Combining multiple trending models for photovoltaics plant output forecasting
A method of predicting an amount of power that will be generated by a solar power plant at a future time includes: forecasting a value of a data variable at the future time that is likely to affect the ability of the solar power plant to produce electricity (S 301 ); computing a plurality of features from prior observed amounts of power generated by the power plant during different previous durations (S 302 ); determining a trending model from the computed features and the forecasted value (S 303 ); and predicting the amount of power that will be generated by the power plant at the future time from the determined model (S 304 ).
1. A process for forecasting power generation by a solar power plant using combined trending models, the process comprising:
forecasting an exogenous vector of weather-related exogenous variables for a future time (h) that is likely to affect the ability of the solar power plant to produce electricity;
generating a plurality of trending models representative of different trending ranges of prior data observations, each trending model generation comprising:
extracting a feature from a trending function, wherein the extracted feature represents a power value at the future time (h) based on the trending function; wherein the trending function is associated with a time period (w) of a duration different than other trending functions and derived from prior observed amounts of power generated by the solar power plant during a corresponding time period (w);
predicting an amount of power that will be generated by the solar power plant at the future time (h) from a combination of the trending models, wherein the combination is selectively a linear combination or a nonlinear combination; wherein
on a condition that the linear combination is selected, combining the feature and the forecasted exogenous vector as a sum of:
the forecasted exogenous vector multiplied by a first weighting coefficient vector, and
a sum of the feature multiplied by a respective feature weighting coefficient, wherein the first weighting coefficient and the respective feature weighting coefficient are adjusted according to the trending range; and
on a condition that the non-linear combination is selected, combining the feature and the forecasted exogenous vector using a Gaussian process that takes the feature and the forecasted exogenous vector as inputs, the Gaussian process having automatic relevance determination such that all the inputs are assigned a weight coefficient nonlinearly automatically via a length scale parameter of the Gaussian process.
2. The method of claim 1 , wherein the data variable indicates at least one of: an amount of cloud cover, an ambient temperature, and an amount of humidity.
3. The method of claim 1 , wherein the extracting of the feature comprises:
selecting part of history data that occurs within the duration associated with the feature;
determining a polynomial that best fits the selected part; and
deriving the feature from the determined polynomial and the future time (h).
4. The method of claim 3 , wherein a highest order power of the polynomial is one.
5. A computer program product for forecasting power generation by a solar power plant using combined trending models, the computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to perform a method comprising:
forecasting an exogenous vector of weather-related exogenous variables for a future time (h) that is likely to affect the ability of the solar power plant to produce electricity;
generating a plurality of trending models representative of different trending ranges of prior data observations, each trending model generation comprising:
extracting a feature from a trending function, wherein the extracted feature represents a power value at the future time (h) based on the trending function; wherein the trending function is associated with a time period (w) of a duration different than other trending functions and derived from prior observed amounts of power generated by the solar power plant during a corresponding time period (w);
predicting an amount of power that will be generated by the solar power plant at the future time (h) from a combination of the trending models, wherein the combination is selectively a linear combination or a nonlinear combination; wherein
on a condition that the linear combination is selected, combining the feature and the forecasted exogenous vector as a sum of:
the forecasted exogenous vector multiplied by a first weighting coefficient vector, and
a sum of the feature multiplied by a respective feature weighting coefficient, wherein the first weighting coefficient and the respective feature weighting coefficient are adjusted according to the trending ranges; and
on a condition that the non-linear combination is selected, combining the feature and the forecasted exogenous vector using a Gaussian process that takes the feature and the forecasted exogenous vector as inputs, the Gaussian process having automatic relevance determination such that all the inputs are assigned a weight coefficient nonlinearly automatically via a length scale parameter of the Gaussian process.
6. The computer product of claim 5 , wherein the computing of the feature comprises:
selecting part of history data that occurs within the duration associated with the feature;
determining a polynomial that best fits the selected part; and
deriving the feature from the determined polynomial and the future time (h).
7. The computer product of claim 6 , wherein a highest order power of the polynomial is one.