Abstract: Addressing the critical challenge that existing statistical analysis and machine learning-based time series forecasting methods struggle to simultaneously capture temporal dependencies and nonlinear characteristics in solar F10.7 cm flux time series—particularly the anomalous fluctuations caused by radio burst events during solar maximum years, which lead to significantly higher prediction errors compared to solar minimum years—this paper proposes a mid-term forecasting method for the solar F10.7 index. The method innovatively integrates Long Short-Term Memory (Long Short-Term Memory network) and fully connected Neural Networks (NN), and incorporates influential factors related to the Sunspot Number (SSN), constructing a hybrid prediction model driven by multiple input variables based on LSTM-NN. Using measured F10.7 data from Solar Cycle 24, seven-day-ahead prediction experiments were conducted. The results demonstrate that the model achieves a prediction correlation coefficient of R=0.95 and a Root Mean Square Error (RMSE) of 11.27 sfu, reducing the prediction error by 7.5% compared to single-input-variable models, with particularly significant improvement in prediction accuracy during solar maximum intervals (error reduction of 8.5%). Through systematic analysis and experimental validation, it is proven that this hybrid model can effectively characterize complex solar activity features, fully leverage the informational value embedded in SSN sequences, and significantly enhance the accuracy and reliability of F10.7 index time series forecasting.