You are on page 1of 8

Journal of Applied Science and Engineering, Vol. 20, No. 1, pp. 55-62 (2017) DOI: 10.6180/jase.2017.20.1.

07

Formulation of Estimation Models for Wind Force
Coefficients of Rectangular Shaped Buildings
Jenmu Wang* and Chii-Ming Cheng
Department of Civil Engineering, Tamkang University,
Tamsui, Taiwan 251, R.O.C.

Abstract
In wind-resistant design of structures, the calculation of wind coefficients is usually based on
data from wind tunnel tests. The process is very time-consuming and expensive. In order to formulate
a model to estimate wind force coefficients of rectangular buildings, various methods including
regression analysis and artificial neural networks (ANNs) were investigated. This paper focuses on
the presentation of the various approaches with emphasis on the detailed result comparisons and
discussions of models developed for alongwind, acrosswind and tortional wind coefficient
predictions.

Key Words: Wind Force Coefficients, Regression, Artificial Neural Networks, Aerodynamic
Database

1. Introduction (Cd) was used as the target of wind coefficient predic-
tion. According to our findings, RBFNN is the most ef-
Artificial neural network (ANN) is an approach to fective method among the five candidates, namely poly-
simulate or predict the results of complex domain by nomial regression, nonlinear regression, BPNN (back
using similar (but highly simplified) models of the bio- propagation neural network), RBFNN (radial basis func-
logical structures found in human brain. Training ANNs tion neural network) and GRNN (general regression
with existing cases with reasonable answers can deduct neural network). The final goal is to trained RBF neural
multi-variable nonlinear models to predict the results of networks to predict alongwind, acrosswind and torsional
similar problems. ANNs are good candidates to be in- wind coefficients all together. Consequently, the trained
corporated in our aero-data based wind resistant design networks can be embedded in a building design wind load
approach [1,2] to perform data mining and knowledge expert system as presented in [2,3] to estimate wind force
discovery from our aerodynamic database. coefficients and calculate wind loads. This paper focuses
It is noted that the purpose of the work reported here on the presentation of the various approaches with em-
is not to replace wind tunnel experiments. The objective phasis on the detailed result comparisons and discussions
is to utilize wind tunnel experiment aerodynamic data- of the final formulation developed for alongwind, acros-
bases to develop an expert system for generating design swind and tortional wind coefficient predictions.
wind loads of tall buildings. It can be used at the pre-
liminary design stage to get reasonably accurate design 2. Scope and Data Processing
wind loads without performing costly and time-consum-
ing wind tunnel tests. The design wind load aerodynamic database de-
At first, alongwind mean coefficient of base shear veloped at the Wind Engineering Research Center of
Tamkang University in Taiwan (WERC-TKU) [4,5] con-
*Corresponding author. E-mail: wang@mail.tku.edu.tw sists of four parts: (i) isolated tall buildings, (ii) interfer-

