Tackling climate change with ML (ICLR Workshop 2020) - day 2

Opportunities and Challenges for Machine Learning in the African Electricity Sector

Nathan Williams (Rochester Institute of Technology)

E-Guide Initiative - eguide.io

Big data for electricity consumption prediction. Sophisticated energy planning tools exist but rely on poor quality demand data.

• Big data from utilities:
  • use geolocated historical billing data and satellite imagery from 100ks of customers
  • Deep learning to make large scale prediction at fine spatial resolution
• Rich data from mini-grids
  • Uses detailed survey and smart meter data from the off-grid sector from thousands of customers
  • ML methods to make better predictions for system design and financial planning
• Using consumption data to improve system planning and operation
  • Estimating subsidies for grid extension
  • Small commercial consumption patterns
• Co-planning of electricity and agricultural infrastructure
  • Goal: create geospatial data products to facilitate cross-sectorial planning of infrastructure planning
• Mapping power quality and reliability with night lights imagery (work done for India)

Opportunities

• Fill data gaps using public available data
• Measure infra (location, condition, reliability)
• Improve system planning (design and efficiency with better demand data)
• Identify opp for productive use
• Facilitating access to credit for unbanked

Challenges:

• Data availability, data quality, data access
• Capacity, skills and infra at data gathering org
• Data standardization
• Privacy and data security

Fireside chat with Jon Bonanno on cleantech entrepreneurship

• Artificial intelligence in climate change solutions:
  • Energy storage: PowerFlex, Energy Vault, EnZinc
  • Green generation / grid management: SolarCity, Solar.com, Ocergy, predictive for solar and wind, Station A, Open EE
  • Mobility / logistics: Tesla, autonomous, ride share, mapping / directions, flying cars, Flux EV
  • Built environment: SkyCool systems, Stasis Energy group, Opower, Nativus power, ZYD energy, correlate, 75F
• Getting to "go":
  • Market: size, cycle, velocity, pain, etc
  • Identify "compelling value hypothesis": features, audience, business model, USPs: "will the dog eat the food"
  • Test: build, measure, learn from potential customers, iterate
  • Intense and active patience: take your time, PMF is the priority - not being first. Iterate quickly based on feedback toward MVP, retest, stay alive
  • Are you there? Growing exponentially without marketing, strong word-of-mouth from delighted customers, voting with dollars
  • Reasons startups fail: no market fit, ran out of cash, team