Conclusions
The variability present in the wind power can be recognized by information theory in a dynamical way. The mutability µ(t) of the recent WEP produc tivity is a valid indicator for the changes in the productivity pattern. When this is combined with the time variation of WEP we can define function alert A(t) which focuses on the increases of WEP only. A calibration procedure can allow to determine the threshold value for A(t), namely AC, which announces a period of high wind energy generation. We have presented a way to do this using the data for all Germany. Similar procedures could be established for local plants. Moreover, seasonal corrections can also 508 be contemplated. In this way the information content of the time series giving the actual electricity generated by wind turbines can be used to predict its favorable periods. This is similar to what has been done in the case of economical variables [25,26] and biomedical data [27,28]. In a way, this instrument can be thought of like a ”thermometer” measuring the positive agitation prior to a potent period of WEP. It is possible to calibrate a protocol according to different local conditions and productivity levels. As explained above wlzip allows tuning of several knobs to optimize its performance. The numerical basis can be chosen in accordance 518 with the range of oscillations of the data: for relatively small oscillations a low numerical basis should be used (In the case of blood pressure data a binary basis was used [27,28]); we settled for a quaternary basis for the data adding all WEP sources in Germany. The field can be tuned so as to begin the data recognition over the most sensitive digit subtle to change within the time window of operation; the third digit in quaternary basis turned out to be appropriate for the present study. The number of significant digits over which the data recognition is to be performed can also be adjusted to each problem; three digits in quaternary basis are enough in the present case. In the case of dynamical analysis like this one the time window over which the data recognition is to be performed is the most delicate choice to balance both precision (long time windows) and anticipation (short time windows).
