An enhancement to the previously developed repetitive neurocontroller (RNC) is discussed and investigated in the paper. Originally, the time-base generator (TBG) has been used to produce the only input signal for the neural approximator. The resulting search space makes the dynamic optimization problem (DOP) of shaping the control signal solvable with the help of a function approximator such as the feed-forward neural network (FFNN). The plant under consideration, i.e. a constant-amplitude constant-frequency voltage-source inverter (CACF VSI) with an output LC filter, is assumed to be equipped with the disturbance load current sensor to enable implementation of the disturbance feed-forward (pDFF) path as a part of the non-repetitive subsystem acting in the along the pass p-direction. An investigation has been undertaken to explore potential benefits of using this signal also as an additional input for the RNC to augment the approximation space and potentially enhance the convergence rate of the real-time search process. It is numerically demonstrated in the paper that the disturbance feed-forward path active in the pass-to-pass k-direction (kDFF) improves the dynamics of the repetitive part as well indeed.
The paper describes a modification to the recently developed plug-in direct particle swarm repetitive controller (PDPSRC) for the sine-wave constant-amplitude constant-frequency (CACF) voltage-source inverter (VSI). The original PDPSRC algorithm assumes that the particle swarm optimizer (PSO) takes into account a performance index defined over the whole reference signal period. Each particle stores all the samples of the control signal, e.g. α = 200 samples for a controller working at 10 kHz and the reference frequency equal to 50 Hz. Therefore, the fitness landscape (i.e. the performance index) is -dimensional ( D), which makes optimization challenging. That solution can be categorized as the single-swarm one. It has been previously shown that the swarm controller does not suffer from long-term stability issues encountered in the classic iterative learning controllers (ILC). However, the convergence of the swarm has to be kept at a relatively low rate to enable successful exploitation in the D search space, which in turn results in slow responsiveness of the PDPSRC. Here a multi-swarm approach is proposed in which we divide a dynamic optimization problem (DOP) among less dimensional swarms. The reference signal period is segmented into shorter intervals and the control signal is optimized in each interval independently by separate swarms. The effectiveness of the proposed approach is illustrated with the help of numerical experiments on the CACF VSI with an output LC filter operating under nonlinear loads.
In this paper two different update schemes for the recently developed plug-in direct particle swarm repetitive controller (PDPSRC) are investigated and compared. The proposed approach employs the particle swarm optimizer (PSO) to solve in on-line mode a dynamic optimization problem (DOP) related to the control task in the constant-amplitude constant-frequency voltage-source inverter (CACF VSI) with an LC output filter. The effectiveness of synchronous and asynchronous update rules, both commonly used in static optimization problems (SOPs), is assessed and compared in the case of PDPSRC. The performance of the controller, when synthesized using each of the update schemes, is studied numerically.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.