15.05.2015 Views

Christoph Haederli - Les thèses en ligne de l'INP - Institut National ...

Christoph Haederli - Les thèses en ligne de l'INP - Institut National ...

Christoph Haederli - Les thèses en ligne de l'INP - Institut National ...

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.

NP Control with Optimal Sequ<strong>en</strong>ce SVM 129<br />

TABLE 51, POSSIBLE MODULATION SCHEMES MAKING USE OF MODIFIED AND VIRTUAL VECTORS<br />

Standard modulation scheme (NTV SVM,<br />

carrier based PWM, etc.)<br />

Predictive optimal<br />

sequ<strong>en</strong>ce SVM<br />

SVM based on<br />

alternate triangles<br />

(a) (b) (c) (d)<br />

Defnition of scheme<br />

Replace individual<br />

vectors directly by<br />

the partner states of a<br />

modified or virtual<br />

vector.<br />

Replace complete<br />

sequ<strong>en</strong>ce within a giv<strong>en</strong><br />

triangle by a<br />

pre<strong>de</strong>termined<br />

sequ<strong>en</strong>ce making use of<br />

modified and virtual<br />

vectors.<br />

Evaluation of<br />

multiple sequ<strong>en</strong>ce<br />

according to the<br />

POS (predictive<br />

optimal sequ<strong>en</strong>ce)<br />

SVM scheme as<br />

proposed in<br />

paragraph 6.2<br />

New small triangles can<br />

be <strong>de</strong>fined based on the<br />

position of the type 1<br />

virtual vectors.<br />

Explanations and comm<strong>en</strong>ts<br />

This works fine for<br />

modified vector and<br />

virtual vectors type 2.<br />

Virtual vectors type 1<br />

would still need to be<br />

implem<strong>en</strong>ted with an<br />

alternate algorithm.<br />

This scheme is likely<br />

to g<strong>en</strong>erate high<br />

switching losses as it<br />

applies CM jumps<br />

without consi<strong>de</strong>ring<br />

the rest of the<br />

modulation<br />

sequ<strong>en</strong>ce.<br />

This approach allows<br />

making a pre-selection<br />

of modified and virtual<br />

vectors to be applied<br />

and <strong>de</strong>termines all<br />

possible sequ<strong>en</strong>ces off<br />

line.<br />

This scheme produces<br />

relatively low switching<br />

losses as the whole<br />

sequ<strong>en</strong>ce can be<br />

optimized and partner<br />

states are not applied<br />

directly one after<br />

another.<br />

There is simple<br />

application at runtime,<br />

once a proper<br />

<strong>de</strong>finition of sequ<strong>en</strong>ces<br />

has be<strong>en</strong> ma<strong>de</strong> offline.<br />

This approach is<br />

really an ext<strong>en</strong>sion<br />

of the POS SVM.<br />

The systematic<br />

inclusion of<br />

modified and virtual<br />

vectors in the<br />

sequ<strong>en</strong>ces leads to<br />

high performance<br />

regarding NP and<br />

FC control.<br />

The calculating<br />

effort is ev<strong>en</strong> higher<br />

than for the pure<br />

POS SVM approach.<br />

This requires new<br />

algorithms to <strong>de</strong>termine<br />

application times of<br />

states in modified and<br />

virtual vectors, as well as<br />

algorithms for the choice<br />

of modulation type.<br />

The implem<strong>en</strong>tation of<br />

the actual nearest three<br />

vector modulation is<br />

equival<strong>en</strong>t to standard<br />

SVM. But an optimal<br />

choice of a set of vectors<br />

to be used is complex. If<br />

multiple modified and<br />

virtual vectors are used in<br />

one triangle, up to 12<br />

states may need to be<br />

applied.<br />

6.1.4 Application of modified or virtual vectors<br />

Modified or virtual vectors can be applied in differ<strong>en</strong>t ways as indicated in TABLE 51. The<br />

most obvious approach is the use of the modified and virtual vectors as vertices in NTV SVM.<br />

However, there is a lot of redundancy. The choice of base vectors to be used is not obvious. An<br />

online optimization is quite <strong>de</strong>manding and does not necessarily result in best performance.<br />

Furthermore, a direct application of virtual vectors without specifically integrating the states within<br />

a sequ<strong>en</strong>ce will result in high switching losses due to the jumps in CM.<br />

An offline sequ<strong>en</strong>ce optimization applying modified and virtual vectors according to TABLE 51<br />

(b) offers high performance at reasonable complexity. This approach has be<strong>en</strong> chos<strong>en</strong> to be<br />

implem<strong>en</strong>ted for simulation and experim<strong>en</strong>tal verification.

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

Saved successfully!

Ooh no, something went wrong!