Two interns developed an innovative algorithm to predict road surface temperature by integrating fleet data and third-party weather sources. The approach combines ambient temperature, sun irradiance, wiper speed (from vehicle sensors), and cloud coverage and humidity (from external data sources). The fleet data, consisting of sensor readings transmitted to our AWS cloud platform, was complemented by ground truth temperature readings from road weather stations to ensure model accuracy. The project led to the development of two algorithms: a physical model and a machine learning-based Extra Trees model. Both models outperformed the current solution. The physical model, which offers superior interpretability, will be implemented to enhance the model’s quality and reliability.
Languages/frameworks: AWS, Python, React, TypeScript, k8s, Airflow, Java, Spring Boot, Go, Flink, Kafka
