A root stone danger prediction method, device, system and storage medium

Abstract

The present invention discloses a root rock hazard prediction method, device, system and storage medium, belonging to the technical field of flood control engineering hazard identification, the method comprises: obtaining multi-source data of root rock hazard influencing factors, and constructing a feature data set based on the multi-source data of root rock hazard influencing factors; inputting the feature data set into a pre-constructed root rock hazard prediction model to perform root rock hazard prediction, and obtaining a root rock hazard prediction result; wherein the multi-source data of root rock hazard influencing factors includes structured data and unstructured data; the root rock hazard prediction model is constructed based on an extreme gradient boosting tree algorithm; when performing root rock hazard prediction, based on a confusion matrix, a grid search method combined with a cross-validation method is used to optimize the parameters of the root rock hazard prediction model. The method can improve the efficiency and accuracy of root rock hazard prediction.

Description

Technical Field

[0001] The invention relates to a root rock hazard prediction method, device, system and storage medium, belonging to the technical field of water-voltage device variable and performance prediction. Background Art

[0002] Root stones are the main component of flood control projects. If early abnormalities or signs of loss of root stones can be predicted, the risk of damage to flood control projects will be effectively reduced.

[0003] The analysis and identification of root stone hazards currently have the following deficiencies and defects:

[0004] (1) The level of intelligence of identification technology is low. Currently, the method of setting alarm thresholds for monitoring equipment is used to identify root rock hazards. However, the threshold alarm cannot distinguish non-hazardous situations such as abnormal and missing monitoring data caused by the impact of water flow on the sensor. Management personnel need to conduct repeated comparisons and rush to the site to conduct investigations, which increases the burden on river section managers.

[0005] (2) Insufficient decision-making ability in identifying dangerous situations. Since root rock hazards have a great impact on the safety of flood control projects, they need to be quickly analyzed and countermeasures proposed once they occur. Threshold methods and other remote identification methods have always attributed all abnormalities to root rock hazards, resulting in a waste of human and material resources.

[0006] (3) Traditional methods are not universal. Due to differences in natural geographical conditions, monitoring element data types, and management personnel experience in different regions, traditional methods are not able to play a practical role in the analysis and prediction of different river sections. Summary of the invention

[0007] The purpose of the present invention is to provide a root stone hazard prediction method, device, system and storage medium, which can improve the efficiency and accuracy of root stone hazard prediction.

[0008] In order to achieve the above object, the present invention provides the following technical solutions:

[0009] In a first aspect, the present invention provides a method for predicting root stone danger, comprising:

[0010] Acquire multi-source data of factors affecting root rock hazard, and construct a feature data set based on the multi-source data of factors affecting root rock hazard;

[0011] Inputting the characteristic data set into a pre-built root rock hazard prediction model to perform root rock hazard prediction and obtain a root rock hazard prediction result;

[0012] Among them, the multi-source data of root stone hazard influencing factors include structured data and unstructured data; the root stone hazard prediction model is constructed based on the extreme gradient boosting tree algorithm; when predicting the root stone hazard, based on the confusion matrix, the grid search method combined with the cross-validation method is used to optimize the parameters of the root stone hazard prediction model.

[0013] In combination with the first aspect, further, the multi-source data of factors affecting root rock danger includes hydrological data and working condition data of monitoring equipment;

[0014] The hydrological data include precipitation, water level, flow, and turbulent energy in the area where the root stone is located;

[0015] The monitoring equipment working condition data includes the function mode, reporting mode, offset angle, detection cycle, reporting frequency, alarm angle, data docking port, communication signal strength, and voltage value of the rootstone sensor;

[0016] Among them, water level, flow, reporting mode, offset angle, detection cycle, reporting frequency, alarm angle, data docking port, communication signal strength, and voltage value are numerical data; precipitation and turbulence energy are image data; and functional mode is text data.

[0017] In combination with the first aspect, further, based on the multi-source data of the root stone danger influencing factors, constructing a feature data set includes:

[0018] Processing the multi-source data of the factors affecting the root rock hazard to obtain characteristic data;

[0019] The characteristic data together constitute a characteristic data set;

[0020] The multi-source data of the factors affecting the root rock hazard are processed to obtain characteristic data, including:

[0021] Performing data aggregation and data fusion on structured data and unstructured data in the multi-source data of the root stone hazard influencing factors, and converting the unstructured data into structured data;

[0022] The structured data is cleaned, the structured data is aligned, missing values ​​and abnormal values ​​in the structured data are deleted, and feature encoding is performed on the structured data to obtain feature data.

[0023] In combination with the first aspect, further, the root stone danger prediction model is:

[0024] ;

[0025] in, The first feature datasets, The output of the root rock hazard prediction model The corresponding root stone danger prediction results, Indicates the first Tree pairs The results of the classification of root stone hazards are as follows: It represents the total number of trees in the root rock danger prediction model;

[0026] During each prediction, the root rock hazard prediction model classifies the root rock hazard based on the importance of the feature data in the feature data set to the root rock hazard, and obtains a root rock hazard classification result;

[0027] The calculation formula of the importance of the characteristic data on the root stone hazard is:

[0028] ;

[0029] in, Indicates the importance of the characteristic data on the root rock hazard. , Represents the first-order gradient of the feature data in the left and right leaf nodes of the root rock hazard prediction model, , Represents the second-order gradient of the feature data in the left and right leaf nodes of the root rock hazard prediction model, represents the first adjustment parameter, Represents the computational complexity control variable.

