AIR POLLUTION – MONITORING MODELLING AND HEALTH
air pollution â monitoring, modelling and health - Ademloos air pollution â monitoring, modelling and health - Ademloos
272 Air Pollution – Monitoring, Modelling and Health Fig. 3. Sub-domains representation, in a two-dimensional projection (x,y), with different emission rates, and black points () represent the sensor position at 10 m height. In order to try some improvement for the inverse solution, the minimum value for the functional (22) is computed using the PSO scheme. This method is a stochastic technique and it deals with a population (candidate inverse solutions). Therefore, the numerical results represent the average from 25 seeds used by the PSO algorithm, working with 12 particles, with the following parameters: w = 0.2, c 1 = 0.1, and c 2 = 0.2. The total error obtained with the PSO was less than inverse solution computed from the quasi-Newton method. It is also important to point out that for the PSO approach no regularization operator was used ( = 0 in Eq. (22)). The inverse problem was also addressed by neural network. The MLP-NN was designed and trained to find the gas flux between the soil and the atmosphere. However, there are two previous steps before the application of MLP-NN. Firstly, it is necessary to determine the appropriated activation function, number of hidden layers for the NN, and the number of neurons for each layer. Secondly, the values of the connection weights (learning process) must be identified. The hyperbolic tangent was used as activation function. Different topologies for MLP-NN were tested with 6 inputs (measured points for 6 different sensors) and 12 outputs (emission for each subdomain). Good results were achieved for two neural networks with 2 hidden layers: NN-1 with 6 and 12 (6:6:12:12), and NN-2 with 15 and 30 (6:15:30:12) neurons for the hidden layers, respectively.
Strategies for Estimation of Gas Mass Flux Rate Between Surface and the Atmosphere 273 Table 3 shows the numerical values computed for the emission rate for each marked cell. Good results were obtained for all methodologies. The cells 12 and 14 had a worst estimation using optimization solution. Inversions with MLP-NN were better, in particular for the emission of the cells 12 and 14. Figures 4 and 5 show the emission estimation for the experiment-1 (5% of noise level) and experiment-2 (10% of noise level), only results obtained with NN-2 (6:15:30:12) are shown (the best results with MLP-NN). Sub-domain Exact Q-N PSO MLP-NN-1 MLP-NN-2 A 2 10 8.97 9.83 10.33 10.11 A 3 10 9.97 10.40 10.11 10.11 A 4 10 12.53 10.79 10.12 10.20 A 7 10 7.99 10.50 10.27 10.20 A 8 10 10.15 12.06 10.24 10.06 A 9 10 11.56 11.28 10.21 10.17 A 12 20 13.85 14.56 21.28 20.97 A 13 20 22.65 22.67 21.32 21.00 A 14 20 14.14 15.85 21.47 20.95 A 17 20 19.99 21.56 21.47 20.99 A 18 20 21.17 20.05 21.38 20.99 A 19 20 24.90 21.74 20.96 20.96 Table 3. Estimation of gas flux emission by the methodologies: quasi-Newton (Q-N), PSO, MLP-NN-1 (6:6:12:12) MLP-NN-2 (6:15:30:12). (a)
- Page 232 and 233: 222 Air Pollution - Monitoring, Mod
- Page 234 and 235: 224 Air Pollution - Monitoring, Mod
- Page 236 and 237: 226 Air Pollution - Monitoring, Mod
- Page 238 and 239: 228 Air Pollution - Monitoring, Mod
- Page 240 and 241: 230 Air Pollution - Monitoring, Mod
- Page 242 and 243: 232 Air Pollution - Monitoring, Mod
- Page 244 and 245: 234 Air Pollution - Monitoring, Mod
- Page 246 and 247: 236 Air Pollution - Monitoring, Mod
- Page 248 and 249: 238 Air Pollution - Monitoring, Mod
- Page 250 and 251: 240 Air Pollution - Monitoring, Mod
- Page 252 and 253: 242 Air Pollution - Monitoring, Mod
- Page 254 and 255: 244 Air Pollution - Monitoring, Mod
- Page 256 and 257: 246 Air Pollution - Monitoring, Mod
- Page 258 and 259: 248 Air Pollution - Monitoring, Mod
