28 فروردین 1403

سجاد احمدیان

مرتبه علمی: استادیار
نشانی: دانشگاه صنعتی کرمانشاه
تحصیلات: دکترای تخصصی / مهندسی کامپیوتر
تلفن: 09188339565
دانشکده: دانشکده فناوری اطلاعات

مشخصات پژوهش

عنوان
Solar irradiance forecasting using a novel hybrid deep ensemble reinforcement learning algorithm
نوع پژوهش مقاله چاپ شده
کلیدواژه‌ها
Solar irradiance forecasting; Deep neural networks; Evolutionary computation; Ensemble strategy; Deep reinforcement learning
پژوهشگران سید محمد جعفر جلالی (نفر اول)، سجاد احمدیان (نفر دوم)، بهاره نکیسا (نفر سوم)، مهدی خدایار (نفر چهارم)، عباس خسروی (نفر پنجم)، سعید نهاوندی (نفر ششم به بعد)، سید محمد شمس اسلام (نفر ششم به بعد)، میعادرضا شفیعی خواه (نفر ششم به بعد)، ژائو کاتالائو (نفر ششم به بعد)

چکیده

Solar irradiance forecasting is a major priority for the power transmission systems in order to generate and incorporate the performance of massive photovoltaic plants efficiently. As such, prior forecasting techniques that use classical modelling and single deep learning models that undertake feature extraction procedures manually were unable to meet the output demands in specific situations with dynamic variability. Therefore, in this study, we propose an efficient novel hybrid solar irradiance forecasting model based on three steps. In the first step, we employ a powerful variable input selection strategy named as partial mutual information (PMI) to calculate the linear and non-linear correlations of the original solar irradiance data. In the second step, unlike the traditional deep learning models designing their architectures manually, we utilize several deep long short term memory-convolutional neural network (LSTM-CNN) models optimized by a novel modified whale optimization algorithm in order to compute the forecasting results of the solar irradiance datasets. Finally, in the third step, we deploy a deep reinforcement learning strategy for selecting the best subset of the combined deep optimized LSTM-CNN models. Through analysing the forecasting results over two real-world datasets gathered from the USA solar irradiance stations, it can be inferred that our proposed algorithm outperforms other powerful benchmarked algorithms in 1-step, 2-step, 12-step, and 24-step ahead forecasting.