Short-term forecasting and diagnostic tools for severe changes of wind power production (power ramps) may provide reliable information for a secure power system operation at a small cost. Understanding the underlying role of the synoptic weather regimes (WRs) in triggering the wind power ramp events can be an added value to improve and complement the current forecast techniques. This work identifies and classifies the WRs over mainland Portugal associated with the occurrence of severe wind power ramps. The most representative WRs are identified on compressed surface level atmospheric data using principal component analysis by applying K-means clustering. The results show a strong association between some synoptic circulation patterns and step variations of the wind power production indicating the possibility to identify certain WRs that are prone to trigger severe wind power ramps, thus opening the possibility for future development of diagnostic warning systems for system operators use.