Adversarial Attacks to Solar Power Forecast
Ningkai Tang, Shiwen Mao, R.M. Nelms · 2021 IEEE Global Communications Conference (GLOBECOM) · 2021
With development of the photovoltaic industry, solar power generation forecasting using weather data has become an important problem. Various machine learning (ML) algorithms have been proposed to handle the random and massive weather data, with considerable recent interest on deep neural networks (DNN). Recent studies show that DNNs are vulnerable to adversarial examples, but most prior work has focused on their impact on the classification problem. In this paper, we investigate the problem of adversarial attacks on solar power generation forecasting, which is a regression problem. We examine the impact of adversarial attacks on both the DNN model and a LASSO-based statistical model proposed in our prior work. Both white-box attack and black-box attack are examined, along with the effect of adversarial training.