,
,
спутника в инерциальной
системе, то
r =
(3)
и вычисление вектора F в инерциальной системе можно свести к формуле
F = -m
= -mg
r. (4)
Выполнение этих операций и обеспечивает блок П/С СИЛА). Поскольку на вход блок 6DoF
следует подать вектор силы через его проекции на связанные с ИСЗ оси, то далее полученный вектор
силы в инерциальной системе преобразуется в вектор проекций этой силы на связанную систему
координат путем умножения на матрицу направляющих косинусов ИСЗ в инерциальной системе.
Вторая задача блока – сформировать силу активного воздействия вдоль оси
связанной
системы для осуществления перехода на другую орбиту. Эта цель достигается посредством нижней
ветви блок-схемы. Здесь формируется постоянная по величине сила, начинающих свое действие в
заданный момент времени и действующая в течение заданного промежутка времени. При этом
использованы следующие обозначения:
- Tn – начальный момент времени действия силы коррекции орбиты;
- tau – промежуток времени, в течение которого действует сила коррекции;
- DF – величина силы коррекции.
Блок СУО в представленной модели отличается от аналогичного блока в предыдущей модели
прежде всего наличием части, вычисляющей угловое отклонение связанной с ИСЗ системы
координат от орбитальной. Это необходимо для обеспечения управления ориентацией спутника
относительно не инерциальной, а орбитальной системы. Кроме того, вместо блока непосредственного
преобразования углов Эйлера в кватернион поворота, используется последовательное преобразование
углов Эйлера в матрицу направляющих косинусов, а затем последней – в кватернион поворота. Это
позволяет избавиться от разрывов в угловых координатах при перехода значений углов через 180°.
В докладе приводятся результаты решения задачи построения сложной неопределенной
системы автоматического управления барабанным котлом (рисунок 1). В частности, математическая
модель сложного неопределенного объекта управления в пространстве состояний описывается
системой интервальных дифференциальных уравнений:
)
(
)
(
),
(
)
(
)
(
t
x
c
t
y
t
u
t
x
t
x
T
B
A
,
,
0
t
t
,
0
0
)
(
x
t
x
, (5)
где
,
0
t
t
– непрерывное время; t
0
– начальное значение;
n
R
t
x
)
(
— вектор состояний
объекта управления;
)
(
,
IR
M
n
n
A
– интервальная матрица, с элементами:
249
A
n
j
i
a
a
a
a
ij
ij
ij
ij
,
1
,
,
,
:
; (6)
R
a
a
ij
ij
,
– соответственно нижняя и верхняя границы значений элементов матрицы A;
)
(
,
IR
M
n
n
– множество матриц размерности
)
(
n
n
, элементами которых являются
вещественные интервалы;
B
)
(
,
R
M
m
n
– матрицы, размерности (n
m) с элементами:
B
,n
i
b
b
b
b
i
i
i
i
1
,
,
:
, (7)
m
R
t
u
)
(
– управление;
y( t) – выходной сигнал.
В рассматриваемом случае, под математической моделью (1) будет пониматься семейство
математических моделей стационарных систем, вида:
t
Bu
t
Ax
t
x
)
(
)
(
)
(
,
,
0
t
t
,
0
0
)
(
x
t
x
, (8)
где
A
A
– точечная матрица, размерности (n
n) с элементами:
n
j
n
i
a
A
ij
,
1
,
,
1
:
, (9)
где
B
B
– точечная матрица, размерности (n
m) с элементами:
n
i
b
B
i
,
1
:
, (10)
остальные обозначения совпадают с (5).
При построении объекта управления (5) использовался алгоритм [4], призванный снизить
вычислительные трудности с интервальными дифференциальными уравнениями за счет специально
сконструированных множеств векторов и матриц. Данный алгоритм основан на формировании
вспомогательного множества [5]:
1
,
1
:
i
n
i
z
R
z
z
,
n
i
,
1
,
и определения q=
1
2
,
1
n
ключевых матриц:
A
a
A
q
ij
q
,
А
(11)
1
если
,
,
1
если
,
j
i
q
ij
j
i
q
ij
q
ij
z
z
a
z
z
a
a
)
,
1
,
(
n
j
i
,
для математической модели (5).
ЛИТЕРАТУРА
1. Методы робастного, нейро-нечеткого и адаптивного управления: Учебник / Под ред. Н.Д. Егупова. -
М.: Изд-во МГТУ им. Н.Э. Баумана, 2001. - 744 с.
2. Аро Х.О. Применение методологии робастного синтеза к системе угловой стабилизации
метеорологической ракеты // Известия вузов – Приборостроение, Санкт-Петербург. - №11. - 2011.
