The main objective is to demonstrate that the Multiple Scenario Cooperative Model Predictive Control (MSC-MPC) algorithms developed under the Advanced Grant OCONTSOLAR can be effectively applied to optimize the production of commercial solar plants. This proposed PoC aims to validate the MSC-MPC algorithms developed in OCONTSOLAR not only for the solar field for the entire solar power plant. The proposal addresses optimal scheduling and control while accounting for the intermittent and stochastic nature of solar radiation and electricity market prices. Specifically, Model Predictive Control based on multiple scenarios will be used to manage uncertainties. The experiments and demonstrations will be conducted on a commercial plant. If successful, the involved company intends to apply these results to other plants. Furthermore, the proponent believes these techniques could be extended to other types of power plants.

This POC will prove that coalitional Model Predictive Control (Co-MPC) can be implemented on the existing distributed control system (DCS) of a real commercial solar trough plant (50MW) and can significantly increase the amount of solar energy collected and significantly reduce maintenance costs. This will be the first time that a Co-MPC is implemented in a real plant with so many dynamically interconnected subsystems (90). We have demonstrated that manipulating the loop HTF flows is fundamental for maximizing the collected solar energy in trough plants. The resulting MPC problem is too difficult to be solved with current control techniques because the number of dynamically coupled systems, up to 3200 collectors and 800 manipulated variables in the biggest solar trough plants and the complexity of the collector dynamics (nonlinear PDEs). The idea of Co-MPC is to divide the resulting complex MPC problem into several simpler MPC problems. Each of the MPC controls a coalition formed by a reduced number of subsystems. The coalitions are dynamically formed by clustering loops that can benefit from cooperation by exchanging the allocated oil flow (manipulated variable for each loop). This is done by using a market-based clustering MPC strategy in which controllers of collector loops (agents) may offer and demand heat transfer fluid in a market. Artificial neural networks will be used to approximate MPC controllers to decrease the computational load. We have shown that these techniques speed up the MPC computation time by a factor of 3000 allowing the implementation of coalitional MPC in the biggest solar trough plants. The PI has long experience in MPC control of solar energy systems and in the control of commercial solar trough plants having designed, implemented and commissioned MPC control systems for 17 commercial solar trough plants. A letter of support/intend of the industrial sponsor (one of the biggest stakeholders in Europe) is included.