Beruflich Dokumente
Kultur Dokumente
Abstract—In this paper, the nonlinear stochastic dynamical ness of batch culture. In this paper, we propose a stochastic
system of batch fermentation is investigated in the process of version of the batch fermentation process which the noise is
glycerol bio-dissimilation to 1,3-propanediol by klebsiella pneu- idealized to be independent in state variables. The process is
moniae. Both the patterns of microbial growth and metabolism
and formation of fermentation products are strongly effected modeled by a stochastic ordinary differential system driven by
by bioreactor conditions and fermentation control. Bioreactor five-dimensional Brownian motion. The global Lipschitz and
conditions and some operations in culture control impose much linear growth conditions of the coefficients of the stochastic
randomness to the process, particularly on the regulation of system are proved to make sure the existence and uniqueness
temperature and controlling rotate speed of stirrer. The stochastic of solution of the stochastic system. Furthermore we investi-
version of the batch fermentation process is more suitable for the
factual fermentation. Stochastic viability of a closed convex set gate Stochastic Viability of a closed convex set under weak
under weak solutions of the nonlinear stochastic kinetic system of solutions of the nonlinear stochastic kinetic system(NSKS) of
batch culture is investigated. Computer simulation results reveal batch culture. Finally stochastic simulation is carried out under
that the peculiar role of randomness in the dynamical responses the Stochastic Euler-Maruyama method. Numerical examples
of the batch culture. confirm that the proposed stochastic system is more suitable
to formulate the dynamics of batch culture.
I. I NTRODUCTION
This paper is organized as follows. In section 2, we present
Stochastic influences play an important role in bioprocess a NSKS of batch fermentation process. In section 3, the
development. Biotechnical treatment of microorganisms is existence, uniqueness of solutions to the stochastic system are
commonly described by systems of nonlinear ordinary differ- proved and the continuity, boundedness and Markov property
ential equations(ODEs) [1]. It includes an idealization of the of solutions are discussed. In section 4, We investigate the
technical system component and a qualitative characterization existence of a viable set under weak solutions of the NSKS
of the biological part. of batch culture. Numerical examples are provided to simulate
The bioconversion of glycerol to 1,3-propanediol has re- the NSKS of batch culture in Section 5. In section 6, we draw
cently been receiving more and more attention throughout the the conclusions and trace the direction for future works.
world due to its environmentally safe, high region specificity,
cheaply available feedstock, and relatively high theoretical II. NSKS OF BATCH CULTURE
molar yield [2]. Among various microbial production methods Mass balances of biomass, substrate and products in batch
of 1,3-PD, dissimilation of glycerol to 1,3-PD by Klebsiella microbial culture are written as follows (see [7]).
pneumoniae has been widely investigated since the 1980s ⎧
[3], including experimental investigations of the multiple- ⎨ ẋ1 (t) = μx1 (t)
ẋ2 (t) = −q2 x1 (t) t ∈ (0, T ) (1)
inhibitions in the fermentation [4] and quantitative analysis of ⎩
this complex bioprocess [5]. Compared with continuous and ẋi (t) = qi x1 (t), i = 3, 4, 5.
feed-batch cultures, glycerol fermentation in batch culture can x(0) = x0 .
obtain the highest production concentration and molar yield
1,3-PD to glycerol [6]. So nonlinear kinetic systems in batch where the specific growth rate of cells μ, specific consumption
culture have been extensively considered in recent years [7]. rate of substrate q2 and specific formation rate of product qi
However, the deterministic model imply some hidden as- are expressed by Eqs. (2)-(4), respectively.
sumptions, different culture states during long term continuous 5
x2 (t) xi (t)
fermentation of glycerol by K. pneumoniae under similar ini- μ = μm ( ) (1 − ∗ ) (2)
tial fermentation conditions are obtained, resulting in random- x2 (t) + ks i=2 xi
dX(t) = F(X(t))dt + σ(X(t))dW(t), t ∈ I, for every t ≥ 0, 0 |Ψ(s, ω)|ds < ∞ a.s. and Lloc 2 denotes
(5) the set of all the real measurable
T process on Ω adapted to Ft
X(0) = ξ.