3. Shiryayeva O., Samigulina Z., Samigulina G., Fourati H. Adaptive Control strategy based reference model for
Spacecraft Motion Trajectory //International Journal of Adaptive Control and Signal Processing. – 2014.
4. Цыпкин Я.З. Робастность в системах управления и обработки данных // Автоматика и телемеханика. -
1992. - № 1. - С.165-169.
5. Бобылев Н.А. О положительной определенности интервальных семейств симметрических матриц //
Автоматика и телемеханика. - 2000. – № 8. – С. 200.
250
REFERENCES
1. Metody robastnogo, nejro-nechetkogo i adaptivnogo upravlenija: Uchebnik / Pod red. N.D. Egupova. - M.:
Izd-vo MGTU im. N.Je. Baumana, 2001. - 744 s.
2. Aro H.O. Primenenie metodologii robastnogo sinteza k sisteme uglovoj stabilizacii meteorologicheskoj rakety
// Izvestija vuzov – Priborostroenie, Sankt-Peterburg. - №11. - 2011.
3. Shiryayeva O., Samigulina Z.,Samigulina G., Fourati H. Adaptive Control strategy based reference model for
Spacecraft Motion Trajectory // International Journal of Adaptive Control and Signal Processing. – 2014.
4. Cypkin Ja.Z. Robastnost' v sistemah upravlenija i obrabotki dannyh // Avtomatika i telemehanika. - 1992. - №
1. - S.165-169.
5. Bobylev N.A. O polozhitel'noj opredelennosti interval'nyh semejstv simmetricheskih matric // Avtomatika i
telemehanika. - 2000. – № 8. – S. 200.
Қуандықова Г.Е., Ширяева О.И.
Аэрокосмостық аппарат қозғалысын робасты басқару жүйесінің математикалық моделі
Түйіндеме. Бұл статьяда интервалды арифметика əдістерінің негізінде күрделі анықталмаған
аэрокосмостық аппарат қозғалысын басқару жүйесін жасаудың нəтижелері көрсетілген. Бұл мəселе сыртқы
орта əсерінен болатын априорлық белгісіз проблемаларды шешумен байланысты болғандықтан өте маңызды
болып саналады. Қазіргі уақытта объектіні сипаттау кезінде параметрлік белгісіздікті есептеу үшін робасты
жүйе теориясы, оның ішінде стохастикалық, анық емес, интервалды-берілген жүйелер теориясы қолданылады.
Бұл статьяда интервалды-берілген сияқты анықталмаған объект сипаттамасы көрсетіледі. Басқарудың
интервалды-берілген объектісін құру кезінде арнайы құрастырылған векторлар мен матрицалар жиыны
көмегімен интервалды дифференциалдық теңдеулерді шығару кезінде болатын қиындықтарды жеңілдететін
алгоритм қолданылды. Бұл алгоритм кілттік матрицалармен қосымша жиындарды құруға негізделген.
Түйін сөздер: Күрделі анықталмаған басқару жүйесі, декомпозиция, децентрализация, аэрокосмостық
объект, интервалды арифметика.
Kuandykova G.E., Shiryayeva O.I.
Mathematical model of robust traffic management system aerospace vehicle
Summary. In this paper the research methods of uncertain control system of satellites on based of interval
arithmetic are considered. The synthesis algorithm of uncertain control system of the satellite moving is developed. This
problem is relevant, as it is connected with the problem of a priori uncertainty arising due to the influence of external
disturbances. To account for parametric uncertainty in the description of the object is currently used theory of robust
systems, including stochastic, fuzzy, interval-specified. This article uses the indefinite description of the object as
interval-specified. When constructing interval- specified control object used algorithm designed to reduce the
computational difficulties with interval differential equations by a specially designed set of vectors and matrices. This
algorithm is based on the formation of a plurality of auxiliary key matrices. Procedures for decentralization considered
structural columns of centralized and decentralized systems.
Key words: Robust control system , decomposition, decentralization, aerospace vehicle , interval arithmetic.
UDK 519.97
Kunanbayeva M.M. , Sadibekov K. S., Aitkulov Zh. S.,Mamyrbayev O.Zh.
Kazakh national technical university after K.I.Satpayev,
Almaty, The Republic of Kazakhstan
jalau@mail.ru
MATHEMATICAL METHODS IN THE SPEECH RECOGNITION SYSTEMS
Abstract. The article discusses the speech acoustic signal , which is a carrier of a very wide range of information .
Subjecting the speech signal complex analysis, we can get information about the system of language; of speech, as the
process of communication through language. The paper presents a model of speech recognition and determining the
energy of speech using the algorithm of duality. Using mathematical models were graphed speech. Examples of the use
of the Gauss method in speech recognition.There are also a number of tasks which after recognition of the speech from
the obtained data, will allow to make the morphological analysis, to define the root and affixes. The assigned task
requires the creation of an algorithm that allows to define the phrase.
