June 18, 2024

Sajad Ahmadian

Academic rank: Assistant professor
Address:
Education: Ph.D in Computer Engineering
Phone: 09188339565
Faculty: Faculty of Information Technology

Research

Title
New Hybrid Deep Neural Architectural Search-Based Ensemble Reinforcement Learning Strategy for Wind Power Forecasting
Type Article
Keywords
Advanced evolutionary algorithm, deep neural architectural search, ensemble reinforcement learning (RL) strategy, hybrid model, wind power forecasting
Researchers Seyed Mohammad Jafar Jalali، Gerardo J. Osório، Sajad Ahmadian، Mohamed Lotfi، Vasco M. A. Campos، Miadreza Shafie-khah، Abbas Khosravi، Joao Catalao

Abstract

Wind power instability and inconsistency involve the reliability of renewable power energy, the safety of the transmission system, the electrical grid stability and the rapid developments of energy market. The study on wind power forecasting is quite important at this stage in order to facilitate maximum wind energy growth as well as better efficiency of electrical power systems. In this work, we propose a novel hybrid data driven model based on the concepts of deep learning-based convolutional-long short term memory (CLSTM), mutual information, evolutionary algorithm, neural architectural search procedure, and ensemble-based deep reinforcement learning (RL) strategies. We name this hybrid model as DOCREL. In the first step, the mutual information extracts the most effective characteristics from raw wind power time series datasets. Second, we develop an improved version of the evolutionary whale optimization algorithm in order to effectively optimize the architecture of the deep CLSTM models by performing the neural architectural search procedure. At the end, our proposed deep RL-based ensemble algorithm integrates the optimized deep learning models to achieve the lowest possible wind power forecasting errors for two wind power datasets. In comparison with fourteen state-of-the-art deep learning models, our proposed DOCREL algorithm represents an excellent performance seasonally for two different case studies.