Screening method, breach peak flow prediction method and related device

Abstract

The invention discloses a screening method, a breach peak flow prediction method and a related device, and the method comprises the steps: training a pre-selected breach peak flow prediction model through employing a training sample corresponding to a to-be-screened feature, determining a contribution value of each to-be-screened feature to a prediction result, sorting the to-be-screened features according to the contribution values, and carrying out the sorting of the to-be-screened features. According to the breach peak flow prediction method, redundant features are eliminated by adopting a feature selection method, the optimal feature subset matched with the pre-selected breach peak flow prediction model is obtained, feature screening is effectively achieved, and when the breach peak flow is predicted, the prediction difficulty is reduced, and the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to a screening method, a prediction method for peak breach discharge and related devices, belonging to the field. Background Art

[0002] Reservoir dams and dikes of polder areas are important infrastructure, which have played a huge role in flood control and disaster reduction, urban water supply, hydropower generation, farmland irrigation and other aspects. However, during the long-term service process, earth-rock dams are prone to disasters such as cracking, landsliding, dispersion, piping, quicksand and erosion. Once breached, it will cause floods. The breach of the dam is sudden, and the development trend of the breach is difficult to predict, which greatly increases the difficulty of emergency rescue and disaster relief. The rapid prediction of peak breach discharge is of great significance for emergency rescue and disaster relief and downstream flood disaster assessment. For a long time, its prediction method has been an important research hotspot.

[0003] At present, many scholars at home and abroad have carried out a lot of research on peak breach discharge prediction models, such as RF (Random Forest), GBM (Gradient Boosting Machine), XGBoost (Extreme Gradient Boosting), LightGBM (Light Gradient Boosting Machine), CatBoost (Categorical Boosting), AdaBoost (Adaptive Boosting), etc. Through research, it is found that peak breach discharge is affected by multiple features, such as dam height, water depth above the breach bottom, reservoir capacity, and breach width, etc. The input of too many features also makes the model complex, increases the prediction difficulty and reduces the prediction accuracy. Therefore, screening appropriate input features for the model is a prerequisite for reducing the prediction difficulty and improving the prediction accuracy, but there is no corresponding method at present. Summary of the Invention

[0004] The present invention provides a screening method, a prediction method for peak breach discharge and related devices, which solves the problems disclosed in the background art.

[0005] According to one aspect of the present application, a feature screening method is provided, where the feature is a feature that affects peak breach discharge, and the method includes: Training a pre-selected peak breach discharge prediction model with the features to be screened to obtain the prediction results during training; Determining the contribution value of each feature to be screened to the prediction result according to the features to be screened and the prediction results, and sorting the features to be screened according to the contribution values; The sorted features to be screened are screened using a feature selection method to obtain an optimal feature subset adapted to the pre-selected breach peak discharge prediction model.

[0006] Further, the contribution value is the SHAP value; the SHAP method is used to determine the contribution value of each feature to be screened to the prediction result.

[0007] Further, the RFE method is used to screen the sorted features to be screened.

[0008] According to another aspect of the present application, there is provided a feature screening device, where the features are features affecting the breach peak discharge, and the device includes: A result prediction module that trains a pre-selected breach peak discharge prediction model using the features to be screened to obtain the prediction result during training; A contribution value determination module that determines the contribution value of each feature to be screened to the prediction result according to the features to be screened and the prediction result, and sorts the features to be screened according to the contribution value; A screening module that screens the sorted features to be screened using a feature selection method to obtain an optimal feature adapted to the pre-selected breach peak discharge prediction model.

[0009] According to another aspect of the present application, there is provided a model screening method, where the model is a breach peak discharge prediction model, and the method includes: Traverse the model set, use the above feature screening method to determine the optimal feature subset adapted to the model, use the test samples corresponding to the optimal feature subset as the input of the model, and test the prediction performance of the model; Select the model with the best prediction performance as the final model for predicting the breach peak discharge.

[0010] According to another aspect of the present application, there is provided a model screening device, where the model is a breach peak discharge prediction model, and the device includes: A traversal module that traverses the model set, uses the above feature screening method to determine the optimal feature subset adapted to the model, uses the test samples corresponding to the optimal feature subset as the input of the model, and tests the prediction performance of the model; A selection module that selects the model with the best prediction performance as the final model for predicting the breach peak discharge.

[0011] According to another aspect of the present application, there is provided a method for predicting the breach peak discharge, including: using a model to predict the breach peak discharge; where the model is the model screened by the above model screening method.

