Robust optimal control for a batch nonlinear enzyme-catalytic switched time-delayed process with noisy output measurements
Introduction
The evolution of a time delayed system is influenced by its state and/or control variables at some past time instants. Many practical real life systems can be modeled as time delayed systems [1], [2], [3], [4], [5]. Time-delays are sometimes deliberately introduced to help stabilize the system or force the system to track real data. However, in most cases, they appear naturally due to the nature of the problem concerned. The presence of time-delays in a system can cause undesirable behavior of the system or even causing it to become unstable. Thus, it has generated a strong interest among control community. For example, an optimization approach is proposed in [6] to the identification of state-delays. In [7], a class of optimal state-delay control problems is considered. Optimal control of switched systems with multiple time-delays and a cost on changing control is studied in [8]. A computational method is developed in [9] to solve time-delay optimal control problems with free terminal time. Delay independent stability criteria are derived for impulsive switched systems with time-invariant delays in [10]. A computation method is proposed in [11] to solve a class of dynamic optimization problems governed by switched time-delay systems with state-dependent switching conditions. There are many other studies on the control of time-delayed systems in the literature, such as those cited in the references of the papers mentioned above.
1,3-Propanediol (1,3-PD) is an important products used in chemical industry [12]. For the production of 1,3-PD, there are two methods: chemical synthesis; and microbial conversion of glycerol by Klebsiella pneumoniae (K. pneumoniae). This paper focuses on the second method because it is environmentally friendly, and high region specificity, and the recycling of feedstock is cheap. For the fermentation process, there are three main operation models: (i) batch culture [13] (all substrate is present at the beginning of the reaction and nothing is added or removed from the fermentor during the reaction); (ii) continuous culture (glycerol and alkali are continuously added during the reaction while removing old medium) [14]; and (iii) fed-batch culture (fresh medium is added discontinuously to the reactor at a constant rate to prevent nutrient depletion, but nothing is removed) [15]. The batch culture is a highly complicated real practically important problem. So the proposed solution approach is applicable to solve real practical problems. Other supporting reasons for the need of studying batch culture [16] are: (1) Batch culture can be expressed as an excessive metabolism process of glycerol. Thus, 1,3-PD yield, which is defined as the ratio between the formation of 1,3-PD and the consumption of glycerol, is high. This means that the high concentration of glycerol will lead to the high formation of 1,3-PD and low formation of by-product in batch culture of glycerol. It is worth noting that the high target product yield in batch culture can only be applied to the bioconversion of glycerol. However, not all the batch cultures are of the high target product yield; (2) Batch culture is a simple and easy operation mode when compared with fed-batch and continuous cultures; (3) Batch culture is basic for the understanding or controlling of fed-batch and continuous cultures. Therefore, batch culture has been extensively studied in the existing literature. There are many studies on the batch fermentation process being reported in the literature, such as bi-objective optimization [17], distributionally robust optimization [18], joint estimation [19], hybrid system [20], two-stage system [21] , strong stability [22], robust bi-objective optimal control [23], stochastic optimal control problem [24] and multiple characteristic time points [25]. However, the presence of time-delays is not considered in these studies.
In a dynamical system of microbial fermentation, the occurrences of time-delays are mainly due to two factors: (i) there exists the process of substrate absorption together with the inhibitions of substrate and multi-products across the cell membrane; (ii) cells are required to go through the growth process before they start to produce products [26]. In a recent paper [27], the identification of parameters of a nonlinear time-delay system in batch culture is under investigation, where the concentrations of intracellular substances are, however, ignored. In [28], a novel dynamic system, which takes into account the changes of the concentrations of both extracellular substances (biomass, glycerol, 1,3-PD, acetic acid and ethnol) and intracellular substances (glycerol, 1, 3-PD and 3-hydroxypropionaldehyde (3-HPA)), is proposed. According to the definition given in [29], the biological robustness is a property of a biological system that is insensitive to internal or external environmental influences. For a biological system, its biological robustness will ensure that its performance is maintained when it is subject to disturbances. For supporting arguments, see [30]. With the rapid development of systems biology, the research on biological robustness has been extensive [31], [32], [33], [34], [35]. In [20], [36], the definition of the quantitative robustness of batch fermentation organisms is given. However, it is only defined based on the expectation of the relative deviation of different system outputs caused by the perturbation of the decision variable in a given domain. The fluctuation of this relative deviation is not taken into account. In addition, the study being carried out in the above literature is based on the assumption that the experimental data is accurate. However, the experimental data cannot be completely accurate due to the influence of external temperature, humidity and pressure [37].
