LoadSEER Satellite Analytics
Transcription
LoadSEER Satellite Analytics
Spatial Forecasting [email protected] 513 479 1235 or 513 762 7621 312 Walnut St Suite 1600, Cincinnati Ohio USA 45202 Copyright 2006-2012 Integral Analytics Integral to Consultants & Utilities Putting advanced analytics into software to manage costs & risks WindStore GridStore SolarSEEK Optimal Demand Dispatching Transformer Aging Analytics/ Load Leveling DRPricer Optimal Demand Response Bill Insure Fixed Bill Copyright 2006-2012 Integral Analytics Outline • • • • • What is spatial forecasting? How is spatial forecasting used? Spatial forecasting issues Methodology How does Integral Analytics perform spatial forecasting? Copyright 2006-2012 Integral Analytics What Is Spatial Forecasting? • Typical load forecast – Utility service area – Project volume and timing • What - When • Spatial load forecast – Small area: a few acres to square mile – Project volume, timing, and location • What - When - Where Copyright 2006-2012 Integral Analytics What Is Spatial Forecasting? • Not all forecasts of load by location are truly spatial forecasts – Small area forecasts are not necessarily spatial • Trending of loads for a substation or an equipment area is not a spatial forecast • Spatial forecast for an area incorporates impacts of economic activity in surrounding areas – Land use type which drives intensity of use – Forecast of land use impacted by growth in surrounding areas Copyright 2006-2012 Integral Analytics How Is Spatial Forecasting Used? • Traditional use: T&D planning – Plan system expansion – Reliably meet new customer loads • Additional Drivers of Spatial Forecasts – Budget constraints – Need for resource optimization – Regulatory requirements – Forecast accuracy – Integration of alternate resources, e.g., DG / renewable Copyright 2006-2012 Integral Analytics Spatial Forecasting Issues • What – When - Where – Level of locational detail required? – Length of forecast horizon? – Methodology? • • • • Small area trends on loads Incorporate economic drivers Utilize GIS: Geographical Information System data Blending of locational economic data with T&D facility areas – Weather impacts? – Uncertainty? Copyright 2006-2012 Integral Analytics Methodology • Level of locational detail – Equipment area – Grid of small areas • Length of forecast horizon – Short-term vs. long-term • Load Class – Customer class level loads by location – End-use detail Copyright 2006-2012 Integral Analytics Methodology Polygon / Equipment Based Area Copyright 2006-2012 Integral Analytics Grid / Small Area Methodology • Customer Classes – Residential – Commercial : e.g., office, retail, small , large – Industrial • Energy use: high, low • Industry group • End Uses – HVAC, water heating, – Motors, cooking, refrigeration, lighting – Process heat, computers, data centers Copyright 2006-2012 Integral Analytics Methodology • Forecast Method Options – Small area forecasts • Areas forecast independently – Trend extrapolation – S – Curves (Gompertz) • Issues: – – – – – Little use of economic and demographic data Difficulty with Green Field development Difficulty with redevelopment of existing space Inability to integrate with activity of surrounding areas Simpler to develop, but less accurate Copyright 2006-2012 Integral Analytics Methodology Copyright 2006-2012 Integral Analytics Methodology S - Curve Copyright 2006-2012 Integral Analytics Methodology • Forecast Method Options – Spatial forecasts • Area forecasts coordinated – – – – Simulation model Uses detailed data: demographic and economic Gravity model for density of economic activity Class locational preferences drive the forecast • Issues: – – – – Acquire and organize detailed land use data Greater detail requires resources Higher complexity, but higher accuracy Facilitates targeting of energy efficiency and load management resources as well as electric vehicles Copyright 2006-2012 Integral Analytics Methodology Customer Classes Copyright 2006-2012 Integral Analytics Methodology Urban Pole A major employment area has a large and wide influence on the areas around it: every one wants to get as close to it as they can. This circular, decreasing value function of distance, centered at the downtown area, represents this influence. Copyright 2006-2012 Integral Analytics Methodology Proximity Score 0 - 100 • ¼ mi Euclidean Distance to an object Methodology • Forecast Method Options – Geographic forecasts • Forecast of a larger area – Econometric model – Uses demographic and economic data • Issues: – Lacks detailed data for smaller areas – Areas too big for T&D planning purposes Copyright 2006-2012 Integral Analytics ™ Spatial Electric Expansion & Risk Space-time calculator. Geographic calculator. We created this tool to give distribution planners the ability to quickly extrapolate complex geographical knowledge across space and time in order to plan for economic growth and system expansion. LoadSEER System - LoadSEER-GIS (Geographic Information System) is a spatial forecasting desktop application that is used by Area Senior