Weather parameters forecasting with time series using deep hybrid neural networks
Sercan Yalçın
Abstract
SUMMARY Weather forecasting has recently gained great importance in terms of predicting potential disasters, taking precautions, and increasing the quality of life around the world. Researchers apply optimization and artificial intelligence techniques to make weather forecasts based on various nature parameters. In this study, it is proposed a weather condition forecasting scheme with time series using deep hybrid neural networks. In the proposed scheme, the essential parameters for weather forecasting, namely, relative humidity, temperature, atmospheric pressure, and wind speed are trained and predicted with long short‐term memory (LSTM)‐convolutional neural networks (CNNs) models in a hybrid way. The values represented by the input neurons are first passed through the CNN layers for a clearer and more accurate estimation of the data. Then, after fine‐tuning, the results are sent to the LSTM block. The proposed hybrid deep method has been compared with both machine learning and deep learning methods According to the results, running the proposed method, the RMSE, MADE, and MAPE values are 1.82, 1.64, 1.21 for temperature, 7.12, 6.58, and 4.63 for relative humidity, 2.61, 2.36, and 1.86 for atmospheric pressure, and 1.06, 0.85, and 0.52 for wind speed, respectively.