In time-varying environments, traditional models only remove some communication crosstalk and have high transmission losses. Therefore, a machine learning channel modeling method for free space optical communication systems in time-varying environments is proposed. The histogram statistical method is used to calculate the peak values of the dataset, select the shortest communication path, obtain the optimal channel parameters through genetic algorithm, calculate the communication impedance based on the optimal parameters through machine learning, obtain capacitance and conductivity values, adjust the weights to remove communication crosstalk, convert the antenna domain into beam domain based on the channel size characteristics, and build a free space optical communication channel in a time-varying environment. The simulation experimental results show that the average transmission loss of the model in this paper reaches 148 dB, which is relatively low and has high application value.