Second-scale atmospheric turbulence predicting with the Kolmogorov-Arnold network

Area of Science:

• Optical Communications
• Atmospheric Physics
• Artificial Intelligence

Background:

• Free-space optical communication systems are vulnerable to atmospheric turbulence.
• Accurate, high-temporal-resolution atmospheric turbulence prediction is crucial for optimizing these systems, particularly satellite-to-ground laser links.

Purpose of the Study:

• To introduce a novel artificial neural network, the Kolmogorov-Arnold network (KAN), for predicting atmospheric turbulence.
• To evaluate KAN's performance in capturing complex correlations between meteorological parameters and atmospheric turbulence strength.

Main Methods:

• Developed a new artificial neural network, the Kolmogorov-Arnold network (KAN), based on the Kolmogorov-Arnold theorem.
• Utilized meteorological parameters to establish a nonlinear mapping for atmospheric turbulence prediction at a 1-second resolution.
• Introduced learnable nonlinear basis functions within KAN to enhance correlation capture.

Main Results:

• KAN demonstrated effectiveness in predicting atmospheric turbulence strength (Cn2) with a 1-second resolution.
• Mean Absolute Percentage Error (MAPE) and Symmetric Mean Absolute Percentage Error (SMAPE) were 31.54% and 31.44% respectively.
• KAN outperformed traditional Multi-Layer Perceptron (MLP) by reducing MAPE by 6.74% and SMAPE by 4.35%.

Conclusions:

• The Kolmogorov-Arnold network (KAN) offers a more effective approach for atmospheric turbulence prediction compared to traditional neural networks.
• KAN's ability to model complex correlations in meteorological data significantly improves the accuracy of atmospheric refractive-index structure constant (Cn2) prediction.
• This advancement provides strategic support for enhancing the reliability and performance of free-space optical communication systems.