| The mirror of condenser is a parabolic structure,which leads to the deformation or even damage of the entire mirror surface under wind load,thus affecting the thermal efficiency of the condenser system.At present,the research on wind load of parabolic condenser mainly focuses on flat ground.However,the research on wind load of parabolic condenser on roof is less.Due to the difference between the wind field environment of the roof and the ground,the influence of wind load on the parabolic condenser placed on the roof will be different.Therefore,it is of great engineering significance to study the wind resistance of parabolic condenser on the roof of multi-storey buildings.In addition,because the wind tunnel experiment is limited by the experimental cost,the selection of experimental conditions is often not perfect,which leads to the incomplete disclosure of the rules behind the experimental phenomena.At the same time,the data missing or abnormal caused by small defects such as pressure tube blockage and sensor failure in the experimental process will also affect the reliability of the data.To solve these problems,it often takes a lot of time to solve them,and the use of supervised learning in machine learning technology to improve the wind tunnel experimental data can be more efficient and scientific to solve the above problems.At the same time,the prediction of machine learning model for the test results of unknown conditions can also help to verify the reliability of wind tunnel test results.Therefore,it is of great significance to improve the wind tunnel experimental data by using machine learning technology.In this paper,the wind tunnel test method is used to measure the pressure of the mirror part of the parabolic condenser model on the flat roof of multi-story building,and the average wind pressure distribution law(including the variation law with the working condition and the maximum wind pressure distribution characteristics),the fluctuating wind pressure distribution law and the extreme wind pressure distribution law are obtained.At the same time,the similarities and differences between the two are analyzed by comparing and discussing the wind pressure distribution characteristics of the mirror surface of the ground parabolic condenser.The results show that the average wind pressure due to the interference effect of the building and the daughter wall is about 30 % smaller than that of the ground,and the fluctuating wind pressure increases by about 40 %.At 30 °,135 °～150 °wind direction angle for wind resistance adverse conditions.According to the distribution characteristics of the maximum wind pressure and the extreme wind pressure,the mirror edge,especially the corner,is most susceptible to wind load.Attention should be paid to the wind-resistant structure design of condenser.The average wind pressure coefficient and fluctuating wind pressure coefficient obtained from wind tunnel experimental data processing are used as the training data set of machine learning model.Regression Tree,Random Forest and XGBoost are used to train the model based on the data set.The machine learning model most suitable for this study is selected through regression evaluation indexes.Finally,the most suitable model is used to predict the average wind pressure coefficient and fluctuating wind pressure coefficient at the typical area of the mirror under unknown conditions,and the predicted results are compared with the existing conclusions of wind tunnel experiments.The results show that the XGBoost model has high reliability in improving the wind tunnel experimental data of the parabolic condenser. |
