2016 Probabilistic Assessment. December 5, 2016 Southwest Power Pool

Transcription

1 2016 Probabilistic Assessment December 5, 2016 Southwest Power Pool

2 Table of Contents 1. Summary...2 a. SPP Planning Coordinator Footprint...2 b. Seasonal Capacity Totals... Error! Bookmark not defined. c. Forecasted Total Internal Demand... Error! Bookmark not defined. d. Net Energy for Load... Error! Bookmark not defined. e. SPP Metric Results...3 f. Simulation Results Software Model Description...4 a. Computational Approach...4 b. Algorithm Usage...4 c. Tiered Simulation...5 d. External Modeling Demand Modeling...6 a. Probabilistic Assessment and LTRA Differences...6 b. Load Modeling...7 c. Load Forecast Uncertainty...7 d. Behind-the-Meter Generation Capacity Modeling...10 a. Probabilistic Assessment and LTRA Capacities...10 b. Determination of Firm and Deliverable Capacity...11 c. Jointly Owned Units...11 d. Capacity Purchases and Sales...11 e. Intermittent and Energy-limited Variable Resources...11 f. Controllable Capacity Demand Response...11 g. Traditional Dispatchable Capacity...12 h. Operating Reserves Transmission Modelling Assistance from External Resources Sensitivity Case...15 SPP 2016 Probabilistic Assessment 1

3 1. Summary The Southwest Power Pool (SPP) 2016 Probabilistic Assessment is a mandatory study requested by North American Electric Reliability Corporation (NERC), under the guidance of the Reliability Assessment Subcommittee (RAS). The objective of this assessment is to provide a common set of probabilistic reliability indices that could be used to supplement NERC s Long-Term Reliability Assessments (LTRA). The SPP Assessment Area is synonymous with the reporting region of the LTRA. This study is based on the 2016 LTRA data, which included entities in the SPP Planning Coordinator (PC) footprint. SPP used GridView version 9.2, an ABB application, to perform the probabilistic analysis for study years 3 and 5, 2018 and 2020, respectively. Two metric results were calculated in this study: annual Loss of Load Hours (LOLH) and Expected Unserved Energy (EUE). a. SPP Planning Coordinator Footprint The 2016 Probabilistic Assessment was performed on the SPP Planning Coordinator (PC) footprint. Figure 1 shows a geographical representation of the SPP PC footprint which includes all or parts of Arkansas, Iowa, Kansas, Louisiana, Minnesota, Missouri, Montana, Nebraska, New Mexico, North Dakota, Oklahoma, South Dakota, and Texas. Figure 1: SPP Planning Coordinator Footprint SPP 2016 Probabilistic Assessment 2

4 b. SPP Metric Results LOLH is the Hourly Loss-Of-Load expectation. This metric provides the hours of resource reliability shortfall per year, which is the time in hours that the demand exceeds the capacity throughout the year. This value is a summation of hourly events for a 24-hour period that a loss of load event occurred. Daily LOLE is the expected daily occurrences of the loss of load throughout the year. Per the SPP Criteria, generation reliability assessments examine the regional ability to maintain a Loss of Load Expectation (LOLE) standard of one day in 10 years (0.1 day/year). EUE is the expected amount of megawatt-hours of load that will not be served in a given year. This is the summation of the expected amount of unserved energy during the time that the demand exceeds the capacity throughout the study year. GridView provides Expected Energy Not Served (EENS), which equates to EUE. c. Simulation Results The values in the table below reflect the results of the GridView LOLH/LOLE simulations for both study years (2018, 2020), as well as the two reserve margins types for the NERC LTRA. Based upon the given LTRA values, no loss of load was shown in the SPP footprint for 2018 and Since the annual results reflect zero loss of load events, the monthly LOLH, LOLE, and EUE values for 2018 and 2020 were zero as well. The sensitivity case for both study years resulted in no loss of load events. Simulation Results SPP Region Forecast Planning Reserve Margin (%) Forecast Operable Reserve Margin (%) Base Case Sensitivity Case % 22.2% 22.0% 17.4% 17.2% 15.0% 14.9% 10.5% LOLE (days/year) LOLH (hours/year) EUE (MWh/year) SPP 2016 Probabilistic Assessment 3

