GL-SD-204 LOAD FORECAST METHODOLOGY AND PROCESSES AN OVERVIEW OF THE PROCESSES AND METHODOLOGY USED BY THE SYSTEM OPERATOR TO PREPARE THE FORECAST OF Doc Reference: GL-SD-204 17/03/2016
Version Date Change 1 28/06/2012 2 23/10/2012 Update to section 3.2.2 to reflect that MV90 data is no longer an input into the LF. Update to section 4.1 to clarify calculation of BLPFs for GXPs with wind generation. 3.0 17/6/14 Reissued, link to EA site conforming and non-conforming GXP s changed. 4.0 17/3/16 Cyclic review completed, reissued with no change 2 IMPORTANT Disclaimer The information in this document is provided in good-faith and represents the opinion of Transpower New Zealand Limited, as the System Operator, at the date of publication. Transpower New Zealand Limited does not make any representations, warranties or undertakings either express or implied, about the accuracy or the completeness of the information provided. The act of making the information available does not constitute any representation, warranty or undertaking, either express or implied. This document does not, and is not intended to; create any legal obligation or duty on Transpower New Zealand Limited. To the extent permitted by law, no liability (whether in negligence or other tort, by contract, under statute or in equity) is accepted by Transpower New Zealand Limited by reason of, or in connection with, any statement made in this document or by any actual or purported reliance on it by any party. Transpower New Zealand Limited reserves all rights, in its absolute discretion, to alter any of the information provided in this document. Copyright The concepts and information contained in this document are the property of Transpower New Zealand Limited. Reproduction of this document in whole or in part without the written permission of Transpower New Zealand is prohibited.
TABLE OF CONTENTS 1. Introduction... 4 2. Overview... 5 3. EMS Load Forecast Application... 6 3.1 Forecast Areas... 6 3.2 Data Sources... 9 3.2.1 SCADA Load Data... 9 3.2.2 3.2.3 Metered Load Data (MV90)... 9 Weather Data Actual and Forecast... 9 3.3 Operations... 10 3.3.1 Medium Term Load Forecast... 10 3.3.2 Daylight Savings... 10 3.3.3 Maintenance... 11 4. Load Forecast and the Market System... 12 4.1 Bus Load Participation Factors... 12 4.2 Market Schedules... 12 4.3 Monitoring and Adjusting the MTLF... 13 5. Glossary... 14 3
1. INTRODUCTION Under a new obligation in the Electricity Industry Participation Code associated with the Electricity Authority s demand-side bidding and forecasting (DSBF) initiative, the System Operator is required to disclose the processes and methodology used to prepare the demand forecast to the Electricity Authority and to the public. 4 This document provides an overview of the Energy Management System (EMS) 1 Load Forecast tool, its major sources of data, processing and output. The output of the EMS Load Forecast tool is used as the basis of the Medium Term Load Forecast (MTLF) within the Market System. This document also describes how information from the MTLF is used in the market schedules. 1 Not to be confused with Energy Market Services
2. OVERVIEW A GXP may be classified as conforming or non-conforming. Under the new DSBF arrangements, this determination is made by the Electricity Authority in accordance with 13.7C of the Electricity Industry Participation Code. This determination is important as it establishes how GXP load is derived in the System Operator s market schedules (see section 4.2 for more detail). The System Operator is required under 13.7A of the Electricity Industry Participation Code to prepare a forecast of demand at conforming GXPs for each trading period in a schedule period. The System Operator uses the EMS Load Forecast (LF) application to produce the forecast of demand. The LF application produces a forecast at a regional level. This regional level forecast is then broken down to GXP level using ratios based on historical metering data. The GXP level forecast is the information used by the System Operator s Scheduling, Pricing and Dispatch (SPD) tool, and is also published in accordance with 13.104(1)(a)(ii). Realtime staff have the ability to view the load forecast information and how it is tracking against actual load. Realtime staff can manually adjust the load forecast using overrides if the forecast is too far from the actual load. 5 The data flow is summarised in the diagram below. Metered data (MV90) SCADA Actuals MDB User override s MTLF Forecast data MOI Weather forecast data EMS load forecast Schedules Historic weathe r data
3. EMS LOAD FORECAST APPLICATION The EMS Load Forecast (LF) application runs in the production EMS environment, and uses aggregate load and weather data to forecast aggregate regional conforming load on a half-hourly basis. This forecast is known as the Medium Term Load Forecast (MTLF) and is used in the Market System to derive GXP load forecasts for specific schedules in SPD. 3.1 FORECAST AREAS 6 As noted above, the LF application produces a load forecast at an area (regional) level. This aggregation is intended to reduce or smooth the variability in the input data in order to improve the forecast accuracy. There are ten forecast areas currently in use. The diagrams below illustrate the North Island and South Island load area designations. Note: Non-conforming GXPs are not included in the area level forecast. A list of nonconforming GXPS as determined by the Electricity Authority can be found at http://www.ea.govt.nz/operations/wholesale/spot-pricing/conforming-and-nonconforming-gxps/.
