Alternative sphere decoding for finite control set model predictive control of power electronic converters

Date
2021-03
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH ABSTRACT: The rapid advance in computational power over the past decade has enabled model predictive control (MPC) to emerge as a promising alternative control technique for fast reacting power electronic applications. Trending in the power electronic system eld is - nite control set MPC, whereby a single multiple-input multiple-output (MIMO) controller can be realised to replace complicated PI control loops and deliver control signals for direct application to the converter. As an extension of basic MPC, long-horizon MPC refers to the evaluation of the optimisation criterion over a prediction horizon of more than one. Long-horizon MPC is desirable for its enhanced closed-loop performance during steadystate operation. Unfortunately, long-horizon MPC requires more computations to nd the optimum control action, especially in FCS-MPC, as the computational burden increases exponentially with an extension of the prediction horizon. The computational burden associated with long-horizon MPC can be reduced through mathematical programming where the underlying MPC problem is re-formulated as an integer least-squares problem. With regard to power electronic systems in particular, a recent adaptation of the Sphere Decoding Algorithm (SDA) has emerged as a very popular choice for solving or decoding of the ILS problem. The main objective of this thesis is to devise an alternative decoding strategy that will solve the ILS problem relating to long-horizon FCS-MPC for multi-level inverters. Exponential space algorithms are investigated and matched to an appropriate power electronic application. From the investigation, a novel preconditioning algorithm or projection algorithm evolved. Characteristics that favour modern processing technology are captured and applied to the well-acclaimed SDA to propose a decoding approach that involves matrix calculations. This also permits the e cient precomputation and storage of system matrices o ine for online use. The proposed algorithms are incorporated into a predictive controller for a power electronic drive and are evaluated in real time.
AFRIKAANSE OPSOMMING: Die dinamiese vooruitgang in rekenaartegnologie die afgelope dekade het modelvoorspellendebeheer (MVB) na vore laat tree as 'n belowende alternatiewe beheertegniek vir drywingselektronika-toepassings. Die neiging in die drywingselektronikaveld is eindige beheerstel modelvoorspellendebeheer, waardeur 'n enkele veelvoudige-inset veelvoudigeuitset beheerder tot stand gebring kan word om ingewikkelde beheerlusse te vervang en beheerseine vir direkte toepassing aan die omskakelaar te genereer. Langhorison-MVB verwys na die evaluering van die optimaliseringskriterium oor 'n voorspellingshorison van meer as een. Dit is veral gesog vir die geslote lusprestasie wat dit tydens bestendige toestande bied. Langhorison MVB benodig egter meer berekeninge om die optimale beheeraksie te vind. Die berekeningslas geassosieer met MVB kan wel verminder word deur wiskundige programmering, waar die onderliggende MVB-probleem geformuleer word as 'n heeltallige-minimumkwadraat probleem. Met betrekking tot drywingselektroniese stelsels, het 'n onlangse aanpassing van die sfeer-dekoderingsalgoritme na vore gekom as 'n baie gewilde keuse vir die oplossing van die heeltallige-minimumkwadraat probleem. Die primêre bydrae van hierdie tesis is om 'n alternatiewe dekoderingstrategie te ontwerp wat die heeltallige-minimumkwadraat probleem met betrekking tot eindige beheerstel MVB vir meervlakomskakelaars sal oplos. Eksponensiële ruimte-algoritmes word ondersoek en gekoppel aan 'n toepaslike drywingselektroniese toepassing. 'n Sekondêre bydrae van die tesis is 'n vooraf-kondisioneringsalgoritme of projeksie-algoritme wat voorgestel word. Eienskappe wat moderne prosesseringstegnologie bevoordeel, word vasgelê en toegepas op die veelbekroonde bol-dekoderingsalgoritme om 'n dekoderingsbenadering voor te stel wat matriksberekeninge insluit. Dit maak voorsiening vir die doeltre ende voorafberekening en berging van stelselmatrikse vir latere aanlyngebruik. Die voorgestelde algoritmes word toegepas in 'n voorspellende beheerder vir 'n induksiemotordryf, en word in werklike tyd geëvalueer.
Description
Thesis (PhD)--Stellenbosch University, 2021.
Keywords
Alternative Sphere Decoding, UCTD, Predictive control, Model predictive control, Electric current converters
Citation