The underwater non-imaging optical Multiple Input Multiple Output-Orthogonal Frequency Division Multiplexing (MIMO-OFDM) communication system adopts intensity modulation-direct detection method without the frequency and phase components. The beams of different light sources superimpose each other at the receiving end, so the channel correlation is strong and the bit error rate (BER) is high. In order to solve this problem, a precoding algorithm based on MIMO-OFDM received signal Euclidean distance is proposed.Assuming that the transmitter anticipates the channel state information,the optimal precoding matrix is solved under the constraints of non-negative optical signal and limited total power,and the minimum Euclidean distance of the received signal is maximized by allocating different power to different light sources. The transmitting end of the system precodes the signal through the precoding matrix, and the receiving end decodes the received signal which maximizes the Euclidean distance by the maximum likelihood method. The simulation results show that compared with the optical MIMO precoding algorithm based on singular value decomposition (SVD), the proposed algorithm can reduce the BER of the non-imaging optical MIMO-OFDM system under the premise of constant total power.