaster early warning method based on machine learning, characterized in that: include: Obtain geological and rainfall information of the target area; Inputting the geological information and the rainfall information into a landslide disaster early warning model to obtain a landslide hazard result; The landslide disaster warning model includes a plurality of learners trained in parallel; the landslide disaster warning model is trained according to a boosting ensemble algorithm; the hyperparameters in the boosting ensemble algorithm are calculated according to a Bayesian optimization method based on a tree structure; the hyperparameters include the number of ensembles of learners and the model parameters of each learner; the hyperparameters evaluate the effect of the previous iteration in each iteration process and infer the selection interval for the next iteration in a manner of maximizing the expected value of probability; The method further comprises: Selecting positive samples and negative samples, wherein the positive samples are landslide points and the negative samples are non-landslide points; Acquire multiple preset disaster-causing factor features, and use the disaster-causing factor features as feature parameters, and use the positive samples and negative samples as output feature parameters to construct an initial sample data set; Extracting and cleaning the attributes of the initial sample data set to obtain a landslide disaster machine learning sample data set; The objective function of the gradient boosting machine model in the pre-set boosting ensemble learning algorithm is optimized by using a tree-structured Bayesian optimization method to obtain a minimum loss function; wherein the gradient boosting machine model includes multiple learners, the input of the objective function is a set of hyperparameters, and the output is an optimal hyperparameter sequence; Classifying and calculating the landslide disaster machine learning sample data set according to multiple learners, and evaluating the classification results of each learner using a preset evaluation index to obtain an evaluation result of each learner; Any learner is used as a meta-model and other learners are used as base models to obtain multiple Stacking integration models; wherein; the characteristics of the meta-model are obtained by weighted average of the prediction results of each base model according to the test accuracy of each base model; According to the evaluation results, the landslide disaster early warning model is determined from a variety of stacking integrated models.

2. The landslide disaster early warning method based on machine learning according to claim 1 is characterized in that: The learner includes but is not limited to at least one of the following: K-nearest neighbors, support vector machines, multilayer perceptron neural networks, and logistic regression.

3. The landslide disaster early warning method based on machine learning according to claim 1 is characterized in that: The hazard factor characteristics include at least one of the following: Slope direction, slope, elevation, landform type, stratum lithology, number of historical disaster points, distance from faults, distance from rivers, average annual rainfall, distance from houses, distance from roads, rainfall on the day, and daily rainfall from 1 to 15 days ago.

4. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the landslide disaster early warning method based on machine learning as described in any one of claims 1 to 3 above are implemented.

5. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the landslide disaster early warning method based on machine learning as described in any one of claims 1 to 3 above are implemented.

Images