satisfying for every T > 0, 0 Φ(x(s))2 ds < ∞ a.s.,
where 3) with probability one, the solution X of (5) and the Brownian
i) The drift vector motion W satisfy
t
t
F(x(t)) = (μx1 (t), −q2 x1 (t), q3 x1 (t), q4 x1 (t), q5 x1 (t))T . (6) X(t) = ξ + F(X(s))ds + σ(X(s))dW(s). (8)
0 0
5 5×5
ii) The dispersion matrix σ : R → R is a known Definition 3.3: If X1 (t) and X2 (t) are arbitrary two solu-
diffusion matrix on R5
tions of (5) with respect to W with the same initial ξ, and
σ(x(t)) = diag(a1 (x∗1 (t) − x1 (t)), a2 (x2 (t) − x∗2 (t)), P {supt∈I |X1 (t) − X2 (t)| = 0} = 1. Then a solution X(t)
a3 (x∗3 (t) − x3 (t), a4 (x∗4 (t) − x4 (t), a5 (x∗5 (t) − x5 (t))) (7) of (5) is called unique .
2
It is clear from the continuity, the functions defined in (6) Theorem 3.3: The unique solution x(t) of system (5) is
and (7) are measurable for t ∈ I, x ∈ E. a Markov process on the interval I whose initial probability
Theorem 3.1: For the functions F(x) and σ(x) defined in distribution at t = 0 is the distribution of ξ and the x(t) has
(6) and (7) , there exist positive constants K and K such continuous paths, moreover
that for t ∈ I the following conditions hold:
(sup0≤t≤T Ex(t))2 < B(1 + Eξ2 )
a) uniform Lipschitz condition, ∀ x1 and x2 ∈ E,
where B depends on K in the proof of Theorem 3.1 and T.
F(x2 ) − F(x1 ) + σ(x2 ) − σ(x1 ) ≤ Kx2 − x1 E ,
IV. V IABLE SET OF THE NSKS
b) growth condition, For the stochastic system, we construct a convex set K
F(x) + σ(x) ≤ K (1 + xE ), ∀ x ∈ E. K = {0 ≤ xi (t) ≤ x∗i , i = 1, 2, 3, 4, 5 and t ∈ I} (12)
Proof. We begin with the uniform Lipschitz condition. By Consider the closed convex set K ⊂ R5 defined by (12) and
Quasi-differential Mean Value theorem, we now obtain stochastic differential equation in system (5).
Definition 4.1: X(t) ˜ is said to be a weak solution for
F(x2 ) − F(x1 ) ≤ JF (x1 + θ(x2 − x1 ))x2 − x1 E . ˜ B(t)),
˜
system (5) if there exists a pair of process ((X(t), F)
where θ ∈ (0, 1) and JF (x1 +θ(x2 −x1 )) denotes Jacobian of on a probability space (Ω, F , P) such that (5) holds, where
F at x1 +θ(x2 −x1 ). Let L = maxx∈S0 JF (x1 +θ(x2 −x1 )), ˜ is not need to be adapted to the filtration Ft and B(t)
X(t) ˜
we can rewrite the above inequality is an arbitrary Ft /Bt (E)-measurable Brownian motion.
It is clear that a solution of the system (5) is a weak solution.
F(x2 ) − F(x1 ) ≤ M x2 − x1 E . (9) Definition 4.2: We say that K is viable relative to the
On the other hand, letting A = maxi∈I5 {|ai |}, by the system (5) if for every x ∈ K there exists a weak solution
definition of the function σ, we have X(·, x) to (5) starting at X(0, x) defined on R+ such that for
all t > 0, X(t, x) ∈ K almost surely.
σ(x2 ) − σ(x1 ) ≤ Ax2 − x1 E (10) To express the viability, we also introduce the differential
1,1 1,1
operator Lk on Cloc (Kc , R+ ) for every f ∈ Cloc (Kc , R+ )
Thus, it follows from (10) and (11) and Let K = M + A, c 5
and x ∈ K = R+ \K where the second derivative f (x) does
then we have the following equality
exist Lk f (x) = 12 T r[σσ ∗ (Πk (x))f (x)] + f (x), F(Πk (x)).
