Optimization of artificial neural network models through genetic algorithms for surface ozone concentration forecasting

Introduction

This study proposes three methodologies to define artificial neural network models through genetic algorithms (GAs) to predict the next-day hourly average surface ozone (O₃) concentrations. GAs were applied to define the activation function in hidden layer and the number of hidden neurons.

Methods

Two of the methodologies define threshold models, which assume that the behaviour of the dependent variable (O₃ concentrations) changes when it enters in a different regime (two and four regimes were considered in this study). The change from one regime to another depends on a specific value (threshold value) of an explanatory variable (threshold variable), which is also defined by GAs. The predictor variables were the hourly average concentrations of carbon monoxide (CO), nitrogen oxide, nitrogen dioxide (NO₂), and O₃ (recorded in the previous day at an urban site with traffic influence) and also meteorological data (hourly averages of temperature, solar radiation, relative humidity and wind speed). The study was performed for the period from May to August 2004.

Results and discussion

Several models were achieved and only the best model of each methodology was analysed. In threshold models, the variables selected by GAs to define the O₃ regimes were temperature, CO and NO₂ concentrations, due to their importance in O₃ chemistry in an urban atmosphere.

Conclusion

In the prediction of O₃ concentrations, the threshold model that considers two regimes was the one that fitted the data most efficiently.     

Artificial neural network-derived trends in daily maximum surface ozone concentrations

Interannual variability in meteorological conditions can confound attempts to identify changes in ozone concentrations driven by reduced precursor emissions. In this paper, a technique is described that attempts to maximize the removal of meteorological variability from a daily maximum ozone time series, thereby revealing longer term changes in ozone concentrations with increased confidence. The technique employs artificial neural network [multilayer perceptron (MLP)] models, and is shown to remove more of the meteorological variability from U.S. ozone data than does a Kolmogorov-Zurbenko (KZ) filter and conventional regression-based technique.     

A comparison of nonlinear regression and neural network models for ground-level ozone forecasting

A hybrid nonlinear regression (NLR) model and a neural network (NN) model, each designed to forecast next-day maximum 1-hr average ground-level O3 concentrations in Louisville, KY, were compared for two O3 seasons--1998 and 1999. The model predictions were compared for the forecast mode, using forecasted meteorological data as input, and for the hindcast mode, using observed meteorological data as input. The two models performed nearly the same in the forecast mode. For the two seasons combined, the mean absolute forecast error was 12.5 ppb for the NLR model and 12.3 ppb for the NN model. The detection rate of 120 ppb threshold exceedances was 42% for each model in the forecast mode. In the hindcast mode, the NLR model performed marginally better than the NN model. The mean absolute hindcast error was 11.1 ppb for the NLR model and 12.9 ppb for the NN model. The hindcast detection rate was 92% for the NLR model and 75% for the NN model.     

Systematic development of an artificial neural network model for real-time prediction of ground-level ozone in Edmonton, Alberta, Canada

Ground-level ozone is a secondary pollutant that has recently gained notoriety for its detrimental effects on human and vegetation health. In this paper, a systematic approach is applied to develop artificial neural network (ANN) models for ground-level ozone (O3) prediction in Edmonton, Alberta, Canada, using ambient monitoring data for input. The intent of these models is to provide regulatory agencies with a tool for addressing data gaps in ambient monitoring information and predicting O3 events. The models are used to determine the meteorological conditions and precursors that most affect O3 concentrations. O3 time-series effects and the efficacy of the systematic approach are also assessed. The developed models showed good predictive success, with coefficient of multiple determination values ranging from 0.75 to 0.94 for forecasts up to 2 hr in advance. The inputs most important for O3 prediction were temperature and concentrations of nitric oxide, total hydrocarbons, sulfur dioxide, and nitrogen dioxide.     

