Forecasting Meeting Minutes for Week of September 13, 2016

Present: Gordon, Brieanna, Jaimie

Saturday, September 10, 2016 (Proposal Presentations)

Tuesday, September 13, 2016

Produce:
- Well-written code
- New python libraries to pull off of
- Build library of functions other groups can use
- Solar Industry
- Calibrate Data
Decisions:
- Normalization Method (Zenith Angle?)
- Prediction time (1 hr?)
- Solar Irradiance doesn't need to be sampled too quickly (1-2 mins ok)
Understanding PV:
- How much in Hawaii
  - Commercial and residential
  - Projection of these
- Solar Thermal
- Solar Farms
- Concentrated Solar
- Energy Production
  - PV Data Sheets
  - 10-15% efficiency w/ lower cost panels
  - 40% efficiency w/ increase in efficiencies, cost, etc.
- Breaks for Federal and State and HECO
  - Larger contracts (PPA)
Forecasting:
- Weather and time series prediction
- Predicting an hour
- Reasonable with just Least Squares
  - Depends on normalization (ex. subtract mean of specific time of day/std, or zenith angle)
  - Take zenith angle: Time of day and day of year in account
- Window (FIR filters)
  - Straight average of data over time
Online Methods:
- Estimate of parameter weight and update estimate based on new data
- New linear estimate
- Recursive Least Squares (RLS)
  - Depends on first and second order (mean and covariance matrixes)
- Least Mean Squares (LMS)
  - Won't perform as well
Supervised and Unsupervised Learning:
- Nearest Neighbor (supervised)
- Support Vector Machine Kernels ()
- Deep Learning ()
- Clustering used for large dimensional data (unsupervised, Lots of factors)
Estimation (regression) and Detection (classification)