Page - 145 - in Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources

5.2. ResultsofSystemIdentification 5.2. ResultsofSystemIdentification Based on the verification results, different system identification al- gorithms have been tested and compared in this section, to identify which one is more accurate to estimate the real dynamics of the HEP- HAISOTS system. Data from practical experiments are used to certify the reliability of each algorithm. The validation of system identifi- cation algorithms contain two different tests. The first test is to use differentalgorithmstolearnthedynamicsof theplantfromtheexper- imentaldata,andchecktheirone-steppredictionaccuracies. Theone- steppredictionaccuracyindicatestheabilitytopredict thefuturetem- peraturesbasedonthecurrent temperaturesandcontrol inputvalues, which is how the system estimator functions in real control process. The second test is the regeneration of temperature curves. Given the same initial temperature value and the whole control input sequence, theestimators trainedbythefirst testareusedtoregenerate theentire temperature sequence. The regenerated data are compared with the real data, to test if individual algorithms really capture the dynamics ofHEPHAISTOS. It isassumedthat thedatausedin twotestsaregivenas D1= {( Ud1(1),Yd1(1) ) , ( Ud1(2),Yd1(2) ) , .. . , ( Ud1(Q1),Yd1(Q1) )} , D2= {( Ud2(1),Yd2(1) ) , ( Ud2(2),Yd2(2) ) , .. . , ( Ud2(Q2),Yd2(Q2) )} , where D1 and D2 are data sets obtained from experiments using the same setup and the same heating environment. In the first test, the objective is to use each algorithm for online system identification (the data pair is presented and estimated one by one), and compare the one-step predicted temperature Ypre and the real temperature Yd1. Theone-stepprediction isexpressedas Ypre(1) =Yd1(1), Ypre(q+1) =fq(Yd1(q),Ud1(q)), 1≤ q≤Q1−1, wherefq represents thefunctionof thesystemestimator (after theup- dateof theq-thdatapair). 145