The system imbalance prediction

Transcription

1 POSTER 2015, PRAGUE MAY 14 1 The system imbalance prediction Štěpán KRATOCHVÍL 1, Jan BEJBL 2 1,2 Dept. of Economy, management and humanities, Czech Technical University, Technická 2, Praha, Czech Republic kratoste@fel.cvut.cz, bejblja1@ fel.cvut.cz Abstract. This paper investigates the most important features of the system imbalance needed for appropriate and accurate forecast. At first, there is described time series of the system imbalance and other inputs, where relevant descriptive statistics are presented. The analysis of the development of imbalances in recent years showed that speculation to achieve imbalance in opposite direction to system imbalance is an interesting business opportunity that balance responsible parties increasingly use. There are used multiple exogenous variables in order to get new information into possible forecast model. All exogenous variables are analysed to evaluate the impact of each input variable on the system imbalance forecast. Keywords System imbalance, prediction, renewables, correlation. 1. Introduction Every trader, generator operator or generally power market participant has several opportunities how to act (gain profit) in the electricity market. There are long-term strategies like futures or forwards enabling sell or buy electricity in year, month and week contracts. Next to them, there are short-term strategies, which consist mainly from the day ahead market serving to close the position in the sense of balance the portfolio to cover all the customers needs and consume all the electricity from the suppliers. Next market in short-term strategies is the intraday market for speculation on the price deviation and compensation of the deviations caused by the change in expected supply and consumption of the electricity. The last part is the balancing market organized by the TSO (Transmission system operator) for buying upward and downward regulation price for the balancing system demand and consumption in the real time. The balancing market is operated by the TSO, who is responsible for balancing electricity supply and demand in the real-time for entire electricity network. This amount of energy is called system imbalance and is created by the summation of the system imbalances of the all balance responsible parties (BRP). BRP is the market participant that has taken the responsibility for balancing the portfolio of generation and consumption. BPR are obliged to submit their energy program in the day before the day of the delivery and are penalized for each deviation from this plan. This so called Imbalance settlement process gives to BRP incentive to balance their portfolio with the highest precision. The time unit for which BRP balance their portfolio is called PTU (Program time unit). Our focus is based on using of the exogenous variables. We will use multiple inputs describing the deviations in the planned supply and the behavior of the other BRP. Inputs describing the deviations in supply will be aimed on the RES, specifically wind and PV power plants. Inputs describing the BRP behavior will be dayahead price, intraday price and balancing energy bought by the TSO. Besides these factors was also analyzed the influence of import and export of the electricity and differential of reality and predictions of their balance. Of course there are other factors that can not be directly quantified. This could include the ability of market participants to learn and respond to the strategy of the other participants, etc. 2. System imbalance definition in the Czech Republic At first, we will define the BRP imbalance. BRP imbalance is defined in Decree 541/2005 Sb., About electricity market rules as the difference between scheduled value of energy supplied into the grid and the actual supply or the difference between scheduled valued of demand and actual demand respectively. Supply is marked by the positive sign and the demand by the negative. For each PTU (program time unit), which is for Czech Republic one hour, is the imbalance set by Operator of the market (OTE) with the precision of the one decimal. System imbalance is defined as the sum of the BRP imbalances. But this is not accurate. To define system imbalance, we need to modify the BRP imbalance equation by adding import and export variables, since these variables are considered while calculating system imbalance, lost in the transmission network and regulation energy. This methodology covers all the areas of the system imbalance and set the value for the each hour.

