Generation of inputs to renewable energy sources using matched-block bootstrap approach with fitness proportionate selection

Abstract

Renewable energy sources produce clean, green energy, but their production is highly variable in time due to changeable weather conditions. Energy management systems are implemented to cope with that problem. Their proper design requires an exhaustive testing. Creation of realistic test cases requires certain amount of test data (such as wind speed and insolation). As the required amount of measured data is usually not available, they have to be generated in a way that preserves the statistical qualities of the real life phenomenon. In this article a matched-block bootstrap with fitness proportionate selection method is presented. It is a non-parametric method that samples data blocks from a real data set and concatenates them into a new data set. To model the seasonal cycles (especially visible for temperature and solar irradiance) the blocks in the bootstrap method are categorized by month, day and time, creating sets of subsequent blocks from different years with the same time and date. To maintain the coherence of data, the fitness proportionate selection methods are introduced. They provide mechanisms to choose better matching blocks with higher probability. Two fitness functions are considered for this, one using an inversion and another using a negation operation. The matched-block bootstrap methods were tested using 9 years of measurements of solar irradiance and 10 years of data of wind speed in central Poland both taken at 10-minute intervals. The generated time series have the same values for the basic statistic factors as the original data and allow for creating test sequences of an arbitrary length.