[0030] In combination with the first aspect, further, when predicting the root stone danger, based on the confusion matrix, a grid search method combined with a cross-validation method is used to optimize the parameters of the root stone danger prediction model, including:

[0031] Based on the parameters to be optimized and the preset parameter range of the root stone hazard prediction model, a parameter grid is constructed, and each point in the parameter grid is traversed using a grid search method, wherein each point in the parameter grid corresponds to a set of parameter combinations;

[0032] Based on the cross-validation method, the feature data set is randomly divided into mutually exclusive subsets, using each subset as a test set in turn, and the remaining subsets as training sets, to test the root stone hazard prediction model. training and testing, and based on the confusion matrix, evaluating the performance of the root stone hazard prediction model, and obtaining the performance evaluation results of the root stone hazard prediction model corresponding to each group of parameter combinations;

[0033] Based on the performance evaluation results, the optimal parameter combination of the root rock hazard prediction model is obtained.

[0034] In combination with the first aspect, further, when the root rock hazard prediction model is trained, the objective function of the root rock hazard prediction model is for:

[0035] ;

[0036] in, The output of the root rock hazard prediction model The corresponding root stone danger prediction results, The first feature data, express The corresponding real result is represents the loss function, represents the total number of feature data in the training set, Indicates the first Tree pairs The results of the classification of root stone hazards are as follows: represents the regularization term, Represents the total number of trees in the root rock hazard prediction model.

[0037] In combination with the first aspect, further, the confusion matrix is:

[0038] ;

[0039] in, , , It indicates the probability that the root rock hazard prediction result output by the root rock hazard prediction model is high, medium, or low when the actual root rock hazard result is high. , , It indicates the probability that the root rock hazard prediction result output by the root rock hazard prediction model is high, medium, or low when the actual root rock hazard result is medium. , , Indicates the probability that the root stone hazard prediction result output by the root stone hazard prediction model is high, medium, or low when the actual root stone hazard result is low. High means that the root stone displacement is greater than the preset threshold and continues to change, or the root stone is lost; medium means that the root stone displacement is greater than the preset threshold and remains stable, or the root stone sensor is abnormal; and low means that the root stone displacement is less than or equal to the preset threshold and remains stable.

[0040] The performance evaluation indicators of the root stone hazard prediction model include accuracy, precision, recall and balanced F score;

[0041] The calculation formula of the accuracy is:

[0042] ;

[0043] in, It indicates the accuracy of the root stone hazard prediction model;

[0044] The calculation formula of the accuracy is:

[0045] ;

[0046] in, It represents the accuracy of the root stone hazard prediction model;

[0047] The calculation formula of the recall rate is:

[0048] ;

[0049] in, represents the recall rate of the root stone hazard prediction model;

[0050] The calculation formula of the balanced F score is:

[0051] ;

[0052] in, represents the balanced F score of the root rock hazard prediction model, Represents the weight factor.

[0053] In combination with the first aspect, further, the parameters to be optimized of the root stone hazard prediction model include the booster type, learning rate, number of iterations, maximum depth of the tree, minimum weight of child nodes, minimum loss reduction required for splitting, subsampling ratio, column sampling ratio, number of threads, L1 regularization term coefficient, and random seed.

[0054] In a second aspect, the present invention provides a root stone danger prediction device, comprising:

[0055] A data acquisition module, used to acquire multi-source data of factors affecting root rock hazard, and construct a feature data set based on the multi-source data of factors affecting root rock hazard;

[0056] A prediction module, used for inputting the feature data set into a pre-built root rock hazard prediction model to perform root rock hazard prediction and obtain a root rock hazard prediction result;

[0057] Among them, the multi-source data of root stone hazard influencing factors include structured data and unstructured data; the root stone hazard prediction model is constructed based on the extreme gradient boosting tree algorithm; when predicting the root stone hazard, based on the confusion matrix, the grid search method combined with the cross-validation method is used to optimize the parameters of the root stone hazard prediction model.

[0058] In a third aspect, the present invention provides a computer system, comprising:

[0059] Storage medium: used to store computer programs;

[0060] Processor: used to execute the computer program to implement the steps of the root stone hazard prediction method described in any one of the first aspects.

[0061] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the root stone hazard prediction method described in any one of the first aspects.

[0062] Fifthly, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the root stone hazard prediction method described in any one of the first aspects.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] The root rock hazard prediction method provided by the present invention constructs a feature data set based on multi-source data of root rock hazard influencing factors, constructs a root rock hazard prediction model based on the extreme gradient boosting tree algorithm, inputs the feature data set into the root rock hazard prediction model, and obtains the root rock hazard prediction result. The efficiency and accuracy of root rock hazard prediction can be improved, the monitoring and early warning capabilities of root rock hazards can be enhanced, decision-making basis can be provided for the reinforcement of root rock hazards, and the stability and safety of root rocks can be guaranteed.