Evaluation of PM10 forecasting based on the artificial neural network model and intake fraction in an urban area: A case study in Taiyuan City,China

Primary fine particulate matters with a diameter of less than 10 µm (PM₁₀) are important air emissions causing human health damage. PM₁₀ concentration forecast is important and necessary to perform in order to assess the impact of air on the health of living beings. To better understand the PM₁₀ pollution health risk in Taiyuan City, China, this paper forecasted the temporal and spatial distribution of PM₁₀ yearly average concentration, using Back Propagation Artificial Neural Network (BPANN) model with various air quality parameters. The predicted results of the models were consistent with the observations with a correlation coefficient of 0.72. The PM₁₀ yearly average concentrations combined with the population data from 2002 to 2008 were given into the Intake Fraction (IF) model to calculate the IFs, which are defined as the integrated incremental intake of a pollutant released from a source category or a region over all exposed individuals. The results in this study are only for main stationary sources of the research area, and the traffic sources have not been included. The computed IFs results are therefore under-estimations. The IFs of PM₁₀ from Taiyuan with a mean of 8.5 per million were relatively high compared with other IFs of the United States, Northern Europe and other cities in China. The results of this study indicate that the artificial neural network is an effective method for PM₁₀ pollution modeling, and the Intake Fraction model provides a rapid population risk estimate for pollutant emission reduction strategies and policies.

Implications The PM₁₀ (particulate matter with an aerodynamic diameter ≤10 μm) yearly average concentration of Taiyuan, with a mean of 0.176 mg/m³, was higher than the 65 μg/m³ recommended by the U.S. Environmental Protection Agency (EPA). The spatial distribution of PM₁₀ yearly average concentrations showed that wind direction and wind speed played an important role, whereas temperature and humidity had a lower effect than expected. Intake fraction estimates of Taiyuan were relatively high compared with those observed in other cities. Population density was the major factor influencing PM₁₀ spatial distribution. The results indicated that the artificial neural network was an effective method for PM₁₀ pollution modeling.     

VOC concentration profiles in an ozone non-attainment area: A case study in an urban and industrial complex metroplex in southern Taiwan

VOCs are important precursors of the atmospheric ozone formation species. This study investigated the airborne concentrations of 52 VOCs at two air quality monitoring stations, Daliao and Tzouying, during wintertime in southern Taiwan. Airborne VOCs samples were taken in stainless steel canisters four times per day and analyzed via gas chromatography/mass spectrometry. Maximum increment reactivity (MIR) was used to evaluate the ozone formation potential in this ozone non-attainment region. Toluene, propane, isopentane, propene, n-butane, n-pentane and isoprene contributed 78–79% of the 52 VOCs in Daliao. Toluene, 1-butene, isopentane, propene, propane, n-undecane, and n-butane contributed 71–77% of the 52 VOCs in Tzouying. The VOCs concentrations were higher in Daliao due to the high toluene emissions from a paint plant and a solvent plant in the nearby industrial district. The 24-h VOC concentrations averaged 25 ppb higher in Tzouying than in Daliao. The ozone formation potential of airborne VOCs was 1687–2730 and 1717–2261 μg-O₃/g-VOCs in Daliao and Tzouying, respectively. Ozone concentrations in Tzouying were 44 ppb higher than in Daliao during the 1200–1600 sampling period.     

Air quality modeling of interpollutant trading for ozone precursors in an urban area

Emission trading is a market-based approach designed to improve the efficiency and economic viability of emission control programs; emission trading has typically been confined to trades among single pollutants. Interpollutant trading (IPT), as described in this work, allows for trades among emissions of different compounds that affect the same air quality end point, in this work, ambient ozone (O3) concentrations. Because emissions of different compounds impact air quality end points differently, weighting factors or trading ratios (tons of emissions of nitrogen oxides (NO(x)) equivalent to a ton of emissions of volatile organic compounds [VOCs]) must be developed to allow for IPT. In this work, IPT indices based on reductions in O3 concentrations and based on reductions in population exposures to O3 were developed and evaluated using a three-dimensional gridded photochemical model for Austin, TX, a city currently on the cusp of nonattainment with the National Ambient Air Quality Standards for O3 concentrations averaged over 8 hr. Emissions of VOC and NO(x) from area and mobile sources in Austin are larger than emissions from point sources. The analysis indicated that mobile and area sources exhibited similar impacts. Trading ratios based on maximum O3 concentration or population exposure were similar. In contrast, the trading ratios did exhibit significant (more than a factor of two) day-to-day variability. Analysis of the air quality modeling indicated that the daily variability in trading ratios could be attributed to daily variations in both emissions and meteorology.     

