The Influence of Demand Resource Response Time in Balancing Wind and Load

Transcription

1 th Hawaii International Conference on System Sciences The Influence of Demand Resource Response Time in Balancing Wind and Load Judith Cardell Smith College Lindsay Anderson Cornell University Abstract The integration of demand response resources into wholesale electricity markets facilitates the growth in wind power integration. Available demand resources have different capabilities in terms of response time, as demonstrated by the variety of programs ranging from day-ahead markets to dynamic pricing. This paper analyzes the importance of selecting the most appropriate demand response resources, DRR, for balancing wind variability at different time scales. The paper first considers the power spectral density of wind power and electric load for ISOne region, and identifies benefits in using demand response programs to mitigate wind variability. This comparison provides insight into the time scales most valuable for balancing wind variability. Next, the paper presents empirical tests using OPF and Monte Carlo simulations to analyze the use of DRR for mitigating wind uncertainty. Results show that an optimal combination of hour ahead and 10-minute DRR eliminates price spikes at low and moderate wind penetrations while significantly reducing price spike events at 30% wind penetration. 1. Introduction Industry and policy makers are increasingly acknowledging that for the power system to successfully evolve into a more efficient and renewable-energy-based system, the power industry as well as customers must adopt a significant amount of demand response. At the federal level, in compliance with the Energy Independence and Security Act, the Federal Energy Regulatory Commission, FERC, and the US Department of Energy, DOE, have developed a National Action Plan of Demand Response, [1], followed by a proposed implementation of the plan [2]. In summarizing state activities, these reports identify patterns within state legislatures for requiring utilities to rely on demand response before turning to construction of new generation to meet demand growth. FERC also highlighted the important role of publicity and education to gain acceptance of the demand response programs. Many of the programs surveyed discussed the use of DRR at different time scales, including the use of day ahead and hour ahead demand response resource scheduling. For shorter time scale demand response, the proposed implementation of the national action plan identifies the DOE SGIG (Smart Grid Investment Grant) program for use in supporting dynamic pricing studies, customer education and data analysis. The use of the SGIG to fund demand response research firmly ties demand response to smart grid development efforts, and encourages the use of modern, enabling technologies for demand response implementation. In Order 745 [3], FERC addressed the need to ensure fair compensation for demand response resources (DRR) and developed a net benefits test to be used by ISOs in determining the cost-effectiveness of DRR in organized wholesale energy markets. Though not required in that order, FERC acknowledged that determining the cost-effectiveness of DRR as part of the dispatch process could be an improvement on the otherwise mandated use of a monthly LMP threshold. To this end, FERC has required each ISO to submit a report, by September 2012, investigating the use of dispatch algorithms to dynamically determine the costeffectiveness of DRR in their system. Most recently, FERC has proposed to adopt the North American Energy Standards Board s methods for the measurement and verification of demand response in energy markets [4]. The inclusion of these standards in federal policy will affect meter data accuracy and reporting as well as telemetry standards, thus upgrading the technical infrastructure as required for increased use of dynamic pricing. These state demand response programs, federal reports and policies variously address the use of day ahead, hour ahead and shorter term dynamic demand response resources, programs and pricing methodologies, without necessarily distinguishing the role of, or benefits from, resources with these different response times. This paper explicitly distinguishes between hour ahead and 10 minute demand response resources and their integration into a power system /12 $ IEEE DOI /HICSS