5 2. Software Model Description a. Computational Approach GridView 9.2 was used to perform the analysis. GridView is a software application developed by ABB Inc. to simulate the economic dispatch of an electric power system, while monitoring key transmission elements for each hour. GridView can be used to study the operational and planning issues facing regulated utilities, as well as competitive electric markets. The key advantage of using the GridView application is having the ability to model a detailed transmission system in the study region, not just a transportation model. The transmission model allows for realistic power delivery based on actual modeled limits on transmission lines imported from powerflow models. Some other features available in this program include contingency constraints, nomograms, and emergency imports. A sequential Monte Carlo simulation was used to perform the analysis of the SPP reliability assessment. Figure 2: Diagram showing different types of probabilistic modelling b. Algorithm Usage Monte Carlo simulation is a method for iteratively evaluating a deterministic model using sets of random numbers as inputs. The goal is to determine how random variation or uncertainty affects the reliability of the system that is being modeled. Monte Carlo simulation is categorized as a sampling method, because the inputs are randomly generated from probability distributions to simulate the process of sampling from an actual population. Within GridView, Monte Carlo simulation allows detailed modeling of the pre-contingency conditions and outages of generation and/or transmission equipment and/or changes in demand, fuel prices, and/or wind generation. GridView can also model the correlation between area load demands and fuel prices. It uses probability distributions for equipment outages during a sequential mode of simulations hour by hour, and typically for a year. The selection of testing conditions is by random sampling. In order to obtain accurate risk indices, many simulations will have to be performed (3,000 simulations/year for this assessment). In SPP 2016 Probabilistic Assessment 4

6 general, the simulations provide the loss of load reliability indices. For the reliability assessment, a linear programming model is applied to the generation dispatch calculation for every hour in each trial in order to compute the amount of load shed in order to eliminate transmission overload problems. c. Tiered Simulation The 2016 Probabilistic Assessment was modeled in a tiered approach that became more and more constraining as the study advanced. This was akin to multiple scenarios for SPP, and allowed the footprint to be analyzed under constricting parameters. The first simulation type was a transmission model both internal and external to the SPP PC footprint. No limits were attached to the transmission model, and any amount of power necessary could flow to solve the case. Next, important flowgates and interfaces within SPP and between SPP and its first-tier areas were added to the model. With these, flow was monitored on these lines and kept within the appropriate transmission line ratings. The final tier was to monitor and enforce ratings on all 100kV and above transmission lines within SPP and crossing interfaces with its first-tier areas. The 100kV distinction was chosen based on the definition for the Bulk Electric System 1. Since SPP models a nodal representation of the transmission system with bus level data, the tired simulation approach was utilized to reduce the amount of modelling errors and address any issues for specific data entries for the study. d. External Modeling GridView allows external areas to be modeled in the same fashion as internal areas. The key difference between the two is that external generation is ignored when selecting random outages and external load is not increased by load uncertainty factors during the Monte-Carlo simulations. External transmission, however, was considered for calculating flow on lines. SPP assumes zero non-firm support from external regions. The external capacity modelled was provided as firm capacity which is reflective of the values provided in the 2016 LTRA. 1 SPP 2016 Probabilistic Assessment 5