acrosswind and tortional wind coefficients respec- tabase [5.25. Initial Investigation rections of earlier models were measured using HFFB. root mean square errors (RMSE) definitions and setting in wind load calculations can be and the maximum errors of the applications in Table 3 found in [10]. The data used include BP (back propagation).0. Data grouping methods for Cd Name Data grouping Application Aspect ratio series 3 terrains and 9 side ratios to form 27 sets Polynomial regression Side ratio series 3 terrains and 5 aspect ratios to form 15 sets Polynomial regression Terrain series 3 terrains to form 3 sets Nonlinear regression & neural networks No grouping All data in a set Nonlinear regression & neural networks .33. (iii) other and aspect ratio (H/B) as inputs. wind. The final formulation at Wind Engineering Research Center of Tamkang Uni. RBF (radial basis used for this research is from the aerodynamic database function) and GR (general regression) neural networks. 0. ANNs were used as well to train. The coefficients investi. cator for selection of the final estimation method. were studied as well.. The neural networks buildings and structures and (iv) CFD based. side ratio (D/B) Alongwind RMS coefficient of base shear Cdd Acrosswind RMS coefficient of base shear Cld Table 1. namely polynomial regression and nonlinear Table 2.0. 3. trained three RBF neural networks to estimate along- versity (WERC-TKU) to construct an aerodynamic da.0. Their The fitting results. for isolated tall buildings. Total of 150-plus building shapes were tively. 5. 4. 4. In Wind coefficient description Abbreviation addition.6]. 6. 0. to investigate the prediction of wind force coefficients.0 moment Aspect ratio (H/ BD ) 3. two regression analysis methods. 1.32. C (a = 0.2.25.0.15) moment Side ratio (D/B) 0. 0. alongwind tronic pressure scanning system. data grouping strategies. Wind tunnel test data selected Alongwind mean coefficient of base Cdm Model cross-section Square and rectangular moment Alongwind RMS coefficient of base Cdmd Terrain exposure A. studied. 3. A total of 135 wind tunnel experiment data sets as de- scribed in Table 1 were used. Two dict wind coefficients and spectra for buildings such as regression methods and three neural network methods [7-9] etc. this research selected pressure measurements were used for the forecast of Cd. At the preliminary stage of this research. B. 5. In order to yield better of several models in the WERC aerodynamic database results.5. Acrosswind RMS coefficient of base Clmd 3. According to the results of the investigation presented The wind tunnel measurements and their analysis of in this paper. simulate and Alongwind mean coefficient of base shear Cd forecast wind coefficients using terrain. 2. were used to compare the results at first. 0. mean coefficient of base shear Cd was used as an indi- Sharing the same goal of similar researches that pre. are summarized in Tables 4 to 6. 0. The wind force coefficients and reduced force spectra in the alongwind. performed using MATLAB’s build-in regression analysis cients of rectangular buildings. RBF neural network is the most effective various generic building shapes have been performed mean to predict wind coefficients. All the analyses were To formulate a model to estimate wind force coeffi. Nomenclature of wind force coefficients regression.1 Regression Analysis gated and their abbreviations are listed in Table 2. acrosswind and torsional di. whereas later were measured with multi-channel elec. 7 RMS coefficient of base torsion Ctd Table 3. as described in Table 3.56 Jenmu Wang and Chii-Ming Cheng ence effects of two adjacent tall buildings.

0095 0.4424 2.0061 0.0508 + 0.1123 3 -0.0424 0.4180 Polynomial regression D/B series RMSE Max.0309 ´ (D/B)-2. error (%) Cd = p1 (D/B) + p2 0.4218 2.7639 ´ (H/B) -0.0.1985 B 3. error (%) Cd = p1 (H/B) + p2 0.16.0109 0.0678 C 5 -0. fitted.9469 ´ (D/B)-0.4195 A 1.5389 0.0186 -0.0475 2.20 -0.0045 0.4170 0.5744 ´ (H/B)-0.4252 2. Nonlinear regression (terrain series.1.1639 B 4 -0.1504 6 -0.4424 2. Nonlinear regression (no grouping.3899 118.098 19. Polynomial regression (needs 27 equations for the H/B series and 15 equations for the D/B series) Polynomial regression H/B series RMSE Max.50 -0. error (%) Cd = -15.4377 2.1149 + 63.54 Table 6.0339 .6457 0.00 0.091 12.1238 0.33 -0.4267 2.1419 + (D/B)-0.0419 0.33 0.131 16.9867 4.50 0.0071 0.7 Terrain Aspect ratio p1 p2 Terrain Aspect ratio p1 p2 3 -0.1622 7 -0.25 0.1879 0.9775 5.46 C Cd = -0.0834 0. 15 equations with different p1 and p2 parameters were in the equations is actually H / BD .17 ´ (a) -0.9185 4.9610 6 -0.1807 Table 5.4031 + (D/B)0. 0.00 0.0118 1. The maximum errors exceed 100% in some cases.1890 2.8512 .22.4290 2.9652 ´ (a) ´ (H/B)0.5133 1.0058 2.0029 0.0234 2.00 -0.8133 ´ (H/B)0.4910 B 0.6082 ´ (H/B)-1.0186 1.0209 3.00 0.00 0.50 -0.6962 ´ (D/B)0. which are not usable.0176 2.3916 2.0799 + 0.0380 1.4236 0.0527 ´ (D/B)-2.0216 2.0099 0.00 0.7635 4. The 1st equation in Table 4.00 -0.7719 0. which Table 4.65 Cd ´ (D/B) .0669 7 -0.0755 0.0334 2.4384 2.1697 5 -0.5221 0. one equation for each terrain) Max error Terrain Nonlinear regression RMSE (%) A Cd = 0.2590 3.7995 2. symboled as H/B.4336 2.1892 A 5 -0. For the side ratio series (the 2nd equation in Table 4).4320 C 1.541 ´ (H/B)0. Note that the aspect ratio.92 B Cd = -62.00 -0.00 0.094 12.4448 2.33 -0.5504 0. aspect and side ratios) Nonlinear regression RMSE Max.0.00 -0.4330 2.79 Terrain Side ratio p1 p2 Terrain Side ratio p1 p2 0.25 -0.00 0.0875 6 -0.00 -0.3295 ´ (D/B) -0.6197 3.1370 3 -0.20 0.8350 5.0824 4 -0.0.9538 ´ (a) + 57.4401 2.7937 0.0195 .00 0.9875 0.4462 2.3987 1.4857 0.204 0.3695 5.0039 ´ (D/B) ´ (H/B) .00 -0.9368 .00 0.25 -0.0163 + 0.0597 1.0458 1.0173 0. Formulation of Estimation Models for Wind Force Coefficients of Rectangular Shaped Buildings 57 functions. one equation for different terrains.0215 2.5736 .1586 7 -0.0003 0.0151 0.0191 0.20 -0.4467 2.1939 B 4 -0.0003.