- Page 260 and 261: 250 Air Pollution - Monitoring, Mod
- Page 262 and 263: 252 Air Pollution - Monitoring, Mod
- Page 264 and 265: 254 Air Pollution - Monitoring, Mod
- Page 266 and 267: 256 Air Pollution - Monitoring, Mod
- Page 268 and 269: 258 Air Pollution - Monitoring, Mod
- Page 270 and 271: 260 Air Pollution - Monitoring, Mod
- Page 272 and 273: 262 Air Pollution - Monitoring, Mod
- Page 274 and 275: 264 Air Pollution - Monitoring, Mod
- Page 276 and 277: 266 Air Pollution - Monitoring, Mod
- Page 278 and 279: 268 Air Pollution - Monitoring, Mod
- Page 280 and 281: 270 Air Pollution - Monitoring, Mod
- Page 284 and 285: 274 Air Pollution - Monitoring, Mod
- Page 286 and 287: 276 Air Pollution - Monitoring, Mod
- Page 288 and 289: 278 Air Pollution - Monitoring, Mod
- Page 290 and 291: 280 Air Pollution - Monitoring, Mod
- Page 292 and 293: 282 Air Pollution - Monitoring, Mod
- Page 294 and 295: 284 Air Pollution - Monitoring, Mod
- Page 296 and 297: 286 Air Pollution - Monitoring, Mod
- Page 298 and 299: 288 Air Pollution - Monitoring, Mod
- Page 300 and 301: 290 Air Pollution - Monitoring, Mod
- Page 302 and 303: 292 Air Pollution - Monitoring, Mod
- Page 304 and 305: 294 Air Pollution - Monitoring, Mod
- Page 306 and 307: 296 Air Pollution - Monitoring, Mod
- Page 308 and 309: 298 Air Pollution - Monitoring, Mod
- Page 310 and 311: 300 Air Pollution - Monitoring, Mod
- Page 312 and 313: 302 Air Pollution - Monitoring, Mod
- Page 314 and 315: 304 Air Pollution - Monitoring, Mod
- Page 316 and 317: 306 Air Pollution - Monitoring, Mod
- Page 318 and 319: 308 Air Pollution - Monitoring, Mod
- Page 320 and 321: 310 Air Pollution - Monitoring, Mod
- Page 322 and 323: 312 Air Pollution - Monitoring, Mod
- Page 324 and 325: 314 Air Pollution - Monitoring, Mod
- Page 326 and 327: 316 Air Pollution - Monitoring, Mod
- Page 328 and 329: 318 Air Pollution - Monitoring, Mod
- Page 330 and 331: 320 Air Pollution - Monitoring, Mod
272<br />
Air Pollution <strong>–</strong> Monitoring, Modelling and Health<br />
Fig. 3. Sub-domains representation, in a two-dimensional projection (x,y), with different<br />
emission rates, and black points () represent the sensor position at 10 m height.<br />
In order to try some improvement for the inverse solution, the minimum value for the<br />
functional (22) is computed using the PSO scheme. This method is a stochastic technique<br />
and it deals with a population (candidate inverse solutions). Therefore, the numerical results<br />
represent the average from 25 seeds used by the PSO algorithm, working with 12 particles,<br />
with the following parameters: w = 0.2, c 1 = 0.1, and c 2 = 0.2. The total error obtained with<br />
the PSO was less than inverse solution computed from the quasi-Newton method. It is also<br />
important to point out that for the PSO approach no regularization operator was used ( = 0<br />
in Eq. (22)).<br />
The inverse problem was also addressed by neural network. The MLP-NN was designed<br />
and trained to find the gas flux between the soil and the atmosphere. However, there are<br />
two previous steps before the application of MLP-NN. Firstly, it is necessary to determine<br />
the appropriated activation function, number of hidden layers for the NN, and the number<br />
of neurons for each layer. Secondly, the values of the connection weights (learning process)<br />
must be identified.<br />
The hyperbolic tangent was used as activation function. Different topologies for MLP-NN<br />
were tested with 6 inputs (measured points for 6 different sensors) and 12 outputs (emission<br />
for each subdomain). Good results were achieved for two neural networks with 2 hidden<br />
layers: NN-1 with 6 and 12 (6:6:12:12), and NN-2 with 15 and 30 (6:15:30:12) neurons for the<br />
hidden layers, respectively.