Deep reinforcement learning (DRL) combines the powerful environmental feature perception capabilities of deep learning with the environmental interaction capabilities of reinforcement learning, often used to handle perception-decision problems. Value-based algorithms are the most basic and common type of algorithm in reinforcement learning. The core idea of this algorithm is to continuously interact with the environment through the agent and optimize the value function according to the feedback, to establish a measure of the cumulative future reward of each policy in the entire action state space, and to take it as the basis for selecting the best policy. Value-based reinforcement learning includes dynamic programming, Monte Carlo, and temporal difference learning, etc. Combined with deep learning technology, value-based deep reinforcement learning has evolved into a series of deep reinforcement learning methods such as deep Q-network (DQN) and its derivative algorithms, which are becoming representatives of this type of algorithms. This chapter is to introduce DQN, double DQN, prioritized experience replay DQN, and dueling DQN.