In numerous fields such as industrial production, intelligent transportation, and aerospace, multi-sensor systems have become critical infrastructure. These systems typically contain hundreds or even thousands of sensors, each generating massive amounts of parameter data over time. The data are interrelated, and even slight changes in the parameter distribution of one sensor can trigger a chain reaction, causing abnormalities in other parameters and even system paralysis. This can lead to significant economic losses and safety risks. Therefore, there is an urgent need for accurate and real-time anomaly detection technology for multi-sensor systems. Focusing on this challenge, our project has developed a deep structural model based on Bi-LSTM AutoEncoder. This model can monitor the parameter status of all sensors in the system in real-time and comprehensively. By capturing the temporal dependencies in time-series data through Bidirectional Long Short-Term Memory (Bi-LSTM) networks and learning and reconstructing normal data features via an AutoEncoder, the system triggers an alarm mechanism immediately when parameter features deviate from the normal range. This notifies relevant personnel for maintenance in time. The system's real-time detection capability allows it to issue alerts at the first sign of anomalies, minimizing their impact on system operation.