Diurnal variations of ozone and other pollutants in an urban area

A theoretical model is used to describe the diurnal variations of primary and secondary pollutants, with emphasis on ozone. This is done for an urban basin with anthropogenic sources of nitrogen oxides and hydrocarbons. We propose a scheme for the decomposition of aromatic compounds. According to this scheme, each aromatic molecule gives rise to six transfers of NO to NO₂ without consumption of odd oxygen. It is concluded that it is not a good approximation to represent urban hydrocarbon emissions by one single species, neither in short term (a few hours) nor multiday simulations. Species with both high and low reactivity ought to be included. We show that the nocturnal minimum in ozone often observed in urban areas, is mainly induced by gas chemistry. It is not a good approximation to omit the chemical development during the night-time in a theoretical analysis of urban photochemical pollution. Such an omission introduces errors also in the day-time chemistry. Application of constant dissociation rate coefficients over the day gives rise to false morning and evening ozone maxima.     

A simple semi-empirical photochemical model for the simulation of ozone concentration in the Seoul metropolitan area in Korea

O₃ concentrations were simulated over the Seoul metropolitan area in Korea using a simple semi-empirical reaction (SEGRS) model which consists of generic reaction set (GRS), photochemical reaction set, and the diagnostic wind field generation model. The aggregated VOC emission strength was empirically scaled by the comparison of the simulated slope of (O₃–2NO–NO₂) concentration as a function of cumulative actinic light flux against measurements on high surface ozone concentration days with the relatively weak easterly geostrophic winds at the 850 hPa level in summer when the effect of horizontal advection was fairly small. The results indicated that the spatial distribution patterns and temporal variations of spatially averaged ground-level ozone concentrations were quite well simulated compared with those of observations with the modified volatile organic compound (VOC) emission strength. The diurnal trend of the surface ozone concentration and the maximum concentration compared observations were also quite reasonably simulated. However, the maximum ozone concentration occurring time at Seoul lagged about 2 h and the ozone concentration in the suburban area was slightly overestimated in the afternoon due to the influx of high ozone concentration from the urban area. It was found that the SEGRS model could be effectively used to simulate or predict the ground-level ozone concentration reasonably well without heavy computational cost provided the emission of ozone precursors are given.     

Meteorologically adjusted trends of daily maximum ozone concentrations in Taipei,Taiwan

The meteorological conditions exert large impacts on ozone concentrations, and may mask the long-term trends in ozone concentrations resulting from precursor emissions. Estimation of long-term trends of ozone concentrations due to the changes in precursor emissions is important for corresponding control strategy. Multiple linear regression (method I), multilayer perceptron (MLP) neural network (method II) and Komogorov-Zurbenko (KZ) filter method plus MLP methodology (method III), are used to estimate the meteorologically adjusted long-term trends of daily maximum ozone concentrations by removing the masking effects of meteorological conditions in this study. The daily maximum ozone concentrations and relative meteorological variables were extracted from six air-monitoring stations in Taipei area from 1994 to 2001. The data collected during 1994–2000 period were used as modeling set and utilized to estimate the meteorologically adjusted trends, and the data of 2001 were used as the validation data. The meteorologically adjusted trends of ozone for these three methods were calculated and compared. The results show that both MLP and KZ filter +MLP models are more suitable than multiple linear regression for estimating the long-term trends of ozone in Taipei, Taiwan. The long-term linear trends of meteorologically adjusted ozone concentrations due to the precursor emissions show an increase trend at all stations, and the percent changes per year range from 1.0% to 2.25% during the modeling period in Taipei area.     

The potential of statistical state space models in urban ozone forecasting

State space models for tropospheric urban ozone prediction are introduced and compared with linear regression models. The linear and non-linear state space models make accurate short-term predictions of the ozone dynamics. The average prediction error one hour in advance is 7 μg/m³ and increases logarithmically with time until it reaches 26 μg/m³ after 30 days. For a given sequence of solar radiation inputs, predictions converge exponentially with a time scale of 8 hours, so that the model is insensitive to perturbations of more than 150 μg/m³ O₃. The slow increase of the prediction error in addition to the uniqueness of the prediction are encouraging for applications of state space models in forecasting ozone levels when coupled with a model that predicts total radiation. Since a radiation prediction model will be more accurate during cloud-free conditions, in addition to the fact that the state space models perform better during the summer months, state space models are suitable for applications in sunny environments.     

