Predictive Analytics for Smart Grid to Make it happen Britta Hilt | MD Dublin, 6th May 2014 www.ispredict.com | Copyright IS Predict & Scheer Group Employees 2010 - 2014 800 400 100 Visionary, researcher and author of standard works for business Turnover (million €) 2010 - 2014 50 Prof. A.-W. Scheer information systems www.ispredict.com | Copyright Member of the council for innovation and growth of the German Government President of the German Association for Information Technology (BITKOM 2007-2011) Ranked as 2nd most important German IT person (of 100) by Computerwoche magazin (after Hasso Plattner / SAP) in 2011 Founder of international software companies IDS Scheer & IMC AG Sole Shareholder of Scheer Group GmbH Locations Germany France Turkey Australia Great Britain Ukraine Austria Rumania Benelux Switzerland 2 Smart Grid´s Complexities Challenge 1 – How to… Challenge 2 – How to… … keep control in volatile grids? … use energy efficiently? Various players High volatile power generation & consumption On grid / Off grid Reliable & affordable “Green” … www.ispredict.com | Copyright 3 How to manage Smart Grid`s challenges Plannable Energy Flows Highly accurate knowledge when how much Wind / sun energy will be generated Energy will be consumed by industry and private households Transparency on Root & Cause Highly accurate knowledge which (hidden) factors increase energy consumption Complex human behavior Complex machine behavior Challenge 1 – How to… Challenge 2 – How to… … keep control in volatile grids? … use energy efficiently? www.ispredict.com | Copyright 4 Comparions of Prediction Tools in Highly Volatile Use Cases Comparison: 24 h prediction gas consumption Example: Difficult month Generic regression Specific forecast External supplier for utilities with focus energy forecasts Deviation Average: 18% Maximum: 47 % Generic „Discovery“ www.ispredict.com | Copyright Resource Intelligence prediction with automatic model generation Deviation Average: 8% Maximum: 26 % 5 Example Utilities Gas Energy Reduced costs for energy via more precise 24 h gas prediction Objective: Problem: How: Data: Plan demand-oriented gas purchase for tomorrow & thus, reduce purchasing costs Standard load profiles too inflexible for dynamic demand of consumer Dynamic load profiles with flexible pattern recognizion Historic gas consumptions, weather (past and forecast); no consumer classification Resource Intelligence Accuracy O 24h (%) Jan Feb Mar Apr May Jun 96 89 91 88 86 88 April State of the Art Solution Accuracy O 24h (%) Jan Feb Mar Apr May Jun 92 90 81 83 67 74 Resource Intelligence ca. twice as precise than state of the art solution with standard load profiles www.ispredict.com | Copyright 6 Example Utilities PV Power Usage Reduced power costs due to optimal usage of own PV power Objective: Run production machinery mostly on PV power, generated by your own Problem: PV power very volatile and difficult to plan; energy demand of machinery also volatile; energy demand does not match energy availability How: Foresighted machinery control via accurate PV power generation prediction Data: Weather (past / forecast) power generation (past) Accurate 24 h PV power generation prediction for 1 individual installation Accuracy Mar Apr May Jun Jul Aug O Month O Day 94 % 91 % 97 % 93 % 94 % 93 % 99 % 96 % 92 % 93 % 95 % 95 % Sep Oct 97 % 92 % 93 % 93 % Resource Intelligence realizes flexible and precise predictions despite high volatility www.ispredict.com | Copyright 7 Example Utilities Water Consumption Less costs for water supply Objective: Cost optimization in drinking water supply Problem: Water demand highly volatile and difficult to plan; Reduction of power costs for pumps but also 100% availability of water supply How: Running pumps when power price is low thanks to precise water consumption prediction Data: Weather (past / forecast), water consumption (past) Accurate 24 h water consumption prediction Accuracy Jan Feb Mar Apr May Jun O Month 98,1 % 97,0 % 99,6 % 99,3 % 99,0 % 98,0 % O Day 97 % 97 % 99 % 94 % 97 % 95 % Resource Intelligence realizes flexible and precise predictions despite high volatility www.ispredict.com | Copyright 8 Example Production Predictive Dispatching Optimized Energy Dispatching Objective: Way: Problem: Data: Efficient energy dispatching and planned energy purchase in steel company Enable planning for large energy consumers despite “not planable” consumption Highly volatile energy demand which does not seem to be caused by production. Energy consumption & limited production (planning) data Steel mass Steel width Required energy Energy Demand Variations in oven Energy forecast for energy planning & efficiency analysis 1 month Time: 24 h Predicted Energy Required Energy www.ispredict.com | Copyright 9 Machinery Control Reduced Operating Costs Energy consumption Which factors cause increased energy consumption per tonne of pellets? Objective: To control two wood dryer in optimal energy usage although product quality stays the same Please note: No permanent usage of dryers required. Problem: Highly varying energy consumption of wood dryer. Therefore, it is unclear which factors increase energy consumption Solution: Discovering influencing factors for energy consumption via pattern recognizion and correlation analysis Data: Consumption of long distance heating, outside temperature, other production data 1st Discovery Despite expert expectation: Air humidity has minimal influence on energy consumption! www.ispredict.com | Copyright 2nd Discovery More than proportional increase when assembly belt speed is increased. 10 Example: Predictive Maintenance Individual demand-oriented maintenance via anomaly analysis Objective: Way: Condition: Problem: Increase efficiency via early information on (future) wear & tear Discover first and hidden signs when machinery does not run efficient anymore Individual & cost-reduced analysis per machine without additional sensors Strongly volatile energy demand, only engine energy data, no production data Discovery of anomalies between 86% - 100%! 10 minutes: Engine run without disturbances 10 minutes: 51 disturbances due to breaks Discover anomalies in machinery behavior i.e. in resource consumption Evaluate anomalies Irregularities with various strengths and frequency Anomaly Details No regularity in variable energy demand during disturbance Early warning Alert for technical service 11 SAP CEO visits IS Predict CeBIT 2014 SAP CEO Jim Hagemann Snabe informs himself about Resource Intelligence Projects Increased efficiency for Man & Machinery thanks to Predictive Control The self-learning & adaptive IT system for cost reduction 23 % in Smart Home Grid Realizing full potential of renewable energy usage 16 % in Smart Utility More precise energy purchase & sale, also for renewable energy 12 – 62 % in Smart Building Foresighted and adaptive building energy control 14 % in Smart Production Resource management, energy dispatching, machinery control, predictive maintenance, process efficiency, capacity planning www.ispredict.com | Copyright Honored with 8 Innovation Awards 13 ff Why we are here: We want to improve Resource Efficiency at Irish Grid / Production / Power Plants, too. We are looking for challenging projects to optimize complex and difficult processes with innovative & self-learning IT solutions. www.ispredict.com | Copyright | Contact: Britta.Hilt@ispredict.com +49 176 – 63 72 92 28 IS Predict GmbH Scheer Tower | Uni Campus Nord D5.1 66123 Saarbrücken | Germany Phone +49 681 – 96777-200 www.ispredict.com 14