[0065] When predicting the root stone hazard, the root stone hazard prediction model is optimized based on the confusion matrix by using the grid search method combined with the cross-validation method, which can reduce the model parameter adjustment time and computing resource loss, and improve the prediction accuracy of the root stone hazard prediction model. After obtaining the optimal parameter combination of the root stone hazard prediction model, the root stone hazard prediction model is solidified based on the optimal parameter combination of the root stone hazard prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 is a flow chart of a root stone danger prediction method provided by an embodiment of the present invention;

[0067] Figure 2 is a schematic diagram of the importance ranking of characteristic factors provided by an embodiment of the present invention;

[0068] Figure 3 Schematic diagram of a confusion matrix provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The technical solution of the present application is further described in detail below in conjunction with specific implementation methods.

[0070] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and should not be construed as limitations on the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments may be combined with each other.

[0071] Embodiment 1:

[0072] Figure 1 This is a flow chart of a root stone danger prediction method provided by this embodiment. This flow chart only shows the logical sequence of the method of this embodiment. Different methods can be used without conflict. Figure 1 The steps shown or described are accomplished in the order shown.

[0073] The root stone hazard prediction method provided in this embodiment can be applied to a terminal and can be executed by a root stone hazard prediction device, which can be implemented by software and / or hardware, and the device can be integrated in a terminal, for example: any tablet computer or computer device with communication function.

[0074] See also Figure 1 The root stone danger prediction method of this embodiment specifically includes the following steps:

[0075] Step 1: Obtain multi-source data on factors affecting root stone hazard, and construct a feature data set based on the multi-source data on factors affecting root stone hazard;

[0076] In this embodiment, the multi-source data of factors affecting root rock hazard conditions include structured data and unstructured data, the structured data include numerical data, and the unstructured data include other data except numerical data, such as image data, text data, etc. Through geospatial image processing technology, feature engineering technology, etc., the multi-source data of factors affecting root rock hazard conditions are uniformly converted into numerical data to construct a feature data set that can be read by the root rock hazard prediction model.

[0077] Specifically, the multi-source data of factors affecting root stone hazards include hydrological data and working condition data of monitoring equipment.

[0078] The hydrological data include precipitation (rainfall), water level, flow, and turbulent energy in the area where the root stone is located.

[0079] The monitoring equipment working condition data includes the function mode, reporting mode, offset angle, detection cycle, reporting frequency (heartbeat cycle), alarm angle, data docking port (equipment port), communication signal strength (signal strength), and voltage value (equipment voltage) of the rootstone sensor. Among them, the offset angle includes the offset angle in the x, y, and z directions (inclination x-axis data, inclination y-axis data, inclination z-axis data).

[0080] In this embodiment, multi-source data of factors affecting root stone hazard are shown in Table 1.

[0081] Table 1 Multi-source data on factors affecting root rock hazards

[0082]

[0083] It can be seen from Table 1 that water level, flow, reporting mode, offset angle, detection cycle, reporting frequency, alarm angle, data docking port, communication signal strength, and voltage value are numerical data, among which the reporting mode is discrete numerical data; precipitation and turbulence can be image data; and the function mode is text data.

[0084] In this embodiment, for precipitation, the Global Precipitation Measurement (GPM) data is selected, the data every three hours is converted into daily data, and the spatial resolution is adjusted according to actual requirements, and precipitation data with a reduced range is extracted according to the spatial position.

[0085] The spatial resolution of global precipitation measurement data is 0.1°×0.1°. The global precipitation measurement data are shown in Table 2.

[0086] Table 2 Global precipitation measurement data

[0087]

[0088] The calibrated precipitation band in the global precipitation measurement data was selected, and the precipitation images of the corresponding time period were filtered out using the EE date object (ee.Date) and getRange in Google Earth Engine. The three-hour precipitation data was reduced through the image to form a daily data set according to EE reduction (ee.Reducer, usually refers to the reduce operation). The image grid was resampled using resample, and 0.1°×0.1° was upsampled to 100m. The image data was cropped using the latitude and longitude information of the root stone area, and the image data was converted into point data using the mean, maximum, etc., and then imported into the feature data set.

[0089] In this embodiment, for the turbulent energy mechanism model image data, the data value of the corresponding root stone segment is extracted according to the image projection position.

[0090] Specifically, the turbulent energy data uses the Geospatial Data Abstraction Library (GDAL) to open the raster image in matrix form, uses the raster X size (RasterXSize) and the raster Y size (RasterYSize) to obtain the pixel size, sets the new image resolution, affine information, and pixel value type through driver.Create, and outputs the adjusted image data. The image is cropped using the longitude and latitude information of the root stone area, and the turbulent energy maximum point data is extracted and stored in the feature data set.