760 9. is added for the networks in Table 8.097 11. error (%) Terrain RMSE what.089 8. Errors of neural network estimations (no cases and cases of aspect ratio 4 and 6 were used to vali- grouping. which solve the problem of polynomial re- B 0.3138 rors tended to increase.9680 8.105~1.0974 14.535 13. error (%) was increased gradually. one network for each terrain) Cd with different aspect ratios. spread of radial basis Radial basis function neural networks (RBFNN) functions (SPREAD) for both RBFNN and GRNN were GOAL = 0. which is 3/5 Back propagation neural networks (BPNN) Neuron center (NC) = 3 of the total 135 cases collected.660 gression. 0.121 terrain. SPREAD = 1.053 5. For the RBFNNs.063 12. Different num. SPREAD = 1. The best results usually oc. and the series of equa. adding terrain exposure C 0. For the BPNNs. Max. one network for all terrain) date the trained ANNs. Errors of neural network estimations (terrain needs 27 equations with different p1 and p2 to forecast series. As shown in Table 6.771 3. 5 and 7 were used as training Table 8. Training Validation A 0. Training Validation bers of neuron centers (NC) for BPNN.519 10. That is.1 better accuracy.2302 error goal (GOAL) for RBFNN.484 5.445 12. RBF (radial basis General regression neural networks (GRNN) function) and GR (general regression) to forecast Cd.727 Three different neural network architectures were C 0. followed by adjusting the SPREAD to achieve SPREAD = 0. All the ANN C 0. a total of 81 wind tunnel experimental cases were used for training.7313 The errors of the three different ANNs are all ac- .836 (a) to the independent variables further reduces the equ- Radial basis function neural networks (RBFNN) ation to one and retains about the same level of preci- GOAL = 0. the GOAL was General regression neural networks (GRNN) set first. error (%) The three nonlinear regression equations in Table 5 Terrain RMSE Training Validation use aspect and side ratio as independent variables to A 0.0648 12.348 estimate Cd.058 6.0464 6.053 6.556 6. 0. The inputs of the networks in A 0.878 4. Cases of aspect ratio 3.042 4. The RMSE is about the same and the maximum er- ror. namely BP (back propagation).58 Jenmu Wang and Chii-Ming Cheng is the aspect ratio series. Back propagation neural networks (BPNN) tions cannot be used for cases that are not covered by the Neuron center (NC) = 2 original 9 side ratios. It sometimes was necessary to change Max. However.2.2 sion. which occurred at terrain A this time. Max.096 14.038 6. the number of center points Max.1009 12. RMSE curred when the neuron center was set to two or three. Training Validation When the center points were over five the validation er. mean squared 0.5415 12. error (%) RMSE networks in order to find the best results. and the results are Terrain RMSE Training Validation shown in Tables 7 and 8. error (%) MATLAB’s neural network toolbox.625 Table 7 are aspect and side ratio and an additional input. The final settings RMSE Training Validation are listed in Tables 7 and 8. error (%) the GOAL to reach a better estimation. rises some.1. it Table 7.399 output is Cd.120 5.1 All the training and testing were performed using Max.2 ANN Wind Coefficient Predictions B 0. Various parameters were adjusted when training the Max.476 used.2 investigated. is very accurate. B 0. SPREAD = 0.382 12.

726 Cdm 6. the Cdd 9. However. The root mean square clusive that the ANN models have clear advantage over errors and the absolute maximum errors of the RBFNN nonlinear regression when only one equation or one net.0067 Cdmd 9.136 0.100 1.550 275.13 0.351 of using four single-output RBFNNs.350 6. However. the errors of the one-net.20 0. the results of the validation the results of BPNN are comparable to nonlinear regres. finally.002 7. error (%) Wind GOAL/ Max.583 5.83 errors are even less than the dedicated networks in Table 0. RBFNN has the the RMS coefficient. Also. Errors of single RBFNN for alongwind RMS At first. The inputs are still terrain. dict all four alongwind coefficients. the fol- previous sections. neural network architectures with multiple outputs were studied. On the RBFNN three-network terrain series. RBFNN was selected as the simula.178 .200 1.908 8. which is good for practical application.12 0. Alongwind coefficient error summary using 4 Table 12. the 4 wind coefficients were divided into 2 coefficients categories. Table 11. Especially. but comparing Table 9 with 12.001 1.314 Cd 6. lowing remarks can be made: (1) The RMSEs and maxi- tion model for all four alongwind force coefficients in mum errors decrease when using one network to predict Table 2. works extremely well.0218 over 55% for both training and validation cases. it is con. The mean coefficients Wind Max.525 0.360 8. However. ment for Cdd. the extension of the prediction scope tends to ANN methods in Table 8. we trained one RBF neural network to pre- Comparing the results in Tables 6 with 8. error (%) aspect ratio and side ratio.031 59. The assumption here is that all the alongwind four alongwind coefficients compared with using four coefficients exhibit similar changing trend with respect to side ratio.043 7. the RBFNN model in Table 8 network to one.907 Cdmd 0.14 0. Errors of RBFNN for alongwind coefficients single-output RBFNNs using a single network Wind Max.968 8. Based on the investigation of prediction of Cd in the Analyzing the RBFNN simulation results. Predictions for Multiple Alongwind original assumption.001 1 0. the work RBFNN are just slightly higher compared with the RMSEs for Cd and Cdm are not that different.941 4. cases are extremely poor.046 6. The final result also proved that the 4. The results are listed in Tables GOAL SPREAD RMSE coef Training Validation 10 and 11.563 31. sion for the terrain series. Errors of single RBFNN for alongwind mean coefficients for the four coefficients. which the errors are coefficient Training Validation all below 9%.1 1.00 Cdmd 7.100 8. tions.1/1. which is that the variation trends of Coefficients all 4 alongwind coefficients are similar therefore RBFNN can be applied similarly. For the mean coefficient Cd and Cdm. drop accuracy.65 Cdd 7. This reduced the number of work is used. error (%) GOAL SPREAD RMSE RMSE coefficient Training Validation coefficient SPREAD Training Validation Cd 0. The errors of the training cases are all below 10% for Table 9.670 Cdd 0. RBFNN and GRNN After extensive experiments of different combina- are clearly better. a large improve- best results in both tables. Wind Max.76 55. Formulation of Estimation Models for Wind Force Coefficients of Rectangular Shaped Buildings 59 ceptable as indicated in Tables 7 and 8. especially for Cdmd. the training and validation re- sults are summarized in Table 9. Referring to Tables 7 and 5. It is better than the other two Usually. the maximum errors for Cdd and Cdmd im- prove dramatically.031 7.415 10. mean and RMS.467 0. are summarized in Table 12. error (%) GOAL SPREAD RMSE Cd and Cdm had the best results.6 9.0174 Cdm 0.329 4. the maximum error of Cdd is Cd 5.84 0. Instead Cdm 4. aspect ratio and terrain. Cdd and Cdmd. Using four RBFNNs Table 10. other hand. is sound.

