Інформаційна технологія пошуку можливих джерел підвищеного забруднення річки з використанням моделі Prophet

Abstract

Climate change has led to many low-water years and, consequently, a decrease of the volume of water to dilution an-thropogenic pollution. Thus, research aimed at identifying the main sources of pollution to regulate them immediately is becoming increasingly important. Moreover, according to the EU Water Framework Directive, which, according to the Asso-ciation Agreement with the EU, Ukraine is obliged to comply with, it is necessary to develop a set of actions soon to achieve or stabilize at least good environmental status in all water bodies. In Ukraine, as in many other European countries, the water quality monitoring system does not provide a sufficient amount of regular observation data for localization in space and time of all, including unregistered, sources of increased pollution, which complicates the implementation of the policy of their regulation. Therefore, it is important to create information technology to find possible sources of increased anthropo-genic pressure on the river according to regular observations of water quality in the basin of a given river. The analysis showed that such data is characterized by a change in the frequency of observations (especially in the long run for dec-ades), there is a practice of one-time observations (once a quarter or six months, each time at different times), many missed data, etc., which makes it impossible to use typical similar problems of multiple regressions and time series models based on autoregression and integrated moving average (ARIMA). It is proposed to use Facebook's Prophet model and package for R and Python, which is devoid of all these short-comings and is optimal for solving this problem. The methodology of its application is developed and characterized, which consists in the modeling of monitoring data with filtering of different types of seasonality and allocation of a linear trend between change points, the first approximation of each of which is set at the beginning of intervals in one or several years, depending from the amount of available data. The identified trends between these points are compared by different indicators at each observation post and a specially developed algorithm reveals the largest increases in trends ("pulses"), which then cause a monotonous increase in pollution up to this time. The detected dates of such "pulses" are scaled and aggregated by different indicators, which allows to determine the date of occurrence of the source of pollution at each section between posts and then, according to other data with the involvement of relevant control services, more accurately identify the source of increased river pollution, at present. A program in Python was devel-oped, which tested the efficiency of the technology to detect such "impulses" on the example of the Southern Bug River from its source to Vinnytsia according to the state water quality monitoring system for 2002-2019 and presents the successful results of its work.