[0091] In this embodiment, based on the multi-source data of factors affecting the root stone hazard, constructing a feature data set specifically includes the following steps:

[0092] Step 1: Process the multi-source data of factors affecting root rock hazards to obtain characteristic data;

[0093] In this embodiment, data processing is performed on multi-source data of factors affecting root rock danger to obtain characteristic data, specifically including the following steps:

[0094] Step ①: Collect and fuse the structured and unstructured data in the multi-source data of factors affecting root rock hazards, and convert the unstructured data into structured data;

[0095] Step ②: Clean the structured data, align the structured data, delete missing values ​​and outliers in the structured data, and feature encode the structured data to obtain feature data.

[0096] In this embodiment, all feature data in the feature data set are labeled, the distribution of feature data is checked using the data analysis library (pandas), and the spatial accuracy of image data is adjusted using upsampling or downsampling. For excessive data, downsampling is used to balance various types of feature data; for too little data, simulated data is used to supplement; for different types of feature data, interpolation, mode, mean, etc. are used to fill; unique hot encoding or label encoding is used to convert and encode feature data in high-dimensional feature space to obtain feature data.

[0097] Step 2: The feature data sets are composed of various feature data.

[0098] Step 2: Based on the extreme gradient boosting tree algorithm, a root rock hazard prediction model is constructed, and the feature data set is input into the root rock hazard prediction model to perform root rock hazard prediction and obtain the root rock hazard prediction result;

[0099] Based on the extreme gradient boosting tree algorithm, a root stone hazard prediction model is constructed. Compared with traditional manual judgment, the advantages of the extreme gradient boosting tree algorithm are: (1) further expanding the model calculation depth, which can ensure the gradient descent rate while ensuring the error accuracy; (2) adding model complexity control when deepening the model calculation can prevent the model from overfitting and enhance the model generalization ability; (3) constructing an algorithm processing strategy for missing information, which can ensure that the model has good performance when dealing with abnormal data;

[0100] In this embodiment, the root stone danger prediction model is:

[0101] ;

[0102] in, The first feature data, The output of the root rock hazard prediction model The corresponding root stone danger prediction results, Indicates the first Tree pairs The results of the classification of root stone hazards are as follows: Represents the total number of trees in the root rock hazard prediction model.

[0103] During each prediction, the root rock hazard prediction model classifies the root rock hazard based on the importance of the feature data in the feature data set influencing the root rock hazard, and obtains the root rock hazard classification result.

[0104] In this embodiment, the calculation formula of the importance of the characteristic data on the root stone hazard is:

[0105] ;

[0106] in, It indicates the importance of the feature data on the root rock hazard, and represents the mean gain when a feature data is used as a split node in all trees of the root rock hazard prediction model. , Represents the first-order gradient of the feature data in the left and right leaf nodes of the root rock hazard prediction model, , Represents the second-order gradient of the feature data in the left and right leaf nodes of the root rock hazard prediction model, represents the first adjustment parameter, Represents the computational complexity control variable.

[0107] In addition, the importance of feature data on the root rock hazard prediction model can also be calculated using weight, cover, total_gain, total_cover, etc., where weight represents the number of times a feature data is used in a split node in all trees of the root rock hazard prediction model, cover represents the mean number of samples covered when a feature data is used as a split node, total_gain represents the total gain of all trees of the root rock hazard prediction model, and total_cover represents the total number of samples that can be covered in all trees of the root rock hazard prediction model.

[0108] In this embodiment, the classification of root stone hazards is shown in Table 3.

[0109] Table 3 Classification of root rock hazards

[0110]

[0111] It can be seen from Table 3 that if the root rock hazard classification is high, it means that the root rock displacement is greater than the preset threshold and continues to change, or the root rock is lost, which has a serious impact on the safety of flood control projects and the flood control projects will be damaged; if the root rock hazard classification is medium, it means that the root rock displacement is greater than the preset threshold and remains stable, or the root rock sensor is abnormal, which has a certain impact on the safety of flood control projects; if the root rock hazard classification is low, it means that the root rock displacement is less than the preset threshold and remains stable, and has no impact on the safety of flood control projects.

[0112] Step 3: When predicting the root stone hazard, based on the confusion matrix, the grid search method combined with the cross-validation method is used to optimize the parameters of the root stone hazard prediction model to obtain the optimal parameter combination of the root stone hazard prediction model;

[0113] In this embodiment, when predicting the root stone danger, based on the confusion matrix, the grid search method combined with the cross-validation method is used to optimize the parameters of the root stone danger prediction model, which specifically includes the following steps:

[0114] Step 1: Based on the parameters to be optimized and the preset parameter range of the root stone hazard prediction model, a parameter grid is constructed, and each point in the parameter grid is traversed using the grid search method, where each point in the parameter grid corresponds to a set of parameter combinations;

[0115] In this embodiment, the parameters to be optimized of the root stone hazard prediction model include the booster type (booster), learning rate (learning_rate), number of iterations (n_estimators), maximum depth of the tree (max_depth), minimum weight of child nodes (min_child_weight), minimum loss reduction required for splitting (gamma), subsampling ratio (subsample), column sampling ratio (colsample_bytree), number of threads (nthread), L1 regularization term coefficient (reg_alpha), and random seed (seed).

