The non-linear characteristic of the light emitting diode can give rise to non-linear distortion of the light signal, aiming at this problem, the artificial neural network is introduced to the receiver end to realize the non-linear distortion suppression for the signals, and then the bit error rate of the visible light communication is reduced. The input electrical signal of the light emitting diode and the transformed electrical signal of the receiver are combined paired-wise data, then the paired-wise data is collected to train the artificial neural network, and learn non-linear distortion characteristic of electric-optical conversion, channel transmission and optical-electric conversion. As a result, the non-linear distortion of signals is estimated and suppressed with the neural network. Besides, EDA is adopted to search the super parameters of the neural network during the training phase, to reduce the training difficulty. Experimental results show that, the proposed method can effectively improve the performance of the visible light communication with different channel conditions.