[0012] According to another aspect of the present application, there is provided a device for predicting the peak discharge of a breach, including a prediction module that uses a model to predict the peak discharge of the breach; wherein, the model is the model selected by using the above-mentioned model screening method.

[0013] According to another aspect of the present application, there is provided a computer-readable storage medium storing one or more programs, and the one or more programs include instructions that, when executed by a computing device, cause the computing device to execute the feature screening method, the model screening method, or the peak discharge prediction method of the breach.

[0014] According to another aspect of the present application, there is provided a computer device including one or more processors and one or more memories. The one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors. The one or more programs include instructions for executing the feature screening method, the model screening method, or the peak discharge prediction method of the breach.

[0015] The beneficial effects achieved by the present invention: The present invention uses the training samples corresponding to the features to be screened to train the pre-selected peak discharge prediction model of the breach, determines the contribution value of each feature to be screened to the prediction result, sorts the features to be screened according to the contribution value, and uses the feature selection method to eliminate redundant features, obtaining the optimal feature subset adapted to the pre-selected peak discharge prediction model of the breach, effectively realizing feature screening. When predicting the peak discharge of the breach, the prediction difficulty is reduced and the prediction accuracy is improved. Description of the Drawings

[0016] Figure 1 It is a flowchart of the feature screening method; Figure 2 It is a schematic diagram of the cross-section of the earth dam breach; Figure 3 It is a schematic diagram of the longitudinal section of the earth dam breach; Figure 4 It is a SHAP summary diagram corresponding to RandomForest; Figure 5 It is a SHAP summary diagram corresponding to GradientBoosting; Figure 6 It is a SHAP summary diagram corresponding to XGBoost; Figure 7 It is a SHAP summary diagram corresponding to LightGBM; Figure 8 It is a SHAP summary diagram corresponding to CatBoost; Figure 9 It is a SHAP summary diagram corresponding to AdaBoost Figure 10It is the SHAP bar chart corresponding to RandomForest; Figure 11 It is the SHAP bar chart corresponding to GradientBoosting; Figure 12 It is the SHAP bar chart corresponding to XGBoost; Figure 13 It is the SHAP bar chart corresponding to LightGBM; Figure 14 It is the SHAP bar chart corresponding to CatBoost; Figure 15 It is the SHAP bar chart corresponding to AdaBoost; Figure 16 It is the comparison chart during the screening process; Figure 17 It is the block diagram of the feature screening device; Figure 18 It is the flowchart of the model screening method; Figure 19 It is the final performance comparison chart of the model; Figure 20 It is the block diagram of the model screening device. Detailed implementation manners

[0017] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and in no way limits the present application and its application or use. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0018] Unless otherwise specifically stated, the relative arrangements, numerical expressions, and numerical values of the components and steps described in these embodiments do not limit the scope of the present application.

[0019] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the accompanying drawings are not drawn in actual proportional relationships.

[0020] For technologies, methods, and devices known to those of ordinary skill in the relevant fields, they may not be discussed in detail, but under appropriate circumstances, the technologies, methods, and devices should be regarded as part of the specification.

[0021] In all the examples shown and discussed here, any specific value should be interpreted as merely exemplary and not as a limitation. Therefore, other examples of the exemplary embodiments may have different values.

[0022] It should be noted that similar symbols and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0023] The embodiment of the present application provides a feature screening method based on the SHAP method and the RFE method. This feature screening method can be executed by a feature screening device, which can be a terminal device or a server. Among them, the terminal device can include, but is not limited to, mobile phones, computers, smart wearable devices, intelligent vehicle-mounted devices, etc., and the embodiment of the present application does not make restrictions; the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, big data, and artificial intelligence platforms, etc., and the embodiment of the present application does not make restrictions. Optionally, this feature screening method can also be executed collaboratively by multiple electronic devices with computing power. For the convenience of description, the subsequent embodiments will be described with the feature screening device executing.

[0024] See Figure 1 , Figure 1 is a flowchart of a feature screening method provided by the embodiment of the present application. This feature screening method can be executed by a feature screening device. The features in this feature screening method mainly refer to the features that affect the peak discharge of the breach. This method can at least include the following steps: Step 1: Train a pre-selected peak discharge prediction model of the breach using the training samples corresponding to the features to be screened, and obtain the trained prediction results.

