Leveraging machine learning and low-cost hardware for economical wind-driven water level prediction

Although the prospect of the Internet of Things (IoT) is almost unlimited, developing IoT techniques can look discouraging, demanding a multidomain expertise and complex web infrastructure. Combining machine learning and data analytics by smart connected devices can enable a large variety of applications including sophisticated predictive maintenance systems, home-grown traffic monitors, and futuristic user goods (such as the Google Nest and the Amazon Echo). This work develops a novel method to estimate tide levels integrating wind records by means of two neural network architectures. We merge past measurements of tide level with wind data that available to be downloaded. All the data are pre-processed by sorting out tidal harmonics and wind-induced surges. Correlations between surge variations and wind stress are then computed. Next, fitting neural network and NARX network are trained after preparing data that include actual tide levels, estimated harmonic tide altitudes, and wind stress elements. We evaluate both models using the error performance and show the deployment of the selected model with MATLAB ThingSpeak visualization environment. The results demonstrated that the performance of input–output Neural Fitting was 0.0592, while it was 0.0039 for the NARX neural network.