A Reinforcement Learning Approach for Defending Against Multiscenario Load Redistribution Attacks