[0116] Step 2: Set the initialization parameters of the root stone hazard prediction model;

[0117] In this embodiment, the initialization parameters of the root rock hazard prediction model include booster type (booster), tree method (tree_method), objective function (objective), and random seed (seed).

[0118] The initialization parameters of the root stone hazard prediction model are set as follows: booster = "gbtree", tree_method = "gpu_hist", objective = "multi: softmax", seed = 10, where gbtree represents the gradient boosting tree, gpu_hist represents the GPU histogram algorithm, and multi: softmax represents the multi-classification softmax objective function.

[0119] Step 3: Based on the cross-validation method, the feature data set is randomly divided into mutually exclusive subsets, and use each subset as the test set in turn, and the remaining subsets as the training set to test the root stone hazard prediction model. The performance of the root rock hazard prediction model is evaluated based on the confusion matrix, and the performance evaluation results of the root rock hazard prediction model corresponding to each parameter combination are obtained;

[0120] In this embodiment, feature data is extracted from the feature data set by random sampling, with an extraction ratio of 8:2, of which 80% is used as a training set and 20% is used as a test set, and the recurrence seed is set to 10. Feature data is continuously extracted from the training set and input into the root stone hazard prediction model for training.

[0121] In this embodiment, when the root stone hazard prediction model is trained, the objective function of the root stone hazard prediction model is for:

[0122] ;

[0123] in, express The corresponding real result is Represents the loss function, which is used to measure the difference between the predicted result and the actual result. represents the total number of feature data in the training set, is a regular term, which represents the first The complexity of each tree is used to control the complexity of the root rock hazard prediction model to prevent overfitting when the model accuracy is improved by increasing the tree depth.

[0124] In order to solve the objective function, the approximate method of Taylor expansion and second-order derivative is used to perform a second-order Taylor expansion of the loss function near the prediction result, and the objective function is iteratively optimized by adding new trees.

[0125] At each iteration, a new tree is found that can minimize the objective function, that is, the solution:

[0126] ;

[0127] in, Indicates The output of the root rock hazard prediction model at the iteration The corresponding root stone danger prediction results, Indicates the first Tree pairs The results of the classification of root stone hazards are as follows: Indicates The loss function at the iteration is: Indicates The regularization term at the iteration.

[0128] For The loss function at the iteration ,exist Perform a second-order Taylor expansion at , and we get:

[0129] ;

[0130] in, , express about The first and second derivatives of , .

[0131] It can be expressed as:

[0132] ;

[0133] in, Represents the total number of leaf nodes in the root rock hazard prediction model. Indicates the first The weight of a leaf node, represents the split threshold of leaf nodes in the root rock hazard prediction model, represents the second adjustment parameter.

[0134] Will exist The second-order Taylor expansion at Substitute the expression into , and ignore the constant term ,get:

[0135] ;

[0136] Will Expressed as Function, let Yes The function mapped to the leaf node is:

[0137] ;

[0138] Will Substitution ,get:

[0139] ;

[0140] Will Rewrite it as a leaf node aggregation form, let Indicates that all mappings to The set of feature data of leaf nodes, then:

[0141] ;

[0142] right Take the derivative and set it to 0, and solve for:

[0143] ;

[0144] Will Substitution , solve The minimum values ​​are:

[0145] ;

[0146] In this embodiment, the confusion matrix is:

[0147] ;

[0148] in, , , It indicates the probability that the root rock hazard prediction result output by the root rock hazard prediction model is high, medium, or low when the actual root rock hazard result is high. , , It indicates the probability that the root rock hazard prediction result output by the root rock hazard prediction model is high, medium, or low when the actual root rock hazard result is medium. , , It indicates the probability that the root stone hazard prediction result output by the root stone hazard prediction model is high, medium, or low when the actual root stone hazard result is low. High means that the root stone displacement is greater than the preset threshold and continues to change, or the root stone is lost; medium means that the root stone displacement is greater than the preset threshold and remains stable, or the root stone sensor is abnormal; and low means that the root stone displacement is less than or equal to the preset threshold and remains stable.

[0149] The accuracy of the root stone hazard prediction results is visualized through the confusion matrix, which is shown in Table 4.

[0150] Table 4 Confusion matrix

[0151]

[0152] In this embodiment, the performance evaluation indexes of the root stone hazard prediction model include accuracy, precision, recall and balanced F score. Among them, the accuracy rate represents the overall accuracy of the root stone hazard prediction model. The higher the accuracy rate, the better the performance of the root stone hazard prediction model; the precision rate represents the ratio of samples with correct prediction results of the root stone hazard prediction model to actual samples; the recall rate represents the proportion of actual samples correctly predicted by the root stone hazard prediction model. The larger the recall rate, the fewer samples that the root stone hazard prediction model fails to correctly identify, and the stronger the ability of the root stone hazard prediction model; the balanced F score is the harmonic mean index of the comprehensive precision and recall rate, which can effectively identify the prediction error caused by sample imbalance. The larger the value of the balanced F score, the better the comprehensive performance of the root stone hazard prediction model.

