Shumilin Alexey Dmitrievich, postgraduate student, Penza State University (440026, 40 Krasnaya street, Penza, Russia), firstname.lastname@example.org
Vershinin Nikolay Nikolaevich, doctor of technical sciences, professor, head of sub-department of technospheric safety, Penza State University (440026, 40 Krasnaya street, Penza, Russia), email@example.com
Vershinin Aleksey Evgen'evich, undergraduate student, Penza State University (440026, 40 Krasnaya street, Penza, Russia), firstname.lastname@example.org
Volkova Alisa Sergeevna, postgraduate student, Penza State University (440026, 40 Krasnaya street, Penza, Russia), email@example.com
Background. It is shown that motor transport generates environmental problems in large cities, since the peculiarity of motor transport as a mobile source of pollution is a low location (at the level of children's breathing), distribution to indeterminate territories, immediate proximity to residential areas. It is necessary to obtain full information about the environmental situation in the city.
Materials and methods. It is shown that the use of artificial neural networks is one of the most promising methods used in conducting studies of sociological, biological, financial, economic and other complex systems. It is suggested to use artificial intelligence technologies for forecasting negative factors of motor transport influence on the urban environment.
Results. The article considers the technology of forecasting using artificial neural networks. The stages of identifying factors and constructing models for the application of neural network forecasting are described. As an example, the process of predicting the impact of vehicles on the ecological situation of the city is considered. Time series are determined to determine the repeatability of the samples characterizing them, depending on the factors. A method for predicting time series by a neural network of back propagation of an error is described. With the help of the generalization error, the effect of retraining was revealed. During the training, there was a constant reduction in the learning error to a minimum, after which training ceased. Possible options for the use of artificial neural networks in the sphere of environmental control are proposed. The capabilities of the analytic platform Deductor Studio, used in training a neural network, are shown.
Conclusions. The results of the presented forecast can be used in practice when forming local forecasts of the number of vehicles in areas with installed surveillance cameras. To improve the accuracy of the forecast, it is advisable to link the number of vehicles to the days of the week, the period of the year. It is shown that a satisfactory model can be constructed with the help of neural networks, even with insufficient data, which can be further refined as new data become available.
neural network, forecasting, neural network training, learning by example, artificial neurons, learning sample, city ecology, ecological safety