Optimization of Convolutional Long Short-Term Memory Hybrid Neural Network Model Based on Genetic Algorithm for Weather Prediction

It is an important task for us to forecast whether it will rain or not. Nowadays, the prediction methods based on neural network have great changes to the final prediction result with the different selection of hyper-parameters. In order to find an appropriate hyperparameter and build a stable neural network model, this paper proposes a GA-CNN-LSTM model which uses genetic algorithm to optimize the parameters of neural network. The model is divided into two parts. One part is CNN-LSTM, which serves as the mapping rule of input and output and plays the role of fitness function. The other part is genetic algorithm (GA), whose function is to find the optimal hyperparameter and its corresponding solution set. Applied to the real Australian weather data to test, the test results show that the accuracy of this model is 1.3527953% higher than that of the single CNN and 0.0934936% higher than that of the unoptimized CNN-LSTM model.