116 Cld 13.4680 1.0221 0.5679 0.6971 0. acrosswind and torsional wind coefficients 1 5 1.0842 2.60 Jenmu Wang and Chii-Ming Cheng single purpose RBFNNs. Table 15. Errors of RBFNN for acrosswind coefficients 6 0. The error 4 0.259 .8814 1.6531 2.5488 2.0408 1.4666 0.308 training or validation cases.4739 2.4670 0. 3 0. Errors of RBFNN prediction for Cd (Terrain A) efficients.041 0.4146 2. 13 and 14. the predictions of the 4 0. The Final Formulation and Verification 3 2.4055 2.4513 0.3515 0. 4 2.3765 0. our final formulation trained three independent RBF neural networks to predict 3 0. Cd and Cdm.9564 0.333 5 2.294 rized in Tables 12.3367 2.3094 2. which may be caused by the in.9524 1.14 0.787 24. 6 2.432 Table 14.2 5 2.6575 5.718 RMSE 5 5 0.234 Table 13.3903 0.2.7323 0. 6 0.8660 1.0035 15.4154 0.251 mum errors of Cld and Clmd were over 15% for either 7 2.3054 2. (2) Among the alongwind co.257 Using 3 individual neural networks for the 3 across.5061 0.5%. Experiment Prediction Error (%) ratio ratio ferent terrain for different wind force coefficients.9974 1.3626 2.1103 1.868 7 0.097 Wind GOAL/ Max.0022 Clmd 14.5 5 1.583 the simulation model for the acrosswind and torsional 3 2.029 4 1. Using 7 0.860 and 0.803 19. the maxi.6972 0.0120 0. Cdd and Cdmd.7327 0.001/1.749 alongwind coefficients in the previous section.923 5.034 coefficient SPREAD Training Validation 6 0.4443 4.5088 2. 0.125 0.4725 0.123 7 2.6070 0.4835 4.8946 0. 0.3451 3.0288 2.3907 0.1567 6.670 tions at terrain B and C are better than terrain A.0972 0.4406 0.6253 2.9926 0. the estima. (3) The worst case scenario happens at dif.744 mean coefficients. 0.2955 2. 0. On the whole.592 lute maximum errors of the three RBFNNs are summa.034 The worse errors tend to happen with side ratio 0.4792 1. 3 0.091 6 0. Therefore. error (%) RMSE 3 0.703 7 1.050 For alongwind coefficients.0438 0.3700 0.6974 0.052 7 0.3727 0.5064 0.0050 1.5041 0. 4 2. Errors of RBFNN for torsional coefficient 3 0.066 Based on the investigation of prediction of Cd and 4 4 2.4106 3.870 18.364 6 2. error (%) 4 0.086 3 5 0.6570 0.699 0. the two acrosswind 3 1.9836 3.1014 0. (4) 3 2.078 wind force coefficients in Table 2.9129 4.755 7 0.3111 6.0133 1. RBFNN 6 2.007 Ctd 0. 6 1.6548 0.594 This improved accuracy.1326 2.0125 1.4633 0.1343 0.4730 0.6975 0.089 coefficient SPREAD Training Validation 4 0.25 5 2.6514 0. 6 0.055 the average prediction errors are around 1.706 4 5 0. are better than the RMS 2 5 0.3471 1.202 crease of training cases.4912 0.4901 0.4425 0.001/1.051 respectively.432 Wind GOAL/ Max.010 coefficients.25 4 2.820 To summarize the results.722 wind and torsional wind force coefficients. The root mean square errors and the abso.2970 0.8980 1.3558 0.1167 2.9006 1.033 coefficients were grouped together using one RBFNN.912 was selected again because of its previous successes as 7 2.0919 2.3065 0.8974 0.4293 4.7022 0.764 alongwind.4507 0. Cd and Cdm can be better estimated than Cdd Side Aspect and Cdmd.3766 0.347 Cd and Cdm (as shown in Tables 15 and 16) as examples.333 building shapes.4511 0. 7 1.9929 0.

