TRANSMISSION DELAY PREDICTION METHOD FOR OPTOELECTRONIC COMMUNICATION SYSTEMS BASED ON TEMPORAL FEATURE ENHANCEMENT

Abstract

Transmission delay prediction is essential for low-latency service assurance and intelligent operation in optoelectronic communication systems. In practical transmission environments, delay is not only determined by physical propagation distance, but is also affected by traffic load, bandwidth utilization, queue status, bit error rate, and optical link quality. To improve prediction accuracy under dynamic system conditions, this paper proposes a transmission delay prediction method based on temporal feature enhancement. The proposed method first analyzes the main factors influencing transmission delay and constructs a prediction variable set including historical delay, traffic load, bandwidth utilization, queue length, bit error rate, and received optical power. Then, lag features, sliding-window statistical features, and difference features are introduced to enhance the temporal representation of the original monitoring data. Based on the enhanced feature sequence, an LSTM-based prediction model is developed to learn the nonlinear temporal relationship between system states and future transmission delay. Experimental results show that the proposed method achieves lower prediction errors than ARIMA, SVR, XGBoost, and standard LSTM models. The ablation analysis further confirms that temporal feature enhancement can effectively improve the model's ability to capture historical dependence, short-term fluctuation, and dynamic variation trends.