Search

To overcome the strikes from the tremendous noise of pseudorange and the integer ambiguity of carrier phase observation, we developed a novel dual-domain filtering method that integrates the traditional observation- and an innovative position- domain filtering method. The novel method employed the conventional Hatch's smoother to improve the precision of pseudorange at the observation-domain filtering part. At the position- domain filtering part, the proposed method constructs an accurate dynamical model for KF (Kalman filter) or ARKF (adaptive-robust KF) that relies on the TDCP (time differential carrier phase) velocity measurement technology. To assess the effectiveness of the novel method, we compared it with other methods under various system integration by measured data from IGS (International GNSS Service) MGEX (Multi-GNSS Experiment) and dynamic data collected at CUMT (China University of Mining and Technology). The static numerical experiments obtained from IGS datasets show that the ARKF improves the precision of 60/84 and 55/84 situations' directions compared with the LS method based on OP (original pseudorange) and SP (smoothed pseudorange), respectively, and Hatch's smother improves the precision of all situations' directions. The experiment results prove that both types of observation- and position- domain filtering methods can improve positioning precision. Also, the novel dual-domain filtering improved the precision of 80/84 situations' directions compared with the LS method. Moreover, the dynamic experiment shows that the novel method can improve all of the situations' directions except the N (North) direction provided by KF base on SP under GC and U (Upon) direction provided by LS base on SP under GR. The average of improved accuracy is 0.212m, 0.346m, and 0.588m at the E (East), N, U direction and its corresponding average percentage of improved accuracy is 30.2%, 60.2%, and 26.7%, respectively, which proves the novel dual-domain filtering method provided the best performance among the tested algorithms.

Load modeling plays an important role in accessing and enhancing the dynamic stability of power systems. Though the Synthesis Load Model Considering Voltage Regulation of Distribution Network ( $\pi $ model) has high accuracy, its parameters are too many. In order to improve the identification efficiency and reduce the difficulty of identification, a simplified model identification strategy based on parameter sensitivity analysis is proposed. Firstly, based on the global sensitivity analysis, the sensitivity analysis of the model parameters is carried out to obtain the First Order Sensitivity Indices (FSI)and Total Sensitivity Indices (TSI). Secondly, the FSI and TSI of each parameter are analyzed, and the effect on the output of model of each parameter is determined by FSI. For less influential parameters, whether the parameter should be fixed as constant is determined by the value of TSI. The parameter whose TSI equal or approximately equal to zero should be fixed as a constant. Finally, the improved genetic algorithm is used to identify the parameter-simplified model, and the effectiveness of the simplified identification strategy is verified by comparing the fitting effects with the measured curve and the residual and the integrated parameter models.

This paper proposes a robust and highly efficient feature-based visual-inertial odometry (VIO) approach. In order to save the computational resource, a simplified stereo visual model is applied to reduce the dimension of visual measurements. Moreover, the speed of feature matching is improved by using prior information from the inertial sensor. And through the marginalization, optimization is limited in two sliding windows, which can meet the need for the real-time application. In addition, a high accuracy initialization method and the homogeneous extraction of image features are introduced to ensure excellent pose tracking. The proposed VIO system is implemented on open datasets to show its merit compared with other state-of-the-art algorithms. In addition, we also perform this system on a low-cost stereo visual-inertial sensor and validate its practicability and superiority. Furthermore, the comparative experiment shows that the proposed algorithm has a higher accuracy than the monocular VIO and a shorter running time than the stereo VIO.

We consider the scenario that a multi-coil transmitter transfers energy to one or more single-coil receiver(s) based on magnetic resonance. The power transfer efficiency for fixed positions is determined by the activation pattern of the controllable transmit coil array, and the optimal activation pattern can be obtained offline. In order to efficiently charge the power receivers, online prediction of the receiver positions is necessary, and for this purpose, we consider two machine learning algorithms, including random forest (RF) and deep neural network (DNN). The prediction accuracy and training duration of the two algorithms are measured and compared. Simulation results indicate that both RF and DNN perform well for the single receiver case, and for the two-receiver case DNN still works well but RF does not.

Acquiring the locations of WiFi access points (APs) not only plays a vital role in various WiFi related applications, such as localization, security and AP deployment, but also inspires the emergence of novel applications. Thus, many efforts have been invested in studying AP localization. Most existing studies adopt the well-known lognormal distance path loss (LDPL) model, which only accounts for large-scale fading but ignores small-scale fading induced by multipath propagation. In this paper, we tackle the problem of AP localization based on the Rayleigh lognormal model which characterizes the influence of both large-scale fading and small-scale fading. In addition, particle filtering is used to sequentially narrow the scope of possible locations of the target AP. In order to label the locations of received signal strength (RSS) measurements in real time, manual configuration or certain indoor and outdoor localization techniques, including GPS, pedestrian dead reckoning (PDR) and WiFi fingerprinting, can be leveraged. Moreover, due to the bias caused by unavailable or inaccurate state space, a particle area dynamic adjustment strategy (PADAS) is designed to improve the AP localization accuracy. Extensive experiments were carried out in typical indoor and outdoor scenarios. It is shown that, if accurate location labels are available, the proposed method is able to achieve an average localization accuracy of 2.48 m indoors and 4.21 m outdoors. In comparison with the LDPL based solutions, the proposed method improves by 23.82% - 70.38% indoors and 14.13% - 35.94% outdoors; more importantly, the proposed method enhanced by using PDR, PADAS, GPS and WiFi fingerprinting can achieve localization accuracy comparable to that with accurate location labels. In addition, an Android application (APP) was developed to demonstrate the feasibility of the proposed algorithm on smartphones.