822 for all three terrains.0594 0.2645 1.0650 12.0754 10.2 5 1.2281 0.5513 0.0965 18.25 7 0.468 Aspect ratio Experiment Prediction Error (%) 3 0.333.1090 13.3857 0. 3 0.049 ratio 0.25 5 1. and terrain A is better than terrain B and C.523 6 0. Worst case scenario of Cld RBFNN predictions 2 5 0. including polynomial regression.2861 0.4247 0. Ctd RBFNN predictions 4 5 0.3998 0.0342 0.870 4 0.2829 1.827 4 0.015 4 1.0546 11.1048 14.529 Table 17.0344 1. Formulation of Estimation Models for Wind Force Coefficients of Rectangular Shaped Buildings 61 Table 16.023 Further study demonstrated that instead of using 4 0.0490 0.446 der study. 7 1.5 5 1.037 6 0.965 5 5 0.2399 1.25 Experiment Prediction Error (%) and 0. The worst scenario for Cld.3856 4.153 Ctd for terrain B and side ratio 0.2546 1.2230 0.2822 1. Table 18 shows the prediction results of Ctd for 0.2363 1. 4 1. the errors are small when the side ratio is greater 3 1.2088 1.4190 0.2594 1.5720 0.2784 0.5967 0.2128 0.1532 0.5719 0.25 and terrain B. and the overall max- 0.971 for the estimation of all the seven wind coefficients un- 7 0. 6 1.5518 3.4136 0.2566 4. In ge- neral. and side ratio 0.374 7 0.2927 0.118 7 0.0877 20.480 For torsional wind coefficients.2917 1.203 6.2726 0.496 7 1.0418 1.320 imal prediction error is over 13%.0787 0.0761 1.032 the best results in terms of accuracy and usefulness.0408 0.3840 0.1446 1.4055 0.2923 0.158 on this finding.907 Cld for terrain B and side ratio 0.0494 0.015 3 0.078 5 0.527 Table 18. is presented in Table 17.4068 1.599 7 0. the same RBFNN architecture was applied 1 5 0.2727 0.1542 1.0670 12.5668 0.0912 0.2815 3.2636 0. which holds 7 1.242 nonlinear regression and ANN.1523 2.047 For acrosswind coefficients.2665 0.4254 1.3156 2.1263 1.0725 0.0545 10.3932 1.101 4 1.1270 1.4059 1. terrain B has the worst 4 1.2554 0.3493 1. Conclusions 3 1. Errors of RBFNN prediction for Cdm (Terrain A) distribution is similar for Cd and Cdm.4049 0.449 Aspect ratio Experiment Prediction Error (%) 3 0.0878 1.2677 0.2920 0.73%.729 6 0. the maximal error is 0.341 .482 below 4% when side ratio is greater than 1.474 studied for the prediction of wind force coefficients.5940 0. However.046 3 0.630 than 2.4191 0.25 at Terrain B.5698 0.1610 1.2763 0.2198 0.0467 14.899 Our investigation showed that using RBFNN yielded 3 0.2112 0.329 cients.673 side ratio 0.2079 2.469 Several methods.2625 3.0501 3.5780 1.269 7 0. The worst errors Side Aspect all occur at cases with aspect ratio 7.377 results with average error of 3. side 3 1.0579 0.1007 1.0397 1.293 6 0. 0.5511 0. have been carefully 6 1. Based 4 0.2666 0.884 4 0.012 6 0.1690 6.2189 0.369 4 0.4034 0.3397 1.2228 0. ratio ratio 3 1.2549 6.2749 2.2280 0.2855 0. the error is 6 1.248 7 1.2777 0.428 3 0.048 4 0.213 5 0.734 over 24% impacted by the small values of the coeffi- 6 1.706 3 5 0.0200 1.333 5 1.1554 0.785 6 0.0266 1.2623 0.0813 0.0681 0.2920 0.0957 0.25 7 0.021 6 0.4050 0.2721 0.

