The reinforcement learning-based multi-agent cooperative approach for the adaptive speed regulation on a metallurgical pickling line