Weather forecasts exert both direct and indirect influences on a nation’s economy and the well-being of its populace. The major goal of this study is to predict the weather parameters using deep learning weather prediction algorithms, such as recurrent neural networks (RNNs), long short-term memory (LSTM), and gated recurrent units (GRU). In this study, we have implemented three deep learning RNN models. The first one is basic (i) RNNs, which is the most classical and traditionally used architecture with time series sequence data forecasting problems, and the second one is (ii) LSTM, which is mainly developed to overcome the vanishing gradient problem of RNN. The third one is (iii) GRU, which is similar to LSTM but it takes less time to train the models compared to LSTM. Here, we have used different time series-based data preprocessing techniques like anomaly detection, interpolation, and Dickyfuller test techniques to filter the time series weather data. We have implemented these three models with weather datasets for univariate analysis. We have done future prediction of temperature features by taking previous 100 days data and predicted 30 days temperature with LSTM and with GRU, we took past 4000 days data and predicted 1000 days temperature. Finally, we evaluate model performance which is good enough than other traditional model accuracy results.