Engineers to model long term load horizons for the System as a whole and each division, planning area, bank, and circuit respectively. LoadSEER-GIS replicates urban development processes based on historical land use change, zoning information from government, customer rate class from utilities, and the energy model of consumption patterns. - LoadSEER-FIT (Forecast Integration Tool) is a load trending web-based application that is used by Distribution Engineers to adjust planning area, bank, and circuit forecasts with their local knowledge and judgment. ESRI Personal Geodatabase Area & Division Forecast Approval Load Adjustments Planning Area Forecast Approval LoadSEER FIT – Web client SQL Server Central Repository: Load Forecasting Database LoadSEER GIS – Desktop client Area Forecast Aggregation Bank/Circuit Peak Load Normalization & Trending System Load Allocation and Optimization Customer class based spatial load growth simulation Confidential and Proprietary GIS Conceptual Program Workflow A. What does my Service Territory look like? I. Land Use II. Transportation III. Large Area Employment B. How does my Service Territory grow? I. Growth Rates a. Corporate Forecast b. Local / State Planning II. Growth Drivers a. Proximity, Surround, Urban Pole factors b. Preference coefficients C. What does Seasonal Peak Load look like? I. 24 hour peak load – one acre density - Data Types & Schema - Points - Lines - Polygons Land Use Polygons Transportation Urban Pole Lines Points Land Use & Customer Load Electric Grid & Customer Load Data Land Use Classes Land Use 1 A C R E C O N V E R S I O N Commercial Residential – Suburban Industrial Vacant Land Travel Time Urban Pole (Center City) Euclidean Distance Activity Centers A major employment area has a large and wide influence on the areas around it: every one wants to get as close to it as they can. This circular, decreasing value function of distance, centered at the downtown area, represents this influence. Customer Preference Survey • • Customers are recruited to participate in an online market research study which is specifically designed to quantify the types of trade-offs customers make when they decide to live or build within specific areas. In all cases, particular attention is paid to nonlinear preferences to different distances from attributes, such as a Central Business District, Hospitals, and Highways, which informs LoadSEER’s specific Regional Factors. Historical Land Change Modeling • • Historical land-use classification maps are established for prior years to consistently represent the Study Area’s geography across time. Using these historical land-use maps, we quantify the spatial land-use changes that historically have occurred within the study area, across time and space. Once modeling is complete, a full set of Proximity and Surround factor Preference Weight Coefficients will be generated for LoadSEER’s long range forecasts. Model calibration Income Segments vs. Distance from Downtown *Avg. Utilities is the Average Preference Score for each Segment. Notice: Preferences Scores change nonlinearly with distance. For example, high income customer show a significant increase in preference between 2 – 5 miles away. Model calibration Income Segments vs. Distance from Hospital *Avg. Utilities is the Average Preference Score for each Segment. Notice: Customer show significant preference for locating within 10 miles of a hospital. 35 Model calibration Income Segments vs. Distance from Highway *Avg. Utilities is the Average Preference Score for each Segment. Notice: The preference to located near highways peaks at 2 – 5 miles; customers want to live close, but not too close to a highway. 36 Tract Size • Minimum number of contiguous acres required to allocate growth for each class. Residential = 1 acre Commercial = 3 acre Heavy Industrial = 10 acre Copyright 2006-2012 Integral Analytics Satellite Image Analysis… 1991 Satellite Image Analysis… 2011 Satellite Image Analysis… 1991 The image cannot be display ed. Your computer may not hav e enough memory to open the image, or the image may hav e been corrupted. Restart y our computer, and then open the file again. If the red x still appears, y ou may hav e to delete the image and then insert it again. Satellite Image Analysis… 2011 The image cannot be display ed. Your computer may not hav e enough memory to open the image, or the image may hav e been corrupted. Restart y our computer, and then open the file again. If the red x still appears, y ou may hav e to delete the image and then insert it again. Satellite Image Analysis… Change detection LoadSEER Satellite Analytics Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Copyright 2006-2012 Integral Analytics Integral Analytics LoadSEER Satellite Analytics Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Copyright 2006-2012 Integral Analytics Enhance with Building Descriptions Enhance with Circuit lines & Nodes Weather and Economic Modeling Within LoadSEER FIT Model LoadSEER FIT Forecasts Are Normalized For Both Weather AND Economy At The Same Time For each feeder