具有噪声的基于密度的空间聚类(Density‑based spatial clustering of applications with noise,DBSCAN)能够发现不同密度和大小的类簇,对噪声也有很好的鲁棒性,被广泛地应用到数据挖掘的任务中。DBSCAN通常需要调整参数MinPts和Eps以...具有噪声的基于密度的空间聚类(Density‑based spatial clustering of applications with noise,DBSCAN)能够发现不同密度和大小的类簇,对噪声也有很好的鲁棒性,被广泛地应用到数据挖掘的任务中。DBSCAN通常需要调整参数MinPts和Eps以达到更优的聚类效果,但往往在搜索最优参数的过程中会影响DBSCAN的性能。本文从两个方面优化DBSCAN,一方面,提出一种无参的方法优化DBSCAN全局参数选择。无参方法利用自然最近邻获得数据集的自然特征值,并将自然特征值作为参数MinPts值。然后,根据自然特征值计算自然特征集合,利用自然特征集合中的数据分布特性,分别采取统计最小值、平均值和最大值3种方式得到Eps值。另一方面,采用集成数据科学实时加速平台(Real‑time acceleration platform for integrated data science,RAPIDS)的图形处理器(Graphics processing unit,GPU)计算加快DBSCAN算法的收敛速度。实验结果表明,本文提出的方法在优化DBSCAN参数选择的同时,取得了与密度峰值聚类(Density peaks clustering,DPC)相当的聚类结果。展开更多
This paper demonstrates the assembly of a servo-controlled platform with two degrees of freedom, empirical methods and a developed closed-loop control found in the system mathematical model. This control aims to stabi...This paper demonstrates the assembly of a servo-controlled platform with two degrees of freedom, empirical methods and a developed closed-loop control found in the system mathematical model. This control aims to stabilize and hold small objects on the platform. We parsed the step response in X and Y axes, hence we found the first and second-order models for each one. We did some further analyses to decide which one would better represent the behavior of the system. The MATLAB software provided step response for the model empirically obtained and latter compared it to experimental data acquired in the trials. Accelerometers and gyro sensors from the MPU-6050 sensor measured the angular position of platform on X and Y axes. In order to improve measurements accuracy and eliminate noise effects, we implemented the complementary filter to the firmware system. We used Arduino to control servomotors through PWM pulses and perform data acquisition.展开更多
To ensure success of precise navigation, it is necessary to carry out in-field calibration for the accelerometers in platform inertial navigation system(PINS) before a mission is launched.Traditional continuous self-c...To ensure success of precise navigation, it is necessary to carry out in-field calibration for the accelerometers in platform inertial navigation system(PINS) before a mission is launched.Traditional continuous self-calibration methods are not fit for fast calibration of accelerometers because the platform misalignments have to be estimated precisely and the nonlinear coupling terms will affect accuracy. The multi-position methods with a "shape of motion" algorithm also have some existing disadvantages: High precision calibration results cannot be obtained when the accelerometer's output data are used directly and it is difficult to optimize the calibration scheme. Focusing on this field, this paper proposes new fast self-calibration methods for the accelerometers of PINS. A data compression filter is employed to improve the accuracy of parameter estimation because it is impossible to obtain non-biased estimation for accelerometer parameters when using the "shape of motion" algorithm. Besides, continuous calibration schemes are designed and optimized by the genetic algorithm(GA) to improve the observability of parameters. Simulations prove that the proposed methods can estimate the accelerometer parameter more precisely than traditional continuous methods and multi-position methods, and they are more practical to deal with urgent situations than multi-position methods.展开更多
文摘具有噪声的基于密度的空间聚类(Density‑based spatial clustering of applications with noise,DBSCAN)能够发现不同密度和大小的类簇,对噪声也有很好的鲁棒性,被广泛地应用到数据挖掘的任务中。DBSCAN通常需要调整参数MinPts和Eps以达到更优的聚类效果,但往往在搜索最优参数的过程中会影响DBSCAN的性能。本文从两个方面优化DBSCAN,一方面,提出一种无参的方法优化DBSCAN全局参数选择。无参方法利用自然最近邻获得数据集的自然特征值,并将自然特征值作为参数MinPts值。然后,根据自然特征值计算自然特征集合,利用自然特征集合中的数据分布特性,分别采取统计最小值、平均值和最大值3种方式得到Eps值。另一方面,采用集成数据科学实时加速平台(Real‑time acceleration platform for integrated data science,RAPIDS)的图形处理器(Graphics processing unit,GPU)计算加快DBSCAN算法的收敛速度。实验结果表明,本文提出的方法在优化DBSCAN参数选择的同时,取得了与密度峰值聚类(Density peaks clustering,DPC)相当的聚类结果。
文摘This paper demonstrates the assembly of a servo-controlled platform with two degrees of freedom, empirical methods and a developed closed-loop control found in the system mathematical model. This control aims to stabilize and hold small objects on the platform. We parsed the step response in X and Y axes, hence we found the first and second-order models for each one. We did some further analyses to decide which one would better represent the behavior of the system. The MATLAB software provided step response for the model empirically obtained and latter compared it to experimental data acquired in the trials. Accelerometers and gyro sensors from the MPU-6050 sensor measured the angular position of platform on X and Y axes. In order to improve measurements accuracy and eliminate noise effects, we implemented the complementary filter to the firmware system. We used Arduino to control servomotors through PWM pulses and perform data acquisition.
文摘To ensure success of precise navigation, it is necessary to carry out in-field calibration for the accelerometers in platform inertial navigation system(PINS) before a mission is launched.Traditional continuous self-calibration methods are not fit for fast calibration of accelerometers because the platform misalignments have to be estimated precisely and the nonlinear coupling terms will affect accuracy. The multi-position methods with a "shape of motion" algorithm also have some existing disadvantages: High precision calibration results cannot be obtained when the accelerometer's output data are used directly and it is difficult to optimize the calibration scheme. Focusing on this field, this paper proposes new fast self-calibration methods for the accelerometers of PINS. A data compression filter is employed to improve the accuracy of parameter estimation because it is impossible to obtain non-biased estimation for accelerometer parameters when using the "shape of motion" algorithm. Besides, continuous calibration schemes are designed and optimized by the genetic algorithm(GA) to improve the observability of parameters. Simulations prove that the proposed methods can estimate the accelerometer parameter more precisely than traditional continuous methods and multi-position methods, and they are more practical to deal with urgent situations than multi-position methods.