To construct an accurate model to describe the fermentation process, an unknown time varying function is required to be identified. However no information about this unknown time varying function can be extracted directly from the data set. Thus, optimization methods are required to be used to identify this time-varying function. In this paper, the batch culture of glycerol bioconversion to 1,3-PD induced by K. pneumonia is modeled as a nonlinear switched time-delayed (NSTD) system with an unknown time-varying function and noisy output measurements. According to the actual fermentation process, the time-varying function is in the form of a four-segment piecewise linear function with unknown kinetic parameters and switching times. To address the issue due to the lack of information on the unknown kinetic parameters and switching times, the quantitative biological robustness, which is inspired by the quantitative description of the biological robustness [29], [30], is proposed. It is expressed as a weighted sum of the expectation and variance of the relative deviation of the intracellular substance concentrations before and after the time-varying function is perturbed. To address the issue in regards of the noisy output measurements, the expectation of the least-squares error between the real system output (measured through experiments) and the predicted system output (generated from the mathematical model) of the extracellular substance concentration is defined. Our objective is to estimate the unknown kinetic parameters and switching times from the noisy experimental data and the quantitative biological robustness. In order to achieve this goal, a robust optimal control problem is proposed, in which the kinetic parameters and switching times are its decision variables, and the cost function is a weighted sum of the quantitative biological robustness of the intracellular substance concentrations and the least squares error of the expectation of the extracellular substance concentrations. This optimal control problem is governed by the NSTD system subject to the continuous state inequality constraints arising from the need that the concentrations of the extracellular and intracellular substances are to be limited to a specified region. By using the penalty function method, novel time scaling transformation and constraint transcription method, this optimal control problem is approximated by a sequence of nonlinear programming subproblems. Considering the complexity of these subproblems, a hybrid parallel algorithm, which is developed based on Nelder–Mead simplex search method (N–M simplex method), simulated annealing and the gradients of the constraint functions, is proposed to solve each of these subproblems. The numerical simulation shows that the optimal kinetic parameters and the switching times obtained are highly satisfactory.
The remainder of this paper is organized as follows. In Section 2, a NSTD system is formulated. In Section 3, a ROCP is proposed. In Section 4, a computational approach is developed. In Section 5, numerical simulations are carried out. In Section 6, some concluding remarks are made and the direction for future research is indicated.
Section snippets
Nonlinear switched time-delayed system
Some symbols are listed and explained in Table 1. Based on the factual experiments, the following assumptions are assumed throughout the paper.
- •
(A1). No medium is added to or withdrawn from the bioreactor during batch culture; and
- •
(A2). The concentrations of the reactants is uniformly distributed in the reactor.
In Figs. 2-5 of [18], on the left of the four figures, the red denotes computed results, and the blue denotes experimental data. Note that the nonlinear dynamic system is an
Robust optimal control problem
To identify these decision variables, we formulate an optimal control problem such that the cost function expressed as a weighted sum of the quantitative biological robustness and the expectation of the least-squares error between the real system output and the predicted system output of the extracellular substance concentrations is minimized.
Computational approaches
In this section, the ROCP is approximated by a sequence of approximate subproblems. Then, a parallelized synchronous algorithm is developed based on simulated annealing and N–M method to solve each of these subproblems.
Numerical results
Some experimental data are listed in Table 4 from [51] and [27]. System (26) is solved by using Euler method with a step size of (h). Those specific parameters appeared in Algorithm 1, Algorithm 2, Algorithm 3 are chosen empirically through numerical experiments to satisfy the required conditions of Algorithm 1, Algorithm 2, Algorithm 3 [50], [52], [53]. is solved by using Algorithm 1, Algorithm 2, Algorithm 3. In (9), to ensure that the expectation and variance are in the same
Discussions and conclusions
In this paper, we considered a NSTD system with an unknown time-varying function in batch culture. This time-varying function was in the form of a piecewise-linear function (4 segments) with unknown kinetic parameters and switching times. The quantitative biological robustness was defined. We formulated a robust optimal control problem for the identification of the unknown kinetic parameters and switching times. The optimal control problem was approximated as a sequence of nonlinear subproblems
CRediT authorship contribution statement
Jinlong Yuan: Writing - original draft. Lei Wang: Conceptualization. Jingang Zhai: Software. Kok Lay Teo: Methodology. Changjun Yu: Methodology. Ming Huang: Software. Jun Xie: Data curation.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (Grant Nos.: 11901075, 51979034, 11771008, 11701063, 71831002, 11871039 and 61773086), the China Postdoctoral Science Foundation (Grant Nos.: 2019M661073 and 2019M651091), the Natural Science Foundation of Liaoning Province in China (Doctoral Start-up Foundation of Liaoning Province in China, Grant No.: 2020-BS-074), the Fundamental Research Funds for the Central Universities, China (Grant No.: DUT20YG125), the
References (53)
Time-delay systems: an overview of some recent advances and open problems
Automatica
(2003)- et al.
A class of optimal state-delay control problems
Nonlinear Anal-Real
(2013) - et al.
A computational method for solving time-delay optimal control problems with free terminal time
Syst. Control Lett.
(2014) - et al.
Delay independent stability criteria of impulsive switched systems with time-invariant delays
Math. Comput. Model.
(2008) - et al.
Advances in bioconversion of glycerol to 1, 3-propanediol: prospects and challenges
Process Biochem.
(2018) - et al.
Multi-objective optimization of a continuous bio-dissimilation process of glycerol to 1, 3-propanediol
J. Biotechnol.
(2016) - et al.
Robust multi-objective optimal switching control arising in 1, 3-propanediol microbial fed-batch process
Nonlinear Anal. Hybrid
(2017) - et al.
Microbial production of 1, 3-propanediol: recent developments and emerging opportunities
Biotechnol. Adv.
(2009) - et al.
Distributionally robust parameter identification of a time-delay dynamical system with stochastic measurements
Appl. Math. Model.
(2019) - et al.
Pathway identification using parallel optimization for a nonlinear hybrid system in batch culture
Nonlinear Anal. Hybrid
(2015)