Northland (NL) 7 Auckland (AK) Bay of Plenty (BP) Hamilton (HM) Napier (NR) Palmerston North (PN) Wellington (WN)
8 West Coast/Nelson(WC) Christchurch (CH) Invercargill (IN)
3.2 DATA SOURCES The LF application uses load and weather data to determine the area level forecast. Load data comes directly from SCADA, where the instantaneous telemetered MW values representing off-take are aggregated to the GXP level within a station (e.g. ALB0331, SFD2201, ADD0661) and then up to the forecast area level aggregates. The instantaneous forecast area aggregates are further aggregated over a 30 minute period within SCADA to produce an average for each half hour. The LF application maintains an online history within the production EMS environment of the 30-minute load averages for one year. 9 The LF previously made use of a secondary source of half-hour metered load from the MV90 metering system. This data used to be loaded into LF to replace the previous day s telemetry based average MW. MV90 data becomes available daily at around 07:00 for the previous day. It is considered to be of higher quality than the telemetered SCADA data, and is less susceptible to glitches, for example, due to remote terminal unit (RTU) failure or equipment outages. However, MV90 metering data incorporates wind generation information by treating it as negative load 2. MV90 data is no longer loaded into the LF due to the distortionary effect of wind on the load forecast. Forecast and actual weather data comes from the MetService via a set of plain text files. There are two forecasts each produced twice daily, plus one set of actual weather data. The information is provided at an area level. Data Produced Increments Horizon AM Forecast 03:00 and 09:00 6-hourly 36 hours PM Forecast 15:00 and 21:00 6-hourly 36 hours Actual Weather Daily Hourly Weather observations for each forecast area for yesterday The weather parameters include ambient temperature and wind speed for each forecast area. However, while all areas utilise temperature, only Wellington (WN), Christchurch (CH) and West Coast (WC) utilise wind speed data. Weather history is maintained within the production EMS environment for the same duration as the average load data. 2 Wind generation at the GXP is subtracted from the actual load as required under the Electricity Industry Participation Code.
