LSTCN Approaches for Self-Adaptive Real-Time Weather Forecasting

Weather observations are recorded using sensor networks in the form of data streams. This data is most often used by predictive models based on machine learning to make weather forecasts. However, weather conditions can vary (or change) gradually or abruptly due to the chaotic nature of the climate. These characteristics mean that conventional machine learning-based weather forecasting models deployed in a real-world environment are inadequate for accurately modeling the dynamics of weather conditions due to their static parameters. This highlights the need for other approaches capable of learning beyond the production phase, which will allow their parameters to be kept up to date in real time in response to changing weather conditions. The objective of our study is first to evaluate the performance of online learning approaches (LSTCN and ARIMA-Online) compared to conventional ML methods (LSTM, CNN, LSTM+CNN), and then to evaluate and compare the performance between these online learning approaches. This evaluation and comparison study will be carried out on univariate and multivariate time series with short- and long-term horizons. The metrics used are MAE, learning time, and testing time. According to the results of our study, the LSTCN model is the most effective compared to conventional machine learning methods in terms of training time, testing time, and accuracy on short- or long-term univariate or multivariate time series. Similarly, this model outperforms the ARIMA OGD (Online Gradient Descent) model in terms of accuracy and testing time. However, its training time is slightly longer than that of the ARIMA OGD model. So, LSTCN model is better suited for real-time weather forecasting.