7 3. Demand Modeling a. Probabilistic Assessment and LTRA Demand and Energy The SPP 2016 LTRA forecasted seasonal values are listed below, along with simulation results from GridView. 50/50 seasonal peak demands Total Internal Demand Summer (MW) Winter (MW) Summer (MW) Winter (MW) LTRA 52,819 41,379 53,409 41,445 Simulation 51,023 39,588 51,593 39,651 Difference 1,796 1,791 1,816 1,794 Net Energy for Load Net Energy (GWh) (GWh) LTRA 252, ,148 Simulation 244, ,405 Difference -8,595-8,743 There are three reasons the reported 2016 LTRA and simulation demand values are different. For the simulation, total internal demand, which excluded the projected available demand response, was used with demand response being explicitly modeled as generation, as described in section 4f. Secondly, GridView only allows for the adjustment of the annual peak demand, which occurs during the summer for SPP. When the annual peak is adjusted, the winter peak and every other hour will be adjusted by a proportional amount, based on the hourly load profile. This functionality prevents the winter peak value from aligning with what is provided in the LTRA. Lastly, the total internal demand reported in the LTRA is the aggregation of multiple peaks from entities within SPP. To produce an SPP coincident peak, a 96.6% peak demand ratio was applied to the non-coincidental peak demand. This diversity factor was derived from six years of historical hourly load data. SPP 2016 Probabilistic Assessment 6

8 The difference between the net energy and the LTRA is also attributable to the proportional adjustment of the hourly load profile. b. Demand Modeling SPP modelled a projected 8,760 hourly demand profile to provide load variability and volatility for chronological hours during simulation. The demand curve is based on the historical hourly data from year 2012, which resulted in the highest duration of load hours compared to its peak hour. Figure 3 shows load duration curves compared to each year s peak for the upper 10 th percentile of Figure 3: Load duration curves for the upper tenth percentile A 96.6% peak demand ratio was applied to the forecasted total internal demand for 2018 and 2020 provided in the LTRA to produce a SPP coincident peak. The 96.6% peak demand ratio was derived from historical hourly load profiles. Each year s non-coincident peak was divided into the coincident peak demand to produce demand ratios. The averaged ratio was applied to the SPP peak load hour for simulation. The total internal demand for 2018 and 2020 is based on a 50/50 forecast and no out-of-region load was modeled in this assessment. c. Load Forecast Uncertainty i. Method GridView allows for two options in dealing with load uncertainty: 1) User defined uncertainty pattern, and 2) probability distribution. For this study, a user-defined uncertainty pattern and a probability distribution were both used to add uncertainty to the load values. A different load uncertainty was created for each month. ii. Uncertainty Components and Applied Probability A load model was used to define the peak-load multipliers used to modify forecasted loads. The daily peak was selected and regressed against historical peak temperatures from Crystal Ball Pro was used to analyze the probability distributions of SPP 2016 Probabilistic Assessment 7

9 temperatures observed at key weather stations throughout the SPP footprint. Multipliers were calculated and were populated in a user defined uncertainty pattern. The userdefined uncertainty pattern allows users to provide seven monthly load patterns. Each area has a different value for each month multiplied by seven probabilities (a total of 84 values). GridView randomly selects the load pattern at the beginning of the simulation hour, and applies it for that trial. The same uncertainty methodology used in the 2012 and 2014 Probabilistic Assessments was used in the 2016 Probabilistic Assessment. The figures below show the SPP load uncertainty multiplier pattern applied for each month and the probability of occurrences. The randomly selected load multipliers were determined by sampling from a uniform distribution, and selecting one of seven possible multipliers. These multipliers have dramatically decreasing likelihoods (e.g., Set 1 is 50% likely, Set 2 is 19% likely, Set 7 is 0.6% likely). Multiplier Set 7 contain the highest multipliers, and are the least likely to occur. As such, this set should be considered SPP s extreme peak. No multipliers decrease the load values in this study. Multiplier Set 1 is the base case multiplier, and effectively multiplies all loads by 1. Sets 2 7 are intended to proportionally increase loads up to SPP expected extreme peaks. SPP only analyzes the positive multipliers of load forecast uncertainty, which increases the load curve during simulation, to produce a conservative approach for adequate reliability planning Probability Standard Deviation iii. Standard Deviation Probability M1 M2 M3 M4 M5 M6 M7 M8 M9 M10 M11 M % % % % % % % Load Forecast uncertainty Factors SPP 2016 Probabilistic Assessment 8

