With the increasing penetration of renewable energy sources in the electrical power grid, demand response via thermostatic appliances such as electric water heaters is a promising storage means to compensate the significant variability in renewable generation power. We propose a multi-stage stochastic optimization model that computes the optimal day-ahead target profile of the mean thermal energy contained in a large population of heaters, given various possible wind power production and uncontrollable load scenarios, where this optimal profile is calculated to make the variable net demand as flat as possible.