A simplified FPGA implementation of neural network algorithms integrated with stochastic theory for power electronics applications

Recently, artificial neural networks (ANNs) have found many promising applications in power electronics. However, the hardware implementation of ANN requires too many resources due to its parallel structure. The unavailability of real-time ANN hardware at an attractive price limits its applications. This paper proposes a simplified ANN hardware architecture using FPGA for the power electronic applications, and applies the proposed structure to a neural network based wind speed sensorless control system for wind turbine driven generators. This approach provides the solution for the low cost FPGA ANN chip. In order to reduce the scale of the digital circuits to implement ANN, stochastic principles are introduced to employ the simplified feed-forward neural network. By using this method which the real numbers are performed using random streams of bits instead of binary number, the normal digital approaches of arithmetic operations such as multiplication are replaced by stochastic arithmetic which could save digital resource significantly. In this paper, VHDL based FPGA techniques are also applied to analyze the performance of the proposed structure.