Experimental investigation on complex-valued neural network for wind prediction

Abstract

Complex-valued signal is widely used in communication engineering and medical applications. With the progression of adaptive array signal processing, there is an increased in demand for neural network. Thus, there is a need to develop efficient complex-valued neural network system. Hence, it resulted in more emphasis on developing complex-valued neural network.

There are other application of complex signal such as for modelling and forecasting of wind which in turn aid the wind turbines to be effectively positioned to cater in as much wind power as possible.

However, the dispute is over the effectiveness of the different algorithms of the complex-valued neural networks. Although there are algorithms that has already been developed, each of them has its own limitation. Thus, in this project, a newly proposed neural network which is the fully complex-valued radial basis function (FC-RBF) network will be evaluated and compare against the split complex-valued networks which is a popular choice of complex-valued neural network.

Moreover, the performance of FC-RBF network would be assessed against other type of complex-valued neural network in the ability of convergence, approximation and performance gain.

In the development of the fully complex-valued radial basis function network, it is noted that the program has to perform faster than the matlab prototype and capable of supporting complex data type. Thus, the program is developed in C programming language.