In a context of stream processing, it is important to guarantee some properties of performance, quality of results and scalability to final users. Adjusting resource usage to processing requirements in order to consume only necessary resources, is a major challenge dealing with Big Data and Green IT. The approach suggested in this article, adapts dynamically and automatically the parallelism degree of operators belonging to a same continuous query. It takes into account the evolution of input stream rates. We suggest i) a metric estimating the activity level of operators in a near future ii) the approach AUTOSCALE which evaluates the gain brought by a set of the parallelism degree modifications at local and global scope iii) thanks to an integration to the solution Apache Storm, we show performance tests comparing our approach to the native solution of this stream processing engine.