Full Download China Satellite Navigation Conference CSNC 2017 Proceedings Volume III 1st Edition Jiadong Sun PDF
Full Download China Satellite Navigation Conference CSNC 2017 Proceedings Volume III 1st Edition Jiadong Sun PDF
Full Download China Satellite Navigation Conference CSNC 2017 Proceedings Volume III 1st Edition Jiadong Sun PDF
com
https://textbookfull.com/product/china-
satellite-navigation-conference-
csnc-2017-proceedings-volume-iii-1st-edition-
jiadong-sun/
textbookfull
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
https://textbookfull.com/product/china-satellite-navigation-
conference-csnc-2018-proceedings-volume-iii-jiadong-sun/
https://textbookfull.com/product/china-satellite-navigation-
conference-csnc-2017-proceedings-volume-i-1st-edition-jiadong-
sun/
https://textbookfull.com/product/china-satellite-navigation-
conference-csnc-2018-proceedings-volume-ii-jiadong-sun/
https://textbookfull.com/product/china-satellite-navigation-
conference-csnc-2019-proceedings-volume-i-jiadong-sun/
China Satellite Navigation Conference CSNC 2019
Proceedings Volume II Jiadong Sun
https://textbookfull.com/product/china-satellite-navigation-
conference-csnc-2019-proceedings-volume-ii-jiadong-sun/
https://textbookfull.com/product/advanced-hybrid-information-
processing-first-international-conference-adhip-2017-harbin-
china-july-17-18-2017-proceedings-1st-edition-guanglu-sun/
https://textbookfull.com/product/artificial-intelligence-and-
security-6th-international-conference-icais-2020-hohhot-china-
july-17-20-2020-proceedings-part-iii-xingming-sun/
https://textbookfull.com/product/wireless-and-satellite-
systems-9th-international-conference-wisats-2017-oxford-uk-
september-14-15-2017-proceedings-1st-edition-prashant-pillai/
China Satellite
Navigation
Conference (CSNC)
2017 Proceedings:
Volume III
Lecture Notes in Electrical Engineering
Volume 439
Editors
123
Editors
Jiadong Sun Shiwei Fan
Academician of CAS China Satellite Navigation Office
China Aerospace Science and Technology Beijing
Corporation China
Beijing
China Wenxian Yu
Shanghai Jiao Tong University
Jingnan Liu Shanghai
Wuhan University China
Wuhan
China
Yuanxi Yang
National Administration of GNSS
and Applications
Beijing
China
Conference Topics
v
vi Preface
vii
viii Editorial Board
Scientific Committee
Chairman
Jiadong Sun, China Aerospace Science and Technology Corporation
Vice-Chairman
Rongjun Shen, China
Jisheng Li, China
Qisheng Sui, China
Changfei Yang, China
Zuhong Li, China Academy of Space Technology
Shusen Tan, Beijing Satellite Navigation Center, China
Executive Chairman
Jingnan Liu, Wuhan University
Yuanxi Yang, China National Administration of GNSS and Applications
Shiwei Fan, China
xi
xii Scientific Committee and Organizing Committee
Organizing Committee
Secretary General
Haitao Wu, Navigation Headquarters, Chinese Academy of Sciences
Vice-Secretary General
Weina Hao, Navigation Headquarters, Chinese Academy of Sciences
Under Secretary
Wenhai Jiao, China Satellite Navigation Office Engineering Center
Zhao Wenjun, Beijing Satellite Navigation Center
Wenxian Yu, Shanghai Jiao Tong University
Wang Bo, Academic Exchange Center of China Satellite Navigation Office
xv
xvi Contents
Daming Bai
Keywords Particle filter Orbit determination Non-Gaussian Estimation value
1 Introduction
D. Bai (&)
Xi’an Satellite Control Center, Xi’an 710043, China
e-mail: 464864443@qq.com
D. Bai
Key Laboratory of Spacecraft in-Orbit Fault Diagnosis and Maintenance,
Xi’an 710043, China
and velocity is not Gauss distribution. In order to solve the nonlinear model and the
estimation of the initial deviation to the negative effects caused by the filtering, this
paper uses particle filtering algorithm for satellite orbit, and its effectiveness is
proved by mathematical simulation.
