A novel genetic LSTM model for wind power forecast

• Published in: Energy (Oxford) Vol. 223; p. 120069
• Main Authors: Shahid, Farah, Zameer, Aneela, Muneeb, Muhammad
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 15-05-2021; Elsevier BV
• Subjects: Electric power demand; Electric power systems; Electricity consumption; Genetic algorithm; Genetic algorithms; Global optimization; Long short-term memory; Machine learning; Performance indices; Predictions; Rank tests; Regression; Renewable energy; Short term; System effectiveness; Wind farms; Wind power; Wind power forecast; Windmills; Long short-term memory; Regression; Wind power forecast; Genetic algorithm; Machine learning
• ISSN: 0360-5442; 1873-6785
• DOI: 10.1016/j.energy.2021.120069

Variations of produced power in windmills may influence the appropriate integration in power-driven grids which may disrupt the balance between electricity demand and its production. Consequently, accurate prediction is extremely preferred for planning reliable and effective execution of power systems and to guarantee the continuous supply. For this purpose, a novel genetic long short term memory (GLSTM) framework comprising of long short term memory and genetic algorithm (GA) is proposed to predict short-term wind power. In the proposed GLSTM model, the strength of LSTM is employed due to its capability of automatically learning features from sequential data, while the global optimization strategy of GA is exploited to optimize window size and number of neurons in LSTM layers. Prediction from GLSTM has been compared with actual power, predictions of support vector regressor, and with reported techniques in terms of standard performance indices. It can be evaluated from the comparison that GLSTM and its variants provide accurate, reliable, and robust predictions of wind power of seven wind farms in Europe. In terms of percentage improvement, GLSTM, on average, improves wind power predictions from 6% to 30% as opposed to existing techniques. Wilcoxon signed-rank test demonstrates that GLSTM is significantly different from standard LSTM.
•LSTM with ability of learning features from sequence data employed on wind dataset.•Genetic algorithm optimizes window size and number of neurons in LSTM layers.•Novel genetic LSTM (GLSTM) is proposed as LSTM with for wind power forecast.•Compared error measures from GLSTM with existing techniques and LSTM.•Wilcoxon Signed-Rank test ensures that GLSTM is significantly different from LSTM.