Different methods are applied to forecast the condition of the atmospheric air – analytical and empirical, numerical, statistical, combined and many others. All these methods have their advantages and faults. For example analytical and empirical methods have gross errors and numerical methods require utilization of the data on the source of pollution and this information is often unknown. That’s why the research of the processes of pollution of the atmospheric air and working out the models of forecasting are urgent tasks of the ecological monitoring [2].

The goal of the master’s work is the research of the processes of pollution of the atmospheric air based on the data received from the automated posts monitoring the condition of the atmospheric air.

The tasks of the master’s work:

• analysis of the methods and means of short-term forecasting of the diffusion of impurities in the atmospheric air;
• analysis of the data on the pollution of the atmospheric air received from the automated posts of monitoring and comparing it with the data received from the nearby stationary posts of the subjects of monitoring;
• eliciting the nature of diffusion of impurity in the spatial and temporal coordinates;
• constructing the model of diffusion of impurity and analysis of its effectiveness through a priori estimation of the basic parameters;
• approbation and introduction of the scientific achievements.

Research methods applied in the master’s work are: statistical, artificial neural networks, spectral analysis.

The estimated scientific novelty of this work lies in the following:

• research of the nature of impurity concentration change based on the data received from the automated monitoring posts located in Donetsk and Makeevka area;
• obtaining mathematical model of ARIMA type estimating the change of concentration of impurity for carbon oxide (CO), sulphur dioxide (SO2) and nitrogen dioxide (NO2);
• forecasting the change of impurities concentration for the short-term period and constructing an a priori estimation of the quality of forecast model.

The results of the work were reported on the V international conference for students, post-graduate-students and young scientists “Computer monitoring and informational technologies” (CMIT-2009) and published in the conference digest.

There are many methods of solving the task of forecasting the atmosphere pollution.

The method of estimation of concentration All-Union Normative Document (AND-86) is generally accepted and applied in some CIS countries. Its goal is to estimate the maximum level of concentration of polluting substance on a certain distance from the source of exhausting. That’s why reflecting the level of atmosphere pollution is a very hard process. Many of the coefficients of the equations were found in empiric way for the climatic features of CIS countries which makes this method non applicable in other countries [3].

In his works M.E. Berland describes different models of forecasting of atmosphere pollution based on generic equation of diffusion. Solving the given equation and for its simplification a number of assumptions and hypotheses were applied. The advantage of this method is the universality of the obtained models and the disadvantage is a high indeterminacy of the initial data, complicated procedures of estimation of authenticity of the models and some other [4].

One of the methods of estimation of atmospheric air pollution applying the data received from the monitoring posts are based on the statistics. Nowadays it is officially applied on the territory of Ukraine and published in [1]. The given method considers the effect of not separate polluting substances but their general negative effect and allows to estimate the field of concentration. This method allows to solve stationary the task of forecasting.

The analysis of the software products in the field of monitoring of atmosphere pollution shows that there are almost no efficient specialized software products fit for forecasting the atmosphere pollution. The following local software products are applied now: EOL, Plener, UPRZA Ecolog, Kedr and some other and also foreign products by CalPuff, Plume, TARM [5].

Most local software products have a number of disadvantages with impossibility to consider the relief features of the area being predominant. Application of foreign software products is impossible through a number of reasons: differences in climatic conditions, systems of submitting the ecological data and some other [6].

Judging by all the above mentioned we can say that the task of finding new methods of forecasting the atmospheric air pollution and working out software products of new generation based on this is the urgent task of monitoring of the environment.

On the territory of Donetsk-Makeevka area we can see today the installation of automated posts of monitoring of atmosphere pollution. One of the posts is located on the territory of DonNTU and is functioning since 2008. The given post monitors the concentration of carbon oxide (CO), sulphur dioxide (SO2) and nitrogen dioxide (NO2) in the air. The process of measuring includes the analysis of the atmospheric air with application of gas analyzer having gauges for CO, SO2 and NO2. The information received from the gauges is averaged and in an interval of 10 minutes is recorded to the database incorporated in the programmed complex Akiam.

The accumulated data represent temporal series of monitoring the concentration of the given substances. For the moment of writing this abstract of thesis the number of available data comprises approximately 20.000 observations. The given information can be applied for working out forecast models of atmosphere pollution. The following models can be applied for the data analysis: interrupted temporal series analysis, exponential smoothing and forecasting, autoregressive moving average model (ARIMA), spectral analysis (Fourier), forecasting with application of neural networks and other. The master’s work supposes application of the latter three methods.

The general structure of the system of forecasting of atmospheric air pollution is represented in fig. 1.

Figure 1 – The structure of the system of atmosphere pollution forecast

(animation: size - 29,1 КB; image size - 560x417px; shots quantity - 9;
delay between shots - 90 ms; delay between last and first shot - 150 ms;
number of repetition cycles - 7)

The following tasks of the master’s work are completed for the moment: data received from the automated post of Akiam system and the given data prepared for the analysis; temporal series analysis done with the application of artificial neural networks and ARIMA method and the attempt to forecast them.

