13.07.2015 Views

Assessment and Future Directions of Nonlinear Model Predictive ...

Assessment and Future Directions of Nonlinear Model Predictive ...

Assessment and Future Directions of Nonlinear Model Predictive ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

8 E.F. Camacho <strong>and</strong> C. Bordonsat every sub-iteration (inside a sampling period) is used <strong>and</strong> a decrease in thecost function is achieved, optimisation can be stopped when the time is over <strong>and</strong>stability can still be guaranteed. It can be demonstrated that it is sufficient toachieve a continuous decrease in the cost function to guarantee stability.The main technique that uses this concept was proposed by Scokaert et al. [43],<strong>and</strong> consists <strong>of</strong> a dual-mode strategy which steers the state towards a terminalset Ω <strong>and</strong>, once the state has entered the set, a local controller drives the stateto the origin. Now, the first controller does not try to minimise the cost functionJ, but to find a predicted control trajectory which gives a sufficient reduction <strong>of</strong>the cost.Simultaneous Approach: In this approach [12] (also known as multiple shooting[6]) the system dynamics at the sampling points enter as nonlinear constraintsto the optimization problems, i.e. at every sampling point the following equalityconstraint must be satisfied:¯s i+1 =¯x(t i+1 , ¯s i , ū i )Here ¯s i is introduced as additional degree in the optimization problem <strong>and</strong> describesthe initial condition for the sampling interval i. This constraint requires,once the optimization has converged, that the state trajectory pieces fit together.Thus additionally to the input vector [ū 1 ,..., u¯N ] also the vector <strong>of</strong> the ¯s i appearsas optimization variables. For both approaches the resulting optimizationproblem is <strong>of</strong>ten solved using sqp techniques. This approach has different advantages<strong>and</strong> disadvantages. For example the introduction <strong>of</strong> the initial states ¯s ias optimization variables does lead to a special b<strong>and</strong>ed-sparse structure <strong>of</strong> theunderlying qp problem. This structure can be taken into account to lead to afast solution strategy. A drawback <strong>of</strong> the simultaneous approach is, that only atthe end <strong>of</strong> the iteration a valid state trajectory for the system is available. Thusif the optimization cannot be finished on time, nothing can be said about thefeasibility <strong>of</strong> the trajectory at all.Use <strong>of</strong> Short Horizons: It is clear that short horizons are desirable froma computational point <strong>of</strong> view, since the number <strong>of</strong> decision variables <strong>of</strong> theoptimisation problem is reduced. However, long horizons are required to achievethe desired closed-loop performance <strong>and</strong> stability (as will be shown in the nextsection). Some approaches have been proposed that try to overcome this problem.In [46] an algorithm which combines the best features <strong>of</strong> exact optimisation<strong>and</strong> a low computational dem<strong>and</strong> is presented. The key idea is to calculateexactly the first control move which is actually implemented, <strong>and</strong> to approximatethe rest <strong>of</strong> the control sequence which is not implemented. Therefore the number<strong>of</strong> decision variables is one, regardless <strong>of</strong> the control horizon. The idea is thatif there is not enough time to calculate the complete control sequence, thencompute only the first one <strong>and</strong> approximate the rest as well as possible.An algorithm that uses only a single degree <strong>of</strong> freedom is proposed in [21] fornonlinear, control affine plants. A univariate online optimisation is derived byinterpolating between a control law which is optimal in the absence <strong>of</strong> constraints

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!