2 2 Š. KRATOCHVÍL, J. BEJBL, THE SYSTEM IMBALANCE PREDICTION From the histogram presented in figure 1 can be seen that the positive system imbalance (system is in the surplus) is more often than the negative system imbalances. system imbalance (1. and 2. cases) and on the figure 3, where BRP imbalance is in the opposite direction as the system imbalance (3. and 4. cases). Figure 1. System imbalances volumes histogram from till System imbalance in the Czech Republic The Czech Republic system imbalance is priced in four basic cases: BRP positive imbalance (surplus) with positive system imbalance (surplus) BRP negative imbalance (shortage) with negative system imbalance (shortage) BRP positive imbalance (surplus) with negative system imbalance (shortage) BRP negative imbalance (shortage) with positive system imbalance (surplus) All of these cases are priced according to the different methodology. payment is calculated by multiplying the system imbalance price and the system imbalance volume. In the rest of the chapter, we will use positive payment prices, for payment paid to BRP and negative payment prices for payment paid from BRP to TSO. For example, if the BRP has negative imbalance and the imbalance price is positive the total payment has negative sign and the BRP pay to the TSO. For the negative imbalance price the TSO pay to BRP. The most often regulation energy comes from primary and secondary regulation services. Prices for regulation power are set by ERU and are 2350 CZK/MWh for upward regulation and -1 CZK/MWh for downward regulation. So in these hours the price for the cases will be following: IP=-1-3,5*SI CZK/MWh IP=2350+5,5*SI CZK/MWh IP=-1 CZK/MWh IP=2350 CZK/MWh The tertiary regulation service changes the price, if it is activated. The prices for BRP imbalances are shown on the figure 2, where BRP imbalance is in the same direction as the Figure 2. BRP imbalances prices histogram from till For the BRP imbalances in the same direction as system imbalance. Figure 3. BRP imbalances prices histogram from till For the BRP imbalances in the opposite direction to the system imbalance. 4. Exogenous variables We will differ from state of art papers by incorporating multiple exogenous variables into our prediction model. The variables can be divided into three main classes: demand variable, supply variables and variables incorporating behavior of the others market participants. 4.1 Demand variable This variable is presented on the web page of the Czech TSO and is the summation of all demand of all consumers in the Czech Republic power grid. The data are presented hourly in one hour delay. The planned amount of demand is presented day ahead in the web system of the TSO. By taking the difference between planed value and actual demand we will get demand contribution to the system imbalance creation. 4.2 Supply variable These variables describe supply side of the energy market. This can be described generally as the summation of the all generators or partially supply The actual supply is also presented on the web page of the Czech TSO in one hour delay data and the planned

3 POSTER 2015, PRAGUE MAY 14 3 value is presented in the web system of the TSO as the demand variable. These data will be handled in the same way as the demand by taking difference between planned value and actual measured. The result will be the contribution to the system imbalance creation RES supply We will aim on the supply from RES, which power cannot be controlled and are very hard to forecast. Here we use the actual production and forecast presented in 01:00 and 13:00 for the next day. The RES used in our work will be the German wind power, German solar power and Czech solar power. We assume that strong connection between Germany and Czech Republic power markets lead to the fact that Czech system imbalance is influenced by the imbalance in the supply of the german RES. Information about German Balancing market can be found in Müsgens et al. (2014). 4.3 Market participant s behavior This variable capture the behavior all the market participants and their own personal forecast of the future development Intraday market price (IMP) The intraday market has the role of last chance to balance BRP position and make a speculation on the system imbalance. When the system imbalance is significantly negative (there is shortage of energy), the price on the intraday market climb up and opposite for the positive system imbalance. Taking in mind the hourly average price, we can observe the behavior of the market participants and speculate in the same direction (direction of the BRP imbalance). The close time of the intraday market is one hour before delivery, so there is enough time to modify the amount of generated electricity Balancing market price (BMP) This is the market, where the TSO buy the balancing power for the next hour (close time is 30 minutes before delivery). So the amount and type of regulation energy (upward or downward) carry the information of the TSO system imbalance forecast for the next hour Day ahead market price (SDAMP) The spot day-ahead market price reflects the position of the generators of the electricity. When the spot price is high, most of the power sources are on the maximum level of generation and there is not enough space to increase the amount of generated electricity. In the situation, that the negative system imbalance occurs, the power sources are not able to increase their generation and lower the system imbalance. The system imbalance is then more stable. The opposite situation is for low spot price, which leads to generation on the minimum level and inability to decrease the amount of generated electricity. Mean Median Std. Dev. Skewness Kurtosis Min Max 25% qtl 75% qtl Quantity IMP ,33 0, SDAMP ,11 0, BMP+ 19,62 15,00 15,56 1,90 5,29 2,00 100,00 10,00 28, BMP- -25,25-20,00 21,89-2,41 8,24-170,5-1,0-31,0-10, BMP -4,39 0,00 20,17-1,57 10,08-170,5 100,0-8,00 0, PV CZ reality 496,0 386,8 444,9 0,63-0,80 0, PV CZ diff 28,27 19,89 156,9-0,40 1,87-737,4 657,48-17,28 107, PV DE reality ,68-0, PV Germany diff ,54 5, W Germany reality ,33 1, W Germany AVD -2,61-23,99 941,29 0,23 2, W Germany diff ,79 1, supply ,00-0, supply AVD -0,82-24,90 267,50 0,50 1, Consumption real ,03-0, Consumption - AVD -0,57-35,07 287,20 0,63 1, export-import real ,06-0, export-import diff -2,94-1,61 44,58-0,05 9,30-378,38 376,84-20,96 16, Tab. 1. Descriptive statistics of the time series of exogenous variables from till Source: own calculations.

