FORECASTING OF FLASH POINT BY MEANS OF KDS 1.0 NEUROPACKAGE ON THE EXAMPLE OF ESTERS OF OLEIC ACID

Abstract

In article topical issue - lack of physical and chemical properties of the new synthesized substances is brought up. These properties will allow workers of supervising activity to develop systems of ensuring fire safety on objects of protection. Operability of such systems is reached by an exception of the combustible environment or a source of ignition. On the example of esters of oleic acid which are used practically in all areas of the industry and are made according to help data in number of more than several tens millions tons per year, it was succeeded to predict flash point, one of the most important fire-dangerous indicators of properties of substance, by means of KDS 1.0 neuropackage developed by us. The Neyropaket KDS 1.0 program allows: to load and look through the databases containing structures of chemical compounds and their properties; to carry out correlation of the entered data; statistically to estimate the received models; to use the received neural network models for forecasting of properties of substances, without carrying out difficult experiment. It's carried out verification of data based on some help data. Besides, flash point of esters of oleic acid data about which are absent in reference books was predicted. It gives the chance to make a start from the received values by development of systems of ensuring fire safety. On the basis of knowledge of flash point it is possible to perfrom calculation of category of the room which is necessary for establishment of requirements of fire safety. Such approach to forecasting of fire-dangerous property of substance is based on the description of structure of a molecule by means of molecular descriptors and establishment of quantitative correlations between the found values by means of artificial neural networks. During research of flash point the artificial neural network allowing to predict values with a margin error, not exceeding 8 %, in comparison with help data was simulated. Besides, the way of forecasting of fire-dangerous properties of substances based on use of molecular descriptors and artificial neural networks allows to draw a conclusion on possibility of application of such way for forecasting of other fire-dangerous properties of organic substances.