Attention! Is Recycling Artificial Neural Network Effective for Maintaining Renewable Energy Efficiency?

Abstract

Modern society interests on the renewable energy and maintaining the efficiency of them. In the case of using solar energy, recognition and response to defectives as soon as possible is recommended because defects in solar panels reduces energy efficiency. In the same context as the interest in renewable energy, it would be better to use a proper lightweight defective detection model than a high-performance heavy model. In order to reduce the computational load in the training procedure, we define statistical features from the solar panel and use those for defective detection. Also, assuming that the attention mechanism that guides the key information, we recycle the pre-trained convolutional neural network that learned MNIST datset to enhance the feature values. Through the extracted statistical features, we achieve both reducing computational load in the training process and 0.831 for defective detection performance. Also, the detection performance is improved to 0.845 via recycling the pre-trained attention mechanism. The above means our approach additionally contributes renewable energy and sustainability via statistical feature extraction and recycling pre-trained neural network.