Constructing covert channels on blockchains has recently become a significant research focus. A major challenge lies in embedding data into blockchain transactions, while maintaining strong concealment. This study focuses on utilizing blockchain transaction values as a carrier, exploring methods to ensure that the distribution of values with embedded data mimics that of standard transactions. Initially, we gathered authentic blockchain transaction data, conducted a frequency analysis on the transaction values, and filtered out abnormal values to establish a baseline normal value dataset. Subsequently, we employed a Restricted Boltzmann Machine (RBM) to generate a simulated value dataset that mirrors the distribution of the normal dataset, facilitating the embedding of data within the simulated value dataset. By fine-tuning the number of bits embedded in the values, we achieved a value dataset with embedded data that closely resembles the normal distribution. We have calculated the Kullback-Leibler Divergence and the Kolmogorov-Smirnov Test between the value dataset with embedded data and the normal value dataset under various embedding scenarios, revealing that the changes to the original dataset are minimal.