[0153] The accuracy is calculated as follows:

[0154] ;

[0155] in, Represents the accuracy of the root stone hazard prediction model.

[0156] The calculation formula for accuracy is:

[0157] ;

[0158] in, Represents the accuracy of the root stone hazard prediction model.

[0159] Specific:

[0160] ;

[0161] in, , , It indicates the accuracy rate corresponding to high, medium and low root rock hazard prediction results output by the root rock hazard prediction model.

[0162] The calculation formula for recall is:

[0163] ;

[0164] in, Represents the recall rate of the root stone hazard prediction model.

[0165] Specific:

[0166] ;

[0167] in, , , It indicates the recall rate corresponding to the high, medium and low root rock hazard prediction results output by the root rock hazard prediction model.

[0168] The calculation formula for the balanced F-score is:

[0169] ;

[0170] in, represents the balanced F score of the root rock hazard prediction model, Represents the weight factor.

[0171] Specific:

[0172] ;

[0173] Step 4: Based on the performance evaluation results, obtain the optimal parameter combination of the root stone hazard prediction model.

[0174] Step 4: Based on the optimal parameter combination of the root stone hazard prediction model, solidify the root stone hazard prediction model.

[0175] In this embodiment, the parameters to be optimized of the root stone hazard prediction model are adjusted through the model parameter tuning technology to improve the prediction accuracy of the root stone hazard prediction model; the prediction ability and generalization ability of the root stone hazard prediction model are evaluated on the test set to obtain the optimal parameter combination of the root stone hazard prediction model to ensure that the root stone hazard prediction model has excellent prediction accuracy and good generalization ability when dealing with different data situations; after obtaining the optimal parameter combination of the root stone hazard prediction model, the root stone hazard prediction model is solidified based on the optimal parameter combination of the root stone hazard prediction model to obtain the optimal root stone hazard prediction model.

[0176] The optimal root rock hazard prediction model is applied to the test set to evaluate the prediction ability of the optimal root rock hazard prediction model, and the importance ranking of each feature factor, confusion matrix, and root rock hazard prediction model evaluation index are obtained. The importance ranking of each feature factor is as follows: Figure 2 As shown, the confusion matrix is Figure 3 The evaluation indicators of the root stone hazard prediction model are shown in Table 5.

[0177] Table 5 Evaluation indexes of root rock hazard prediction model

[0178]

[0179] Depend on Figure 2 It can be seen that among the 15 features, the z-direction offset angle, y-direction offset angle, and water level are significantly more important than other features. The corresponding field survey report and research results show that the cause of the root stone danger is that the flow velocity in front of the dam is too large, the force of the water flow on the root stone block exceeds the resistance of the root stone itself, and the dam foundation is washed away by the water flow, scouring pits are formed around the dam body, the root stone collapses downward, and the root stone is lost due to scouring, resulting in root stone danger. In addition, water flow conditions are an important factor affecting the stability of the dam bank root stone, especially in the wandering river section, where the water flow is prone to directly hit the dam bank project, the dam foundation is severely washed away, and the root stone on the surface of the dam pile is directly washed away, causing the dam bank project to be unstable and the dam bank root stone is prone to be lost. For high-sediment-content rivers, the rheological characteristics of high-sediment-content water flow have changed, and the sand-carrying capacity and scouring intensity are greater than those of general sand-carrying water flow. In addition, the narrowing effect of the river channel increases the flow rate per width of the water flow, the water flow is concentrated, and the scouring force of the water flow is enhanced. Deep scouring pits are prone to appear at the span angle and the front of the dam, and root stones are prone to disappearance. The z-direction offset angle and the y-direction offset angle are direct monitoring factors for monitoring the displacement of root stones. At the same time, the water level is closely related to the flow velocity, so the importance of the water level feature is higher than that of other features, which shows that the root stone hazard prediction model is reasonable in predicting root stone hazard.

[0180] In the subsequent root rock hazard prediction, the solidified root rock hazard prediction model, that is, the optimal root rock hazard prediction model, can be directly used to perform root rock hazard prediction.

[0181] The root rock hazard prediction method provided in this embodiment processes and converts multi-source data of root rock hazard influencing factors through multi-source data processing technologies such as image processing technology (including raster image accumulation method, raster image upsampling method, pixel-level image data extraction method, etc.) and text mapping technology to construct a feature data set; automatically mines the correlation between multi-source data of root rock hazard influencing factors and root rock hazard through the extreme gradient boosting tree algorithm, identifies the weights of key factors affecting root rock hazard, can achieve rapid and accurate prediction of root rock hazard, and provide technical support for flood control decision-making; quickly determines the model parameter range and corresponding threshold value through a combination of cross-validation and grid search, can save model optimization time and computing resources, and reduce unnecessary loss of manpower and material resources.

