Table 3 The experimental results of learning rate adjustment.

From: Channel equalization in ultraviolet communication based on LSTM-DNN hybrid model

Learning rate

The number of training epochs required to reach convergence

The final accuracy of the test set

Stability of loss reduction

0.0001

400

91.2%

Significant fluctuations, occasional rebound

0.00005

450

92.5%

Stable decline, slightly fluctuating

0.00002

500

93.1%

Steady decline without significant rebound