AIIA - Charting the Path to Intelligent Operations with Machine Learning - Atakan Cetinsoy
2 likes1,354 views
This document discusses how machine learning and predictive analytics can help utilities address challenges around sustainability, productivity, and customer engagement. It covers trends in the utility industry like the transition to distributed generation and demand response. It also discusses how sensor data and IoT can be used with machine learning to gain insights from time series and unstructured data. Examples are given of predictive applications for utilities around load forecasting, outage prediction, demand response optimization, and more. The document promotes the use of an end-to-end machine learning platform to build interpretable models for data-driven decision making.
AIIA - Charting the Path to Intelligent Operations with Machine Learning - Atakan Cetinsoy
- 1. Charting the Path to Intelligent Operations with Machine Learning Atakan Cetinsoy VP - Predictive Applications
- 2. 21st Century Megatrends As the world population is headed to 10 billion: • Intensifying scramble for scarce resources • Growing urbanization and diversity • Social media and the shifting balance of power SUSTAINABILITY PRODUCTIVITY ENGAGEMENT
- 3. Utility Industry Trends • Evolving energy portfolio • Transition to distributed generation schemes • Efficiency as a “New” energy resource • Growing smart meter infrastructure • Dynamic pricing and demand response
- 5. The Industrial Internet SOURCE: General Electric • Hypothetical 1% efficiency gain via IoT technology. Savings(inBillionsUSD)
- 6. Sensor Data and Predictive Apps SOURCE: Forrester
- 8. Case Study: Digital Cows SOURCE: Fujitsu.com
- 9. IoT Time Series Data Sensor Time +7 +35 +50 BLOB 101 15:00 N/A N/A N/A {…} 102 15:00 N/A N/A N/A {…} 102 15:01 N/A N/A N/A {…} 103 15:01 11 20 N/A {…} 103 15:02 N/A N/A 33 {…} 1 Minute Time Window Offset in Seconds • Wide row structure with possibly 1000s of measurements • 100M to 1 billion data points per second can be processed! • Compacted into BLOB format stored as a single value SOURCE: MapR
- 10. Big Data or Big Hype? • Data that is • Too big to fit on a single server • Too unstructured to fit into rows and columns • Too continuos to fit into an EDW • “Size matters” but actionable insights take the prize.
- 11. Data Driven Decision Making
- 12. Evolution of Analytics Attribute Traditional Analytics Analytics 2.0 Data Type Rows and Columns Unstructured Volume Up to TBs Up to PBs Flow Static Pool Continuos Technology EDW + SQL Open Source + Machine Learning Analysis Descriptive, Hypothesis-based Predictive, Machine Learned Purpose Internal Decision Support Data-driven Products/ Services SOURCE: Thomas H. Davenport Includes everything in Traditional Analytics plus the following.
- 13. Machine Learning? • “Machine Learning is the field of study that gives computers the ability to learn without being explicitly programmed.” — Prof. Arthur Samuel
- 14. The Need for Machine Learning • Can you find any pattern in this tiny data set? • Now imagine millions of rows and thousands of columns of it!
- 15. The Need for Data-driven Decisions • Human intuition is poor • Human judgement is biased • Human reasoning is causal and not statistical • Machine Learning is a tool to help people make smarter, unbiased, more effective data-driven decisions.
- 16. What is a Data Scientist? Industry Subject-matter Expertise Computer Science and/or Hacking Skills Math and Statistics Knowledge Machine Learning Traditional Research Data Science SOURCE: Drew Conway
- 17. Future of Machine Learning • “Machine Learning is becoming a new abstraction layer of the computing infrastructure.” Tushar Chandra, Principal Engineer — Google Research
- 18. BigML An end-to-end machine learning platform that is • Builds interpretable machine learning models that address the vast majority of predictive tasks. • Accessible to the entire organization to make data- driven decisions. • Provides a public API so that application developers can build predictive applications. • Cloud-born solution that provides instant access and instant scale. CONSUMABLE PROGRAMMABLE SCALABLE
- 19. Predictive Modeling Best Practices • Business objective and predictive model alignment • Proof of concept based on sampled data • Model validation with proper accuracy measures • Transparent vs. “Black Box” algorithms
- 22. Model Evaluation
- 23. Predictive Apps for Utilities • Operational • Accurate and Granular Load Forecasting • Network Outage Predictions • System Failure Predictions • Demand Response Optimization • Marketing • Customer Churn Prediction • Pricing Response Prediction • Energy Efficiency • Household Level Predictive Analytics