The particle filter is used for target tracking, and its principle is that the particle filter
(Particle Filter PF) is a kind of approximate Bayesian filtering algorithm [1, 4]
based on Monte Carlo simulation; its core idea is to use some discrete random
sampling points (particle) probability density function to approximate the system
random variable, instead of integral to the sample mean to obtain estimation of the
state minimum variance. A mathematical description for the stationary random
process, assuming that k − 1 moment system of the posterior probability density for
pðxk1 jzk1 Þ random samples selected n according to certain principles, k time
measurement information, through state prediction and time update process, n
particle of the posterior probability density can be approximated as pðxk jzk Þ. As the
number of particles increases, the probability density function of the particle is
gradually approaching the probability density function of the state. Particle filter
can be applied to any nonlinear stochastic system. With the improvement of
computer performance, the method has been paid more and more attention to [4].
The basic problem of the orbit determination is as follows: The dynamic process
of a differential equation is not exactly known, and the optimal estimation of the
motion state of the satellite in a certain sense is solved using the observation value
with error and the not accurate initial state. When the system can be described as a
linear model, and the system and the measurement error are white noise, the
unbiased optimal estimation in statistical sense can be obtained by Kalman filtering
[5]. However, in practical applications, the target motion model and the measure-
ment model are mostly nonlinear; the noise is non-Gauss; and the traditional
Kalman filter application has been limited. Relative orbit determination problem is
usually based on the Hill equation using extended Kalman filtering algorithm to
estimate the state of the target aircraft. Because of the need to linearize the state
equation and the Hill equation which is only applicable to the relative distance
filtering calculation, when the relative distance or large initial errors are easy to
cause filter divergence and its application scope has been greatly constrained,
therefore, the application to the relative distance between the target orbit determi-
nation method will be an important direction for future research.
With the development of space technology, the requirement of spacecraft orbit
precision becomes higher and higher. The data processing methods of satellite orbit
determination can be divided into two categories: (1) batch processing. Using all
data to get track and related parameter of the epoch is a method of post-processing,
needs to store a large number of observation value, and demands high processing
Orbit Determination Based on Particle Filtering Algorithm 5
ability of computer. (2) Sequential recursive method. When the observation value is
reached, the corresponding processing is performed, and the next moment is no
longer observed by the previous moment. The particle filter algorithm is a
sequential recursive method; the data processing only uses the current value of
observation data and the previous data storage, and requires lower capacity of the
computer. The particle filter algorithm for solving nonlinear and non-Gauss prob-
lems has obvious theoretical advantages, such as the algorithm can be used to set
orbit spacecraft, which is bound to improve the accuracy of orbit determination.
However, no one shows any interest in the research model, algorithm, and appli-
cation of the precise orbit of the spacecraft.
The satellite is subjected to various forces in the course of the movement around the
earth; these forces can be divided into two categories: one is conservative forces;
the other is the divergent force [6]. Conservative forces include earth gravity, day,
month, the gravity of the planet to the satellite, and the gravitational field changes
caused by the earth tide phenomenon. For conservative force systems, it can be
described by the “bit function”. The divergence force includes the atmospheric
resistance, the earth infrared radiation, and the power of satellite attitude control.
The divergence force does not have the position function, so the expression is
directly used. The expressions of the above forces are very complex. Besides the
two-body problem, it is difficult to obtain the analytical solution of the motion
equation of the satellite, which is usually based on the approximate analytical
solution.
On orbit determination, dynamic model is not very accurate. The main point is as
follows: Under certain accuracy, dynamic model omits some perturbed factors,
even without omission, it also made some simplifications and approximations. In
the considered perturbation factors, the model parameters are approximate. The
initial state of general satellite r0 and v0 could not be precisely known; only the
reference value r0 and v0 needs to continue to refine the satellite observation ref-
erence value in order to obtain the initial motion state r0 and v0 of satellite.
