Sensor network-based wind field estimation using deep learning

Abstract

The incorporation of wind fields, or movement of clouds, significantly improves the accuracy of time-series-based solar irradiance prediction models. To resolve problems regarding the cost and accuracy of current wind field estimation methods, there are the challenges in estimating wind fields using only solar irradiance sensor networks and evaluating the performance of models. We propose a cost-effective and reliable method to estimate wind fields through the application of Deep Learning and computational geometric algorithms. Using a realistic cloud simulator, validation datasets for the proposed model were generated, accounting for various complex factors including topology of sensor placement, changing wind speed and direction, and cloud density that directly impact sensor data. Preliminary qualitative and quantitative results indicate promising potential for practical deployment as an estimation model.