A New Hybrid Direct/Specialized Approach for Generating Inverse Neural Models

In the literature the most common proposed solutions for training inverse neural models are the direct (or general) and specialized methods. The second one being considered as more reliable to produce correct inverse models has nevertheless some drawbacks in the implementation. The present paper introduces a hybrid solution that copes with the problems and limitations of both solutions. The hybrid solution merges the training and evaluation stages for producing an inverse model, ensuring the production of a correct inverse which is goal directed and does not need iterative algorithms to be produced. This new solution was developed within the preparation of an automated procedure for creating models for control purposes, which consists in a combination of Genetic Algorithms and Neural Networks. Key-Words: Feedforward Neural Networks, Direct Inverse Control, Internal Model Control, Measurement Noise, Genetic Algorithm, Inverse Modelling, Early Stopping, Direct Training and Specialized Training