class combination, it is typically the case that there are summer MWH histories, possibly by customer class. In the case of the MWH model, we use CDD or cooling degree days, in conjunction with specific econometric variables. Following is one possible set of variables used, to control for both summer weather and the economy, modeled across sets of feeders. Real Gross Regional Product Non-Manufacturing Employment Manufacturing Employment Real Total Retail Sales Real Income: Total Personal Real Income: Earnings - Farm Proprietors Profits Agricultural Employment CDD = cooling degree days FIT Model LoadSEER methods employ not 1, but 3 key sets of models For increased accuracy in load forecasting For increased defensibility with regulators and management To enhance convergent validity (triangulation) 1 Traditional area regression forecast used by distribution planners is based upon historical MW peaks (for past X years). 2 Spatial GIS forecast using LoadSEER algorithms (GIS) and satellite history 3 A MWH model to forecast economic growth, controlled for weather This 3rd model is used to increase accuracy and defensibility, and serves to enhance the GIS based spatial forecasts. This includes normalization for BOTH weather and economy • • MWH Model uses CDD/HDD and Regression FIT uses hourly temps LoadSEER employs a forecast blending process • Blend GIS with MWH Model • Adjusted forecast – align with Corporate aggregate forecast • T&D planner can blend traditional trend model with the adjusted GIS/MWH model Copyright 2006-2012 Integral Analytics LoadSEER View 1 in 10 Year Weather Risk Weather Normal 1 in 2 Year Capacity Economic Predictor Variable Weather Predictor Variable Your Final Forecast appears in red and blends Global Corporate vs. Local Regression Economy Now Trumps Weather Risk LoadSEER uses multiple methods to forecast loads. Spatial analytics from satellite data for growth trending. MWH model with load factor conversions, consistent with Corporate Forecast. Local Regression models with: Weather Impact 150 Economic Variables Recently, economic risk is trumping weather risk. Need for “economic normalization” vs. weather normal. Added complexities of renewable, solar, and EV loads added as layers. Copyright 2006-2012 Integral Analytics Increased Accuracy in New LoadSEER model Several sources of increased accuracy in LoadSEER Forecasts BOTH temperature and economic data (Moody’s) Economic downtown caused poor model fits on temp alone Spatial forecasts more accurate at a local level vs. corporate Accommodates impact of special load additions at specific locations Added the MWH model to enhance accuracy, and improve economic forecast Uses load research per load area to estimate weather risk Enables accurate consistency check to overall Corporate Forecast Incorporates a pooled model for forecasting circuits for MWH model Forecasts feeder growth more accurately than if done individually Enables automatic key driver identification, so LoadSEER finds best fit Enables layering in of new classes (e.g., solar, DG, EV, etc.) More localized IRP strategy Possibility of adding gas end uses (e.g., gas air conditioning). Solar forecast, EV forecasts, other DG/ Smart Grid programs. LoadSEER Satellite Analytics Land Use Forecast (2011) Land Growth Classes 2 = Residential - Rural 3 = Residential - Suburban 4 = Residential - Multi/Dense 5 = Residential - High Rise 6 = Retail Commercial (Including Parking Lots) 8 = Business Parks 12 = Light and Medium Industrial 15 = Institutional (Schools, Churches) Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Land Use Forecast (2012) Land Growth Classes 2 = Residential - Rural 3 = Residential - Suburban 4 = Residential - Multi/Dense 5 = Residential - High Rise 6 = Retail Commercial (Including Parking Lots) 8 = Business Parks 12 = Light and Medium Industrial 15 = Institutional (Schools, Churches) Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Land Use Forecast (2014) Land Growth Classes 2 = Residential - Rural 3 = Residential - Suburban 4 = Residential - Multi/Dense 5 = Residential - High Rise 6 = Retail Commercial (Including Parking Lots) 8 = Business Parks 12 = Light and Medium Industrial 15 = Institutional (Schools, Churches) Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Land Use Forecast (2017) Land Growth Classes 2 = Residential - Rural 3 = Residential - Suburban 4 = Residential - Multi/Dense 5 = Residential - High Rise 6 = Retail Commercial (Including Parking Lots) 8 = Business Parks 12 = Light and Medium Industrial 15 = Institutional (Schools, Churches) Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Land Use Forecast (2022) Land Growth Classes 2 = Residential - Rural 3 = Residential - Suburban 4 = Residential - Multi/Dense 5 = Residential - High Rise 6 = Retail Commercial (Including Parking Lots) 8 = Business Parks 12 = Light and Medium Industrial 15 = Institutional (Schools, Churches) Copyright 2006-2012 Integral Analytics LoadSEER Satellite Analytics Land Use Forecast (2028) Land Growth Classes 2 = Residential - Rural 3 = Residential - Suburban 4 = Residential - Multi/Dense 5 = Residential - High Rise 6 = Retail Commercial (Including Parking Lots) 8 = Business Parks 12 = Light and Medium Industrial 15 = Institutional (Schools, Churches) Copyright 2006-2012 Integral Analytics Copyright 2006-2012 Integral Analytics Copyright 2006-2012 Integral Analytics Copyright 2006-2012 Integral Analytics Copyright 2006-2012 Integral Analytics Copyright 2006-2012 Integral Analytics Copyright 2006-2012 Integral Analytics Copyright 2006-2012 Integral Analytics Uncertainty • While spatial forecasts improve T&D planning, it makes sense to test the plan under multiple scenarios to improve robustness of the plan • Example scenarios – New plant location – Major plant expansion – Faster or slower load growth Copyright 2006-2012 Integral Analytics Incorporating EV adoption forecasts into Grid Planning Risk Market Research and GIS Modeling Copyright 2006-2012 Integral Analytics Web Based Choice Example Copyright 2006-2012 Integral Analytics Web Based Choice Example Copyright 2006-2012 Integral Analytics Market Simulation Example Share Forecast For Simple Market of 4 Cars Shares Vary With Competition, Gas Prices, Fueling, Etc. Example Type Price Fueling time Fueling duration Impact on Reducing Global Warming Share of Preference at Gasoline Price Level $4 / Gallon Share of Preference at Gasoline Price Level $5 / Gallon Share of Preference at Gasoline Price Level $6 / Gallon Share of Preference at Gasoline Price Level $8 / Gallon Share of Preference at Gasoline Price Level $10 / Gallon Standard $25,000 Any time 2 Minutes No impact 87.37% 79.98% 71.86% 54.16% 38.52% PHEV $40,000 Any time 1 hour High Impact 8.36% 14.44% 20.82% 38.15% 54.66% EV $50,000 No fueling between 3 – 6 pm 5 hours High Impact 2.05% 3.36% 5.61% 6.30% 5.93% Scooter < $7,000 Any Time 5 hour Medium Impact 2.22% 2.21% 1.71% 1.39% 0.89% Scenario applied here uses $4 gasoline, 10 cent electricity, 4 cars (over simplified here) and an expensive EV option at $50,000 versus the hybrid option at $40,000. Share forecast varies significantly by demographics and market segments. This is simply an average. Fueling convenience benefits and image factors are just as important as costs and economics. Copyright 2006-2012 Integral Analytics Prius Vehicle Prius Type Vehicle The popularity of the Toyota Prius presents an interesting dichotomy in customer segment preferences. Where we see increased likelihood in the purchase of this type of vehicle among customer segments which are _____________________, we see a marked decrease in likely popularity among __________________(note the negative sign on the Vehicle parameter estimate. Scoring Function for Prius Type Vehicle = - 0.2862+1.544 * VAR1 (combination of 3 demographics) + 2.1171 * Economic VAR - 2.2901 * Specific Vehicle type. Copyright 2006-2012 Integral Analytics Adding EV to Load Growth Segments Tend To Cluster (“Birds of a Feather Flock Together”) Forecasted Adoption of Car Type C If segment clusters are large enough, or charge rates are high enough, local circuits and transformers may be at risk Page 6 Adding EV to Distribution Loads All Segments (T&D focus) Page 10 Adding EV to Distribution Loads Scoring Process 30 50 All customers scored using clustering and regression, to get Prob Value,35eventually replaced with actual purchase (Prob=100%) 65 40 45 70 85 35 65 80 30 30 60 70 20 Page 11 Fine Tuning Within Hot Spots LoadSEER Spatial Forecasting Identifies “Hot Spots” Smart Grid Analytics Identifies Which Service Transformers Are At Risk From Load Growth + EV Copyright 2006-2012 Integral Analytics Service Transformer Analytics Averaged, Load Research profiled load shapes are nice and smooth…. …. AND MISLEADING. In reality, service transformers experience much more load volatility. Load At Risk analytics accurately models loads. Copyright 2006-2012 Integral Analytics Service Transformer Analytics Load At Risk is defined more precisely than “at 95 degrees” since more than temperature drives the loads. Load Normalization is more accurate than Weather Normalization. September 4pm Load At Risk Each hour has its own distribution Every other year, 1 in 2 years 1 in 5 years Graphic representation of full Load At Risk Simulations 1 in 10 years 1 in 100 years 10% Percentiles 1st 5th 10th 50th 5% 1% Load at Risk 90th 95th 99th (avg) Copyright 2006-2012 Integral Analytics Percentiles Service Transformer Analytics Perhaps Under-loaded or “Over”-Sized For Cold Load Pickup At Risk Adding Electric Vehicle Copyright 2006-2012 Integral Analytics At Risk Now What Does It All Mean? • Simple trend based forecast methods may be simpler to implement, but have lower accuracy. • Spatial forecast methods incorporate economic and demographic growth impacts from adjacent locations. – Increases accuracy • Spatial forecasting methods can add capability to include the impacts of solar, EV, storage, or other localized effects. Copyright 2006-2012 Integral Analytics QUESTIONS ? [email protected] 513 479 1235 or 513 762 7621 312 Walnut St Suite 1600, Cincinnati Ohio USA 45202 Copyright 2006-2012 Integral Analytics