Keywords
machine learning, atmospheric clearness index, forecasting
Start Date
1-7-2010 12:00 AM
Abstract
We have developed a framework that integrates statistical and machine learning models for the short-term forecasting of a climatic parameter know as the atmospheric clearness index. We have used a multivariate regression to establish the most significant variable amongst all the previous values for the clearness index series. The value of this variable was used to divide all the available data in several intervals. Three different models were used to select the model that best fits the observations for each of these intervals. With this set of models, one for each interval, we have built a framework that enables the short term forecasting of the clearness index. In this framework, the best model is used for each prediction taking into account the current value of this parameter. Data from 10 Spanish locations have been used to build the framework and to check the forecasting accuracy. The results show that prediction involving different models is better than when only one model is used.
Integration of Statistical and Machine Learning Models for Short-term Forecasting of the Atmospheric Clearness Index
We have developed a framework that integrates statistical and machine learning models for the short-term forecasting of a climatic parameter know as the atmospheric clearness index. We have used a multivariate regression to establish the most significant variable amongst all the previous values for the clearness index series. The value of this variable was used to divide all the available data in several intervals. Three different models were used to select the model that best fits the observations for each of these intervals. With this set of models, one for each interval, we have built a framework that enables the short term forecasting of the clearness index. In this framework, the best model is used for each prediction taking into account the current value of this parameter. Data from 10 Spanish locations have been used to build the framework and to check the forecasting accuracy. The results show that prediction involving different models is better than when only one model is used.
