Neural network modeling of the gust effects on a grid-interactive wind energy conversion system with battery storage

Abstract

This paper presents a simplified way to predict accurately the dynamic responses of a grid-linked Wind Energy Conversion System (WECS) to gusty winds using a Recurrent Neural Network (RNN). The RNN is a single-output feedforward backpropagation network with external feedback. High winds, which are stochastic and turbulent by nature, create a quasi-permanent transitory environment for the WECS, putting the system in a seemingly non-equilibrated state. For instance, the generator current at the system output, besides being dependent upon the current excitation signal at the system input, will also greatly be dependent upon the past history of the system. The feedback in the RNN allows past responses to be fed back to the network input. In this model, the WECS parameters need not be known. After proper training, the known and unknown dynamics of the WECS are captured by the network structure and stored in the connection weights between consecutive layers. For that reason, the neural modeling allows us to circumvent two majors problems faced by conventional linear mathematical modeling of system dynamics: parameter uncertainty and system non-linearity, which could be significant in a difficult dynamic environment as in high winds. Moreover, neural modeling is universal, and involves less computational effort. The viability of the battery supported system as dispatchable unit is also discussed. The simulated values are compared with actual values; excellent results have been achieved.