Simplified model for the prediction of ozone generation in polluted urban areas with continuous precursor species emissions

In this paper we study the transient ozone generation process in polluted urban areas, using a simplified set of chemical reactions for the NO_x and hydrocarbon photolysis. We obtained a reduced kinetic mechanism and it is solved using the boundary layer theory due to the appearance of two time scales in the problem associated with the photo-chemical reactions of the NO₂ and unburned fuel, respectively. Assuming a temporal addition of ozone precursor species, we obtained in closed form, the temporal evolution of the ozone concentration as a function of the physico-chemical parameters.     

A holistic approach for optimal design of air quality monitoring network expansion in an urban area

This paper presents an objective methodology for determining the optimum number of ambient air quality stations in a monitoring network. The methodology integrates the multiple-criteria method with the spatial correlation technique. The pollutant concentration and population exposure data are used in this methodology in different ways. In the first stage, the Fuzzy Analytic Hierarchy Process (FAHP) with triangular fuzzy numbers (TFNs) is used to identify the most desirable monitoring locations. The network configuration is then determined on the basis of the concept of sphere of influences (SOIs). The SOIs are dictated by a predetermined cutoff value (r_c) in the spatial correlation coefficients (r) between the pollutant concentrations at the monitoring stations identified from first step and the corresponding concentrations at neighboring locations in the region. Finally, the optimal station locations are ranked by using combined utility scores gained from the first and second steps. The expansion of air quality monitoring network of Riyadh city in Saudi Arabia is used as a case study to demonstrate the proposed methodology.     

A model for predicting maximum and 8 h average ozone in Houston

The objective of this research was to develop a statistical model to predict one day in advance both the maximum and 8 h (10 am–5 pm) average ozone for Houston (TX). A loess/generalized additive model (GAM) approach was taken to model development. Ozone data (1983–1991) from ten stations in the immediate Houston area were used in the study. The meteorological data came from the Houston International Airport. The models were developed using data for April through October for 1983–1987 and 1989–1990. Forecasts were developed for 1988 and 1991. The final model, which was multiplicative in nature, contained three interaction terms for the west/east and south/north wind components (average of hourly values from 8 pm to 5 am, 6 am to 9 am, and 10 am to 5 pm). Opaque cloud cover (averaged over the period 10  am to 5 pm), yesterday’s maximum ozone, today’s maximum temperature and morning mixing depth were also important variables in the model.Individual forecasts were generated for all ten stations in the Houston area using observed meteorology. In addition forecasts were produced for three measures of the network as a whole. The root-mean-square prediction error for the 8 h average forecasts ranged from 13.2 to 16.3 ppb (with R² ranging from 0.66 to 0.73) for the individual stations and from 18.5 to 22.0 ppb (with R² ranging from 0.61 to 0.68) for maximum ozone. A detailed examination was undertaken for a day on which the forecast was much too low.     

Measurement of surface ozone and its precursors in an urban area in South Brazil

Weekly and seasonal variations of surface ozone and their precursors – nitrogen oxides, carbon monoxide-associated with meteorological parameters (wind direction, temperature, solar radiation) – are reported. Measurements were performed continuously during 2006 at two sampling stations located in the metropolitan area of Porto Alegre, Brazil. Results have shown that O₃ concentrations remained almost constant between weekdays. Levels of NO_x precursors decreased especially on Sundays, due to lighter traffic. The seasonal variation has shown a maximum O₃ concentration during summer and spring while NO_x and NO₂ have maxima at the colder months. The daily cycle of highest ozone concentrations reveals a lower nightly level and an inverse relation between O₃ and NO_x, evidencing the photochemical formation of O₃. There are seasonal variation and source heterogeneity.     

Elemental background concentration in the atmosphere of an industrialized country

Concentrations of heavy elements in air were determined at a rural site in Belgium during periods of very low pollution levels. These low levels were appraised by means of a sensitive SO₂ monitor. Observed mean background concentrations are high or very high as compared with levels observed in remote or very remote areas of continental zones. However, extrapolating the observed concentrations to very low sulfur levels gives results comparable to those observed in remote areas.     

