Deep Learning Based Prediction Of Weather Using Hybrid_stacked Bi-Long Short Term Memory

Abstract

Climate and environment forecast is under control of very high dimensionality as well as interactions on various scale of temporal and spatial factors and disorganized dynamics resulting numerous problems and complications in the field. Furthermore cutting edge mathematical models, in spite of their immense computational expenses are not adequate for some applications. In this way, it is interesting to utilize arising new innovations like Artificial Intelligence or computerized reasoning to handle these issues. This work illustrates the utilization of deep learning models to imitate the full dynamics of improved general circulation model, provide improvised results in the weather prediction as well as accurate and much stable long-term climate time series. Combinations of different techniques of deep learning are used in this work for prediction of weather. Hybrid\underscore Stacked Bi-LSTM model is proposed which comprises of both LSTM and Bi-LSTM to train our model and before training our model the data used for this has been pre-processed using standard scaling technique to make it accurate and in desired format. An improvised weather prediction technique is presented here, using historical data from various climate stations to prepare Deep Learning models, which helps to predict futuristic weather conditions within a very short time span. The performance of proposed model is computed using various regression metrics and the results shows that the model is performing better than the present state-of-the-art techniques.