April 2010 – March 2012 VRL Analysis DRAFTED BY Southwest Power Pool David Gray, Market Engineer Justin Cochran, Market Engineer James Lemley, Market Monitor Barbara Stroope, Market Monitor PUBLISHED: 08.01.2012 Copyright © 2012 by Southwest Power Pool, Inc. All rights reserved Contents Executive Summary .............................................................................................................................. 3 1. Background ...................................................................................................................................... 3 2. Analysis of OC Violations Characteristics in the Market ................................................................. 8 2011 VRL Analysis Page 2 of 10 Executive Summary The resource dispatch engine applied within the SPP EIS market includes Violation Relaxation Limits (VRLs) to assist in quickly arriving at the best possible dispatch solution when it is infeasible to simultaneously enforce all of the defined constraints. The categories of VRL include: 1) Operational constraint (OC) 2) Resource ramp rate limit 3) Generation-to-load balance 4) Resource maximum/minimum output The resource maximum/minimum VRLs experienced no violation occurrences during the period of April 2010-March 2012. The generation-to-load balance VRL experienced no violation occurrences during the period of April 2010-March 2012. These hours were re-priced. The system-wide ramp rate shortage VRL was applied a total of 356 times during the period. Instead of using the ramp VRL value directly, all LIPs during up-ramp violations were set to the highest cleared offer and downramp violations were set to the lowest cleared offer. The OC VRL most directly impacts the OC (flowgate) violations, and has been quantified from a single block of $1,500 /MW to five blocks of $500/MW for 100% loading, $750/MW for 101% loading, $1000/MW for 102% loading, $1250/MW for 103% loading, $1500/MW for 104%+ loading as of March 25, 2011. From April 2010 – March 2011, OC violations occurred for 5.07% of all congested instances with an average shadow price of $1391.92. From April 2011 – March 2012, OC violations occurred for 3.77 % of all congested instances with an average shadow price of $788.60. For the purposes of this document, the term VRL will represent the Operating Constraint (OC) VRL unless otherwise stated. The data examined and analysis conducted by SPP of the use of block VRL results in the following preliminary conclusions and recommendations: The use of block VRL has resulted in a small decrease in violated instances, and at the same time transmission usage has increased greatly and has reduced congestion cost and pricing volatility. SPP recommends keeping the current block VRLs of 100%, 101%, 102%, 103%, and 104% loading with the OC penalties of $500/MW, $750/MW, $1000/MW, $1250/MW, and $1500/MW. SPP will continue to monitor and analyze VRL occurrences, related market parameters and circumstances going forward. 2011 VRL Analysis Page 3 of 10 1. Background The resource dispatch engine used by the SPP EIS market is named the Scheduling, Pricing & Dispatching module, or “SPD”. It is a linear programming (LP) optimization algorithm that includes an optimization equation and a series of constraint equations. The optimization equation for the SPD engine is structured to minimize the total ‘cost’ of dispatching resources (based on offered prices) to meet loads in the market net of scheduled interchanges. SPD has a companion module named the Simultaneous Feasibility Test (SFT) engine which is a security-constrained economic dispatch (SCED1). The LP constraint equations reflect the following categories of constraint for the dispatch solution: MW loading limit of each activated Operational Constraint (OC) (flowgates and other transmission network branches) MW per minute ramp rate limitation for each resource offering into the market as defined in market participant resource plans; Balancing of generation (resource) MW to load MW (including interchange) within the market footprint; Maximum and Minimum MW output levels for each resource as defined in market participant resource plans. The majority of LP optimization algorithms include some form of Violation Relaxation Limits (VRLs2) to assist in quickly arriving at a ‘best possible’ solution when it is mathematically impossible to strictly enforce every constraint. The algorithm will arrive at a final solution more quickly if instructed that constraint violations translate to significant ‘indirect’ costs. This is true even though the specific ‘$ per unit’ value assigned to VRLs generally is not specifically part of the final dispatch solution. Rather, the VRL signals the dispatch solution engine to stop trying to find a lower cost solution. The SPD market simulation reflects a large-scale optimization. The conflict in constraint requirements is resolved through relaxing calculation constraints in the VRL process. The levels at which the VRLs are quantified signal to the LP algorithm a ‘comparable cost penalty’ for each category of violation. This essentially provides a prioritization to the LP algorithm, since the optimization then interprets that a MW of violation within one constraint category is ‘more costly’ or ‘less costly’ than violation of another constraint category. ________________________________ 1 Various aspects of the SCED process applied within the EIS market dispatch are discussed within Section 4.1 of Attachment AE to the SPP OATT. The SFT engine provides updated parameters to the SPD engine for each five-minute EIS market interval. 