2 with up to 30% wind penetration by energy. The analysis presented here investigates the benefits of using a portfolio of demand response resources to mitigate the variability in wind power generation. Sections 2 and 3 of this paper discuss the power spectral density, PSD, in wind resource and demand time series data, with section 2 presenting the analysis methodology and section 3 presenting the PSD results. Sections 4 and 5 present the empirical system analysis, with section 4 discussing the modeling framework and section 5 presenting results of the OPF-plus-Monte Carlo simulations. Section 6 summarizes findings and concludes the paper. 2. Power Spectrum Analysis Background Fourier Analysis is one of the most common methods for time series analysis. Representation of the time series in the frequency domain provides unique insights into the various underlying cycles of the time series [5]. The resulting periodogram is a representation of the contribution to the power variance at each frequency of the time series. Examining the periodogram can give us a sense of the frequencies at which the power density or variance is greatest. Subsequently, this may provide an indication of the relevance and/or value of demand response resources at various timescales. In this paper, we use the power spectral density analysis to motivate empirical system studies of responsive demand applications in various scenarios of wind penetration and system conditions. The various scenarios considered are outlined in section 4. The power spectral density analysis was first proposed by [6], though was not widely used until the growth of computational power made the calculation of discrete Fourier transforms feasible for realistic problems through the Fast Fourier transform (FFT) algorithms. The method was first successfully applied the time series analysis of power system loads in [7]. The most recent application of power spectra analysis in power systems is [8]. The analysis provided in [8] considers the spectrum of power from wind turbines at sampling frequencies of 1-second and 1-hour from two locations, concluding that lower frequency spectra contain more significant power variance than the higher frequencies, though the literature seems more concerned with the higher frequency events. Averaged Periodogram site neg2 Averaged Periodogram site neg5 Averaged Periodogram site neg7 Averaged Periodogram Wind Data aggregated Wind data Figure 1: Periodogram for (a) New England, site 2, (b) New England, Site 5, (c) New England Site 7, and (d) All New England sites aggregated output

3 The study in [8] concludes that it is important to consider the lower order frequencies when considering the type of ramping capabilities required from traditional generators on the system. The work discussed here applies a similar analysis to a number of potential wind sites in the ISOne power system, with the objective of determining the type of responsive demand resource that would be most valuable to the system. The simulated wind data for the ISOne region is obtained from the Eastern Wind Integration Study (EWITS) from the National Renewable Energy Lab (NREL) [9]. Simulated wind output data is collected for seven potential wind sites in New England and the discrete Fourier transform (DFT) is applied, via FFT, to transform the data from the time domain to frequency domain. The periodogram points are calculated to provide an overall estimate of the contribution of each frequency to the power of the time series [5], [8]. The periodogram is constructed for each individual site, the aggregated wind output for all sites, as well as the regional system load for the ISOne [10]. Wind data is collected at a ten minute sample rate, and load data is provided at a five minute sample rate. 3. Spectral Analysis Results The power spectral density (PSD) for three of the seven studied wind sites are shown in Figure 1, along with the PSD for the wind output from all sites combined. Sites shown in Figure 1(a) and 1(c) are the most geographically distant from each other. Figure 1(b) shows results from a simulated site located between the sites represented by Figures 1(a) and (c). Examining the plots in Figure 1, we can see that the geographic location of the sites does not appear to have significant effect on the spectral density, or the important frequencies of fluctuations. This is despite the fact that correlation between wind outputs at these sites is in the range of 0.4, and the standard error of the time series decreases by nearly 20% when the wind output is combined from the two sites. It is important to note that the frequencies that contribute most to power variance in this system are lower order than those typically believed to be important for power system operation. In Figure 2, the PSD for aggregated wind output is compared with that of the system load (for May/June 2012). It is clear from this plot that the decay of the power density occurs at a similar rate for both load and aggregated wind output on the ISOne system. Peaks in the spectra occur consistently at frequencies coinciding with a daily cycle, indicating that these cycles are particularly important. The hourly cycles of wind output occur at frequencies near 3e -04 Hz, which