[0025] It should be noted that research shows that the dam-break flow first forms an initial scour pit on the downstream slope of the dam and gradually scours upstream. When the scour develops to the upstream edge of the dam crest, the breach process becomes very rapid and violent, accompanied by intermittent instability and collapse of the breach slope, and finally a relatively stable breach is formed. The typical cross-section and longitudinal section of the breach are as Figure 2 shown. Fully considering the influence of the upstream reservoir and the characteristics of the breach shape, 6 important features are selected here as the research objects, that is, the features to be screened. The physical meanings and detailed descriptions thereof are shown in Table 1.

[0026] Table 1 Physical meanings and detailed description table of features to be screened

[0027] It should be noted that the training samples include the above-mentioned features and the corresponding peak discharge of the breach. Training the peak discharge prediction model of the breach with the training samples can obtain the trained prediction results, that is, the prediction results output by the model.

[0028] Step 2: According to the features to be screened and the prediction results, the contribution value of each feature to be screened to the prediction result is determined, and the features to be screened are sorted according to the contribution value.

[0029] In some implementations, the contribution value adopts the SHAP value. Therefore, the SHAP method is used here to determine the contribution value of each feature to be screened to the prediction result, which can effectively explore and understand the feature importance of complex models and the relationship between features, thereby improving the interpretability of the model.

[0030] The advantage of the SHAP (Shapley Additive Explanations) method is that it can reflect the feature importance of each sample, and can further explain how each feature variable affects the simulation value. The basic principle of the method is: Explanation model g is a linear function of the feature properties and can explain both individual predictions and the overall model prediction based on the average feature properties across all sample points.

[0031] ; In the formula, is the model for the training sample z The prediction result is decomposed into the baseline value and the sum of the contributions of each feature. M is the number of features, is the baseline prediction, that is, the output of the model when all features take reference values (such as missing or default values), It is i The SHAP value of a feature reflects the contribution of the feature to the prediction result (positive value means positive contribution, negative value means negative contribution). It is a feature existence variable, usually binary (1 means the feature exists, 0 means the feature is missing or takes the reference value).

[0032] It should be noted that the feature sorting here can be sorted from large to small according to the SHAP value, or from small to large, depending on the actual situation.

[0033] Taking the existing six models as an example, the SHAP method is used to calculate the contribution of each feature to the prediction results to explain the output of the final model.

[0034] See also Figures 4 to 9 , in the figure: Pink dots: indicate that the feature value has a positive impact on the prediction in this model; Blue dots: Indicates that the feature value has a negative impact on the prediction in this model; Horizontal axis (SHAP value): It shows the impact magnitude of each feature on the prediction result. The farther the point is from the center line (zero point), the greater the impact of the feature on the model output. A positive SHAP value indicates a positive impact, and a negative SHAP value indicates a negative impact; Vertical axis (feature arrangement): The features arranged vertically in the figure are sorted from top to bottom according to their influence. The features above have a greater overall impact on the model output, while the features below have a smaller impact; Feature influence explanation: Topmost feature: It shows a large number of positive and negative impacts, indicating that its different feature values have very different impacts on the model prediction results; Middle feature: It also shows points of two colors, but the distribution of the points is more concentrated and the influence is relatively small; Bottom feature: It has the smallest impact on the model, and most of the impacts are relatively close to the zero value, indicating that these features contribute less to the model prediction.

[0035] The SHAP plot can not only show which features are important, but also reveal how each feature affects the peak flow. See Figures 10 to 15 , which shows the importance ranking of these features in affecting the peak flow in 6 models. In all models, V w 、 B ave and h d make the greatest contribution to the prediction of peak flow.

[0036] Step 3: Use the feature selection method to screen the sorted features to be screened, and obtain the optimal feature subset adapted to the pre-selected breach peak flow prediction model.

[0037] It should be noted that the feature selection method here adopts the RFE method, that is, the RFE method is used to screen the sorted features to be screened.

[0038] The RFE method, that is, the recursive feature elimination method, is a "greedy" algorithm aimed at searching for the optimal feature subset. The steps are summarized as follows: 31) Establish a training classifier; 32) Calculate the importance measure of the features; 33) Eliminate the irrelevant features with low importance measures; 34) Repeat steps 31) to 33) for the remaining features until the best feature subset is selected.

[0039] Taking the existing 6 models as an example, see Figure 16 , taking the XGBoost model in the figure as an example, in the feature selection process, the coefficient of determination R 2For reflecting the prediction performance, when the first four features ( V w , B ave , h b and h d ) are selected, the R 2 of XGBoost is the largest, that is, the performance is the best, that is, V w , B ave , h b and h d constitute a set that is the optimal feature subset adapted to XGBoost.