the fol. M. Cairns. T. June 3-4. Paper ings. 723- [2] Wang.. dam (1999). Y. No. Vol. “Aero-Data Based Wind Expert System Based on Aerodynamic Databases.” Proceedings 381-1 of Fifth International Symposium on Computational [9] Krian. M. C. C.. Sep. J.” Journal of Wind & Engineering. Acknowledgements [7] Bitsuamlak. Grisogono... H. 5. Gašparac. ference Preprints.. Kozmar.. 359-366 (2007). C. Korea.A. J. Engineering and Industrial Aerodynamics. Antoniƒ. J. is better than acrosswind.” Wind ROC and Wind Engineering Research Center at Tam. H. No. [8] Chen. J.. Wang. 1. and torsional. 423-441 (2003). Australia. (3) Interference Effects of Adjacent Tall Buildings.” Journal of Wind [1] Wang. and Structural Engineering (ICICSE’15).. “The Aerodynamic Database for the Cdm can be better estimated than Cdd and Cdmd. July 1-6. and Chang. Cdd. Lin.62 Jenmu Wang and Chii-Ming Cheng seven RBFNNs for the seven wind coefficients. M.. 91.. C. and Surry. 12th International Conference on Analyzing the RBFNN simulation results. Y.. 1-6. M. J. C.. H. Engineering into the 21st Century. Cdm and Cdmd). J.347 and Applied Climatology. 10-15. [3] Wang.1016/S0167-6105(02)00 Rectangular Cross-Section Buildings. Rotter- kang University Taiwan are gratefully acknowledged. J.” Pro- ceedings of the 6th Asia-Pacific Conference on Wind Manuscript Received: Apr. 14. O.” Theoretical 35. and Cheng. another for the Chang. 12th International Conference on are greater than 1. North Ca. Y. and acrosswind is better than [5] Cheng. Jul. Jul. Wang. Vol. Wang. Y. M. Y. Turkey (2015). P.5359/jawe.S. 359-366 (2007).” Resistant Design of Rectangular Shaped Tall Build- ICWE 13. 329-339 (2005). Wind Engineering. (2) Among the alongwind coefficients. Sensitive Buildings and Structures. “Web-Enabled Design Istanbul.. M. The aforementioned formulation and ANN program [6] Cheng. C.” Con- Acrosswind predictions are better when the side ratios ference Preprints. A. Balkema. “The Aerodynamic Database for the In- acrosswind coefficients (Cld and Clmd) and the other for terference Effects of Adjacent Tall Buildings. “Prediction of References Pressure Coefficients on Roofs of Low Buildings Using Artificial Neural Networks. May 23-27 (2010).. Seoul. Vol. lowing remarks can be made: (1) Alongwind prediction pp. 2. and Cheng. Vol. “The Role of Artificial 736 (2015). N. Wu. 3. ficial Neural Networks to Predict Wind Spectra for pp. 120. Cairns. and Wind Engineering (CWE2010). ing three RBF neural networks.. “Designing Laboratory Wind Simula- rolina. pp. 1. C. pp.” International Conference on Innovations in Civil #191 (2011). and Cheng. tions Using Artificialneural Networks. 2016 Engineering (APCWE VI). J. 20. doi: 10. “Application of Cascade-correlation Learning Network Fordetermina- Financial supports for this research from MOST of tion of Wind Pressure Distribution in Buildings. and Godbole. B. doi: 10. Australia.” Con- the tortional coefficient Ctd is adequate. Chapel Hill. Kopp. C. 12~14. 2016 . train. one for the alongwind [4] Cheng. Accepted: Oct. C. J. Lin. C. Wu. (4) Torsional predictions are better Wind Engineering. pp.. Cd and Chang.. G. and Cheng. U. “The Application of Arti. J. C. C. 50-59 (2008). “e-wind: an have been used in later developed programs for wind Integrated Engineering Solution Package for Wind load calculations with decent results [10]. Vol. Wind Load Expert System for Tall Buildings. D.. doi: 10. pp.. G.. G.1007/s00704-014-1201-4 Neural Networks in a Building Design Wind Load [10] Wang. H. Netherlands. Amsterdam. and coefficients (Cd. when the side ratios are greater than 2.. M.