has a much lower spectral density. To clarify the location of these cycles at daily, weekly and annual frequencies, annotations on Figure 2 highlight the relevant frequencies. Averaged Aggregated Wind Data vs Averaged Load Data 1/f 1 month 1/f 1 week 1/f 1 day Load Data Averaged Aggregated Wind Data 10 2 Figure 2: Annotated Periodogram comparison for aggregated wind output and system load. Intuition provided from Figure 2 indicate that the daily and weekly cycles are likely more significant in wind output variability than the shorter term, 10-minute fluctuations that occur at ~2e -02 Hz. Though hourly cycles have greater power density than 10-minute cycles. This indicates that there is significant value in earlier-dispatch demand side resources (hourly), and that improvements in longer term forecasting could provide significant system benefits in the integration of renewables. 4. Power System Modeling Framework The power spectrum density analysis explored the underlying spectral patterns in the wind generation and electrical demand time series data. The results of these sections indicate that for mitigating wind variability it is at least as important to mitigate the slower moving variations, from a day to a month, as it is to mitigate the shorter 10-minute and real time market variability. This section discusses the use of the wind and demand time series data in simulating the integration of wind power into power system operation. The objective of the simulations is to investigate system performance with high wind penetration and responsive demand. The system performance is reported in terms of locational marginal price (LMP), the amount of demand response used and the amount

4 of wind energy spilled, as required, for system convergence. The analysis is performed using the IEEE 39-bus test system with 10%, 20% and 30% wind penetration, shown in Figure 3. The red lines indicate the approximate geographic regions in New England, with the generation technology mix shown distributed across the region, with five hypothetical wind farms are located as shown. The yellow lines indicate transmission constraints, used in half of the system simulations. Figure 3: IEEE 39-bus test system. Figure 4: Network Re-Dispatch in a Monte Carlo Framework The simulation proceeds in two stages, as shown in Figure 4, where stage 1 represents the hour ahead market and stage 2 represents the ten-minute market. In stage 1, an hour ahead dispatch schedule for the 39-bus test system is determined assuming a specific hour ahead wind forecast, and using the MATPower OPF program [11]. In stage 2, Monte Carlo simulation is used to draw multiple realizations of the expected forecast error between the forecasts made one hour and 10-minutes ahead of real time. The uncertainty in wind generation is represented through a wind generation forecast error based on NREL-EWITS wind data [9] and a forecasting algorithm defined in [12] At this 10-minute (stage 2) time step, the system is re-dispatched using the OPF model, and the system thus requires a response to the wind forecast error from the remaining generators, as well as demand resources. On the supply-side, generators are constrained to respond within their 10-minute ramping limits. For the demand-side, all resources are assumed to be interruptible at a cost of $10,000/MWh. In order to analyze the benefits of using demand response resources with different response capabilities, the following demand response options are defined: No demand response is available (other than $10,000/MWh interruptible load). Hour ahead demand response dispatched to meet an optimal percentage of the expected shortfall in wind generation, due to forecasting errors [12]. 10-minute demand response made available, at a cost of $100/MWh, to meet the expected 10-minute wind generation shortfall. Both hour ahead and 10-minute ahead demand response resources made available. For the analysis of overall system performance with wind integrated into the system, the following two extreme scenarios are defined: Constrained system: Transmission constraints are binding into two load pockets as shown with the yellow lines in Figure 3; no demand response is available. Flexible system: Transmission constraints are relieved; both hour ahead and 10-minute demand response resources are available. Finally, in order to better examine the influence of different demand response resources on system performance, two additional scenarios, either including or removing the hour ahead demand response option as relevant, are added. Thus the final set of options examined include: (1) Constrained system (2) Constrained system with hour ahead DRR (3) Flexible system (4) Flexible system without hour ahead DRR Each scenario set also includes the three levels of wind penetration, 10%, 20% and 30%, as well as three load levels, defined in terms of reserve margins, 10%, 15% and 30%. Table 1 summarizes these scenarios

