This paper reviews the current status and challenges of Deep Reinforcement Learning (DRL)-based algorithm optimisation and risk management for high-frequency trading. By analysing the potential application of Deep Reinforcement Learning in high-frequency trading, its unique advantages in algorithm optimisation, trading decision-making and risk management are discussed. Although DRL demonstrates the ability to make self-adaptive and dynamic decisions in complex market environments, it still faces many challenges such as insufficient real-time algorithmic performance, data sparsity, model overfitting, and risk management complexity in practical applications. This paper summarises the main findings of the current research and proposes directions for future research, suggesting that the application of DRL in high-frequency trading can be further enhanced by improving the algorithmic structure, dealing with data sparsity, and optimising risk management strategies.