Data-driven Prediction Model Selection Considering Time Series Characteristics of Wind Speed

As a clean and renewable energy source, wind power generation is becoming increasingly significant in smart microgrids. However, the inherent intermittency and randomness of wind energy significantly impact power quality and grid stability. Accurately forecasting future wind speeds is essential for maintaining stable operations in smart microgrids. This paper analyzes the ARIMA time series model, BP (back propagation) neural network, and LSTM (long short-term memory) deep learning model by comparing their prediction results, considering the varying volatility and stability of wind speeds across different months. The ARIMA model is well-suited for short-term predictions, achieving high accuracy during periods of stable wind speeds, which is particularly important for medium and long-term forecasts. The LSTM model, on the other hand, is especially effective for data with large fluctuations. The BP neural network model exhibits a good performance when focusing on the wind trend prediction. Thus, a suitable model should be chosen for wind speed prediction considering the specific characteristics of the current wind speed and the prediction needs.