Aiming at the problems of loose group organization, high cost and overdue tasks caused by unplanned individual decision-making in open environments, the work aims to study the open crowdsourcing task optimization based on reinforcement learning. Firstly, the self-organizing decision-making behavior of individual selection whether to participate in the task and contribution module was analyzed and modeled by the utility theory. Then, according to the Markov decision process, the state and action in the crowdsourcing task were constructed, and the reward was set to shorten the task time and reduce the task cost. Finally, based on the DDPG reinforcement learning algorithm, the strategy and value network were constructed to dynamically set the bonus and ability index requirements. The simulation results showed that this method could effectively shorten the task time and reduce the cost, which provided an effective solution for open crowdsourcing task optimization. Using the deep reinforcement learning method to dynamically and intelligently regulate the bonus and capability index requirements in crowdsourcing realizes the shortening of task time and the reduction of task cost, and provides a reference for crowdsourcing decision-making in complex open environments in the future.