2 VRL is specifically a term applied by SPP for the SPD engine of the EIS market; Operations Research professionals will often refer to these parameters as the “cost of slack”, meaning the same as “cost of relaxation of the limit”. 2011 VRL Analysis Page 4 of 10 The four categories of VRLs applied in the Real-Time EIS Market are presently quantified as summarized in Table 1 below: Table 1: Categories of VRLs applied in the SPP Real-Time EIS Market Violation Category VRL $ level and unit of measure Operational Constraint Limit (i.e., flowgate or other network branch $500 per MW of loading at 100% for each OC loading) $750 per MW of loading at 101% for each OC $1000 per MW of loading at 102% for each OC $1250 per MW of loading at 103% for each OC $1500 per MW of loading at 104%+ for each OC Resource Ramp Rate Limit Generation – to – Load Balance Resource Maximum/Minimum Limit $5,000 per MW for each resource $50,000 per MW in each deployment solution $100,000 per MW for each resource Quantifying the OC (flowgate) constraint VRL to exhibit the comparatively lowest $/MW magnitude of the four categories, and the resource ramp rate limit VRL as next in magnitude, is largely based on recognized and unavoidable trade-offs of the violation categories. While SPP pays great attention to minimizing all the violations, short-term OC violations can be and are managed in coordination with the transmission loading relief (TLR) procedures and other established operating procedures implemented in coordination with Transmission Owners and other regional parties such as a Congestion Management Event (CME). Altering the relative magnitude of the VRLs such that the resource ramp rate limit VRL were to be the ‘lowest cost’ would result in significantly (and at times dramatically) larger amounts of dispatch outside of defined ramp rate limits, which had been discussed by the Market Working Group as an undesired trade-off. The same would be true of reordering the magnitude of the generation– – to– – load balance VRL or the resource maximum/minimum MW limit VRL. It would result in undesired overall trade-off of the violations experienced. Thus, all else being equal, the LP dispatch engine will inherently interpret that: A MW of OC violation presents a lower indirect penalty than a MW of resource ramp rate violation A MW of ramp rate violation presents a lower indirect penalty than a MW violation of market generation to market load balance A MW of generation to load balance violation reflects a lower indirect penalty than a MW violation of resource max to min limit. To further understand the VRL parameter, it is useful to address the concept of the operational constraint shadow price. The OC Shadow Price is effectively the marginal cost of a re-dispatch in the market impacting an operating constraint that has reached its MW loading limit (expressed as $ per MW of loading). The equation immediately below reflects a generalized formulation of the OC Shadow Price when a single operational constraint has reached its limit within the dispatch solution. 2011 VRL Analysis Page 5 of 10 Because the shadow price is affected by the inverse of the marginal impact of re-dispatch on constraint loading (i.e., the shift factors are in the divisor of this equation), the OC Shadow Price can reach a relatively high level before the offer prices are of a similar magnitude. Shadow Price C = (OP MR1 – OP MR2) / (Shift Factor C, MR1 - Shift Factor C, MR2) Where, MR = Marginal resource OP = Offer price OP MRX = Marginal offer price of marginal resource X, and Shift Factor C, MRX = Shift factor for marginal resource X applying to C (shadow price) As an example of the above formulation, let us assume that one of two marginal resources has a shift factor of +0.01 and the other marginal resource has a shift factor of -0.01. This means that a redispatch in which one of the resources increases output by 50 MW and the other resource decreases output by 50 MW would impact loading on the constraint by 1 MW.3 If the difference of the offer prices for these two marginal resources (OPMR1 – OPMR2) is $20 per MWh, the net cost impact (C) of re-dispatching these resources expressed per MW of OC loading impact is $20 / 0.02 = $1,000 per MW per hour. (= the OC Shadow Price if that OC is binding within the dispatch solution). It is useful to note that because the locational price solutions are based on the marginal resources dispatched, the details of the OC VRL are applied somewhat differently within the locational price solution than the other three VRL categories, as summarized below: The OC VRL effectively places a cap on the marginal cost of a market dispatch impacting the OC loading, and may also be characterized as a cap on the shadow price of the constraint. Thus, a larger OC VRL effectively instructs the SPD dispatch engine to seek higher-cost redispatches to remove loading violations from the dispatch solution, typically resulting in lower levels of flowgate overload. If a constraint continues to exceed its loading limit within the dispatch solution, a marginal re-dispatch cost