4 4 Š. KRATOCHVÍL, J. BEJBL, THE SYSTEM IMBALANCE PREDICTION Where: Diff refers to difference between reality and prediction, AVD refers to adjacent values differences. It is obvious that the descriptive characteristics of the individual inputs differ. This fact is not an obstacle to their use in modeling the expected system imbalance. Different characteristics suggest that each entry must be approached in another way and one generalized approach is not suitable. 5. Correlation analysis We calculate the correlation between used inputs and system imbalance to assess their relevance. First of all we calculate autocorrelation coefficient of the system imbalance because its autocorrelation prediction promises high accuracy. Autocorrelation coefficients are shown on the figure 4. Sufficient autocorrelation achieve only first values of delay. The highest value achieves one hour delay with the value 0,78. Two hours delay achieves the value 0,52 and three hours delay 0,34. This analyses shows that useable will be only one or two hours delay data. We gain the best results for the Czech PV data, intraday market price, balancing market price and differences between prediction and reality of the export minus import. Generally are the values relatively low, which is the reason why we focus on the other way of how to predict system imbalance target interval prediction. In this way we sorted all the input data into two and five intervals. Two intervals only serves to separate the positive and negative system imbalance. Dividing into 5 intervals brings some improvements. Firstly, there is the middle interval to prevent speculation on the system imbalance close to zero (there is too much risk changing direction of the system imbalance by our potential speculation). Furthermore, both positive and negative system imbalance is divided into two intervals allowing to control the level of risk to our speculation (for the lower absolute value of the imbalance is greater risk than for the higher values). Figure 4. Autocorrelation of the system imbalance in the period from to with the delay 1 to 176 hours. Moreover we calculate correlation coefficients between the system imbalance and others inputs. The results shows, that German RES inputs are not suitable for the prediction model. By the prediction of the system imbalance to maximize the profit it is not necessary to know the precise value, but we only need to know the interval in which the future value will fluctuate. Delay 0 hours 1 hour 2 hours Delay 0 hours 1 hour 2 hour IMP -0,429-0,496-0,536 wind GER real. 0,067 0,062 0,058 SDAMP -0,174-0,161-0,137 wind GER AVD -0,022-0,026-0,020 BMP+ -0,328-0,397-0,276 wind GER diff -0,013-0,002 0,001 BMP- -0,384-0,502-0,407 supply real. -0,114-0,109-0,093 BMP -0,477-0,556-0,519 supply AVD -0,061 0,021 0,079 solar CZ real. 0,202 0,184 0,144 cons. real. -0,140-0,125-0,105 solar CZ diff -0,443-0,399-0,297 cons. AVD -0,080 0,055 0,074 solar DE real. 0,069 0,062 0,051 export-import real. 0,005 0,019 0,019 solar DE diff -0,095-0,096-0,090 export-import diff 0,575 0,380 0,236 Tab. 2. Correlation coefficients of input data. Source: own calculations

