With the advent of the big data era, real-time data stream processing technology has gradually become a key technology in various industries, especially in the fields of finance, logistics, social networks and intelligent manufacturing. As an efficient distributed stream processing framework, Apache Flink is widely used in large-scale data stream processing with its low latency, high throughput and high scalability. This paper designs and optimizes the real-time big data stream processing model based on Apache Flink, focusing on the balance between throughput and latency, and proposes optimization strategies to improve the system’s throughput, reduce latency and enhance fault tolerance. Through experimental verification, the optimized Flink system has increased throughput by 50%, reduced latency by 40%, and in the system fault simulation test, the recovery time has been shortened by 40%, and the data consistency has been maintained at more than 90%. The experimental results show that the optimized Flink system can run efficiently and stably in a large traffic and high concurrency environment, providing an efficient and scalable solution for real-time data stream processing.