higher (perhaps dramatically higher) than the assumed VRL would be necessary to elicit additional relief and the solution allows the limit to be temporarily exceeded, rather than calculating a higher cost dispatch. For the other constraint types (ramp rate limit, generation-to-load balance, and resource max/min limits), the level of penalty or ‘slack’ reflected in the dispatch solution is ‘frozen’ for purposes of the marginal computations, , which effectively removes these VRL ‘violation costs’ from the final LIP computations by effectively pricing based upon the marginal resources dispatched.4 Thus, establishing the $$ /MW value of the OC VRL involves an inherent trade-off between the frequency and MW levels of violations experienced and the magnitudes of price separation experienced. __________________________________________________________________________ 3 (50 MW * 0.01) – (50 MW * - 0.01) = 1 MW of Δ in constraint loading This mechanism provides a further perspective on why the specific values used for these three categories of VRLs are not directly relevant to the final price solution 4 2011 VRL Analysis Page 6 of 10 The Locational Imbalance Price (LIP) is the price administered to the resource. The System Marginal Price “SMP” is the average reference price for the system calculated directly within the dispatch engine. The Shift factor is the reflect impact of delivering generation to the loading constraint k relative to the delivery of the reference in a sum weighted average. Shadow Price represents the marginal cost impact on incrementally loading a constraint. LIPi SMP k [Shift Factori,k Shadow Price k ] LIP= Locational Imbalance Price SMP = System Marginal Price I = Resource K= Constraint The shadow price component affects the resources Location Imbalance Pricing (LIP) only during congestion. When congestion occurs it gives the flowgate a negative Shadow Price and signals the dispatch engine to increase or decrease generation on the constraint tangent upon what’s the resource shift factor to the relative constraint. 2011 VRL Analysis Page 7 of 10 2. Analysis of OC Violation Characteristics in the Market Due to the changing of OC VRL values in March 2011 to the block VRLs from the single $1500 value, the data has been separated between April 2010 through March 2011 and April 2011 through March 2012. Figure 1 and 2 shows the total OC Constraint Violation Instances by the percentage the OC was beyond the OC in an increasing percent format below the graph. A violation is any time the flowgate loading is above the flowgate limit within the dispatch solution. There are a total of 11,608 flowgate violation instances from April 2010 through March 2012. For April 2010 through March 2011 ($1500 single VRL), there were a total of 5,990 violation instances representing 5.07% of all congested instances. The average shadow price of the violated instances for this time range was $1391.92 and an average violation of 5.27%. For April 2011 through March 2012 (block VRL), there were a total of 5,618 violation instances representing 3.77% of all congested instances. The average shadow price of the violation instances for this time range was $788.60 and an average violation of 3.12%. Figures 1 and 2 show the number and percent of OC violates by percent category for the two time frames studied. FIGURE 1 – 5,990 Total OC Violation Instances – Single $1500 VRL 2011 VRL Analysis Page 8 of 10 FIGURE 2 – 5,618 Total OC Violation Instances – Block VRL Figure 3 and 4 shows the distribution of 255,539 Shadow Prices instances for Binding intervals from April 2010 through March 2012. Binding intervals are defined as 5-minute intervals where the flow across the flowgate exactly equals the limit of the flowgate. Price separation occurs and is reported because the algorithm is looking for the next lowest cost segment of the next available unit. The cost associated with the displacement of the initial marginal unit is the incremental cost to calculate a penalty free LIP. For April 2010 through March 2011 ($1500 single VRL), there were a total of 112,062 binding instances with an average shadow price of $98.57. For April 2011 through March 2012 (block VRL), there were a total of 143,477 binding instances with an average shadow price of $62.57. Figure 3 illustrates that 97.27 percent of all binding instances had a shadow price below $500/MW. Shadow prices between $500/MW and $1,000/MW accounted for 2.06 percent. While only 0.67 percent of all binding instances occurred above $1,000/MW. Figure 3 shows that shadow price instances over $500/MW (2,277 or 2.73% of total binding instances) additional intervals might have violated if the OC were changed to VRL Block format with a tradeoff of lowering the shadow price for these binding instances. Fig Figure 4 illustrates that 97.77 percent of all binding instances had a shadow price below $300/MW. Shadow prices between $300/MW and $400/MW accounted for 1.56% percent. While only 0.67 percent of all binding instances occurred above $400/MW. 2011 VRL Analysis Page 9 of 10 FIGURE 3 – 112,062 TOTAL BINDING INSTANCES – Single $1500 VRL $1000-$1500 SP 0.67% $500-$1000 SP 2.06% $0-$500 SP 97.27% FIGURE 4 – 143,477 TOTAL BINDING INSTANCES – Block VRL $400+ SP 0.67% $300-$400 SP 1.56% $200-$300 SP 5.09% $100-$200 SP 14.32% 2011 VRL Analysis $0-$100 SP 78.36% Page 10 of 10