forecast_mcm: Do intra-hourly Markov-chain mixture distribution forecast

In kdayday/solarbenchmarks: Intra-Hour and Intra-Day Probabilistic Solar Forecast Benchmarks

Description

Valid for intra-hourly forecasts. This generates a clear-sky index transition matrix based on the previous 20 days of data before each hourly issue time. Using that transition matrix, clear-sky index is forecasted over the next D time steps (i.e., up to 12 for a 5-minute resolution forecast.)

Usage

forecast_mcm(GHI, lead_up_GHI, percentiles, sun_up, lead_up_sun_up,
  clearsky_GHI, lead_up_clearsky_GHI, ts_per_hour, num_days,
  numBins = length(percentiles) + 1, numSamples = 1000,
  h_per_day = 24)

Arguments

GHI	A vector of the telemetry
lead_up_GHI	A vector of out-of-sample telemetry for the days leading up to the start of the sample period. Number of days must be >= num_days
percentiles	A vector of the percentiles corresponding to the desired forecast quantiles
sun_up	A vector of logicals, indicating whether the sun is up
lead_up_sun_up	A vector of logicals, indicating whether the sun is up for the days leading up to the start of the sample period, corresponding to lead_up_GHI
clearsky_GHI	a vector of clear-sky irradiance estimates
lead_up_clearsky_GHI	A vector of out-of-sample clear-sky irradiance estimates for the days leading up to the start of the sample period, corresponding to lead_up_GHI
ts_per_hour	Time-steps per hour, e.g., 12 for a 5-minute resolution forecast
num_days	Number of days of training data
numBins	(optional) Number of bins to use in the MCM matrix M. Defaults to number of percentiles + 1.
numSamples	(optional) Number of samples to take from MCM model to generate empirical CDF. Defaults to 1000.
h_per_day	(optional) Hours per day = 24 (useful for testing)

Details

Modified from the main.R script at https://github.com/SheperoMah/MCM-distribution-forecasting, Reported in: J. Munkhammar, J. Widén, D. W. van der Meer, Probabilistic forecasting of high-resolution clear-sky index time-series using a Markov-chain mixture distribution model, Solar Energy vol. 184, pp. 688-695, 2019.

See Also

Other forecast functions: forecast_CH_PeEn, forecast_Gaussian_hourly, forecast_Gaussian_intrahour, forecast_NWP, forecast_PeEn_hourly, forecast_PeEn_intrahour, forecast_climatology

kdayday/solarbenchmarks documentation built on May 22, 2020, 10:33 p.m.