Abstract: Dynamic and uncertain optimization problems pose a serious challenge to many intelligent optimization algorithms. A continuous action learning automaton (CALA) is introduced that can solve such problems. The automaton uses a variable interval as its action set, and generates actions with uniform distribution over this interval. The endpoints of the interval are updated according to the best historical action within a sliding window. Simulation results are presented to show the performance of the automaton in two time-varying stochastic environments. It is shown that the new algorithm exceeds three of the traditional CALA algorithms in the accuracy of learning, the rapidity of response, and the behaviors in the worst case.