BP Neural Network for the Signal Recognition of Micro-Energy Devices

Abstract

The recognition of electrical energy generated by different micro-energy devices is a key step for effective energy management in the integrated microsystems. In this paper, we present an improved Backpropagation Neural Network (BPNN) model in conjunction with principle component analysis (PCA) and Levenberg-Marquadt (LM) algorithm. The open-circuit voltage curves containing the noise characteristics from three different types of micro-energy device including radio frequency energy harvester (RFEH), solar cell (SOR), vibration energy harvester (VEH) are used as input vector for classification model training and validation. The classifier of our model achieves a recognition accuracy of 100%, 91.8%, 83.1% for RFEH, SOR and VEH, respectively. These results indicate that our model is valid for energy recognition for these micro-energy devices and may further apply for hybrid energy management in smart systems such as wireless sensor networks (WSN).