System Identification of a Solar Tower Power Plant for Model Based Control

Kurzfassung

In this study, thermal models of solar tower power plant components for the purpose of model-based control are set up and fit to operational data of the solar tower in Jülich, Germany. Solar tower power plants provide renewable energy by concentrating solar radiance and converting the heat into electrical power in a power block. The technology inherently enables the use of thermal storages allowing the decoupling of power production and solar radiance. The volatile nature of the primary energy source, the sun, and the use of a thermal storage set challenges for the control of these systems compared to conventional power plants. Model-based control methods like model predictive control are promising for these power plants. For that a comprehensive model of the plant is needed. In future studies the performance of model based control is to be compared for physics based models and for data driven models based on neural networks. This study aims to provide physics-based models of the power plant’s components for a modular design of a solar tower power plant. Model reduction efforts are conducted in order to keep computational expenses down for optimization processes and to mitigate numerical issues when solving the system of equations. The models are fit to operational data by adjusting a chosen set of parameters. Moving horizon estimation and sequential quadratic least square programming are used to identify these sets of parameters. The systems of equations describing the dynamic behavior of the components is simulated using the do-mpc framework based on Casadi. For the automatic differentiation performed in Casadi, differentiable functions describing the thermodynamic properties of the fluids are needed. The thermodynamic properties have been approximated using polynomial functions. Their accuracy is evaluated for different degrees and domains. The results show that the methods for parameter identification are suitable and the simulations are able to reproduce the measured operational data. Sets of parameters for each simulated component are found and their accuracy is evaluated. The results show that the use of polynomials for steam property approximation as implemented in this study is very restricted for the simulation and optimization of large scale systems of differential algebraic equations. A comprehensive knowledge of the process and the accuracy of the polynomials in various domains is needed for their application. Simplifications made for model reduction efforts lower the model’s accuracy but enhance their computational performance. The effects of simplifications are analyzed and evaluated. This study aims to provide physics-based models of the power plant’s components for a modular design of a solar tower power plant. Model reduction efforts are conducted in order to keep computational expenses down for optimization processes and to mitigate numerical issues when solving the system of equations. The models are fit to operational data by adjusting a chosen set of parameters. Moving horizon estimation and sequential quadratic least square programming are used to identify these sets of parameters. The systems of equations describing the dynamic behavior of the components is simulated using the do-mpc framework based on Casadi. For the automatic differentiation performed in Casadi, differentiable functions describing the thermodynamic properties of the fluids are needed. The thermodynamic properties have been approximated using polynomial functions. Their accuracy is evaluated for different degrees and domains. The results show that the methods for parameter identification are suitable and the simulations are able to reproduce the measured operational data. Sets of parameters for each simulated component are found and their accuracy is evaluated. The results show that the use of polynomials for steam property approximation as implemented in this study is very restricted for the simulation and optimization of large scale systems of differential algebraic equations. A comprehensive knowledge of the process and the accuracy of the polynomials in various domains is needed for their application. Simplifications made for model reduction efforts lower the model’s accuracy but enhance their computational performance. The effects of simplifications are analyzed and evaluated.