Grigor'eva D.R.Branch of Kazan (Volga) Federal University in Naberezhnye Chelny, Naberezhnye Chelny, Republic of Tatarstan, Russian Federation d.r.grigoreva@mail.ru
Faizullina A.G.Branch of Kazan (Volga) Federal University in Naberezhnye Chelny, Naberezhnye Chelny, Republic of Tatarstan, Russian Federation dlya_pisem_t@mail.ru
Subject Consumption by the population and enterprises depends on many factors: temperature, day time, weather conditions. The constant need to supply electricity to households and enterprises generates the need for protection against serious damage. Termination of energy supply to vital objects may cause accidents and lead to huge financial losses; therefore, in most cases it is advisable to spend money on crisis prevention. A possibility to solve this problem is to predict energy consumption. As a rule, experts are aware of the information about potential electricity consumption, and predicting potential damage by an expert will be sufficient. This entails little financial cost and considerable time to collect representative samples. As a rule, the information needed to predict is going to power sensors or utilities. The article discusses the application of the STATISTICA system to forecast energy consumption at enterprises. In addition to risk mitigation, forecasting is another important application. For businesses, the cost of energy bills accounts for a significant budget item. Forecasting enables to apply a more balanced approach to the formation of expenditure items. Objectives The main objective of the work is to analyze statistical methods and forecasting of electricity consumption volume at a plant under the method of researching the time series with the help of the STATISTICA package using neural networks. In this paper, we consider the ARIMA model, which represents a whole class of stochastic processes used for time series analysis. In addition, the paper discusses the application of neural networks possibilities in the problem of forecasting. We set the following tasks: to forecast energy consumption under the ARIMA method; to make a forecast of electricity consumption using neural networks; to check the quality of the model for the recent year; to make a forecast for a longer period. Methods The paper discusses the theoretical basis of the analysis and forecasting, the basic concepts of time series and their types, the theory of the neural network model. We also used the ARIMA model, the indicators of the analysis of dynamics, and performed a comparative analysis of the dynamics of electricity consumption volume for the period of 2009-2013. Results We have made a forecast of a time series under the ARIMA method using the neural networks. Discussion Neural networks have many important properties, but the key one is learning capability. Training of neural networks involves, first of all, changing the 'strength' of synaptic connections between neurons. The classification of neural networks in terms of the nature of learning divides them into neural networks using supervised learning and neural networks using unsupervised learning. The supervised learning implies that for each input vector there is a target vector representing the desired output. Together they are called a training pair. Typically, the training of a neural network is based on a certain number of training pairs, where the output vector is presented, the network output is calculated and compared to appropriate target vector. Further, the weight changes in accordance with the algorithm seeking to minimize the error. The vectors of the training set are presented sequentially, then errors are computed and the weights are adjusted for each vector as long as the error for the entire learning array reaches an acceptable level. The unsupervised learning is a much more plausible model of learning in terms of biological roots of artificial neural networks. Teuvo Kohonen and many other scholars developed the model. It does not need a target vector for output and, therefore, does not require any comparison with predetermined ideal responses. The training set only consists of input vectors. The learning algorithm adjusts the network weights so that to achieve agreed output vectors, i.e. the presentation of sufficiently close input vectors would produce the same outputs. The learning process is, therefore, highlights the statistical properties of the training set and groups similar vectors into classes. Conclusions and Relevance We conclude that when predicting energy consumption in conditions of limited resources and rising prices, the methods based on the use of neural networks turn out to be the most effective. This method allows detecting patterns of events, hidden levers pushing the studied parameters in either direction. Despite the apparent simplicity, this analysis has a number of nuances, which, if disregarded, lead to serious errors in the interpretation of results.
Avtomatizirovannye sistemy upravleniya v energosberezhenii (opyt razrabotki) [Automated control systems in energy saving (development experience)]. Chelyabinsk, SUSU Publ., 2010, 228 p.
Aksenov S.V., Novosel'tsev V.B. Organizatsiya i ispol'zovanie neironnykh setei (metody i tekhnologii) [Organization and the use of neural networks (methods and technologies)]. Tomsk, NTL Publ., 2006, 128 p.
Afanas'ev V.N., Yuzbashev M.M. Analiz vremennykh ryadov i prognozirovanie [Time series analysis and forecasting]. Moscow, Finansy i statistika Publ., 2001, 228 p.
Beer A.S. Kibernetika i upravlenie proizvodstvom [Cybernetics and Management]. Moscow, Nauka Publ., 1965, 392 p.
Box G., Jenkins G. Analiz vremennykh ryadov. Prognoz i upravlenie [Time Series Analysis: Forecast and Control]. Moscow, Mir Publ., 2009, 604 p.
Borovikov V.P. Neironnye seti. STATISTICA Neural Networks: Metodologiya i tekhnologii sovremennogo analiza dannykh [Neural networks. STATISTICA Neural Networks: methodology and technology of modern data analysis]. Moscow, Goryachaya liniya-Telekom Publ., 2008, 392 p.