[0182] Embodiment 2: This embodiment provides a root stone danger prediction device, comprising:

[0183] A data acquisition module is used to acquire multi-source data of factors affecting root stone danger, and to construct a feature data set based on the multi-source data of factors affecting root stone danger;

[0184] A prediction module is used to input the feature data set into a pre-built root rock hazard prediction model to perform root rock hazard prediction and obtain a root rock hazard prediction result;

[0185] Among them, the multi-source data of root stone hazard influencing factors include structured data and unstructured data; the root stone hazard prediction model is constructed based on the extreme gradient boosting tree algorithm; when predicting the root stone hazard, based on the confusion matrix, the grid search method combined with the cross-validation method is used to optimize the parameters of the root stone hazard prediction model.

[0186] The root stone hazard prediction device provided in the embodiment of the present application can execute the root stone hazard prediction method provided in Example 1 of the present application, and has the corresponding functional modules and beneficial effects of the execution method.

[0187] Embodiment 3:

[0188] This embodiment provides a computer system, including:

[0189] Storage medium: used to store computer programs;

[0190] Processor: used to execute computer programs to implement the steps of the root stone hazard prediction method in Example 1.

[0191] Embodiment 4:

[0192] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the root stone hazard prediction method in Embodiment 1 are implemented.

[0193] Embodiment 5:

[0194] This embodiment provides a computer program product, including a computer program, which implements the steps of the root stone hazard prediction method in Example 1 when executed by a processor.

[0195] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0196] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0197] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0198] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0199] The above are only preferred implementations of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A root stone danger prediction method, characterized in that: include: Acquire multi-source data of factors affecting root rock hazard, and construct a feature data set based on the multi-source data of factors affecting root rock hazard; Inputting the characteristic data set into a pre-built root rock hazard prediction model to perform root rock hazard prediction and obtain a root rock hazard prediction result; The multi-source data of factors affecting root stone danger include structured data and unstructured data; the root stone danger prediction model is constructed based on the extreme gradient boosting tree algorithm; when predicting root stone danger, the parameters of the root stone danger prediction model are optimized based on the confusion matrix, using the grid search method combined with the cross-validation method; The multi-source data of factors affecting root rock hazards include hydrological data and monitoring equipment working condition data; The hydrological data include precipitation, water level, flow, and turbulent energy in the area where the root stone is located; The monitoring equipment working condition data includes the function mode, reporting mode, offset angle, detection cycle, reporting frequency, alarm angle, data docking port, communication signal strength, and voltage value of the rootstone sensor; Among them, water level, flow, reporting mode, offset angle, detection cycle, reporting frequency, alarm angle, data docking port, communication signal strength, and voltage value are numerical data; precipitation and turbulence can be image data; and function mode is text data; For precipitation, the calibrated precipitation band in the global precipitation measurement data was selected, and the precipitation images of the corresponding time period were screened out. The three-hour precipitation data was reduced to form a daily data set through image reduction. The precipitation images were upsampled, and the images were cropped using the latitude and longitude information of the root stone area. The image data was converted into point data using the mean or maximum value and imported into the feature data set. For turbulent energy, the turbulent energy mechanism model image is selected, and the turbulent energy mechanism model image is opened in matrix form using the geospatial data abstraction library. The pixel size is obtained using the grid X size and the grid Y size. The new image resolution, affine information, and pixel value type are set through the driver program, and the adjusted image is output. The image is cropped using the longitude and latitude information of the root stone area, and the turbulent energy maximum value point data is extracted and stored in the feature data set.

2. The root stone danger prediction method according to claim 1, characterized in that: Based on the multi-source data of the factors affecting the root stone danger, the characteristic data set is constructed including: Processing the multi-source data of the factors affecting the root rock hazard to obtain characteristic data; The characteristic data together constitute a characteristic data set; The multi-source data of the factors affecting the root rock hazard are processed to obtain characteristic data, including: Performing data aggregation and data fusion on structured data and unstructured data in the multi-source data of the root stone hazard influencing factors, and converting the unstructured data into structured data; The structured data is cleaned, the structured data is aligned, missing values ​​and abnormal values ​​in the structured data are deleted, and feature encoding is performed on the structured data to obtain feature data.

3. The root stone danger prediction method according to claim 1, characterized in that: The root stone danger prediction model is: ； in, The first feature datasets, The output of the root rock hazard prediction model The corresponding root stone danger prediction results, Indicates the first Tree pairs The results of the classification of root stone hazards are as follows: It represents the total number of trees in the root rock danger prediction model; During each prediction, the root rock hazard prediction model classifies the root rock hazard based on the importance of the feature data in the feature data set to the root rock hazard, and obtains a root rock hazard classification result; The calculation formula of the importance of the characteristic data on the root stone hazard is: ； in, Indicates the importance of the characteristic data on the root rock hazard. , Represents the first-order gradient of the feature data in the left and right leaf nodes of the root rock hazard prediction model, , Represents the second-order gradient of the feature data in the left and right leaf nodes of the root rock hazard prediction model, represents the first adjustment parameter, Represents the computational complexity control variable.