3.3 OPERATIONS The data sources described in section 3.2 are used by the LF application to produce the Medium Term Load Forecast (MTLF). The MTLF is computed automatically every 30 minutes shortly after the half-hour 3 in order to guarantee that the most recent load averages have been updated. Overall the forecast spans 15 days, including the current day, and the forecast for each area is made up of a number of components: 10 Forecast Component Long Term Short Term Weather Dependent Refined Description Longer term seasonally dependent component of load. The long term component takes the previous 4 weeks load data and applies a weighting factor. Data older than two weeks has a lower weighting factor than more recent data. Data can be segmented to group days with similar load profiles together to improve the load forecast. For example, Tuesday and Thursday can be treated as effectively the same day. Short term component representing localised variations. The short term component looks at the last 2 days of load data and applies a correction based on short term load trends. This data is updated every half hour. Component of load that correlates with observed weather. The weather based component uses a linear relationship between load and weather data to further refine the load forecast. Refined component produced for the current day only, based on the observed divergence of the actual load from the forecast. The refined component compares the actual load and the forecast load for a given hour and refines the subsequent hourly loads based on the difference. This component can only be calculated once the first set of actual data for the day has been retrieved (i.e. after 00:30) 4. LF builds and maintains the correlation coefficients necessary for each of the components above based on its one year online history. Weather profiles are also constructed based on the weather history, and are used to fill the periods between the forecast points 5 and extend the weather forecast data beyond the window provided by the MetService. Daylight savings is implemented in the sense that there is explicit handling of the extra/missing hour on the daylight savings transitions. However, due to the fact that load behaviour changes after a daylight savings transition, it takes a while 6 before the change in observed load behaviour significantly affects the forecast. 3 Usually 90 seconds after the half-hour 4 This results in a bump in the forecast as it transitions through midnight, and usually means that the first trading period of the day is computed with an unsatisfactory load forecast unless a co-ordinator intervenes. 5 There are 11 points to fill between each of the official forecast values. 6 Up to two weeks
The following maintenance is performed for the EMS LF application: Event Public Holidays Abnormal Load Events Maintenance Description Holidays generally have a significantly different load shape and need to be identified in advance. The LF application allows a schedule of forward-looking holidays to be maintained for each forecast area, with the schedule specifying how the area should be modelled for each holiday. Generally, holidays are specified to be treated as if they were either a Saturday or Sunday. Occasionally, widespread outage events can have a significant impact on the actual load measured for one or more forecast areas which affects the calculation of the correlation coefficients. It is possible to flag such days and forecast areas as excluded from the model. Examples of abnormal load events which were excluded from the model include: 13 December 2011 AUFLS event; and 30 October 2009, loss of supply to Northland and parts of Auckland following the tripping of one of the OTA-HEN 220 kv circuits while the other circuit was removed from service. 11
4. LOAD FORECAST AND THE MARKET SYSTEM The Market Systems interface delivers updated MTLF forecast data to the Market Systems Database (MDB) every half hour. Forecast values are graphically presented to realtime staff who have a number of options for modifying and locking in modifications to the individual trading periods. Note that the interface to the Market System is staged that is, the updated MTLF forecast is sent directly to the Market System staging tables, and the MDB then takes care of operator entries and other local overrides. 12 4.1 BUS LOAD PARTICIPATION FACTORS The area forecast information is broken down to GXP level for use by SPD using Bus Load Participation Factors (BLPFs). These BLPFs are only used in the schedules that use MTLF information (see section 4.2 below). BLPFs are calculated and stored in the MDB. The BLPF is derived from the actual bus load from last week s MV90 data as a proportion of last week s total area load. BLPF = Actual bus load from last week s MV90 Actual conforming total area load from last week X Forecast area load for today s MV90 Note that BLPFs for GXPs with wind generation are calculated using SCADA data from the previous week rather than MV90 data. BLPFs can be modified in planning time 7 to account for situations where the BLPF does not represent the expected load, for example, due to outages or load shifting. A BLPF can be zeroed for the duration of an outage to remove deficit generation infeasibilities from the market schedules where the outage results in the GXP being disconnected. A BLPF can also be modified for situations where a load is shifted from one GXP to another, e.g. from RFN1102 to RFN1101 during an outage of DOB-RFN- IGH-2. 4.2 MARKET SCHEDULES The table below describes how the load forecast is used in the market schedules. Note that RTD, RTP and Final/Interim/Provisional Pricing schedules do not use load forecast information. Week-ahead dispatch schedule (WDS) Non-Response Schedule (NRS) Price-Responsive Schedule (PRS) Conforming GXPs Load Forecast Load Forecast Load Forecast +/- cleared difference bids Non-conforming GXPs Sum of nominated bid quantities Sum of nominated bid quantities Cleared nominated bids 7 Up to two weeks in advance of real-time.
