EcoSta 2018: Registration
View Submission - EcoSta2018
A0343
Title: A transfer learning approach for modeling and monitoring in landslide sensor systems Authors:  Ke Zhang - The Hong Kong University of Science and Technology (Hong Kong) [presenting]
Zhenli Song - The Hong Kong University of Science and Technology (Hong Kong)
Fugee Tsung - The Hong Kong University of Science and Technology (Hong Kong)
Abstract: Landslides are common geographical activities that result in large quantities of rock and debris flowing down hill-slopes, leading to thousands of casualties and billions of dollars in infrastructure damage every year around the world. To detect such abnormal geographical behavior, on-site sensor systems are widely applied for data collection and many existing SPC methods can be adopted for modeling and monitoring. However, the conventional methods may fail to perform well for newly set-up sensors with small data collected. To make use of the new sensors effectively right after any scale-up of the system, we proposed a transfer learning based approach to jointly model the sensor data streams thus getting better understanding on new sensors by the information transferred from old sensors. In the approach, the parameters within auto-regressive models for individual sensors are connected using a Gaussian prior and certain regularization terms. An iterative updating scheme has been developed for parameter estimation in the integrated model. After modeling, a residual-based monitoring scheme is proposed accordingly. Various Monte Carlo simulations have been conducted to illustrate the performance of our transfer learning method over conventional ones. Real data example also shows that the proposed method can be effectively applied in real landslide monitoring applications.