For signal processing related to localization technologies, non line of sight (NLOS) multipaths have a significant impact on the localization error level. This study proposes a localization correction method based on convolution neural network (CNN), which extracts obstacle features from maps to predict the localization errors caused by NLOS effects. A novel compensation scheme is developed and structured around the localization error in terms of distance and azimuth angle predicted by the CNN. Four prediction tasks are executed over different building distributions within the maps for typical urban scenario, resulting in CNN models with high prediction accuracy. Finally, a thorough comparison of the accuracy performance between the time difference of arrival (TDOA) localization algorithm and the results after the error compensation reveals that, generally, the CNN prediction approach demonstrates great localization error correction performance, improving TDOA accuracy by 75%. It can be observed that the powerful feature extraction capability of CNN can be exploited by processing surrounding maps to predict the localization error distribution, showing great potential for further enhancement of TDOA performance under challenging scenarios with rich multipath propagation.
A series of fast protection methods for multi-terminal DC (MTDC) grid have been proposed. However, most of these methods simply use voltage change rate or current change rate, but do not make full use of transient voltage and current information to identify fault regions. This paper proposed a protection method based on the cosine similarity of the transient voltage and transient current. Firstly, the characteristics of transient voltage and current in the case of short-circuit fault are analyzed. Then, the cosine similarity of the transient voltage and the current is introduced, and a protection method is constructed based on the cosine similarity difference of the internal faults and external faults. Finally, a simulation model is built in PSCAD, and the validity of proposed protection method is verified.
In a number of cutting-edge application fields, including image identification, speech and natural language processing, and autonomous driving, machine learning has had considerable success recently. In this study, we use the western black-crowned gibbon song data as the research object and the Kubernetes-based OpenPAI machine learning platform to categorize five environmental sounds in the western black-crowned gibbon habitat in the Mourning Mountains with an accuracy of 97.24%.
. The objective of this paper is to identify loads on suspenders in bridges. The load value at each time was selected as the parameter of identification, and the objective function was built based on measured and relevant simulated data. The differential equation of suspenders vibration was employed, and the relationship between load and acceleration response was given by superposition method. In addition, regularization technology was used in calculations to alleviate ill-posedness of the problem. Taking one tied arch bridge as an example, the load on one suspender of the bridge was obtained. The identification results indicated that the simulated responses were basically consistent with the measured data, which showed effectiveness of the method.
Visual-Inertial Odometry (VIO) is a new technology to determine position and attitude of the object which it is mounted on. Because of its small size and powerful performance, VIO system is widely used in virtual and augmented reality, autonomous driving and robot navigation. Usually, we assume the timestamps of the Inertial Measurement Unit (IMU) and the camera is synchronized, and there is no temporal misalignment (time offset) in VIO system. However, in practice, the different triggering time and propagation delays cause the sensors to be temporal misaligned. If we calibrate the temporal misalignment, the performance of VIO system can be greatly improved. To this end, we propose a novel method to calibrate the time delay between visual and inertial devices. This method accurately estimates the IMU state, time offset and camera pose in the VIO system by using Adaptive Kalman Filter. Simulation experiments show that our method can quickly and accurately estimate the time delay. The results compared against other methods proves that the proposed approach can significantly boost the performance of VIO system.