Measurement data itself are not very accurate; the equipment itself and a variety of
factors affect measurement data with random error and system error. Therefore, the
orbit estimation problem is a dynamic process which is not known exactly for the
differential equation, and the best estimate of the satellite motion state is obtained
using the observational data with random errors and the inaccurate initial state.
6 D. Bai
The particle filter algorithm is used to calculate the satellite motion. The state of the
next moment is calculated according to the current state of the moment. In addition,
it is necessary to obtain the real observations by an observation equation
(1) Equation of state
Aircraft dynamics equation can be expressed as
9
x_ ¼ vx >
>
>
y_ ¼ vy >
>
>
>
z_ ¼ vz >
=
x_ ¼ x2 þ yu2 þ z2 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
x
þ f a
T x þ f px ð1Þ
x þy þz
2 2 2
>
>
y_ ¼ x2 þ yu2 þ z2 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
y
þ f a þ f >
>
T y py >
>
x2 þ y2 þ z2 >
>
z_ ¼ u
x2 þ y2 þ z2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi þ f a þ f ;
z
T z pz
x2 þ y2 þ z2
In the formula, x, y, and z are the position vector components as the target in
inertial system; vx , vy , and vz are the velocity vector component of target in inertial
coordinate system; u is the earth gravity parameters; fpx , fpy , and fpz are the per-
turbation accelerations, and said fT for thrust acceleration; ax , ay , and az for thrust
acceleration vector components. Considering the impact of the J2 perturbation
potential function for earth oblateness,
u 2 1
UJ2 ¼ R J2 ð3 sin2 f 1Þ: ðð2ÞÞ
r3 e 2
In the formula, Re is for the equatorial radius; sin / ¼ rz; J2 for the two-order
harmonic coefficient; and / for the spherical coordinate system defined position
angle. J2 perturbation in Cartesian coordinates for the component, respectively:
The target state model can be divided into two types: non-maneuvering and
maneuvering model. The difference is the influence of thrust acceleration.
(2) Observation equation analysis
Generally, the discrete time system filter is a mature linear estimation theory in
mathematical theory and mathematical methods [5]. However, the orbit determi-
nation is exactly the state equation and observation equations are nonlinear, which
is the main content of this paper.
Orbit Determination Based on Particle Filtering Algorithm 7
The relationship between the measured observations and the state vector of the
satellite is obtained by the tracking and measuring system, and the satellite
observations are oblique distance, slant distance change rate, azimuth, and elevation
angle. The determination of satellite orbit is the observation value obtained by
real-time measurement, and the satellite position is estimated sequentially. The
particle filter algorithm used in this paper is to solve the optimal estimation of
satellite position. n-dimensional linear dynamic systems and m-dimensional linear
observation system equations are described as follows:
)
Xk ¼ f k;k1 Xk1 þ Wk1
ð3Þ
Zk ¼ H k Xk þ Vk
In the formula, Xk is for the n-dimensional state vector system in the moment k,
k ¼ 1; 2; . . .; f k;k1 , called the state transition matrix; the reaction system sampling
time state from k 1 to k is the sampling time state transform; Wk 1 is for the
random disturbances acting on the system at the time k, called the model for noise;
Zk is for the m-dimensional observation vector; H k is for the observation the m n
order of the matrix, for the conversion from the state Xk to the measurements Zk ; Vk
is for the m-dimensional measurement noise.
In the horizontal coordinate system of the TT&C station, the measured is slant
distance R, slant distance change rate R, _ azimuth angle A, and elevation angle E.
The rectangular coordinate in the horizontal coordinate system of the TT&C station
is ðx; y; z; x_ ; y_ ; z_ Þ, in that way:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 9
R¼ x2 þ y2 þ z 2 > >
>
>
þ þ >
>
R_ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi >
x_
x y_y z_z
>
>
x2 þ y2 þ z 2 > >
>
>
8 y =
>
< arctan x[0 : ð4Þ
x >
>
A¼ >
>
: arctan y þ p x\0 >
> >
>
>
>
x z > >
>
>
;
E ¼ arcsin
R
Through the tracking and measuring system, the relationship between the measured
observations and the state vector of the satellite is nonlinear, and the satellite
observations are oblique distance, slant distance change rate, azimuth, and elevation
angle. Satellite orbit determination means sequential estimation of the location of
the satellite based on the observed value of a real-time measurement; the particle
filter algorithm used in this paper is to solve the problem of the best estimation of
satellite position.