Borovikov V.P. Populyarnoe vvedenie v programmu STATISTICA [A popular introduction to STATISTICA software package]. Moscow, Komp'yuterPress Publ., 2008, 267 p.
Galushkin A.I. Neironnye seti. Osnovy teorii [Neural networks. Basic theory]. Moscow, Goryachaya liniya-Telekom Publ., 2010, 496 p.
Callan R. Osnovnye kontseptsii neironnykh setei [The Essence of Neural Networks]. Moscow, Vil'yams Publ., 2001, 287 p.
Kendall M. Vremennye ryady [Time Series]. Moscow, Finansy i statistika Publ., 1981, 199 p.
Krug P.G. Neironnye seti i neirokomp'yutery [Neural networks and neurocomputers]. Moscow, MPEI Publ., 2002, 176 p.
Kudrin B.I., Zhilin B.V., Lagutkin O.E., Oshurkov M.G. Tsenologicheskoe opredelenie parametrov elektropotrebleniya mnogonomenklaturnykh proizvodstv [Cenologic determination of parameters of electrical energy consumption by multiproduct manufactures]. Tula, Priokskoe knizhnoe izdatel'stvo Publ., 1994, 122 p.
Kuznetsov Yu.A., Perova V.I., Vorob'eva E.V. Neirosetevoe modelirovanie finansovo-ekonomicheskoi deyatel'nosti krupneishikh kompanii Povolzh'ya [Neural network modeling of financial and economic activity of largest companies in the Volga region]. Ekonomicheskii analiz: teoriya i praktika = Economic Analysis: Theory and Practice, 2011, no. 35, pp. 25–36.
Neironnye seti [Neural networks]. Available at: Link. (In Russ.)
Neironnye seti i ikh ustroistvo [Neural networks and their organization]. Available at: Link. (In Russ.)
Osnovy iskusstvennykh neironnykh setei [Fundamentals of artificial neural networks]. Available at: Link. (In Russ.)
Pervye shagi v STATISTICA Automated Neural Networks (SANN) [First steps in STATISTICA Automated Neural Networks (SANN)]. Available at: Link. (In Russ.)
Prognozirovanie prodazh polufabrikatov [Forecasting the sales of semi-finished products]. Available at: Link. (In Russ.)
Prognozirovanie s pomoshch'yu neironnykh setei [Forecasting by using neural networks]. Available at: Link. (In Russ.)
Rossiev A.A. Modelirovanie dannykh pri pomoshchi krivykh dlya vosstanovleniya probelov v tablitsakh. V kn.: Metody neiroinformatiki [Data modeling by using curves to repair gaps in tables. In: Methods of neuroinformatics]. Krasnoyarsk, KSTU Publ., 1998, pp. 6–22.
Rutkovskaya D. Neironnye seti, geneticheskie algoritmy i nechetkie sistemy [Neural networks, genetic algorithms, and fuzzy systems]. Moscow, Goryachaya liniya-Telekom Publ., 452 p.
Tarkhov D.A. Neironnye seti. Modeli i algoritmy [Neural networks. Models and algorithms]. Moscow, Radiotekhnika Publ., 2005, 256 p.
Urazbakhtin I.G., Rykov N.I. Algoritm predobrabotki dannykh dlya tselei prognozirovaniya na baze tekhnologii neironnykh setei. V kn.: Sovremennye instrumental'nye sistemy, informatsionnye tekhnologii i innovatsii [Data preprocessing algorithm for forecasting purposes based on the neural networks technology. In: Modern tooling systems, information technology and innovation]. Kursk, KurskSTU Publ., 2003, pp. 64–67.
Uskov A.A., Kuz'min A.V. Intellektual'nye tekhnologii upravleniya. Iskusstvennye neironnye seti i nechetkaya logika [Intelligent control technologies. Artificial neural networks and fuzzy logic]. Moscow, Goryachaya liniya-Telekom Publ., 2004, 143 p.
Haykin S. Neironnye seti. Polnyi kurs [Neural Networks and Learning Machines]. Moscow, Vil'yams Publ., 2006, 1104 p.
Khalafyan A.A. STATISTICA. Statisticheskii analiz dannykh [STATISTICA. Statistical analysis of data]. Moscow, Binom-Press Publ., 2007, 512 p.
Shabonin P.V. Metody nauchnogo prognozirovaniya i ikh prakticheskoe primenenie [Methods of scientific forecasting and their application]. TOP-MANAGER, 2006, no. 3, pp. 15–19.
Yuan Y., Shaw M.J. Induction of Fuzzy Decision Trees. Fuzzy Sets and Systems, 1995, no. 69, pp. 125–139.
Wang L.X., Mendel J.M. Generating Fuzzy Rules by Learning From Examples. IEEE Transactions on Systems,Man and Cybernetics, 1992, vol. 22, no. 6, pp. 1414–1427.
