Research on Weather Prediction based on Long Short-Term Memory with Elite Firefly Algorithm

The weather prediction is a usage of science and technology for forecasting the conditions of weather in a particular. The precise and accurate prediction in rainfall, particularly of heavy rain is challenging for weather prediction models. Hence, this research proposed a Long Short-Term Memory with Elite Firefly Algorithm (LSTM-EFA) for weather prediction. The LSTM overcome the vanishing and exploding gradients to capture both short and long-term dependencies of the weather data inputs. In EFA, the elite replacement strategy supports to replaces all the weaker solutions with the better one which uses in the later iterations to construct a more accurate prediction model. Moreover, its chaotic sequences or other dynamic processes for enhancing the variety of the search and avoiding the troubled convergence too early. The Root Mean Square Error (RMSE), Mean Squared Error (MSE), Mean Absolute Error (MAE) and coefficient of determination (R2) are estimated to show the effectiveness of LSTM-EFA performance. The LSTM-EFA obtained 0.80 RMSE, 0.67 MSE, 0.72 MAE and 0.98 R2 which is better than other existing algorithms.