In order to seek the optimal light source optimization scheme in different indoor environments, a deep learning based indoor visible light comminication light source configuration strategy is proposed. Introducing indoor obstacles, natural light, and human movement interference separately, with indoor signal-to-noise ratio uniformity as the objective function, using the flower pollination algorithm (FPA) to optimize the light source power and half power angle under different room conditions. Using the obtained room state and light source configuration as the training set, a convolutional neural network (CNN) is used for training.The resulting computational model can predict the optimal light source settings in different room states, achieving dynamic adjustment of light source configuration. The simulation results show that the qualification rate of the light source parameters of this strategy reaches 88%, and the predicted results meet the requirements in terms of room signal power and lighting intensity.