General solutions for nonlinear differential equations: a rule-based self-learning approach using deep reinforcement learning