Very Short-Term Wind Power Forecasting Based on SVM-Markov

Abstract

Abstract. Very short-term forecasting of wind power is important to scheduling staff's planning and wind turbine control. This paper has established a combined forecasting model based on Markov chain and support vector machine (SVM). Firstly, the SVM is used to model for wind power. Then, transition probability matrix is made based on Markov chain to modify for SVM prediction. Finally, the prediction confidence interval of combination forecasting model is given by method of fluctuation confidence interval. Verified by an example of a wind farm indicating that the combination forecasting model is better than a single SVM model on a variety of error indicators.

Short-term wind power generation point and interval prediction based on a novel seasonal multivariable grey model

“…The statistical forecasting method is based on the current local observational data, and extracts the typical features through parameters, data training, curve fitting and other methods to establish a prediction model. The commonly used methods are include time series method (Zhang and Li 2016;Xu et al 2022), support vector machine method (Gendeel et al 2021;Jiang et al 2015), neural network method (Wang et al 2023;Yuan et al 2019), grey model method (Dang et al 2023;Ofosu-Adarkwa et al 2020;Ren et al 2023) and so on. Zhang and Li (2016) used the ARIMA model of time series to forecast short-term wind load, assumed that the prediction error followed the normal distribution, and gave prediction intervals using the Monte Carlo method.…”

“…Lin and Weng (2004) gave competitive prediction intervals using the out-of-sample residuals of the support vector machine model (SVM) for all types of samples, using the zero-mean Gaussian and Laplace families as noise models. Jiang et al (2015) used SVM and Markov chain correction for wind power prediction, and then gave prediction intervals using the fluctuation confidence interval method. Gendeel et al (2021) proposed a VMD decomposition and weighted least squares support vector machine (LSSVM) to build a deterministic wind power forecasting model, which quantifies the potential risk of wind power sequences using the lower upper bound estimation (LUBE) method for probabilistic interval prediction.…”