The authors have been developing a teaching support system by using neural network/deep learning technologies. This paper deals with detecting persons with or without a raised hand, tracking, and identifying individuals in a large classroom by using a camera. The algorithm consists of object detection by a deep learning algorithm YOLO and individual identification by cascading several neural networks. An experiment of automatically calling on a person among those with raised hands is also conducted in a real setting to avoid concentration to a very limited number of students. PTZ camera is shown to be essential in this system. In the real-time experiment, Jetson AGX Xavier, a portable GPU machine, is used to meet the requirement of computationally expensive task, and our system is shown to be useful.

In recent years, the occurrence of local torrential rain has increased, and an accurate prediction model is required. Atmospheric water vapor measurement based on the zenith total delay (ZTD) produced by the precise point positioning processing employed in the Global Navigation Satellite System (GNSS) is effective for forecasting. Recently, there has been a lot of research into applying deep learning to forecasting, however, it could not be practical. In this paper, we introduce the Long Short-Term Memory (LSTM) built by a neural network algorithm to effectively model the discrete time series of rain rate, the ZTD and the meteorological sensing data work as explanatory variables. A key message from this analysis is that a deep learning model has the capability to follow the climate variation as long as a short-term event even though it exists spatial locality.

In order to give a theoretical guarantee for frequency response measurement in self-oscillated amplitude estimation problem for thermoacoustic systems, this paper deals with a stability analysis problem of steady-state oscillation control system which automatically maintains the amplitude of plant output at a reference value for frequency response measurement with a resonance in the test frequency range. First, based on a simple second-order oscillation model, it is shown that the stability region on the PI(proportional and integral)-gains plane strongly depends on the excitation frequency in numerical simulation, and almost the same stability region is obtained by numerical analysis of the linear approximation problem of the closed-loop system. Then, the same problem is further analyzed in theoretical way which shows as a result that the stability region on the PI-gains plane is given by a frequency-dependent hyperbolic boundary, and there exists a common PI-gains region for the whole test frequency range to be considered. A setting method of PI-gains is also given.

Microgrids are new power supply systems expected to expand the introduction of Distributed Energy Resources (DER) using Renewable Energy Sources (RES). There are some challenges in a microgrid. First, DERs are power sources with no mechanical inertia. Second, synchronous generators have risks at step-out because of the perturbation of DER using RES. To suppress the angular frequency variation of power sources is necessary, it is valid to introduce a Virtual Synchronous Generator (VSG) and smooth power flow using batteries. For a safe operation in a microgrid with VSG, differential games-based analysis and control are used. The reachable set-based transient analysis represents sufficient conditions for the system to become unsafe. The control based on Iterative Linear Quadratic (ILQ) Games derives an optimal trajectory avoiding unsafe sets of the system. Finally, the numerical solution shows that reachable set-based analysis for microgrids and ILQGames-based control of power sources keep microgrids safe.

We present in this article a pursuit controller with simultaneous data-based 3D target motion prediction and switching estimation by a 2D camera when a moving target may switch between multiple motion patterns. The switching target motion is modelled by a new Gaussian Process model for rigid body motions that can predict velocity fields based on observed target motion data and an online switching estimator. We further prove that the proposed methods for motion prediction and visual pursuit ensure stability and demonstrate the increased performance of pursuing a target by a camera-equipped drone in a Digital Twin simulator.