Performance analysis and implementation of an adaptive real-time weather forecasting system

Recently several real-time weather forecasting systems based on Internet of Things (IoT) have been developed to provide short-term real time forecasts also referred to as Nowcasts. The main challenge in these systems is to use appropriate prediction algorithms that can predict different weather parameters with the highest possible accuracy. In this work, an IoT based weather forecasting system has been implemented to provide short-term weather forecasts on a University campus at intervals ranging from 20 min to one hour. Several adaptive forecasting algorithms based on variants of the Multiple Linear Regression technique as well as K-Nearest Neighbors (K-NN) have been experimented. Moreover, three adaptive selection criteria for selecting the most appropriate prediction algorithm for a given Nowcast have been developed and tested. The parameters analysed are: temperature, humidity, atmospheric pressure, rainfall, luminosity, wind-speed, and wind direction. The best adaptive schemes are able to predict these parameters with an overall percentage error of 6.58% as compared to non-adaptive ones which predicted with a worst case percentage error of 13.59%. (C) 2018 Elsevier B.V. All rights reserved.