MQTT-Based Architecture for Real-Time Data Collection and Anomaly Detection in Smart Livestock Housing

Abstract

This study designed a message queuing telemetry transport (MQTT)-based communication framework to acquire environmental data with stable, low-latency response (soft real-time capability) and detect anomalies in smart livestock housing. We validated the performance of the proposed framework using actual sensor data. It comprises environmental sensor nodes, a Mosquitto MQTT broker, and a GRU-based anomaly detection model, with data transmission via a WiFi-based network. Comparing quality of service (QoS) levels, the QoS 1 configuration demonstrated the most stable performance, with an average latency of ~150 ms, a data collection rate ≥ 99%, and a packet loss rate ≤ 0.5%. In the sensor node expansion experiment, responsiveness (≤200 ms) persisted for 10-15 nodes, whereas latency increased to 238.7 ms for 20 or more nodes. The GRU model proved suitable for low-latency analysis, achieving 97.5% accuracy, an F1-score of 0.972, and 18.5 ms/sample inference latency. In the integrated experiment, we recorded an average end-to-end latency of 185.4 ms, a data retention rate of 98.9%, processing throughput of 5.39 samples/s, and system uptime of 99.6%. These findings demonstrate that combining QoS 1-based lightweight MQTT communication with the GRU model ensures stable system response and low-latency operation (soft real-time capability) in monitoring livestock housing environments, achieving an average end-to-end latency of 185.4 ms.

Keywords: GRU; MQTT; QoS; anomaly detection; low-latency monitoring; smart livestock housing.

Figure 1

System Overview (Adapted from [22]).

Figure 2

MQTT message transmission and processing flow. Solid arrows indicate actual message transmission paths, while dashed arrows represent internal processing flows.

Figure 3

QoS-Level Comparison of MQTT Communication Performance.

Figure 4

Performance Comparison of Anomaly Detection Models.