A Lagrangian stochastic model for predicting concentration fluctuations in urban areas

A combined Lagrangian stochastic model with a micromixing sub-model is used to estimate the fluctuating concentrations observed in two wind tunnel experiments. The Lagrangian stochastic model allows fluid trajectories to be simulated in the inhomogeneous flow, while the mixing model accounts for the dissipation of fluctuations using the interaction by exchange with the mean (IEM) mechanism. The model is first tested against the open terrain, wind tunnel data of Fackrell, J.E. and Robins, A.E. [1982. Concentration fluctuations and fluxes in plumes from point sources in a turbulent boundary layer. Journal of Fluid Mechanics 117, 1–26] and shows good agreement with the observed mean concentrations and fluctuation intensities. The model is then compared with the wind tunnel simulation of a two-dimensional street canyon by Pavageau, M. and Schatzmann, M. [1999. Wind tunnel measurements of concentration fluctuations in an urban street canyon. Atmospheric Environment 33, 3961–3971]. Despite the limitations of the k–ε turbulence scheme and the IEM mixing mechanism, the model reproduces the fluctuation intensity pattern within the canyon well. Overall, the comparison with both sets of wind tunnel experiments are encouraging, and the simplicity of the model means that predictions in a complex, three-dimensional geometry can be produced in a practicable amount of time.     

Neural network-based study for predicting ground-level ozone concentration in large urban areas, applied to the Sao Paulo metropolitan area

Events of high concentration of ground-level ozone constitute a matter of major concern in large urban areas in terms of air quality, and public health. In the Sao Paulo Metropolitan Area (SPMA), air quality data generated by a network of air quality measuring stations have been used in a number of studies correlating ozone formation with different variables. A study was carried out on the application of neural network models in the identification of typical sceneries leading to high ground-level ozone concentrations in the SPMA. The results were then applied in the selection of variables, and in the definition of neural network-based models for estimating ozone levels from meteorological variables. When combined with existing weather prediction tools, the models can be applied in the prediction of ozone levels in the SPMA     

Model for forecasting expressway fine particulate matter and carbon monoxide concentration: application of regression and neural network models

The Borman Expressway is a heavily traveled 16-mi segment of the Interstate 80/94 freeway through Northwestern Indiana. The Lake and Porter counties through which this expressway passes are designated as particulate matter < 2.5 microm (PM2.5) and ozone 8-hr standard nonattainment areas. The Purdue University air quality group has been collecting PM2.5, carbon monoxide (CO), wind speed, wind direction, pressure, and temperature data since September 1999. In this work, regression and neural network models were developed for forecasting hourly PM2.5 and CO concentrations. Time series of PM2.5 and CO concentrations, traffic data, and meteorological parameters were used for developing the neural network and regression models. The models were compared using a number of statistical quality indicators. Both models had reasonable accuracy in predicting hourly PM2.5 concentration with coefficient of determination -0.80, root mean square error (RMSE) <4 microg/m3, and index of agreement (IA) > 0.90. For CO prediction, both models showed moderate forecasting performance with a coefficient of determination -0.55, RMSE < 0.50 ppm, and IA -0.85. These models are computationally less cumbersome and require less number of predictors as compared with the deterministic models. The availability of real time PM2.5 and CO forecasts will help highway managers to identify air pollution episodic events beforehand and to determine mitigation strategies.     

Comparison of artificial neural network and regression models in the prediction of urban stormwater quality.

Urban stormwater quality is influenced by many interrelated processes. However, the site-specific nature of these complex processes makes stormwater quality difficult to predict using physically based process models. This has resulted in the need for more empirical techniques. In this study, artificial neural networks (ANN) were used to model urban stormwater quality. A total of 5 different constituents were analyzed-chemical oxygen demand, lead, suspended solids, total Kjeldahl nitrogen, and total phosphorus. Input variables were selected using stepwise linear regression models, calibrated on logarithmically transformed data. Artificial neural networks models were then developed and compared with the regression models. The results from the analyses indicate that multiple linear regression models were more applicable for predicting urban stormwater quality than ANN models.