PENTA: Parameter Estimation Network for Target Approximation

In this paper, a novel neuro-fuzzy system has been introduced for the use of approximation and prediction of complex plants. The proposed system uses a parameter estimation network to provide coefficients of a linear combination, dynamically and then identifies a plant by approximating its target signal. So we have called it PENTA which stands for Parameter Estimation Network for Target Approximation. PENTA spans over five layers of a neuro-fuzzy network, composing two main parts of the system: a parameter estimation network and a dynamic linear combination. The first component provides the parameters (coefficients) required by the second component which is a dynamic linear combination. The overall output of the network is a synthetic function that is linear combination of the estimated parameters and a set of primitive functions. This architecture results in a very flexible model for learning complex plants. Experimental results show that in comparison with current neuro-fuzzy systems like ANFIS (Adaptive Neuro-Fuzzy Inference System), this novel system can have a better adaptation to complex plants. The results also show the unique capability of this system, in consecutive prediction of chaotic systems.