5 POSTER 2015, PRAGUE MAY Interval data distribution In this chapter we will show results of distribution to the intervals. In two interval distribution we set the boundary of the system imbalance on zero value. For five intervals distribution we marked intervals by numbers -2, - 1 for negative imbalance, for the middle interval 0 and 1 and 2 for the positive imbalance. For the middle interval we chose values between -20 / 20 MW. For values of the intervals separating the positive and negative imbalance, we chose values for defining the extreme values of 10 % and thus limits are -80 respectively 115 MW. Intervals of the system imbalance are thus defined: Interval number -2: <-, -80) Interval number -1: <-80, -20) Interval number 0: <-20, 20) Interval number 1: <20, 115) Interval number 2: <115, ) Individual input data differ in the quantity. This is due to the need of filtering the data with zero real trades by business data and the photovoltaics data, which were filtered hours with zero production. The most important are boldfaced values indicating the accuracy of the individual input prediction system imbalance in both absolute and relative terms. First are the numbers of hours in which the interval system imbalance assign the relevant interval input data. Furthermore is given the precision with which the use of 5 intervals determine the correct direction of the imbalance (if the imbalance is negative, the interval input variables come -2 or -1, and vice versa for positive imbalance). Last bold figure shows the accuracy correctly stated direction of the system imbalance. To optimize the interval limits were used mathematical tool Mathematica, which enabled setting such limits, which cause the highest possible accuracy of the model. This means that was maximized the number of cases when the values of the system imbalance of the given limits are the same to the values of other input factors attributable to this. The number of correctly stated values indicates the annexed table. 6. Conclusion For the prediction of the system imbalance have been identified factors of demand, supply and the factors determining the behavior of participants on the electricity market. Autocorrelation analyses shows, that sufficient results achieve only values of delay of maximum two hours. This increases the correlation between system imbalance and other input variables, however the greatest correlation coefficient reached maximum 0,6, so we divided the data into two and five intervals. This dividing allows controlling the risk incurred. The lower absolute value of the system imbalance implies greater risk level. In the case five interval distribution are the best results for intraday market prices, day ahead spot market and for the differences between prediction and real PV production in the Czech Republic. For two intervals based on the best by BMP-, difference between prediction a real export import and for the intraday market price. By properly adjusted weights for each factor we will get prediction model taking into account the most important "variables" affecting it. Acknowledgements Research described in the paper was supported by the Czech Grant Agency under grant No. SGS14/138/OHK5/2T/13. References [1] Olsson, M.; Soder, L., "Modeling Real-Time Balancing Power Market Prices Using Combined SARIMA and Markov Processes," Power Systems, IEEE Transactions on, vol.23, no.2, pp.443,450, May 2008 doi: /TPWRS [2] Van der Veen, R. A C; Hakvoort, R.A., "Balance responsibility and imbalance settlement in Northern Europe An evaluation," Energy Market, EEM th International Conference on the European, vol., no., pp.1,6, May [3] Reinier A.C. van der Veen, Alireza Abbasy, Rudi A. Hakvoort, A comparison of imbalance settlement designs and results of Germany and the Netherlands, Young Energy Engineers & Economists Seminar (YEEES), 8-9 April 2010, Cambridge, UK [4] Garcia, M.P.; Kirschen, D.S., "Forecasting system imbalance volumes in competitive electricity markets," Power Systems, IEEE Transactions on, vol.21, no.1, pp.240,248, Feb About Authors Štěpán KRATOCHVÍL is the Ph.D. student of the Czech Technical University at the Faculty of electrotechnic at the department of the Economy, management and humanities. He works on his dissertation on the topic of the modeling and forecasting electricity spot prices and its volatilities. He is the student of the 3. Grade and prepare for achieving the final examination. Jan BEJBL was born in Benešov. in Management of power engineering (Ing.) at CTU. Since 2011 he is a PhD. student at the same department. In his dissertation he aims at algorithms pricing of energy commodities.

6 6 Š. KRATOCHVÍL, J. BEJBL, THE SYSTEM IMBALANCE PREDICTION Appendix Interval distribution input data IMP SDAP BMP+ BMP- BMP solar CZ real. solar CZ diff solar DE real. solar DE diff wind DE real. wind DE RSH wind DE diff supply real. supply RSH consum ption rea consump tion. RSH exportimport real exportimport diff Quantity interval boundary-2/ interval boundary -1/ interval boundary 0/ interval boundary 1/ correctly stated intervals 5 intervals ,84% 39,95% 26,71% 18,68% 23,39% 21,28% 48,03% 17,82% 29,42% 31,86% 28,38% 34,83% 16,49% 24,40% 35,76% 31,01% 18,85% 28,38% Correctly stated -2. interval Correctly stated -1. interval Correctly stated 0. interval Correctly stated 1. interval Correctly stated 2. interval correctly stated imbalances 5 intervals ,91% 61,22% 69,01% 83,57% 2,39% 39,49% 33,78% 42,50% 36,77% 36,69% 40,96% 53,65% 38,78% 7,83% 46,30% 59,04% 33,61% 20,41% interval boundary -1/ correctly stated intervals 2 intervals ,49% 61,93% 71,92% 82,85% 40,78% 55,04% 66,17% 52,17% 55,78% 61,43% 57,42% 56,46% 56,44% 46,96% 60,82% 55,16% 61,40% 71,38% Correctly stated -1. interval Correctly stated 1. interval