5 Transmission Constraints No Transmission Constraints No Demand Response HA demand response (1) (2) 10- minute demand response Both HA and 10min (4) (3) Table 1: Simulation Scenarios, 10%, 20% and 30% wind penetration; 10%, 15% and 30% reserve margin analyzed with all 4 scenarios. With this range of scenario attributes and Monte Carlo draws the total number of OPF simulations per scenario exceeds 200,000. The results of these simulations are presented in section Power System Simulations One important measure of power system performance is the price of electricity, or LMP. Charts of the LMP for the set of Monte Carlo simulations in Scenarios (1) and (3), the constrained and flexible systems, are shown below in Figures 5 and 6. The x- axis for most of these charts is seen to extend to $10,000/MWh. This spiking of LMP to $10,000 indicates that use of the expensive, interruptible demand was required in order to ensure system energy balance. For most of these scenarios, there are relatively few price spike events, resulting in a vanishingly small bar on the charts. However, the fact that the x-axis does extend to $10,000 shows that there were some price spike events in the indicated scenario. Significantly, these figures show that the only system configurations that do not require the use of this expensive demand to balance the system are those for Scenario (3), the flexible system, and only for 10% (and 20%, not shown) wind penetration. As Scenario (3) is the one with access to both categories of demand response (in an unconstrained transmission system), these results indicate that both categories of demand response hour ahead and 10-minute resources are in fact required to avoid price spikes. This is seen in the top of Figure 6, in which the x-axis does not extend to $10,000/MWh, but rather is limited to a maximum LMP of $80/MWh. (The x-axis scale maximum is set at $100 rather than the $10,000 of the other charts to allow the actual LMP distribution to be seen.) These results also show that even with an unconstrained transmission system, and access to both categories of demand response, in a system with 30% wind penetration there will still be some price spikes. The extent of these events is investigated next. Figure 5: Scenario 1 LMP, for 10% (top) and 30% (bottom) wind penetration Comparing the results across all scenarios at the 30% wind penetration level allows further examination of the role of the hour ahead demand response in system performance. For this analysis, the comparison of system performance is made based on the likelihood that compensating for the wind variability will require the system to call upon the $10,000/MWh interruptible demand resources. Looking at Table 2, the results of the Monte Carlo simulations show that for the most constrained configuration (Scenario 1), the system must dispatch expensive load 4% of the time in order to maintain the energy balance. This requirement is decreased by one quarter, to a 3% likelihood, with the inclusion of modest amount of hour ahead demand response modeled in these simulations. (A moderate level of hour ahead demand response resources, ranging from 2% to 6% of total load, are assumed to be available.)

6 Figure 6: Scenario 3 LMP, for 10% (top) and 30% (bottom) wind penetration Scenario 10% Wind 20% Wind 30% Wind Table 2: Percent likelihood of price spike with 30% wind penetration; LMP $10,000 Looking at the unconstrained system configuration (Scenario 3), and again comparing the system at the 30% wind penetration level with and without the availability of hour ahead demand response resources, Scenarios (3) and (4) reveal even more dramatic results. For the most flexible system (Scenario 3), the use of expensive demand to maintain the energy balance is required with a modest 0.4% likelihood. These events lead to the x-axis extending to $10,000 in Figure 6(c). However, removing access to the hour ahead demand resources (Scenario 4), even with no transmission constraints and maintaining access to 10- minute demand resources, significantly increases the likelihood of requiring the use of expensive demand resources. As seen in the bottom row of Table 2 for Scenario (4), the likelihood of needing expensive resources to balance the system increases to 4% - the same likelihood as for the constrained system of Scenario (1). Looking at all the cases presented, Table 2 shows the percent likelihood that the system will require the use of expensive interruptible load to balance the system. For the 10% wind penetration, this likelihood is extremely remote, at 10-3 %. However, these rare events remain causes of concern for the market. At 30% wind penetration, though 4% likelihood does not represent a significant