4. The root stone danger prediction method according to claim 1, characterized in that: When predicting the root stone hazard, based on the confusion matrix, the grid search method combined with the cross-validation method is used to optimize the parameters of the root stone hazard prediction model, including: Based on the parameters to be optimized and the preset parameter range of the root stone hazard prediction model, a parameter grid is constructed, and each point in the parameter grid is traversed using a grid search method, wherein each point in the parameter grid corresponds to a set of parameter combinations; Based on the cross-validation method, the feature data set is randomly divided into mutually exclusive subsets, using each subset as a test set in turn, and the remaining subsets as training sets, to test the root stone hazard prediction model. training and testing, and based on the confusion matrix, evaluating the performance of the root stone hazard prediction model, and obtaining the performance evaluation results of the root stone hazard prediction model corresponding to each group of parameter combinations; Based on the performance evaluation results, the optimal parameter combination of the root rock hazard prediction model is obtained.

5. The root stone danger prediction method according to claim 4, characterized in that: When the root rock hazard prediction model is trained, the objective function of the root rock hazard prediction model is for: ； in, The output of the root rock hazard prediction model The corresponding root stone danger prediction results, The first feature data, express The corresponding real result is represents the loss function, represents the total number of feature data in the training set, Indicates the first Tree pairs The results of the classification of root stone hazards are as follows: represents the regularization term, Represents the total number of trees in the root rock hazard prediction model.

6. The root stone danger prediction method according to claim 4, characterized in that: The confusion matrix is: ； in, , , It indicates the probability that the root rock hazard prediction result output by the root rock hazard prediction model is high, medium, or low when the actual root rock hazard result is high. , , It indicates the probability that the root rock hazard prediction result output by the root rock hazard prediction model is high, medium, or low when the actual root rock hazard result is medium. , , Indicates the probability that the root stone hazard prediction result output by the root stone hazard prediction model is high, medium, or low when the actual root stone hazard result is low. High means that the root stone displacement is greater than the preset threshold and continues to change, or the root stone is lost; medium means that the root stone displacement is greater than the preset threshold and remains stable, or the root stone sensor is abnormal; and low means that the root stone displacement is less than or equal to the preset threshold and remains stable. The performance evaluation indicators of the root stone hazard prediction model include accuracy, precision, recall and balanced F score; The calculation formula of the accuracy is: ； in, It indicates the accuracy of the root stone hazard prediction model; The calculation formula of the accuracy is: ； in, It represents the accuracy of the root stone hazard prediction model; The calculation formula of the recall rate is: ； in, represents the recall rate of the root stone hazard prediction model; The calculation formula of the balanced F score is: ； in, represents the balanced F score of the root rock hazard prediction model, Represents the weight factor.

7. The root stone danger prediction method according to claim 4, characterized in that: The parameters to be optimized in the root stone hazard prediction model include booster type, learning rate, number of iterations, maximum depth of the tree, minimum weight of child nodes, minimum loss reduction required for splitting, subsampling ratio, column sampling ratio, number of threads, L1 regularization term coefficient, and random seed.

8. A root stone danger prediction device, characterized in that: include: A data acquisition module, used to acquire multi-source data of factors affecting root rock hazard, and construct a feature data set based on the multi-source data of factors affecting root rock hazard; A prediction module, used for inputting the feature data set into a pre-built root rock hazard prediction model to perform root rock hazard prediction and obtain a root rock hazard prediction result; The multi-source data of factors affecting root stone danger include structured data and unstructured data; the root stone danger prediction model is constructed based on the extreme gradient boosting tree algorithm; when predicting root stone danger, the parameters of the root stone danger prediction model are optimized based on the confusion matrix, using the grid search method combined with the cross-validation method; The multi-source data of factors affecting root rock hazards include hydrological data and monitoring equipment working condition data; The hydrological data include precipitation, water level, flow, and turbulent energy in the area where the root stone is located; The monitoring equipment working condition data includes the function mode, reporting mode, offset angle, detection cycle, reporting frequency, alarm angle, data docking port, communication signal strength, and voltage value of the rootstone sensor; Among them, water level, flow, reporting mode, offset angle, detection cycle, reporting frequency, alarm angle, data docking port, communication signal strength, and voltage value are numerical data; precipitation and turbulence can be image data; and function mode is text data; For precipitation, the calibrated precipitation band in the global precipitation measurement data was selected, and the precipitation images of the corresponding time period were screened out. The three-hour precipitation data was reduced to form a daily data set through image reduction. The precipitation images were upsampled, and the images were cropped using the latitude and longitude information of the root stone area. The image data was converted into point data using the mean or maximum value and imported into the feature data set. For turbulent energy, the turbulent energy mechanism model image is selected, and the turbulent energy mechanism model image is opened in matrix form using the geospatial data abstraction library. The pixel size is obtained using the grid X size and the grid Y size. The new image resolution, affine information, and pixel value type are set through the driver program, and the adjusted image is output. The image is cropped using the longitude and latitude information of the root stone area, and the turbulent energy maximum value point data is extracted and stored in the feature data set.

9. A computer system, characterized in that: include: Storage medium: used to store computer programs; Processor: used to execute the computer program to implement the steps of the root stone hazard prediction method described in any one of claims 1 to 7.

Images