Now we consider in details every method planned to be applied and also the results of the conducted analysis.

1. Spectral analysis

The goal of the spectral analysis is to analyze the number in the functions of sines and cosines of various frequencies to determine the main functions appearance of which is very significant and important. The methods of spectral analysis play a significant role for determining the latent periods in the data. They can be also used for verifying the adequacy of ARIMA model in terms of analysis of residuals [7].

2. Artificial neural networks

The method of neural networks is rather popular and effective for solving the tasks of forecasting. It implies creating an optimal neural network based on initial data, its training on various algorithms (back propagation algorithm, method of conjugate gradients, method of Levenberg-Marquardt, method of Quasi-Newton and other), constructing a forecast [8]. The advantage of this method is the possibility to construct forecasts for any number of periods [9].

Statistical Neural Network program was applied for neural network forecast of temporal series but due to the stochastic nature the given method did not give the anticipated result.

3. Method of auto regression and integrated moving average (ARIMA)

This method is based on the application of auto regress processes and moving average. There are 3 types of model parameters: parameters of auto regress (p), the number of differencing passes (d), parameters of moving average (q). In generally applied parameters the model is recorded as ARIMA (p, d, q) and can be represented with the following equation (1):

(1)

with Y_t - current figure of temporal series; Y_t-i - previous figures of temporal series; a₀ - free term; a₁, a₂, a₃... (-1<a_i<1) - parameters of auto regress; b₀ - free term of moving average; b₁, b₂, b₃... (-1<b_i<1) - parameters of moving average; - casual component on the current pass of modeling; - casual component on the previous pass of modeling.

The estimation of the parameters of the model is done after the identification of the model. The obtained estimations of the parameters are applied during forecasting [7].

The results of model construction with the method of ARIMA in STATISTICA 6.0 for the temporal series of NO₂ concentration in the atmosphere are given in table 1.

Table 1 – Identification of the model of dynamics of NO₂ temporal pass

The given table has the following data: p1, p2 – parameters of auto regress, d – differencing pass, q1, q2 – parameters of moving average, MS residuals – average square of residuals, M_x – mean of residuals, D_x – variance of residuals.

The analysis of significance of coefficients pi, qi, general error of the model (average square of residuals), probability residuals has shown that the models of ARIMA (0, 1, 2), ARIMA (0, 1, 1), ARIMA (2, 1, 2), ARIMA (1, 1, 1) are the most optimum, of which the model ARIMA (1, 1, 1) was chosen.

Analyzing the residuals we made a conclusion that the given model is an adequate one. The equation of ARIMA (1, 1, 1) model is like:

(2)

with C(t) – concentration of NO₂ at the current moment, C(t-1) – concentration at the previous moment, a(t-1) – value of white noise at the previous moment, a(t) – value of white noise at the current moment. Parameters of a(t) are given in the table 1.

Similar research was conducted also for other substances under observation – carbon oxide (CO) and sulphur dioxide (SO₂).

The following new steps are planned to be achieved in this work:

• further search for optimum model of forecasting with application of ARIMA method;
• verification of adequacy of the obtained model through spectral analysis, neural networks and other methods;
• working out software products implementing the forecasting model;
• working out an ecological portal for Makeevka containing the modulus of forecasting of atmospheric air pollution.

The given master’s work is aimed at a very urgent problem of forecasting of the atmospheric air pollution. After making this work a model of the process of atmosphere pollution will be obtained and the appropriate software products will be produced. The feature of this work is that the data for the research are received from the automated posts of Donetsk and Makeevka. The model can be adapted also for other nearby cities since the climatic conditions do not undergo substantial changes in the space.

As a conclusion it’s worth mentioning that the problem of atmospheric air pollution is one of the most critical ecological problems. That’s why the forecasting of atmosphere pollution should be applied to reduce the exhausts and as a sequence reduce the level of atmospheric air pollution.

1. Впровадження нового механізму видачі дозвільних документів з викидів забруднюючих речовин в атмосферне повітря / Під. ред. С.В.Третьякова. — Донецьк: Державне управління екології та природних ресурсів у Донецькій області, Донецька філія Державного закладу «Державний екологічний інститут Мінприроди України», 2006. — 196 с.

2. Сонькин Л.Р. Синоптико-статистический анализ и краткосрочный прогноз загрязнения атмосферы. – Л.: Гидрометеоиздат.1991. – 250с.

3. Методика расчета концентраций в атмосферном воздухе вредных веществ, содержащихся в выбросах предприятий. ОНД-86.: Госкомгидромет. 1986. – 68с.

4. Берлянд М.Е. Прогноз и регулирование загрязнения атмосферы. – СПб.: Гидрометеоиздат. 1985. – 272с.