CN-122021471-A - Method and system for predicting downburst leveling and average wind speed based on regression random forest

CN122021471ACN 122021471 ACN122021471 ACN 122021471ACN-122021471-A

Abstract

The invention relates to the technical field of wind engineering and artificial intelligence, and discloses a method and a system for predicting downburst flow average wind speed based on regression random forest, wherein a CFD numerical simulation model of mobile downburst flow is built based on an impact jet flow theory; the method comprises the steps of carrying out uniform sampling in a two-dimensional parameter space formed by the environmental wind speed and the storm center moving speed by utilizing a Harton sequence to generate a parameter combination sample, constructing a training data set comprising a plurality of groups of input features and target output by means of batch CFD calculation, constructing a regression random forest model, optimizing the number and the maximum depth of decision trees by utilizing parameter sensitivity analysis, and finally carrying out rapid prediction on any working condition to be predicted by utilizing the trained model. The method can accurately capture the double peak characteristics of the wind speed of the movable downburst, greatly reduce the calculation cost while ensuring high prediction accuracy, and realize the real-time and rapid prediction of the downburst wind field.

Inventors

LI ZHIPENG
MAO JIANFENG
LIU QING
YU ZHIWU
YAN DONGDONG
WEI YONGLIANG
XU LEI
Tan sui

Assignees

中铁科学研究院集团有限公司
高速铁路建造技术国家工程研究中心

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. A method for predicting the average wind speed of a downburst leveling based on a regression random forest is characterized by comprising the following steps: s1, based on an impact jet theory, establishing a computational fluid dynamics numerical simulation model of mobile down-stroke storm flow; S2, determining the change range of the movement speed of the ambient wind speed and the storm center so as to construct a two-dimensional parameter space, and uniformly sampling in the two-dimensional parameter space by using a Harton sequence sampling method to generate parameter combination samples containing different ambient wind speeds and different movement speeds of the storm center; S3, inputting the parameter combination sample into the computational fluid dynamics numerical simulation model for calculation, and constructing a training data set comprising a plurality of groups of input characteristic variables and target output variables according to wind speed time interval data of the calculated monitoring points, wherein the input characteristic variables comprise ambient wind speed, storm center moving speed and time points, and the target output variables are downstriking storm flow average wind speeds at corresponding time points; S4, constructing an initial regression random forest model, training by using the training data set, and optimizing the number of decision trees and the maximum depth of the decision trees by adopting a parameter sensitivity analysis method in the training process to obtain an optimal super-parameter combination, so as to obtain a trained regression random forest model as a down-storm flow average wind speed prediction model; S5, acquiring the environmental wind speed, the storm center moving speed and the time sequence to be predicted of the working condition to be predicted, and inputting the down-storm leveling and averaging wind speed prediction model to obtain a down-storm leveling and averaging wind speed prediction result of the corresponding time sequence under the working condition.
2. The regression random forest based method for predicting the average wind speed of the downstorm flow according to claim 1, wherein the S1 specifically comprises: S101, constructing a three-dimensional cylindrical calculation domain by adopting a three-dimensional impact jet model, and arranging a jet inlet at the top of the calculation domain, dividing the calculation domain into a static region and a movable nozzle region aiming at movable downward storm flow, and processing a region interface by utilizing a sliding grid technology; s102, dispersing a calculation domain by adopting a hexahedral structured grid, carrying out grid encryption processing on a near-ground area, and constructing a sliding grid interface between the static area and the mobile nozzle area by utilizing an arbitrary coupling grid interface; s103, adopting an unsteady state Reynolds average Navier-Stokes equation as a fluid control equation, and adopting SST-k The turbulence model analyzes the flow separation phenomenon and adopts SIMPLIC algorithm to solve the pressure-speed coupling problem; S104, setting a velocity component and a moving speed of the jet inlet, setting an ambient wind speed inlet, setting a ground non-slip wall surface condition and setting boundary conditions of the top and the side of a calculation domain, and obtaining a computational fluid dynamics numerical simulation model of moving type downburst.
3. The regression random forest based method for predicting the average wind speed of the downstorm flow according to claim 1, wherein the step S2 specifically comprises: s201, selecting an ambient wind speed Storm centre movement speed As key control parameters and respectively determining the variation intervals of the ambient wind speed Variation interval of storm center moving speed ; S202, selecting two mutually different prime numbers as the cardinalities of the environment wind speed dimension and the storm center moving speed dimension respectively, calculating Ha Erdu sequence values of each sample point in two dimensions by utilizing Ha Erdu sequence generation algorithm, and generating N uniformly distributed dimensionless sample points; S203, mapping the dimensionless sample points back to a change interval of the ambient wind speed and a change interval of the storm center moving speed by using an inverse normalization method, and calculating to obtain corresponding actual physical parameter values; S204, taking each group of calculated actual physical parameter values as an independent working condition sample, and summarizing to generate a parameter combination sample set for driving calculation fluid dynamics numerical simulation.
4. The method for predicting the average wind speed of a downstorm flow based on a regression random forest according to claim 3, wherein in S202, the calculation process of Ha Erdu sequence values is as follows: For the nth sample point, firstly, the index value n is digitally spread under the base b: ; then calculate Ha Erdu sequence value of the sample point under radix b : ; Wherein, the Number of k bits under radix b for n M is the highest digit, the selected base number of the dimension of the ambient wind speed is 2, and the corresponding Ha Erdu sequence value is The storm center moving speed dimension is selected to be 3, and the corresponding Ha Erdu sequence value is ; In S203, the calculation formula of the actual physical parameter value is: ; ; Wherein, the For the ambient wind speed corresponding to the nth sample, For the storm center movement speed corresponding to the nth sample, And Respectively minimum and maximum values of ambient wind speed, And Respectively, a minimum value and a maximum value of the storm center movement speed.