Key words : speech signal , model of speech recognition, method in speech recognition, Gauss method.
Introduction. The aim of this study is to develop methods of recognition of the speech signal in the
Kazakh language modeling and information system that implements these methods.
In this regard, there are the following:
251
Explore and develop new algorithms for uniform recognition of the speech signal;
To develop new methods of phonetic-acoustic classification of the speech signal;
Explore the phonetic-acoustic structure of the Kazakh language;
Develop a set of applications for recognition of the speech signal in the Kazakh language.
The subject of this study is the process of recognizing the speech signal. The objects of the study are the
model of the speech signal detection and methods of modifying it. The study used the methods of the following
areas of expertise: theory analysis and digital signal processing, systems analysis and systems theory,
linguistics and phonetic analysis. Models have been developed and studied in this paper can be used in the
implementation of human-machine interfaces in the various information systems. The recommendations
contained in the work may be useful in constructing unified recognition of the speech signal.
Presentation of the signal. As we recall, it is necessary to compute the speech parameters in short time
intervals to reflect the dynamic change of the speech signal. Typically, the spectral parameters of speech are
estimated in time intervals of 10ms. First, we have to sample and digitize the speech signal. Depending on
the implementation, a sampling frequency between 8kHz and 16kHz and usually a 16bit quantization of
the signal amplitude is used. After digitizing the analog speech signal, we get a series of speech samples
where
or, for easier notation, simply. Now a preemphasis filter is used to eliminate the -
6dB per octave decay of the spectral energy:
Then, a short piece of signal is cut out of the whole speech signal. This is done by multiplying the
speech samples
with a windowing function
to cut out a short segment of the speech signal,
starting with sample number
and ending with sample number
. The length N
of the segment (its duration) is usually chosen to lie between 16ms to 25 ms, while the time window is
shifted in time intervals of about 10ms to compute the next set of speech parameters. Thus, overlapping
segments are used for speech analysis. Many window functions can be used, the most common one is the so–
called Hamming-Window:
where N is the length of the time window in samples. By multiplying our speech signal with the time
window, we get a short speech segment
:
As already mentioned, N denotes the length of the speech segment given in samples while m is
the start time of the segment [1]. The start time m is incremented in intervals of 10ms, so that the
speech segments are overlapping each other. All the following operations refer to this speech segment
ϑ
m
(k),k=m…m+N-1. To simplify the notation, we shift the signal in time by m samples to the left, so that our
time index runs from
again. From the windowed signal, we want to compute its discrete power
spectrum. First of all, the complex spectrum V(n) is computed. The complex spectrum V(n) has the
following properties:
The spectrum V(n) is defined within the range from
to
.
V(n) is periodic with period, i.e.,
Since
is real-valued, the absolute values of the coefficients are also symmetric:
To computer the spectrum, we compute the discrete Fourier transform (DFT, which gives us the
discrete, complex-valued short term spectrum
of the speech signal (for a good introduction to the DFT
and FFT, see [1], and for both FFT and Hartley Transform theory and its applications see [2]):
The DFT gives us N discrete complex values for the spectrum
at the frequencies
where
252
Remember that the complex spectrum V(n) is defined for
, but is periodic with
period N. Thus, the N different values of
are sufficient to represent V(n). One should keep in mind
have to interpret the values of
ranging from n=N/2 to n=N-1 as the values for the negative frequencies
of the spectrum
. One could think that with a frequency
range from
we should have N+1 different values for V(n), but since V(n) is periodic
with period N(4), we know that
So nothing is wrong, and the N different values we get from
are sufficient to describe the
spectrum V (n). For further processing, we are only interested in the power spectrum of the signal. So we can
compute the square of the absolute values,
.
Due to the periodicity
and symmetry
of V(n),
only the values
are used for further processing, giving a total number of N/2+1 values.
It should be noted that
contains only the DC-offset of the signal and therefore provides no
useful information for our speech recognition task.
The non-linear warping of the frequency axis can be modeled by the so-called mel-scale. The
frequency groups are assumed to be linearly distributed along the mel-scale. The so-called mel-frequency
can be computed from the frequency f as follows [3]:
Figure shows a plot of the mel scale (Figure 1).
Figure 1
The criterium of optimality we want to use in searching the optimal path P
opt
should be to minimize
Fortunately, it is not necessary to compute all possible paths P and corresponding distances
to find the optimum.
Out of the huge number of theoretically possible paths, only a faction is reasonable for our purposes.
We know that both sequences of vectors represent feature vectors measured in short time intervals.
