A Wind Power Forecasting Model Incorporating Recursive Bayesian Filtering State Estimation and Time-Series Data Mining

Abstract

To enhance the precision of wind power forecasting and the integration of renewable energy, a wind power prediction model, synthesising recursive Bayesian filtering state estimation with time-series data mining, was developed. Initially, the Autoregressive Integrated Moving Average Model (ARIMA)-Fractionally Integrated Generalized Autoregressive Conditional Heteroskedasticity (FIGARCH) model was utilised for mining historical wind power data and establishing a model. Subsequently, the double-parameter t-distribution was employed to fit the prior estimation error and observation error, which integrated observational information with prior estimates through a sophisticated recursive Bayesian filtering approach, culminating in the formulation of a robust predictive model. Validation of this model was conducted using a diverse dataset, encompassing wind farms with varying capacities and distinct time intervals. Simulation outcomes reveal that this model's forecasting accuracy markedly surpasses that of conventional methodologies. Notably, an enhanced predictive precision is observed in wind farms with larger capacities, particularly when shorter intervals of observational data are employed. This model demonstrates significant potential for advancing the accuracy and efficiency of wind power forecasting, a critical element in the optimization of renewable energy utilization.

Keywords

• data mining
• information fusion
• probability distribution fitting
• recursive bayesian filtering
• wind power forecasting