On the Design of the DISO Controllers for a Neuro-Fuzzy Vector Control The training data processing and some results

摘要

After some good previous results with neuro-fuzzy controllers SISO type for the speed or for the current loops in a vector control system, now the design is orientated to the DISO units. Also, the coexistence of both controllers type as neuro-fuzzy units are considered: the speed controller - with a slow bias and the current - with a high commutation rate for driving directly the inverter. The aim is also to provide a performing and integral intelligent-based solution for applications were the mechanical shock must be controlled (such as trams-trains, personal elevators). The design of such controllers is focused on three aspects: the conditions of a good data acquisition for training by simulation of a well tuned standard system, the data pre-processing for the training of the neural networks and the tuning of the synthesized controllers. The ANFIS methods is used for generating Sugeno fuzzy controllers. Different design details and tuning procedures are taken into account. The paper is continuing a previous work.