5. The regression random forest based down-storm leveling wind speed prediction method according to claim 3, wherein the S3 specifically comprises: S301, sequentially taking a parameter combination sample set as an input condition, respectively assigning motion attributes of an environmental wind speed inlet boundary and a moving area of a computational fluid dynamics model, and starting a solver to perform unsteady state computation until a preset simulation total duration is reached; S302, presetting monitoring points in a calculation domain, recording speed vectors at the monitoring points in real time in the simulation process of each group of working conditions, extracting horizontal radial components of the speed vectors as downstorm flow average wind speed, and outputting a time course curve of the wind speed of the monitoring points along with time change under each group of working conditions, wherein the monitoring points are positioned on the axis of a storm moving path, and the downstorm flow average wind speed is the radial speed in a downstorm flow wind field Curve of change over time t ; S303, discretizing the wind speed time course curve, discretizing continuous time into a plurality of time points, extracting an instantaneous wind speed value corresponding to each time point, correlating each time point with an environment wind speed and storm center moving speed corresponding to simulation, constructing an input feature vector, and taking the corresponding instantaneous wind speed value as a target output scalar; s304, traversing all working conditions and corresponding time points thereof, summarizing all generated input feature vectors and target output scalar quantities, and constructing a training data set containing a plurality of groups of input feature variables and target output variables.
6. The regression random forest based down-storm flow average wind speed prediction method of claim 5, wherein in S303, the input feature vector is: ; The target output scalar is: ; Wherein, the Is the input characteristic vector of the kth time point under the nth set of working conditions, For the ambient wind speed corresponding to the first set of operating conditions, For the storm center moving speed corresponding to the nth set of working conditions, Is the kth discrete point in time; a scalar is output for the target at the corresponding point in time, Radial wind speed value for downburst at the kth discrete time point.
7. The regression random forest based method for predicting the average wind speed of the downstorm flow according to claim 1, wherein the step S4 specifically comprises: S401, based on an ensemble learning theory, adopting a regression random forest algorithm as an infrastructure of a prediction model, and adopting a Bootstrap sampling method to generate a plurality of sub-data sets from the training data set, wherein the sub-data sets are respectively used for training each decision tree; S402, adopting the number of decision trees And maximum depth of decision tree As key super parameters, respectively setting a value range of the number of the decision trees and a value range of the maximum depth of the decision trees, and constructing a two-dimensional parameter grid; S403, traversing the two-dimensional parameter grids by adopting a parameter sensitivity analysis method, constructing and training a corresponding regression random forest model by utilizing the training data set for each group of parameter combinations in the grids, and calculating a decision coefficient under the parameter combinations As an evaluation index of the prediction accuracy; S404, selecting a decision coefficient And when the maximum value is reached, the corresponding parameter combination is used as an optimal super parameter combination, the regression random forest model is retrained by utilizing the optimal super parameter combination and the training data set, and the trained regression random forest model is used as a down-storm-leveling average wind speed prediction model.
8. The regression random forest based method for predicting the average wind speed of the downstorm flow according to claim 1, wherein the step S5 specifically comprises: S501, obtaining the environmental wind speed to be predicted of the working condition to be predicted And the storm centre movement speed to be predicted And determining the time sequence to be predicted ; S502, respectively combining the environmental wind speed to be predicted and the storm center moving speed to be predicted with each time point in the time sequence to construct an input feature matrix of the working condition to be predicted, wherein the formula of the input feature matrix is as follows: ; Wherein, the As an input feature matrix for the condition to be predicted, For the ambient wind speed to be predicted, For the storm centre movement speed to be predicted, P is the total number of prediction time steps, For the time series Each time point of (a); s503, inputting the input feature matrix into the down-storm flow average wind speed prediction model, calculating the input feature matrix according to the mapping rule learned in the model, and outputting a down-storm flow average wind speed prediction value at a corresponding moment; s504, arranging a plurality of predicted values output by the down-storm-level average wind speed prediction model according to the sequence of the time sequence, and reconstructing to obtain a down-storm-level average wind speed time course curve under the working condition to be predicted.
9. A regression random forest based down-storm leveling and averaging wind speed prediction system, based on the regression random forest based down-storm leveling and averaging wind speed prediction method as claimed in any one of claims 1-8, characterized in that the system comprises: the building module is used for building a computational fluid dynamics numerical simulation model of the movable downward-impact storm flow based on the impact jet theory; The generation module is used for determining the change range of the movement speed of the ambient wind speed and the storm center so as to construct a two-dimensional parameter space, and uniformly sampling in the two-dimensional parameter space by using a Harton sequence sampling method to generate parameter combination samples containing different ambient wind speeds and different movement speeds of the storm center; The calculation module is used for inputting the parameter combination sample into the computational fluid dynamics numerical simulation model for calculation, and constructing a training data set comprising a plurality of groups of input characteristic variables and target output variables according to wind speed time interval data of the calculated monitoring points, wherein the input characteristic variables comprise ambient wind speed, storm center moving speed and time points, and the target output variables are downstriking storm flow average wind speeds at corresponding time points; the training module is used for constructing an initial regression random forest model, training by utilizing the training data set, optimizing the number of the decision trees and the maximum depth of the decision trees by adopting a parameter sensitivity analysis method in the training process to obtain an optimal super-parameter combination, and obtaining a trained regression random forest model as a down-storm flow average wind speed prediction model; the prediction module is used for acquiring the environmental wind speed, the storm center moving speed and the time sequence to be predicted of the working condition to be predicted, inputting the down-storm leveling and averaging wind speed prediction model, and obtaining a down-storm leveling and averaging wind speed prediction result of the corresponding time sequence under the working condition.
10. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1-8 when the computer program is executed.