8 D. Bai
Bias filtering provides a way to describe the tracking problem. The tracking
problem can be considered as at the present time t, the state of the target is Xt . Given
all observation states Z t fZ1 ; . . .; Zt g, the posterior probability distribution pðxt jzt Þ of
the target state is obtained. In the tracking process, the target is often divided into
multiple regions (assuming for n) tracking, and this will describe the state of the
target for a joint state Xt , fXit gni¼1 , and at the moment t, the target tracking of the
posterior probability distribution formula is as follows:
Z
PðXt jZ Þ ¼ cPðZt jXt Þ PðXt jXt1 ÞP Xt1 jZ t1 dXt1 :
t
ð5Þ
In the formula, PðZt jXt Þ is for the probability of the observed value Zt at the
given moment t state Xt ; PðXt jXt1 Þ for the motion model represents the probability
that the state Xt1 of the previous moment t 1 is predicted to the current moment t
target state Xt .
In the actual calculation, obtaining the integral formula (5) is difficult, so this
paper uses the sequential importance resampling (SIR) particle filter [7], to get the
integral value by sampling. It is assumed that the moment t 1 of the posterior
probability distribution PðXt1 jZ t1 Þ can be approximated as number for
n oN
ðr Þ ðr Þ ðr Þ
N-weighted particle, i.e., PðXt1 jZ t1 Þ Xt1 ; pt1 ; pt1 is the weight of the
r¼1
first r particle, and the integral value of the formula (5) can be approximated by the
following Monte Carlo method:
X
ðr Þ ðr Þ
PðXt jZ t Þ cPðZt jXt Þ pt1 P Xt jXt1 : ð6Þ
r
Through the LEO satellite and geosynchronous satellite launch phase tracking
and GPS, the simulation data based on this model has shown that particle filter
algorithm can be used to solve the nonlinear problem of satellite orbit determina-
tion, and will have a great effect on satellite orbit determination to raise the level of
valuation.
5 Conclusions
The estimation of satellite orbit is usually based on the assumption that the dis-
tribution of state variables is approximated by Gauss distribution, so as to estimate
the mean and covariance of the state variables. When the distribution of state
variables is obviously not Gauss distribution, the performance of the usual esti-
mation algorithm will be reduced (such as extended Kalman filtering algorithm). In
addition, even though the initial distribution of position and velocity is Gauss
distribution, the process noise and measurement noise are Gauss white noise.
Because of the nonlinearity in the model, the actual distribution of position and
velocity is not Gauss distribution. In order to solve the negative influence brought
by the nonlinearity of the model and the large estimation of the initial bias on the
rail estimation, this paper applies particle filtering algorithm to estimate satellite
orbit, and its effectiveness is proved by simulation. At the same time, the algorithm
will also have a profound impact on the estimation of the state of satellite fault
diagnosis.
References
Abstract In order to improve the accuracy of satellite clock bias (SCB) prediction,
a combined model is proposed. In combined model, polynomial model is used to
extract the trend of SCB, which can enhance relevance of data and improve effi-
ciency of ensemble empirical mode decomposition (EEMD). Simultaneously,
residual data is decomposed into several intrinsic mode functions (IMFs) and a
remainder term according to EEMD. Principal component analysis (PCA) is
introduced to distinguish IMFs using frequency as a reference, and high-frequency
sequence is the sum of IMFs with high frequency, low-frequency sequence is the
sum of IMFs with low frequency and the remainder term. Meanwhile, LSSVM
model is employed to predict the high-frequency sequence, and other sequence is
predicted by GM(1,1) model. The final consequence is the combination of these
two models and the SCB’s trend. SCBs from four different satellites are selected to
evaluate the performance of this combined model. Results show that combined
model is superior to conventional model both in 6- and 24-h prediction. Especially,
as for Cs clock, it achieves 6-h prediction error less than 3 ns, and 24-h prediction
error less than 8 ns.
