China's architectural lighting design mainly adopts the method of utilization factor with relatively com- plicated calculation and it takes a lot of calculation time in the actual design. There may appear large errors when some domestic companies complete lighting design by getting room sizes non-automatically and checking the specification repeatedly. The electrical lighting design automation system uses the average integral projection function and morphological operation to extract the load-bearing walls and improve seed filling algorithm to i- dentify common walls, which can effectively extract all the walls in the building and extract the size of each room. By extracting the Gabor feature and using the Bayesian formula which integrates the classifier of Chinese character rude classification with the classifier of Chinese character particular classification, it can accurately obtain room parameters including the types of rooms. Then, it applies dialux to collect the lighting scheme in the rectangular room that meets the lighting design requirements as a training sample to train the generalized regres- sion neural network (GRNN). In order to improve the prediction accuracy, the 4-fold cross-validation method is used to optimize the smoothing factor and obtain the best input and output values. Experiments show that the op- timized GRNN network has faster convergence speed. The case simulation proves that the proposed electric lighting design automation system can quickly and automatically obtain the lighting layout plan that meets the specifications.