number of market time periods, any price spike events are cause for concern in ISO wholesale energy markets. If such events are associated with increased wind power, they could hinder the inclusion of intermittent renewable resources into the power grid. These results suggest though that use of hour ahead and 10-minute demand resources optimized together can reduce and even eliminate price spike events. Continuing with the comparison of the high LMP events, Table 3 shows the average of the demand response used, both hour ahead and 10-minute, during the price spike events. Note that the system makes full use of the available demand response resources for Scenario 3, leading to the highest average MW amount for the system in this least constrained configuration. Scenario 10% Wind 20% Wind 30% Wind Table 3: Average of Demand Response Use (MW) During Price Spike Events For the high LMP events discussed in Tables 2 and 3, the power system is unable to find sufficient generating resources to both meet demand and cover the wind power shortfall. In these cases, expensive interruptible demand is called upon to maintain the energy balance. At the 10% and 20% wind penetration levels, this is the only action required calling upon interruptible load. The final observation of interest in this analysis is the system s occasional need to spill wind even when generating resources are limited. At the 30% wind penetration level, with transmission constraints (Scenarios 1 and 2), there are a limited number of situations where the system is sufficiently constrained that it must spill otherwise available wind power in order to obey Kirchhoff s laws. These events occur

7 only 0.6% of the time in Scenarios (1) and (2), only at the 30% wind penetration level. (It is also interesting to observe that for Scenario (4), the system can only find a solution during high LMP events if it spills wind power 0.06% of the time.) 7. Acknowledgement This material is based upon work supported in part by the U.S. Department of Energy Consortium for Electric Reliability Technology Solutions (CERTS). 6. Conclusions The analysis of the power spectral density in the wind power and electrical demand time series data shows that fluctuations evolving over the one hour to one month time periods have a higher power density than do faster cycles, though these faster fluctuations tend to receive the greatest attention. The analysis presented here indicates that attention should also be paid to mitigating the longer term fluctuations. Following up on these observations, Monte Carlo simulations with the IEEE 39-bus test system show that slower moving hour ahead demand response resources are in fact required to dramatically reduce the number of price spike events. Reliance on the ramping capabilities of conventional generation along with short-term demand response, or relying on generation ramping with only hour ahead demand response will not eliminate the price spikes. Both categories of demand response resources, along with conventional generator ramping, are required. The possible identification of wind energy as the cause of price spikes in electricity markets contributes to the reluctance to increase wind penetration in the power grid. The results discussed above show that the hour ahead demand response resources, even at a modest level, play a significant role in moderating price spikes and maintaining system energy balance to mitigate wind uncertainty. 8. References [1] FERC, National Action Plan on Demand Response, Accessed May, vol. 27, p. 2011, [2] FERC, Implementation Proposal for The National Action Plan on Demand Response, pp. 1 34, Jul [3] FERC, Demand Response Compensation in Organized Wholesale Energy Markets, pp , Mar [4] FERC, Notice of Proposed Rulemaking:, 139th ed. no FERC 61,041, 2012, pp [5] W. T. Vetterling, Numerical Recipes in C. Cambridge Univ Pr, [6] A. Schuster, On the investigation of hidden periodicities with application to a supposed 26 day period of meteorological phenomena, Terrestrial Magnetism, vol. 3, no. 1, pp , [7] D. J. Trudnowski, W. L. McReynolds, and J. M. Johnson, Real-time very short-term load prediction for power-system automatic generation control, Control Systems Technology, IEEE Transactions on, vol. 9, no. 2, pp , [8] J. Apt, The spectrum of power from wind turbines, Journal of Power Sources, vol. 169, no. 2, pp , [9] M. Brower, Development of Eastern Regional Wind Resource and Wind Plant Output Datasets: March 3, March 31, 2010, National Renewable Energy Laboratory, Dec [10] I. N. England, Acquiring Operational Data from ISO New England Inc. Website, pp. 1 85, May [11] R. D. Zimmerman, Matpower 4.0b4 User s Manual, pp , May [12] C. L. Anderson and J. B. Cardell, A Decision Framework for Optimal Pairing of Wind and Demand Response Resources, IEEE Systems Journal, pp