769 Simulating time-series winds using genetic programming

Wednesday, 26 January 2011
Ethan Cook, Univ. of Oklahoma, Norman, OK; and M. L. Morrissey and J. S. Greene

Genetic programming is an algorithm-discovery heuristic employing mechanisms of biological evolution. In particular, algorithms or functions are represented by coded strings and array entries, or chromosomes, composed of smaller segments, or genes, which correspond to operators, variables and coefficients. A population of chromosomes evolves as members are selected or terminated based on mathematical fitness criteria and as new chromosomes are produced and modified via genetic crossover and mutation.

Genetic Programming has been examined for its usefulness in predicting temperature time series, sea surface temperature anomalies, monthly rainfall, stream flow and offshore winds. Initial results are promising. In this study, genetic programming is used to simulate land-based winds by developing predictor functions of several indicator variables at various distances and time-lags. In particular, functions are evolved to mimic the statistics of winds collected using Oklahoma Wind Power Initiative (OWPI) tall-tower instrumentation.

A simpler, preliminary demonstration is also given, applying genetic programming to synthetic data consisting of realizations of pseudorandom variables (indicator variables) and the values of various known functions of the indicator variables (the target variables).

Functions produced in this manner, applied to the appropriate meteorological and geographic predictor variables, have potential uses in a variety of wind-power applications including measure-correlate-predict (MCP) studies, wind resource mapping and plant output forecasting.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner