Estimate of Short-Term Atmospheric Turbulence Intensity by BP Neural Network

Free-space optical communication is strongly affected by atmospheric turbulence, which causes optical intensity fluctuations and beam wander and reduces system reliability. To improve the accuracy and efficiency of atmospheric refractive index structure constant Cn2 estimation, a backpropagation neural network (BPNN) method was used combining principal component analysis (PCA) and the mind evolutionary algorithm (MEA). Optical intensity fluctuation experiments were conducted in the near-surface atmosphere of Lanzhou from March 6 to 16, 2025, during the spring monsoon transition period. PCA was used to optimize the input features, while MEA was applied to optimize the parameters of the BP network. The results show that the correlation coefficient between the estimated and observed values exceeds 0.76. PCA compresses the original 80-dimensional spot features into 10 principal components while retaining over 95% of the cumulative variance explained. The correlation coefficient for PCA-MEA-BPNN reaches 0.7627, which is 0.3375 higher than that of the conventional BPNN (0.4252)—representing an increase of 33.75 percentage points. The proposed method can effectively reproduce short-term variations in atmospheric turbulence intensity. These results provide technical support for turbulence sensing and adaptive control in free-space optical communication systems. In this paper, we conduct research in the following three aspects: (1) constructing an 80-dimensional feature vector from measured spot images using wavelet transform; (2) compressing this feature vector into 10 principal components via PCA, while retaining over 95% of the cumulative variance explained; (3) employing MEA to identify optimal initial weights and thresholds for BPNN, and validating the estimation performance of the combined method for Cn2 using 700 m near-ground link data.