Turkish electricity market short term market clearing price forecasting

Considering ever-changing technological infrastructure and the effects of this change to daily lives; adaptation process in the last twenty years continued with high momentum. In this regard, the notion of "energy" has become much more prominent with in such globalizing and developing world. Increase in electricity demand, parallel with the evolution mentioned above, makes itself apparent more effective especially after the liberalization of the electricity market. Turkish Electricity Market (TEM) entered into the liberalization process with the Law no.4628 in 2001. In this specialized and liberalized market context, it is aimed to form a market based on transparency and capability to reflect true Market Clearing Price to investors. In time, parallel to developing market dynamics, correct "Day Ahead Market - Market Clearing Price" forecasting has become essential for companies to manage their operations. Balancing Power Market is another sub-electricity market structure in Turkish Electricity Market under existing circumstances. However, within the context of ancillary services instructions, directive of constraints cause Balancing Power Market to have more complicated electricity market structure compared to Day Ahead Market. In Turkish Electricity Market, especially, activity of source based electiricty power-plants and programmed or non-programmed maintenance-breakdown details of electricity power-plants can not be shared by system operator transiently. Considering this situation, forecasting System Marjinal Price which occurs in Balancing Power Market hourly is difficult in Turkish Electricity Market which has unsettled market structure and data quality. Electricity generation and supply investors are willing to participate in a market which is transparent and volatile. Therefore, investment grade of electricity market would increase for investors as the market structure become more liberal. The reason behind this tendency is that, considering "energy" as a political decision making element composing states' primary strategies both in home and foreign affairs, investors wish to have less regulation effect of state on such liberalized electricity market. In this context, significant number of reforms implemented for liberalization of the electricity market in Turkey since 2001. Parallel to these reforms, number of investors both in generation and supply part increased excessively during this period. One of the most important necessities for electiricity market comes out with investors -and volatility in turn- is the accurate Reference Price forecasting. Based on the structure of Reference Price forecasting, investors take short, mid and long term positions aiming to maximize their profit and minimize the risks in financial management of electricity contracts. Electricity generation and consumption activities are the two main factors that effect formation of Reference Price in electricity market. However, detailed hourly data of electricity generation and consumption activities not accessible publicy in Turkish Electricity Market (TEM) which has not completed its liberalization process yet. Moreover, even Turkish Electricity Market completes its liberalization process, data sets subject to electricity energy would not be allowed to share publicly.The reason behind this restriction is that data sets includes shareholders' strategic commercial activities and informations with primary importance. Becasue it is formed as a result of electricity generation & consumption interaction and carrying the trends of these intaractions in its time series, Market Clearing Price which is published publicly by Market & Financial Settlement Center, used in thesis work. Furthermore, Market Clearing Price notion is also known as Reference Price in commercial activities of electricity market. Especially, mid season, day type diversity and socio-economic factors effect electricity generation and consumption behavior. Despite those factors were not used as data type in a mathematical model of this thesis execution, it is aimed to measure the effect of Market Clearing Price to Turkish Electicity Market in the model by utilizing Artificial Neural Networks' learning and dynamic output capability, and testing maturity level of Turkish Electicity Market which has not completed liberalization process yet. Besides, another purpose of the study is to give guidance to electricity investment companies on their short and mid-term operations, by concentrating on short term electricity market price forecasting. For the "Day Ahead Market - Market Clearing Price" forecasting, there are two approaches generated from analytical studies based on electricity market simulation and mathematical models. In this thesis study, Turkish Electiricity Market short-term Market Clearing Price forecasting analyzed by using Levenberg-Marquardt Back Propagation Artificial Neural Network (ANN) algorithm containing different neuron amounts with 3,5,7 and 9 hidden layers. The study was carried out through sampling both terms –Winter and Summer- by one week for each. January 2012 – July 2014 time period was set for the study. January 1st, 2012- January 5th, 2014 time period for Winter and January 1st, 2012 – July 6st, 2014 time period for Summer was used as forecasting training data set; January 6th-12th, 2014 time period for Winter, July 7th-13th, 2014 time period for Summer was used as testing data set. Market Clearing Prices of January 2012 - July 2014 time period disclosed by Market & Financial Settlement Center forms main data set of the model. As a consequence, Artificial Neural Network models which have 7 and 9 neurons in their hidden layer succeed hourly average 4,7% MAPE value in Winter term and an Artificial Neural Network models which have 5 and 9 neurons in their hidden layer succeed hourly average 6,3% MAPE value in Summer term. On the other hand, for all neuron types, hourly 4,9% and daily 2% average MAPE for Winter term and hourly 6,4% and daily 3,9% average MAPE for Summer term estimated in this thesis work. These results show that approach of "Increasing number of neuron supports having more successful and sensitive results in forecasting." can not be assumed as a concrete approach. Additioanlly, difference in estimated error value between Winter and Summer terms indicates that seasonal factors effect Reference Price forecasting. These factors also effect increase or decrease in MAPE values. In future works, to achieve better results with Artificial Neural Network, forecasting method may divided into groups based on season, month, week, day and hour separately and modelled with different numbers of neurons. The reasons of estimated MAPE deviations are; In Daily Head Market, there is only one Market Clearing Price occured for each hour in national base, therefore Market Clearing Price which has used as a data set in this thesis work could not represent regional constraints accurately. In addition to this, the mathematical model of this thesis work does not have both source and stakeholder based hourly generation and programmed or non-programmed maintenance-breakdown data sets of power-plants. Moreover, even those data sets involved in the mathematical model of this thesis work, because regulatory structure in electricity energy operations have high incidence of influence, the Reference Price forecasting deviation in MAPE results can not be dropped below acceptable levels. Electricity market participants have financial obligations of hourly base settlement results because of Day Ahead Market activities and imbalance of demand forecasting in present conditions of Turkish Electricity Market structure. In this context, accurate Reference Price forecasting has critical importance for investment party - whether generation or supplier company of electricity market -, to manage its near future financial operations. Turkish Electricity Market will be head for recorganization in near future. Especially, energy exchange notion will come through by implementation of Energy Market Operation Incorporated Company (EPİAŞ), Intraday Market which is going to be another sub-electricity market in Turkish Electricity Market will contribute more controlled supply and demand operations in market structure and the increasing Base Load Electricity Futures transaction volume will contribute Turkish Electricity Market to be more powerful and dynamic in line with its liberalization objectives. In this context, this thesis work shows that Artificial Neural Network provides indicative short term Reference Price forecasting in Turkish Electricity Market for weekly financial operations of market participants. In thesis sections, following topics were examined: Importance of the study, reason why it was chosen and the aim of the thesis were mentioned in the first part. Besides, through scanning literature studies conducted up to date, studies that may be adapted to Turkish Electricity Market were analyzed. In the second part, starting from formation of the notions of 'value' and 'price', generation of the 'market' concept for electricity, structure of electricity market in Europe and Turkey were examined. In the third part, along with applicability of Artificial Neural Networks to electiricity market; general structure, algorithms and mathematical models were investigated. In the fourth section, after searching factors affecting formation of Reference Price, selected Artificial Neural Netwok was applied to Turkish Electricity Market. The data was detailed for analysis. In the fifth and final part, findings of the thesis and topics that will shed light on the future were mentioned.