FUZZY Q-LEARNING IN SVD REDUCED DYNAMIC STATE-SPACE