4.3 MONITORING AND ADJUSTING THE MTLF Realtime staff can access graphical and tabular views of the load forecast for up to 15 days ahead via the Market Operator Interface (MOI). SCADA actual loads are also saved in the MDB and provide realtime staff with a view of how the load forecast is tracking against SCADA actual load. The load forecast can be manually adjusted by realtime staff if the forecast is too far from the actual load. Each day the MV90 metered loads for the previous day are received and overwrite the SCADA actual values. Realtime staff monitor the MTLF every half hour (i.e. when an updated MTLF is available). Other triggers for realtime staff to monitor the MTLF and adjust it where required include: Constraints binding in the schedule but no associated violations in the Realtime Contingency Analysis (RTCA) tool; Violations in RTCA but no associated constraints binding in the schedule; HVDC not running to schedule; Inaccurate Non-Response Schedule Short (NRSS); and Weather or load patterns significantly different from normal. 13 Actions that realtime staff can take to correct the MTLF include: Correcting the area level trend; Correcting the island level trend; and Adjusting the BLPFs. If demand starts tracking above or below expectations, the load forecast information will adjust as soon as an override is entered by a user. Realtime staff have the ability to scale load forecast information at the area level and island level. The area/island level load forecast can be scaled up or down by a specified number of MW or by a percentage of the existing forecast. Realtime staff also have the ability to adjust the BLPFs described in section 4.1. However, this has the effect of redistributing the load within the entire load area. Realtime staff cannot adjust the load forecast information at the GXP level to a specific value. The updated information will be picked up in the latest schedule (manual or automatic). If realtime staff do not intervene, it can take a couple of trading periods for the load forecast to adjust. If the actual load changes markedly due to some event or other then it may have an impact on the upcoming forecast if the change happens early enough in the trading period (e.g. within the first 10 minutes) that it affects the average for the period. Otherwise the refined forecast may not see a large enough disparity in actual load vs. forecast load to force a significant change in the output forecast. If this is the case (say load change occurs at 25 minutes after the half hour) then the next trading period forecast will not see the change. The trading period after that will see a change 8, but this typically will not be visible in the Market System until about halfway through the trading period. So an event that begins at 04:20 may not result in an updated forecast until 05:15 or so. The raw forecast and any adjustments are archived in the System Operator s data warehouse on a daily basis. 8 Assuming it persists
5. GLOSSARY 14 Acronym EMS BLPF GXP LF MDB MTLF MV90 RTU SCADA SPD Definition Energy Management System A collection of computer software applications used to monitor and control a power system. Bus Load Participation Factor Ratios used by the System Operator to break the area level forecast produced by the Load Forecast application down to GXP level for use in the market schedules. Grid Exit Point A point of connection where electricity may flow out of the grid. Load Forecast application The application which produces the Medium Term Forecast (forecast of demand) using historical load information, historical weather information, and weather forecast information. The Load Forecast application runs in the EMS environment. Market Database The Market Database provides live data storage of the inputs and outputs of the Market System model and its related transactional data. Medium Term Load Forecast The Medium Term Load Forecast is an output of the Load Forecast application. It is the forecast of demand prepared by the System Operator for use at conforming GXPs in accordance with 13.7A of the Electricity Industry Participation Code. Metered load data Remote Terminal Unit The processor at the outstations of a SCADA system. It transmits or receives data to or from an area operating centre or service centre. Supervisory Control and Data Acquisition The monitoring and remote control of equipment from a central location using computers. Scheduling, Pricing and Dispatch A computer system used to calculate the schedules and prices for the New Zealand Electricity Market and the reserves market. It also calculates dispatch instructions for realtime dispatch.