Using Data Science to Distribute Off Grid Solar Power
1 like52 views
The document outlines a vision for using data science at Fenix to enhance operations and improve customer interactions, focusing on remote monitoring and analysis of solar panel performance and system operations. It highlights challenges related to data scalability and integration with existing systems, as well as the need for efficient data collection and analysis. The data science team aims to support various functions, including product diagnostics, financial operations, and market identification through geospatial data.
Using Data Science to Distribute Off Grid Solar Power
- 2. OUR VISION dirty dangerous dim 10-20 cents per day cleaner safer brighter 10-20 cents per day
- 4. DATA SCIENCE AT FENIX Data collection Prediction (credit scoring, repayment & sales models) Classification (automatically diagnose technical issues) Experimentation & pilots
- 5. DATA SCIENCE PROBLEMS Financial Customer vetting Credit scoring Targeting interventions Upgrade eligibility (ex. fuel-efficient stoves) Operations Call center staffing (predicting language needs) Last mile distribution optimization System Performance Product analytics System malfunction prediction and classification Fraud/Tamper detection
- 6. GSM FOR REMOTE DATA COLLECTION Before: Gather data through experiments, technical repair diagnostics
- 7. GSM FOR REMOTE DATA COLLECTION Now: Remotely send/receive near real-time + historical data over GSM network. Use cloud services (Amazon S3 + Spark data pipeline) to store and aggregate millions of data points daily. Devices send data on measurements (e.g. power, energy, temperature), user interactions (e.g. button pressed), system events (e.g. battery full) Data feeds databases, APIs, dashboards, automated diagnostic systems Insights consumed by analytics team, operations, customer support, managerial & business teams
- 9. GSM TO MONITOR SYSTEM USAGE (GENERATION, USAGE) Ideal
- 10. TIME SERIES ANALYSIS TO CLASSIFY PANEL ISSUES Time Power 1 day
- 11. Hour of the Day SOLAR PANEL PERFORMANCE VARIES WITHIN A DEVICE 23 2 5 8 11 14 17 20 23 PowerGeneratedbyPanel(mW)
- 12. GSM DATA CHALLENGES Scalability from thousands of data points, to millions, to billions…..in a single year as customer base grows potentially in parallel Pressure to get the design right the first time, to avoid tedious backfills Interaction with legacy systems already in place e.g. choice of database technology Overloaded analysts: - other departments (data consumers) growing much faster than data team Establishing partnerships and/or budget for relevant sources of supporting data
- 13. DATA SCIENCE PROBLEMS Financial Customer vetting Credit scoring Customer repayment Financial Operations System Performance Field Work
- 14. OTHER DATA SCIENCE PROBLEMS Financial Customer vetting Credit scoring Customer repayment Targeting interventions Financial Operations System Performance Field Work
- 15. OTHER DATA SCIENCE PROBLEMS Operations Track product from supply to end-of-life Last mile distribution optimization Financial Operations System Performance Field Work
- 17. Other cool things the DS team dabbles in - Aggregate information for other teams - Product diagnostic suite - Inform product design (GSM data led to design of new lights) - Track call centre productivity - Identify new markets via geo-spatial data