پیش بینی قیمت روز بعد برق در بازار انرژی اسپانیا با استفاده از شبکه های عصبی مصنوعی
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|15311||2008||10 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Engineering Applications of Artificial Intelligence, Volume 21, Issue 1, February 2008, Pages 53–62
In this paper, next-day hourly forecasts are calculated for the energy price in the electricity production market of Spain. The methodology used to achieve these forecasts is based on artificial neural networks, which have been used successfully in recent years in many forecasting applications. The days to be forecast include working days as well as weekends and holidays, due to the fact that energy price has different behaviours depending on the kind of day to be forecast. Besides, energy price time series are usually composed of too many data, which could be a problem if we are looking for a short period of time to reach an adequate forecast. In this paper, a training method for artificial neural nets is proposed, which is based on making a previous selection for the multilayer perceptron (MLP) training samples, using an ART-type neural network. The MLP is then trained and finally used to calculate forecasts. These forecasts are compared to those obtained from the well-known Box–Jenkins ARIMA forecasting method. Results show that neural nets perform better than ARIMA models, especially for weekends and holidays. Both methodologies calculate more accurate forecasts—in terms of mean absolute percentage error—for working days that for weekends and holidays. Agents involved in the electricity production market, who may need fast forecasts for the price of electricity, would benefit from the results of this study.
This paper analyses the behaviour of energy price and demand variables on the Spanish electricity production market, and then proceeds to calculate price forecasts. The methodology used to obtain these forecasts is based on artificial neural networks (ANNs), which have been used successfully in recent years in many forecasting applications (Zhang et al., 1998). Forecasts obtained using this methodology are tested against forecasts calculated with Box–Jenkins (BJ) ARIMA models. The paper focuses on short-term forecasting, as the time series to be forecast is the next-day hourly energy price. Such forecasts could be of key importance for the agents involved in the Spanish electricity production market. In the last few years, many papers have applied ANNs to short-term electricity demand forecasting (Pack et al., 1991b; Ho et al., 1992; Chen et al., 1992; Peng et al., 1992; Chow and Leung, 1996; Vermaak and Botha, 1998). Other studies calculate forecasts for temperature variables as a necessary step towards good electricity demand forecasting (Khotanzad et al., 1996; Gonzalez and Zamarreño, 2002). Yet, there is a dearth of studies on electricity markets: Nogales et al. (2002) and Contreras et al. (2003) have analysed the Spanish electricity market to achieve next-day electricity price forecasts. In particular, Contreras et al. (2003) use the ARIMA methodology to analyse Spain and California electricity markets. The paper is structured as follows: Section 2 analyses the workings of the electricity market in Spain; Section 3 provides an overview of ANNs forecasting methodology; Section 4 describes the time series that are analysed in this study and the structure of the ANNs used to forecast. Section 5 discusses the results of forecasts compared with actual data. The paper is rounded off by Section 6, which draws relevant conclusions from the study.
نتیجه گیری انگلیسی
This paper describes forecasts for a time series from the Spanish electricity production market, which is a spin-off of the liberalisation process in the electricity sector in Spain. Energy producers and energy purchasers both attend this stock exchange, they present and match purchasing and selling prices for electricity, establish prices, and settle the business of negotiating quantities to be bought and sold by the market's different agents. When forecasts for the HP series are calculated using ANNs, the problem of having too many observations arises. As this market has been working for several years, the HP time series has thousands of observations. This abundance of data makes the ANN training phase very slow, taking hours or even days, which contradicts the initial purpose of making forecasts for the next 24 h of a given day. This problem led us to design a training method in which it is possible to calculate forecasts in a few minutes. There are many other methods that try to accelerate the training process, but the advantage of the proposed method resides in the use of an ART neural net to select training examples. This kind of net has been widely used with great success to solve complex classification problems, which is why we included it in our system. This process culminates in the selection of a few samples (around 200) that are very similar to each other and to the one to be forecast. This makes the MLP training process much faster. The final results suggest that ANN forecasts with the proposed training method outperform those calculated with the BJ methodology, particularly when the days to be forecast are weekend days or national holidays. As stated in Section 2, the Spanish electricity production market was partially regulated for its first two years of functioning, and sometimes prices were established externally. Energy price variables can therefore have anomalous values for some periods during these early years. The proposed selection and training method discards the examples presenting an anomalous behaviour, so that forecasts are not affected. Finally, two points have to be highlighted: first, having accurate and fast forecasts as those calculated for the electricity price with the method proposed in this paper, could be a key factor for any of this market agents’ industrial strategies. Both energy producers and energy purchasers could benefit from having such forecasts. The second and final point to be made is that the training method proposed in this paper can be applied to forecast any time series composed of too many data, and when fast forecasts are required.