Seite - 146 - in Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources

5. ExperimentalResults After the first test, the second test aims to regenerate the whole data set T2by Yreg(1) =Yd2(1), Yreg(q+1) =fQ1−1(Yreg(q),Ud2(q)), 1≤ q≤Q2−1, The regenerated temperature Ypre will also be compared with the real data Yd2 to determine the regeneration error. Compared with the one-step prediction, the test of regeneration can more comprehen- sively reflect the estimation accuracy of each system identification al- gorithm, and hence, is an important factor to be taken into account for the determination of the final implemented system identification algorithm. 5.2.1. Grey-boxApproaches For the grey-box modeling approach, four different system identifica- tionschemeshavebeentested, includingthenonlinearMISOEKF,the linearMISORKF,thelinearMISORLSandthelinearMIMORKF.Dif- ferences among these algorithms can be found in chapter 3. The data used in these tests are obtained from experiments with eight heating sources and two controlled temperatures. Results of the different al- gorithmsareshowninfigures 5.13,and 5.14. In order to get more reliable evaluations to the performance of each algorithm, each algorithm was tested for 20 times and the averaged performance is shown in table 5.1. In following tests, the sampling (discretization)periodis∆t= 1s,whichmeanstheprocessingtimeis equivalent to thenumberofdatapairs (1datapairpersecond). Theresults showninthefigurescangenerallyreflect theperformance of individual algorithms in practical heating experiments. Several brief conclusionsareobtainedfromaboveresults. • Fromtheconvergingspeedpointofview,thenonlinearMISOEKF algorithmhasthefastestconvergingspeedamongthesefouralgo- rithms, using less than 100 seconds to reach the general MSE level of0.1. Incomparison,bothMISORKFandMIMORKFtookabout 100 seconds to reach that MSE level. MISO RKF has the slowest convergingspeed,withabout450secondstoreachthaterror level. 146