10 The load-uncertainty probability took into consideration stochastic temperature within the different areas, in addition to recognizing the structural affects that holidays, weekends, quarters, and previous hour s load have on load expectations. Other sources of uncertainty reasonably independent of temperature are modeled were considered to be sufficiently small in magnitude and not necessary at this time to model independently. A random error term was created to incorporate variability that could occur from uncertainty types such as economic industrial/commercial health and dew point. d. Behind-the-Meter Generation Behind-the-meter generation is generally netted and modeled with customer load. If the behindthe-meter generation is not netted, then it was modeled as regular generation. If the behind the meter generation was not tied to its own bus, then the capacity was divided between its associable generation units within the power flow model. SPP 2016 Probabilistic Assessment 9

11 4. Capacity Modeling a. Probabilistic Assessment and LTRA Capacities Capacity data modeled in the 2016 Probabilistic Assessment was derived from the 2016 LTRA. The table below summarizes the makeup of the capacity categories and amounts used in the assessment by study year (2018, 2020). Seasonal Capacity Totals Category Controllable and Dispatchable Demand Response -Available Summer (MW) Winter (MW) Summer (MW) Winter (MW) On-Peak Expected Capacity (Coal) 24,991 25,068 24,832 24,918 On-Peak Expected Capacity (Gas) 30,512 30,621 30,056 30,671 On-Peak Expected Capacity (Petroleum) 1,375 1,438 1,356 1,404 On-Peak Expected Capacity (Hydro) 4,848 4,558 4,848 4,558 On-Peak Expected Capacity (Nuclear) 2,029 2,031 2,029 2,031 On-Peak Expected Capacity (Wind) 1,550 2,083 1,550 2,078 On-Peak Expected Capacity (Solar) On-Peak Expected Capacity (Biomass) On-Peak Expected Capacity (Other/Unknown) Sales 3,754 3,931 3,754 3,838 Purchases 2,525 2,394 2,525 2,395 Total Capacity 65,669 65,103 65,113 65,073 Demand Response values reported in the 2016 LTRA were modelled as generating resources available during daily on-peak hours instead of reducing the Total Internal Demand. Tier 1 wind SPP 2016 Probabilistic Assessment 10

12 resources included in the LTRA with wind interconnection agreements that have not obtained firm transmission service were not included in the 2016 Probabilistic Assessment. b. Determination of Firm and Deliverable Capacity All generation, future or otherwise, must have firm transmission service in the form of a service agreement or an OASIS transaction to be modeled. c. Jointly Owned Units Jointly owned units were modeled as thermal generators, with the percentage of ownership calculated as capacity for that specific unit. GridView allows the unit s capacity to be split up between multiple owners, using the Reserve Capacity Distribution feature. If an entity within SPP owned a portion of a generator that was located in SPP, the reserve capacity distribution would show a portion going to that entity and the remainder being allocated to another entity external to the SPP footprint. The same concept was applied for external resources being allocated to entities within SPP. d. Capacity Purchases and Sales For a sale or purchase between a SPP and an outside entity, a generator was placed on the outside entity swing bus for the amount of the transaction. The generator area was changed to SPP, so that it would be included in the study. If the transaction was a sale to the outside entity, the capacity would be negative. If a purchase, the capacity would be positive. This methodology allows for capacity correction in accordance with sales and purchases modeled in SPP. e. Intermittent and Energy-limited Variable Resources Wind generation was modeled as an hourly resource using 2012 data from a combination of actual energy generation and estimated energy generation, based on National Renewable Energy Laboratory (NREL) wind-shape data, as required. The year 2012 was utilized to keep consistency with the load shape. The hourly wind shapes consist of 8,760 hourly values for year 2018 and 8,784 values for year If a wind resource was not in service for any part of a year, NREL data was used to supplement where needed. The hourly wind shapes were imported into GridView to provide an accurate profile of the wind generation output for each hour of the year applied for each wind resource. f. Controllable Capacity Demand Response SPP has controllable-capacity demand in the form of Interruptible (Curtailable) demand. In areas that reported controllable-capacity demand, equivalent thermal units were added to the model with high fuel costs assumed, so those units would be dispatched last to reflect demand-response operating scenarios. If the units were dispatched, they were limited to six consecutive on-peak hours. SPP 2016 Probabilistic Assessment 11

