Traditional weather forecasting methods are often difficult to cope with the complex relationships and temporal dependencies between multiple meteorological factors. Therefore, this study proposes a neural network model based on multivariate short term memory (LSTM) to explore an innovative approach to improve the accuracy of weather forecasting. We use a multivariable LSTM model with multiple meteorological variables as inputs to better capture the timing characteristics of weather systems. After training and verification on the actual meteorological data, the results showed that the accuracy of the multivariate LSTM model improved significantly: RMSE decreased from 0.0543 to 0.0137; MSE decreased from 0.0544 to 0.0140. Compared with traditional methods, this model can predict weather conditions more accurately and capture the complex relationship between different meteorological factors.