Online Optimization - Cleaning strategies for CSP plants

Abstract

Due to the high solar irradiation Concentrated Solar Power plants are predominately located in arid areas, where the problem of soiling is often encountered. Dust and sand deposits on the mirror surfaces reduce the reflectivity and thus mirrors must be cleaned frequently in order to maintain the efficiency of the CSP plant. As excessive water consumption is a problem in arid areas, this work investigates a method to reduce consumption by scheduling cleaning in an optimal way. The thesis outlines a new methodology to derive an online algorithm that computes a water-saving cleaning schedule for CSP plants. The optimization algorithm is based on a Markov decision process, which evaluates the sequence of various cleaning options and derives a cleaning policy based on the forecasted soiling rate for a time horizon of up to ten days. A new aspect implemented within this algorithm are space resolved cleanliness values that are used to assess the need for cleaning certain parts of the solar field and thus maximize the energy yield of the whole plant. A technical overview of the plant is presented first, as well as an assessment of characteristics and requirements to the algorithm. Also the definition of a Markov decision process is given and the fundamental algorithm is described. The second part of this thesis is devoted to the formulation of the problem as a Markov decision process and the developed algorithm. Finally this algorithm is evaluated, by applying a quantitative analysis of multiple simulations with a variation of the parameters. The results of this study indicate that a Markov decision process is an appropriate mathematical formulation of the problem at hand. Moreover, the algorithm is able to achieve the following two improvements, compared to a constant cleaning schedule which is currently the common practice. Either a reduced water consumption by up to 20% could be achieved, while maintaining the energy output, or a stronger water reduction of 40% to 70% could be achieved with a minor energy yield reduction of 1% to 2%. Furthermore the use of a single parameter was found to be suitable for regulating the extent to which the optimizer reduces the number of cleaning actions and thus the consumption of water.