[0040] The above method uses the training samples corresponding to the features to be screened to train the pre-selected peak breach discharge prediction model, determines the contribution value of each feature to be screened to the prediction result, sorts the features to be screened according to the contribution value, and uses the feature selection method to eliminate redundant features, obtaining the optimal feature subset adapted to the pre-selected peak breach discharge prediction model, effectively realizing feature screening. When predicting the peak breach discharge, it reduces the prediction difficulty and improves the prediction accuracy.

[0041] See Figure 17 , Figure 17 which is a block diagram of a feature screening device provided by an embodiment of the present application. Figure 17 The embodiment of Figure 17 is a virtual device that can be loaded and executed by a computer device, and the computer device may include the above feature screening device. The device of

[0042] may include a result prediction module, a contribution value determination module, and a screening module. When used to execute the above feature screening method, it can:

[0043] The result prediction module uses the training samples corresponding to the features to be screened to train the pre-selected peak breach discharge prediction model and obtains the trained prediction result; wherein, the feature is a feature that affects the peak breach discharge.

[0044] The above-mentioned device trains a pre-selected peak breach discharge prediction model with training samples corresponding to the features to be screened, determines the contribution value of each feature to be screened to the prediction result, sorts the features to be screened according to the contribution value, and uses a feature selection method to eliminate redundant features, so as to obtain an optimal feature subset adapted to the pre-selected peak breach discharge prediction model, effectively realizing feature screening. When predicting the peak breach discharge, the prediction difficulty is reduced and the prediction accuracy is improved.

[0045] See Figure 18 , Figure 18 which is a flowchart of a model screening method provided by an embodiment of the present application. This model screening method can be executed by a model screening device. The model screening device is similar to the feature screening device and will not be repeated here. The model in this model screening method is a peak breach discharge prediction model. This method may at least include the following steps: S1) Traverse the model set, use the above-mentioned feature screening method to determine the optimal feature subset adapted to the model, use the test samples corresponding to the optimal feature subset as the input of the model, and test the prediction performance of the model.

[0046] It should be noted that the models in the model set may include 6 existing models, namely RF, GBM, XGBoost, LightGBM, CatBoost, and AdaBoost. In order to screen out the optimal model, the test samples corresponding to the optimal feature subset of each model are used as the model input, and the root mean square error RMSE, mean absolute error MAE, and coefficient of determination R 2 are used to reflect the prediction performance.

[0047] S2) Select the model with the optimal prediction performance as the final model for predicting the peak breach discharge.

[0048] Taking the existing 6 models as an example, the final performance evaluation results are as Figure 19 shown. Compared with other machine learning models, XGBoost has the best performance, with the lowest RMSE and MAE and the highest R 2 , that is, XGBoost is used as the final model.

[0049] The above method determines the optimal feature subset of each candidate model based on the feature screening method, evaluates the prediction performance of each candidate model based on the optimal feature subset, and selects the model with the optimal prediction performance as the final model for predicting the peak breach discharge, which can further improve the prediction accuracy.

[0050] See Figure 20 , Figure 20 which is a block diagram of a model screening device provided by an embodiment of the present application. Figure 20An embodiment is a virtual device that can be loaded and executed by a computer device, which may include the above-mentioned model screening device. Figure 20 The device may include a traversal module and a selection module. When used to execute the above-mentioned model screening method, it can: The traversal module traverses the model set, uses the above-mentioned feature screening method to determine the optimal feature subset adapted to the model, takes the test samples corresponding to the optimal feature subset as the input of the model, and tests the prediction performance of the model; wherein, the model is a breach peak flow prediction model.

[0051] The selection module selects the model with the optimal prediction performance as the final model for predicting the breach peak flow.

[0052] The above device determines the optimal feature subset of each candidate model based on the feature screening method, evaluates the prediction performance of each candidate model based on the optimal feature subset, and selects the model with the optimal prediction performance as the final model for predicting the breach peak flow, which can further improve the prediction accuracy.

[0053] An embodiment of the present application also provides a method for predicting the breach peak flow. This method for predicting the breach peak flow can be executed by a prediction device. The prediction device is similar to the model screening device and the feature screening device, and will not be repeated here. This method can at least include using a model to predict the breach peak flow; wherein, the model is the model screened by the above-mentioned model screening method.

[0054] The above method predicts the breach peak flow based on the model with the optimal prediction performance, and can obtain a prediction result with higher accuracy.