Therefore, we might want to restrict the time warping to reasonable boundaries: The first vectors of and
253
should be assigned to each other as well as their last vectors. For the time indices in between, we want to
avoid any giant leap backward or forward in time, but want to restrict the time warping just to the ”reuse” of
the preceding vector to locally warp the duration of a short segment of speech signal. With these restrictions,
we can draw a diagram of possible ”local” path alternatives for one grid point and its possible predecessors.
We will soon get more familiar with this way of thinking. As we can see, a grid point (i, j) can have the
following predecessors [3]:
(i-1, j) : keep the time index j of while the time index of is incremented
(i-1, j-1) : both time indices of and are incremented
(i, j-1) : keep of the time index i of while the time index of is incremented
All possible paths P which we will consider as possible candidates for being the optimal path P
opt
can be
constructed as a concatenation of the local path alternatives as described above. To reach a given grid point
(i, j) from (i-1, j-1), the diagonal transition involves only the single vector distance at grid point (i, j) as
opposed to using the vertical or horizontal transition, where also the distances for the grid points (i -1, j) or (i,
j-1) would have to be added. To compensate this effect, the local distance
is added twice when
using the diagonal transition [4].
Let’s assume we measure only one continuously valued feature x. If we assume the measurement of this
value to be disturbed by many statistically independent processes, we can assume that our measurements will
assume a Gaussian distribution of values, centered around a mean value m, which we then will assume to be
a good estimate for the true value of x. The values we measure can be characterized by the Gaussian
probability density function. As we recall from school, the Gaussian PDF
(x) is defined as:
Where m represents the mean value and
denotes the variance of the PDF. These two parameters
fully characterize the one–dimensional Gaussian PDF.
The Gaussian PDF
can characterize the observation
probability for vectors generated by a single Gaussian process. The PDF of this process has a maximum value at
the position of the mean vector ~m and its value exponentially decreases with increasing distance from the mean
vector. The regions of constant probability density are of elliptical shape, and their orientation is determined by
the Eigenvectors of the covariance matrix [5]. However, for speech recognition, we would like to model more
complex probability distributions, which have more than one maximum and whose regions of constant probability
density are not elliptically shaped, but have complex shapes like the regions we saw in Figure 2.
To do so, the weighted sum over a set of K Gaussian densities can be used to model p( ) [6]:
Where
is the Gaussian PDF as in. The weighting coefficients
are called the mixture
coefficients and have to fit the constraint:
The process parameters are:
C
1
=0.3 ;
C
2
=0.4 ;
C
2
=0.3 ;
REFERENCES
1. E. G. Schukat-Talamazzini. Automatische Spracherkennung. Vieweg Verlag, 1995.
2. J.J. Odell S. Young and P.C. Woodland. Tree-based state tying for high accuracy acoustic modeling. Proc.
Human Language Technology Workshop, Plainsboro NJ, Morgan Kaufman Publishers Inc., pages 307–312, 1994.
254
3. K. F. Lee X. D. Huang and H. W. Hon. On semi-continuous hidden Markov modeling. Proceedings ICASSP
1990, Albuquerque, Mexico, pages 689–692, April 1990.
4. F. Alleva X. Huang, M. Belin and M. Hwang. Unified stochastic engine (use) for speech recognition.
Proceedings ICASSP 1993, II:636–639, 1993.
5. S. J. Young. The general use of tying in phoneme-based hmm speech recognisers. Proceedings of ICASSP
1992, I(2):569–572, 1992.
6. S. Young. Large vocabulary continuous speech recognition: A review. IEEE Signal Processing Magazine,
13(5):45–57, 1996.
ЛИТЕРАТУРА
1. Е. Г. Скукат-Таламазини.Автоматическое распознавание речи.. Вьювег Верлаг, 1995.
2. Ж.-Ж. Оделл С. Янг и Р.С.Хон. Дерево на основе состояния привязки к высокой точности
акустического моделирования. Человеческий язык семинар, посвященный технологиям. Плейнсборо Н,Ж.
Издательство Морган Кауфмана, стр. 307-312, 1994.
3. К. Ф. Ли X. Д. Хуан и H. В. Хон. На полунепрерывно скрытом Марковском моделировании. Материалы
ICASSP 1990, Альбукерке, Мексика, стр. 689-692, апрель 1990 года.
4. Ф. Aллева X.Хуан , М. Белин и М. Хван. Единый стохастический двигатель (использование) для
распознавания речи. Материалы ICASSP 1993, стр. 636-639, 1993.
5. С. Ж. Янг. Общее использование привязки в фонем на основе hmm устройств речи. Материалы
ICASSP 1992, I (2): 569-572, 1992.
6. С. Янг. Большой словарь распознавания слитной речи: обзор. Журнал обработка сигнала, IEEE, 13 (5):
45-57, 1996.
Кунанбаева М.М., Садибеков С.С ., Айткулов Ж.С., Мамырбаев О.Ж.
Достарыңызбен бөлісу: |