ЗАСТОСУВАННЯ НЕЙРОННИХ МЕРЕЖ МГУА ДЛЯ ВИЯВЛЕННЯ ЗМІНИ УМОВ РОБОТИ НАСОСНОЇ СТАЦІЇ ВОДОПОСТАЧАННЯ, ЗУМОВЛЕНИХ ВПЛИВОМ СЕЗОННИХ ЧИННИКІВ

Abstract

Purpose. Development of a mechanism identification of changes in the actual operation conditions of the water supply facility depending on the season. Methodology. Methods of intellectual analysis of profiles of mode indicators were used to identify regularities in the formation of the operation mode of the facility. The mathematical apparatus of pattern recognition algorithm with training was used to classify the profiles of mode indicators. Methods of self-organization of models of complex systems were used for structural and parametric identification of the classifier model. Results. The necessity of the analysis of the characteristics of operation mode of the pumping station of water supply, obtained from the monitoring system, to identify regularities in the formation of water supply was substantiated. The daily graph of water consumption from the water supply network was used as a mode indicator of the water supply process. Indicators of water supply volumes and graph unevenness were used to describe the water supply graph. The expediency of application of self-organization methods for solving the problem of classification and construction of the classifier model was substantiated. Structural and parametric identification of the classifier model for daily water consumption graphs was performed using GMDH Neural Networks. The search for the optimal model was performed in three classes of neural networks. Better neural network structure was chosen on the basis of criterion of regularity. The K-block cross-validation strategy was used to test the models. The results of the verification of the classifier model showed the high quality of the classification. Originality. A method for identifying changes in the operation conditions of the water supply facility due to the influence of seasonal factors, based on the usage of the classifier of profiles of daily water consumption graphs from the water supply network, was proposed. Practical value. The constructed model of the classifier allows defining of belonging of a profile of the daily water consumption graph, received from monitoring system of the water supply mode, to one of typical classes. The fact of class change indicates a change in the actual operation conditions of the water supply facility. Conclusions. Analysis of daily graphs of water consumption using the pattern recognition algorithm makes it possible to establish the change in the actual operation conditions of the water supply facility caused by the influence of seasonal factors. The usage of neural networks of GMDH makes it possible to perform automatically structural and parametric identification of the classifier model. Application of the offered principles is a basis of effective planning of operation modes of pumping station of water supply. References 16, tables 3.