AI Applications
Up AI Applications Basics



Module “Source” of AIR-ECO (information technology for real-time atmospheric air quality assessment)  provides the assessments of pollutant emissions from power plants with discretization of several hours. Some principles of artificial intelligence are used for obtaining numeric values of emissions.

This slide shows how the emission functions are reconstructed. Originally, we have a certain annual average value of emission of a specific power plant. Then sequentially we start to apply some additional information about this source. What is this information? As a rule, it is technological information as fuel consumption, output power, information on monthly average load, and etc.

This partial information is grouped into production rules of type IF … THEN, which are sequentially applied to reconstruct the emission function of specific source.

As a result, we have a reconstructed function of emission with given discretization. Naturally, average value of this function corresponds to the original annual average emission of that source.

This slide presents the results of applying incomplete and partial knowledge about the dynamics of the transportation flows for reconstructing the emission functions into the atmospheric air. 

The knowledge of the flow dynamics represents the expert information that in a given temporal period the emission of pollutants makes up a certain percentage of the annual average value.

For example, it is obvious that

bulletin winter the emission is less than in summer;
bulletat night it is less, than during the day, and in work days - it is higher, than on a weekend, and etc.  

In addition to the expert information, we used the statistical data on the total annual pollutant emission from transportation into the atmospheric air.

As you may see in this slide, transportation flows have a discrete representation. The granularity of discretization of piece-wise linear segments of pathways depends on the requirements to the precision of modeling.

Let us consider a temporal concept of "Evening". It is obvious that this is a very subjective concept. It is impossible to tell precisely when the evening evening starts or finishes. Such a definition will depend on the opinion of a given experts and different experts will most likely have different opinions. 

Some expert might tell that from 7pm to 10pm it is definitely evening; 5pm to 7pm is some transitive ohase between a day and an evening; and 10pm to 12 midnight is a transitive phase between a evening and a night. Such a definition describes a membership function of concept "Evening". 

Here X it is measured in hours and X = [0,24]. Function F(X) is determined on interval X and its values vary from 0 to some maximum (no more than 100 %). 

In the given example for simplification it is possible to say that on the interval [17,19] F(X) changes linearly from 0 to 1, and on the interval [22,24] F(X) changes linearly from 1 to 0. 

Assume that some expert has determined the concepts of seasons and emissions from transport in some city as it is shown on slide.

Here we see some overlap of intervals for each season: winters, spring, summer and fall.

The expert thinks that the average amount of emission from transportation in winter is 70% of the 12-month average emission, 110 % - in the spring, 150 % - in the summer, and 100 % in the fall accordingly.

In order for this information to be acceptable, it must be normalized. In other words, it is necessary, that the sum of the areas under all the trapezoids, divided by 12, must be equal to 100. This measure is those 100% that we assume to be the 12-month average emission.


horizontal rule

This Web site is © 2003 - 2012 by Anatoly Kurkovsky

Last updated: May 30, 2012