Richard Faber

Dynamic optimization with simulated annealing

In this paper we present an approach for the dynamic optimization of energy and chemical engineering processes with the simulated annealing (SA) algorithm. The aim of this work is to develop an optimization methodology which finds optimal control strategies requiring a minimum of user input. The methodology we propose uses rigorous dynamic Simulink models based on first principles in a black-box approach. If any, only slight modifications have to be done for existing Simulink models to be able to be optimized. The presented approach based on the SA algorithm has the potential to find the global optimum and it does not need any additional information than the dynamic model itself. Both control profiles and time-invariant parameters, such as PI control parameters can be optimized simultaneously. The simulated annealing algorithm is implemented in a set of MATLAB script files together with a graphical user interface (GUI) as an independent module of the OptControlCentre (OCC). All parameters of the simulated annealing algorithm and of the model can be specified in the GUI as well as the course of the optimization itself is made visible at runtime. The proposed methodology is tested on a set of example problems and the influence of several simulated annealing parameters on the optimization is shown.

Data reconciliation for real-time optimization of an industrial coke-oven-gas purification process

Abstract Online optimization is more and more used in the chemical industry to run a process near its optimum operating condition by providing real-time computed optimal set-points to the distributed control system. Process measurements are necessary for these applications to determine the actual state of the process and to increase the accuracy of the model with parameter estimation techniques. However, these measurements usually contain random as well as gross errors which have to be identified and eliminated before the measurements are used for online optimization. In this contribution, a data reconciliation approach was integrated into an online optimization framework for the ammonia hydrogen sulfide circulation scrubbing, a common industrial coke-oven-gas purification process. We used a rigorous rate-based model to describe this reactive absorption and desorption process. To increase the accuracy of the model, we estimated several process parameters using a sequential parameter estimation approach. Data reconciliation was performed based on simple component balances to achieve model-consistent data and to identify measurement biases. The model was then validated online on a pilot plant by connecting the estimation package through the process control system. Based on the online measured data, operating cost minimization was carried out and the computed optimal set-points realized real-time. A satisfactory agreement between measured data and optimization was achieved.

A hybrid optimization approach to parameter estimation

Abstract Many parameter estimation problems in chemical or biochemical engineering lead to ill-conditioned and nonconvex optimization problems. For bad starting values the use gradient based result in local optimal solutions. To overcome this drawback, a global optimization approach, Simulated Annealing, has been coupled with a gradient-based SQP approach. To improve the accuracy of the parameter estimates, sensitivity information has been included into the objective function by iteratively adjusting the weighting matrix with the variancecovariance matrix of the model prediction. The hybrid approach has been applied to a case study of biochemical nonlinear parameter estimation problem.

Three-phase distillation in packed towers: Short-cut modelling and parameter tuning

Abstract A short-cut model is developed for the predictive simulation of a three-phase distillation process in packed towers. The model is taking into account the mass transfer resistance between the vapour phase and both liquid phases whereas equilibrium is assumed between the two liquid phases The development of the model was strongly connected with a systematic experimental investigation study of three-phase operated packings. Thus, it was possible, to estimate unknown parameters occurring in the model. Therefore, a powerful parameter estimation technique which is able to handle a large number of experimental data sets was applied. Both the simulation model and the parameter estimation method are discussed in this contribution. The first results are in good agreement with the experimental data of the packed column.

Improving observability of large-scale systems by iterative weighting adjustment

Abstract An optimization-based approach is proposed to improve the observability of large-scale systems with iterative adjustment of the weighting matrix. The approach is based on a rigorous process model making it applicable to nonlinear systems. The result of the state estimation is improved by introducing sensitivity information into the weighting of the objective function. The needed sensitivity information is iteratively computed and adjusted during the optimization run. The approach has been applied to a large-scale nonlinear industrial process to estimate the unknown feed composition from scarce measurement data.