New way to boost performance in changing networks

A new method uses neural policies to improve networked systems during plug-and-play operations.

Researchers developed a way to improve the performance of pre-stabilized networked systems. These systems undergo plug-and-play (PnP) operations, which involve the addition or removal of subsystems.

The method uses Performance Boosting controllers based on the Internal Model Control (IMC) principle. This principle reduces closed-loop stability to a finite ℓ2\ell_2-gain requirement on an auxiliary operator.

The distributed setting uses local neural operators that exchange information only between neighboring subsystems. Two parametrizations were developed: a neighbor-based compositional parametrization and a network-norm parametrization.

The neighbor-based compositional parametrization uses local coupling bounds. After a PnP event, only subsystems with changing bounds need to update their gain parameters.

The network-norm parametrization uses a bound on the overall coupling matrix. This can reduce conservatism but requires network-wide information.

In a PnP DC-microgrid case study, the distributed neural controller improves voltage transients. It does this during topology changes and load variations while preserving nominal closed-loop stability.

Key facts

  • The controller uses local neural operators that exchange information between neighboring subsystems.
  • The system can be trained by unconstrained gradient-based optimization without stability constraints.
  • The method improves voltage transients in a PnP DC-microgrid case study.

All sections

Edition