[0055] An embodiment of the present application also provides a device for predicting the breach peak flow. The embodiment is a virtual device that can be loaded and executed by a computer device, which may include the above-mentioned prediction device. The device may include a prediction module. When used to execute the above-mentioned method for predicting the breach peak flow, it can: The prediction module uses a model to predict the breach peak flow; wherein, the model is the model screened by the above-mentioned model screening method.

[0056] The above device predicts the breach peak flow based on the model with the optimal prediction performance, and can obtain a prediction result with higher accuracy.

[0057] The present application also relates to a computer-readable storage medium. The computer-readable storage medium stores one or more programs. The one or more programs include instructions that, when executed by a computing device, cause the computing device to execute the feature screening method, the model screening method, or the method for predicting the breach peak flow.

[0058] The present application also relates to a computer device, including one or more processors and one or more memories. One or more programs are stored in the one or more memories and configured to be executed by the one or more processors. The one or more programs include instructions for executing the feature screening method, the model screening method, or the breach peak flow prediction method.

[0059] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0060] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to generate a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more of the processes or multiple processes and / or blocks Figure 1 one or more of the blocks or multiple blocks.

[0061] These computer program instructions can also be stored in a computer-readable memory capable of guiding the computer or other programmable data processing devices to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in Figure 1 one or more of the processes or multiple processes and / or blocks Figure 1 one or more of the blocks or multiple blocks.

[0062] These computer program instructions can also be loaded onto the computer or other programmable data processing devices, such that a series of operation steps are executed on the computer or other programmable devices to generate computer-implemented processing. Thus, the instructions executed on the computer or other programmable devices provide steps for implementing the functions specified in Figure 1 one or more of the processes or multiple processes and / or blocks Figure 1 one or more of the blocks or multiple blocks.

[0063] The above are only embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A feature screening method, characterized in that, The feature is a feature that affects the peak discharge of the breach, and the method includes: Training a pre-selected peak discharge prediction model of the breach using training samples corresponding to the features to be screened, and obtaining a trained prediction result; Determining the contribution value of each feature to be screened to the prediction result according to the features to be screened and the prediction result, and sorting the features to be screened according to the contribution value; Screening the sorted features to be screened using a feature selection method to obtain an optimal feature subset adapted to the pre-selected peak discharge prediction model of the breach.

2. The method according to claim 1, characterized in that, The contribution value is the SHAP value; the SHAP method is used to determine the contribution value of each feature to be screened to the prediction result.

3. The method according to claim 1, characterized in that The RFE method is used to screen the sorted features to be screened.

4. A feature screening device, characterized in that, The feature is a feature that affects the peak discharge of the breach, and the device includes: A result prediction module that trains a pre-selected peak discharge prediction model of the breach using training samples corresponding to the features to be screened, and obtains a trained prediction result; A contribution value determination module that determines the contribution value of each feature to be screened to the prediction result according to the features to be screened and the prediction result, and sorts the features to be screened according to the contribution value; A screening module that screens the sorted features to be screened using a feature selection method to obtain an optimal feature adapted to the pre-selected peak discharge prediction model of the breach.

5. A model screening method, characterized in that, The model is a peak discharge prediction model of the breach, and the method includes: Traversing the model set, using the method described in any one of claims 1 to 3 to determine an optimal feature subset adapted to the model, using the test samples corresponding to the optimal feature subset as the input of the model, and testing the prediction performance of the model; Selecting the model with the optimal prediction performance as the final model for predicting the peak discharge of the breach.

6. A model screening device, characterized in that, The model is a peak discharge prediction model of the breach, and the device includes: A traversing module that traverses the model set, uses the method described in any one of claims 1 to 3 to determine an optimal feature subset adapted to the model, uses the test samples corresponding to the optimal feature subset as the input of the model, and tests the prediction performance of the model; A selection module that selects the model with the optimal prediction performance as the final model for predicting the peak discharge of the breach.

7. A method for predicting the peak discharge of a breach, characterized in that, Including: Using the model to predict the peak discharge of the breach; wherein, the model is the model screened by the method described in claim 5.

8. A peak discharge prediction device for a breach, characterized in that, Including a prediction module that uses the model to predict the peak discharge of the breach; wherein, the model is the model screened by the method described in claim 5.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, and the one or more programs include instructions that, when executed by a computing device, cause the computing device to execute the method described in any one of claims 1 to 3, 5, 7.

10. A computer device, characterized in that, Including: One or more processors and one or more memories, the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the method described in any one of claims 1 to 3, 5, 7.

Images