The paper presents aspects of model-free learning control initialization. Model-free learning has several advantages as general purpose approach or adaptive capability. However, practical implementation is not intuitive in each step. Choice of time scale in order to provide the necessary reactivity or level of granularity of states or control actions is not an easy problem. Before it learns environment it may disturb process control in an unacceptable way. This is the motivation to find rules for easer initialization and controller set-up.