1 Introduction
system continuously, namely SCB may not be acquired in real time. Therefore,
predicting SCB plays a significant role in the service of satellite [1–4].
At present, for improving the performance of predicting SCB, several models
have been proposed, such as gray model, least square support vector machine
(LSSVM) model, autoregressive model, etc. In [1], Cui et al. proposed a prediction
model based on gray theory, which had negative performance to predict the SCB
with poor stability; and in [2], Liu et al. used LSSVM to predict SCB, which
received higher precision. However, the optimal parameters of LSSVM cannot be
certain. Autoregressive model was also used to the predict SCB, nevertheless, this
model not only based on large-scale data, but also was sensitive to stability of
original data.
In order to enhance the performance of SCB prediction model, a combined
prediction model is proposed. In the combined model, polynomial model is used to
extract the trend of SCB, which can improve not only relevance of data, but also
efficiency of ensemble empirical mode decomposition (EEMD). Then, EEMD is
used to decompose the residual data into several intrinsic mode functions (IMFs)
and a remainder term. The two sequences with high and low frequencies are
reconstructed according to principal component analysis (PCA). Finally, LSSVM
model is employed to predict the high-frequency sequence. As well, the
low-frequency sequence is predicted by the GM(1,1) model. The final consequence
is the combination of these models.
In this paper, we use the EEMD combined with LSSVM model and GM(1,1) model
to predict the SCB, and these three models are presented as follows, respectively.
Empirical mode decomposition (EMD) can decompose the complicated signal into
intrinsic mode functions (IMFs), which bases on the local characteristic timescales
of a signal [5–7]. However, the problem of mode mixing exists in EMD. Mode
mixing is defined as a single IMF including oscillations of dramatically disparate
scales, or a component of a similar scale residing in different IMFs. So ensemble
empirical mode decomposition (EEMD) is introduced to eliminate the mode mixing
phenomenon and obtain the actual time–frequency distribution of seismic signal.
EEMD adds white noise to the data, which distributes uniformly in the whole time–
frequency space. The bits of signals of different scales can be automatically
designed onto proper scales of reference established by the white noises. EEMD
can decompose the signal f(t) into the following style:
Another random document with
no related content on Scribd:
Äkkiä selvisi Jussille, mitä oli tehtävä.
*****
*****
»Ja viisas ja hoksaava mies olit sinäkin, Joonas!» sanoi hän sitten.
»Lähde nyt käymään siellä Jussin pirtillä ja käske tulla tänne, että
saan kuulla…»
»Ne ovat vieneet kaikki tavarat Keskitalosta Korven Israeliin», tiesi
Joonas.
»Aivan kaikki!»
Patruuna jäi yksin mietteisiinsä. Tavaroita oli kyllä nyt saatu poikki,
mutta vielä oli vaara tarjona, ja mahdollisimman pian ne pitäisi saada
sisämaahan rantatietä ja jokivartta pitkin. Sillä tullimiehet kyllä
koettaisivat parastansa…
V
Silloin hän ymmärsi, mitä peliä tullimiehet nyt pitivät. Mutta hän ei
ollut mitään huomaavinaan, vaikka suksi töksähtikin lakanaan, vaan
hiihteli menemään. Siitä rupesi hän asiaa miettimään ja kertoi
patruunallekin.
»Mutta mitä sanovat ‘hurtat' silloin, kun saavat tietää, kuinka heitä
Tirkassa on petetty?» sanoi patruuna.
Jussi naurahti.
Mutta sitten hän yhtäkkiä unohti, että hänen piti pysyä levollisena,
ja sanoa rokasi:
Jussia nauratti.
Nyt oli kerännyt ja säästänyt rahaa niin paljon, että hänellä nyt oli
enemmän kuin koskaan ennen eläissään. Jos Ruhmulainen tietäisi!
Nyt olisi paras aika yrittää… Ostaisi koko kuorman konjakkia ja
lähtisi ylimaahan…
*****