Since power line channel noise interference is large, noise modeling is complicated, and it is difficult to reconstruct noiseless signal samples. A Noise2Noise neural network optimization model for a one-dimensional time series is proposed, and the feasibility of Noise2Noise algorithm to suppress power line noise is verified and analyzed. First, the principle of the Noise2Noise algorithm is analyzed and theoretically deduced. Second, by selecting the appropriate network structure and improving network input and output, the neural network is suitable for processing experimental data. Sine wave data is used to test the network and verify the model. Then, by building an orthogonal frequency division multi-plexing (OFDM) modulation model, OFDM symbols are generated, and specific noise is added to form noisy samples. Finally, based on the improved network structure, the noise sample test of an OFDM leader sequence of power line received signals is realized. This enhances the feasibility and effectiveness of Noise2Noise method and has good practicability.