Description

Method and system for predicting downburst leveling and average wind speed based on regression random forest Technical Field The invention relates to the technical field of wind engineering and artificial intelligence, in particular to a method and a system for predicting downhill storm leveling and average wind speed based on regression random forests. Background Downburst (Downburst) is a strong wind weather phenomenon formed by strong sinking airflow at the bottom of thunderstorm cloud impacting the ground and then diffusing to the periphery. The method has the characteristics of strong burst, short duration and extremely high destructive power, and becomes a great threat to the safety of engineering structures such as power transmission lines, large-span bridges, high-rise buildings and the like. In particular, the moving speed of the storm center is superposed, so that the structure of the wind field near the ground is more complex, and the impact on the structure often exceeds that of the static type under-storm flow, so that the accurate prediction of the wind speed time course characteristic of the moving type under-storm flow has important significance for wind resistance design and disaster prevention and reduction of engineering structures. Currently, simulation and prediction of a downburst wind field mainly adopt methods such as field actual measurement, wind tunnel test and Computational Fluid Dynamics (CFD) numerical simulation. Although the field actual measurement is the truest, the probability of capturing a complete downburst event is extremely low, the data is extremely deficient, the wind tunnel test and the CFD numerical simulation are long in period and high in cost and are difficult to finely simulate the storm moving effect although the wind field characteristics of downburst can be reproduced, and the CFD numerical simulation is high in accuracy and is often required to consume huge calculation resources and time cost. For example, for unsteady simulation of mobile downburst, single calculation often takes hours or even days, and it is difficult to meet the requirements of rapid evaluation or real-time early warning of massive working conditions in actual engineering. Over the years, with the rapid development of artificial intelligence technology, machine learning algorithms have shown great potential in the field of wind speed prediction. However, most of the existing wind speed prediction models are relatively less researched for the conventional atmospheric boundary layer wind field or steady-state wind field, and are used for the extreme wind field prediction of the mobile downburst type strong nonlinear and non-stationary characteristics. On the premise of ensuring the prediction precision, the calculation cost is greatly reduced, and the rapid prediction of the downburst wind speed time course under the combined working condition of any moving speed and ambient wind speed is realized, so that the method is a technical problem to be solved currently urgently. Disclosure of Invention The invention provides a method and a system for predicting the downburst flow average wind speed based on a regression random forest, which can replace CFD numerical simulation with high time consumption, improve the calculation efficiency by tens of millions of times on the premise of ensuring the prediction precision, and realize real-time and accurate prediction of the movable downburst flow wind speed time course under any working condition. The invention provides a method for predicting downburst leveling and average wind speed based on regression random forests, which is characterized by comprising the following steps: s1, based on an impact jet theory, establishing a computational fluid dynamics numerical simulation model of mobile down-stroke storm flow; S2, determining the change range of the movement speed of the ambient wind speed and the storm center so as to construct a two-dimensional parameter space, and uniformly sampling in the two-dimensional parameter space by using a Harton sequence sampling method to generate parameter combination samples containing different ambient wind speeds and different movement speeds of the storm center; S3, inputting the parameter combination sample into the computational fluid dynamics numerical simulation model for calculation, and constructing a training data set comprising a plurality of groups of input characteristic variables and target output variables according to wind speed time interval data of the calculated monitoring points, wherein the input characteristic variables comprise ambient wind speed, storm center moving speed and time points, and the target output variables are downstriking storm flow average wind speeds at corresponding time points; S4, constructing an initial regression random forest model, training by using the training data set, and optimizing the number of decision trees and the maximum depth of the decision trees by ado