1,1
F(x2 ) − F(x1 ) + σ(x2 ) − σ(x1 ) ≤ Kx2 − x1 E . Where the Cloc (Kc , R+ ) denotes the set of continuously
differentiable functions from Kc into R+ whose derivative
Next, we will show the growth condition of the function F
is locally Lipschitz on Kc , and d(x) mink∈K |x − k| the
and σ. Note that, because of (6), we see that
distance from x ∈ R5+ to K and ΠK (x) denotes the projection
1,1
F(x) = (μ)2 + (−q2 )2 + (q3 )2 + (q4 )2 + (q5 )2 x1 (11) of x on K. It is well known (see e.g.[11]) that d ∈ Cloc (Kc , R)
x−Πk (x)
and for all x∈ ¯ ∂K we have d (x) = |x−Π (x)| .
k
From Eqs.(2)-(4), we can conclude that there exist positive It is obvious that the convex set K is a polyhedron.
constants C1 , C2 , C3 , C4 , C5 such that Theorem 4.1: Suppose convex set K defined by (12), then
5 the following viability condition holds true
x2 (t) xi (t)
|μ| = |μm ( ) (1 − ∗ )| ≤ |μm | ≤ C1
x2 (t) + ks i=2 xi σ(ΠK (x))∗ d (x) = 0, LK d(x) ≤ 0, f or a.e. x ∈ K c .
μ Proof. ∀X ∈ Kc and X = (x1 , x2 , x3 , x4 , x5 ) we discuss
|−q2 | = |m2 + | ≤ C2 , |qi | = |mi +μYi | ≤ Ci , i = 3, 4, 5.
Y2 the problem in the following 3 cases :
x∗2 +x∗2 +x∗2 ∗2 ∗2 1) ΠK (X) is on the face of the polyhedron but not on the
3 +x4 +x5
Referring to (11), Let L2 = 1 2 x . It is edge of the polyhedron.
clear from the definition of the function σ No loss of generality, suppose the face of the polyhedron is
σ(x) ≤ ALxE . {x1 (t) = x∗1 , xj (t) = 0, j = 2, 3, 4, 5 and t ∈ I}, then
ΠK (X) = (x∗1 , x2 , x3 , x4 , x5 ) and it is easy to verified that
Therefore, letting K = AL + C12 + C22 + C32 + C42 + C52 , 1
we can complete the proof by σ(ΠK (x))∗ d (x) = 0, T r[σσ ∗ (Πk (x))d (x)] = 0,
2
F(x) + σ(x) ≤ K (1 + xE ).
d (x), F(Πk (x)) = μx∗1 (x1 − x∗1 ) 0
Based on the T heorem 5.2.1(see [9]) and T heorem 3.1, 2) ΠK (X) is on the edge of the polyhedron but not on the
we can prove the following theorem. vertices of the polyhedron. Suppose the edge of the polyhedron
Theorem 3.2 (Existence and Uniqueness): The system (5) is ∩i=1,2 {xi (t) = x∗i , xj (t) = 0, j = i, t ∈ I}, then ΠK (X) =
has a unique solution X(t) satisfying the initial ξ on I. (x∗1 , x∗2 , x3 , x4 , x5 ) and it is clear that
According to the proof in Theorem 3.1 and Theorem 1
5.4([10]), we can prove the following theorems. σ(ΠK (x))∗ d (x) = 0, T r[σσ ∗ (Πk (x))d (x)] = 0
2
3
TABLE I 10
PARAMETERS VALUES OF EACH REACTANT IN THE STOCHASTIC SYSTEM
biomass
5
Reactant μm ks mi 1 Yi 0
0 100 200 300 400 500 600 700
1, 3 − P D
glycerol
400
– – -2.69 67.69
200
Acetic acid – – -0.97 33.07 0
Ethanol – – -0.97 33.07 300
0 100 200 300 400 500 600 700
1,3−PD
200
100
0
0 100 200 300 400 500 600 700
Fig. 2. The future concentration on time interval [0, 650s] and the number
of sample trajectory is 50.