Exchange rate fluctuations have significant impacts on the global economy. In order to predict exchange rate trends more accurately, this paper explores a method for exchange rate prediction using multi-country exchange rate data, combining Long Short-Term Memory (LSTM) and Transformer models. LSTM demonstrates outstanding performance in handling time series data, effectively capturing long-term dependencies, while the self-attention mechanism of the Transformer model excels in handling both local and global relationships within sequences, allowing for parallel computation and enhancing computational efficiency. By integrating these two models, it can better reflect the trends of exchange rate fluctuations in various countries. Due to the complex interactions between the global economy and financial markets, analyzing exchange rate data from 23 countries can more comprehensively uncover the internal relationships between exchange rates of various countries, thereby predicting a country’s exchange rate more accurately. Empirical studies have shown that the use of the LSTM-Transformer fusion model performs remarkably well in exchange rate prediction. This research provides new directions and methods for further exploration in the field of exchange rate prediction.