ADAPTIVE SHORT-TERM FORECASTING OF ELECTRICITY CONSUMPTION BY AUTONOMOUS POWER SYSTEMS OF SMALL NORTHERN SETTLEMENTS BASED ON CORRELATION ANALYSIS METHODS

Abstract

One of the main problems in building energy-efficient and non-resource-intensive decentralized power supply systems in the Arctic zone and regions of the Far North is forecasting the consumption of electrical energy by small northern settlements. Among the existing methods that give an acceptable result in terms of accuracy, one can single out approaches based on the ARIMA econometric method. A method based on the Wold decomposition and correlation functions of the stochastic process is considered, an adaptive model of the difference equation is constructed, which makes it possible to predict energy consumption of the active power of autonomous systems of a small settlement in the interval of 0–4 hours, by reducing the stochastic process to a stationary Markov process with short memory. The purpose of the work is to build a methodology for short-term forecasting of electric energy consumption by autonomous energy systems of small northern settlements, taking into account the specifics of energy consumption in the conditions of the Arctic zone and the Far North, based on a stochastic series of data on electric energy consumed by the settlement over the previous period. Methods: approach to obtaining a short-term forecast of electricity consumption by autonomous energy systems of small northern settlements based on the ARIMA econometric method. In this case, a difference equation is constructed for the deterministic and random components of the available stochastic series of energy consumption; Wold decomposition and correlation functions of the energy consumption are used. To adapt the model, the studied stochastic process is reduced to a Markov process with a short memory. To do this, it is necessary to use a difference operator, which reduces the relative contribution of consumption deterministic component. Results. Based on Wold decomposition and correlation functions, it was possible to obtain a model that gives a short-term forecast of active power consumption for a lead time of 4 hours. Conclusions. Based on the Wold decomposition and the correlation functions of the stochastic process, the authors of the work managed to obtain an adaptive model of the difference equation, which makes it possible to predict active power consumption of autonomous systems of a small settlement with a lead time of 4 hours. The stationarity of the random process was carried out by introducing a difference operator of the first order, which makes it possible to reduce the relative contribution of the deterministic component of the stochastic series. The authors managed to reduce the process to a stationary Markov process with a short memory. The expansion coefficients of the difference equation were estimated by solving a nonlinear equation, which consists in finding the global maximum of the likelihood function. The constructed 90 % probabilistic boundaries allow us to talk about a satisfactory adjustment of the adaptive parameters of the difference equation for predicting the system. The result of forecast modeling with four-hour lead time shows good agreement with experiment.

Keywords

• stochastic series
• wold decomposition
• autocorrelation function
• partial correlation function
• autoregression
• moving average
• maximum likelihood function