Stack layer & Bidirectional Layer Long Short - Term Memory (LSTM) Time Series Model with Intermediate Variable for weather Prediction

Abstract

Weather forecasting has been a difficult problem for researchers for many years and continues to be today. The development of new and fast algorithms aids researchers in the pursuit of better weather forecast approximations. This problem attracts researchers because of the changing behavior of the environment, the increase in earth's temperature, and the drastic changes in ecosystem. Almost everywhere in the world is currently experiencing a slew of natural disasters, including storms on land and sea that are destroying infrastructure and taking the lives of many people. Machine learning and deep learning algorithms gave researchers and the general public hope that they would be able to develop fast applications and predict weather alarms in real time. Because of the combination of deep learning and the large amount of weather data that is available, researchers are motivated to investigate the hidden patterns of weather in forecasting. In this paper, the proposed model will be used to analyze intermediate variables, as well as variables associated with weather forecasting. Long Short-Term Model (LSTM) accuracy is affected by the number of layers in the model, as well as the number of layers in the stacked layer LSTM and the number of layers in Bidirectional LSTM. Because of the inclusion of an intermediate signal in the memory block, the methods proposed in this paper are an extended version of the LSTM. The premise is that two extremely connected patterns in the input dataset can rectify the input patterns and make it easier for the model to search for and recognize the pattern from the trained dataset by building a stronger connection between the patterns. In every trial, it is necessary to comprehend a long-lasting model for learning and to recognize the weather pattern. It makes use of predicted information such as visibility, as well as intermediate information such as temperature, pressure, humidity, and saturation, among other things. In bidirectional LSTM, the highest accuracy of 0.9355 and the lowest root mean square error of 0.0628 were achieved.