Transferring Multiple Policies to Hotstart Reinforcement Learning in an Air Compressor Management Problem