Near real-time estimation and optimization of microalga photobioreactor system for productivity improvement

Abstract

This thesis has presented the near real-time optimization procedures for productivity improvement of microalgal photobioreactor system under mixotrophic cultivation. Microalgae have been suggested as a promising feedstock for producing biofuel because of their potential for lipid production. However, the development of large-scale algal biodiesel production has been limited by the high production cost of algal biomass. Therefore it is necessary to improve the economic feasibility by reducing costs or increasing productivity. In order to have an economically sound algal bioprocess, this thesis tries to optimize the operating conditions by manipulating nutrient (carbon and nitrogen sources) flow rates and light intensity. For this purposes, it is need to develop a dynamic model that describes algal growth and lipid accumulation in order to support the development of algal bioprocesses, their scale up, optimization and control. However, there are some difficulties in applying model-based control strategies to microalgal cultivation systems. Microalgae cultivation systems are network of complex biochemical reactions manipuated by enzyme kinetics. Modelling of these complex biological systems accurately is difficult task since metabolism inside the cells makes systems have uncertainties. In addition to model uncertainties arising from complex biosystem dynamics, on-line measurement of important variables, especially in lipid is limited and difficult to realize in practice, which makes optimal bioreactor operation a challenging task. To cope with such problems, this thesis focused on the modelling, estimation of lipid concentration, and optimization of photobioreactor systems. At first, the model was developed based on the Droop model, and the optimal input design using D-optimality criterion was performed to compute the system input profile, to estimate parameters more accurately. From the experimental observations, the newly defined yield coefficient was suggested to represent the consumption of lipid and nitrogen within the cell, which reduces the number of parameters with more accurate prediction. Furthermore, the lipid consumption rate was introduced to reflect the experimental results that lipid consumption is related to carbon source concentration. The model was validated with experiments designed with different initial conditions of nutrients and input changes, and showed good agreement with experimental observations. After that, estimation of lipid concentration from other measurable sources such as biomass or glucose sensor was studied. Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), and Particle Filter (PF) were compared in various cases for their applicability to photobioreactor systems. Furthermore, simulation studies to identify appropriate types of sensors for estimating lipid were also performed. Finally, to maximize the biomass and lipid concentration, various optimization methods were investigated in microalgal photobioreactor system under mixotrophic conditions. Lipid concentration was estimated using UKF with other measurable sources and used as lipid data for performing model predictive control (MPC). In addition, maximized biomass and lipid trajectory obtained by open-loop optimization was used as a reference trajectory for traking by MPC. Simulation studies with experimental validation were performed in all cases and significant improvement in productivities of biomass and lipid was obtained when MPC applied. However, it was observed that lag phase occurs while manipulating feed flow rate, which considered to come from large amount of inputs introduced suddenly. This is important phenomena can make model-plant mismatches and needs to be researched more for the optimization of microalgal photobioreactor in reality.