13 g. Traditional Dispatchable Capacity i. Ratings The maximum capacity ratings were based on 2016 LTRA data, as developed by the SPP member s capability testing. The capability testing procedure and requirements are described in SPP Criteria section ii. Forced Outage Modelling Forced outage modeling within GridView consists of using the Equivalent Forced Outage Rate demand (EFORd) values for forced outage rates. Other factors included forced outage durations and scheduled maintenance requirements taken from Energy Velocity, using data from the NERC Generating Availability Data Systems (GADS) and supplemented by Ventyx Advisors staff based on generator age. iii. Planned Outage Modeling Planned outages for thermal units were modeled by using the scheduled maintenance function in GridView to take units offline for a specified period of time based on start time, end time, and duration. Once the outage duration elapsed, the unit was placed back online in the model. The maintenance start date and outage duration were sourced from Control Room Operations Window (CROW) software SPP members use to plan maintenance outages. If generators did not have a designated outage planned, previous planned outage times were modeled. Historical planned outages were also considered when modelling the appropriate outage timeframe for resources within SPP. h. Operating Reserves Base-case simulations included foregoing any operating reserves within SPP. No additional operating procedures were included in the analysis. 2 SPP 2016 Probabilistic Assessment 12

14 5. Transmission Modelling System Topology was drawn from the 2017 SPP Integrated Transmission Planning Near Term series summer models for the 2018 and 2020 study years. GridView allows importing of transmission data from a PSS/E power flow model. For each study year (2018, 2020), separate ITPNT models were imported to represent the latest representation of the SPP transmission grid. Internal and crossing interfaces were implemented using the SPP Operation s OASIS flowgate list. Interfaces are key groups of transmission lines that should be all observed as one group, such as key transmission lines between one area and another. All 100kV and above transmission lines within SPP, as well as crossing interfaces with SPP s first-tier areas, were monitored to not exceed transmission ratings. The 100kV distinction was chosen based on the definition for the Bulk Electric System 3. Internal and crossing flowgates for years 2018 and 2020 were implemented using the Outage Transfer Distribution Factor (OTDF) and Power Transfer Distribution Factor (PTDF) flowgates outlined in the SPP OPS OASIS flowgate list. Transmission additions and retirements were captured in the ITPNT models built with SPP member input and modeling. Transmission additions were modeled and retirements were removed from the GridView models. If retirements were announced after the LTRA posting, they were retired in the GridView case. 3 SPP 2016 Probabilistic Assessment 13

15 6. Assistance from External Resources For this Probabilistic Assessment, it was assumed that SPP does not rely on non-firm assistance from resources outside of the SPP assessment area footprint, consistent with the LTRA report s values. SPP assumes zero non-firm support from external regions. The external capacity modelled was provided as firm capacity which is reflective of the values provided in the 2016 LTRA. SPP 2016 Probabilistic Assessment 14

16 7. Sensitivity Case The sensitivity case simulations required an increase of forecasted demand and energy from the base case. A 2% increase in forecasted demand and energy was applied for the 2018 study year. A 4% increase in demand and 2% increase in energy were applied for the 2020 study year. The increases were applied to the forecasted peak hour for the SPP region. All other assumptions remained unchanged from the base-case assumptions. The sensitivity case for both study years resulted in no loss of load events. Since the annual results for the sensitivity case reflect zero loss of load events, the monthly LOLH, LOLE, and EUE values for 2018 and 2020 were zero as well. SPP 2016 Probabilistic Assessment 15