Wind power plant power ultra-short-term prediction method based on feature selection and recurrent neural network

Abstract

The invention discloses a wind power plant power ultra-short-term prediction method based on feature selection and a recurrent neural network. The method comprises the following steps: serving meteorological factors influencing wind power plant power as candidate features, and collecting historical data corresponding to the candidate features to obtain a candidate feature data set F; carrying outfeature selection on the candidate features in the candidate feature data set F to obtain a selected feature data set S; serving the selected feature data set S as input data of a prediction model based on the recurrent neural network; carrying out training and test on the prediction model based on the recurrent neural network by utilizing the input data of the prediction model; and carrying out power prediction on a wind power plant and outputting a wind power plant power prediction result. The beneficial effects are that generation power can be known in advance, so that detection of generating capacity can be finished conveniently, problems can be found and solved in time, and the grid can be deployed in advance; and the method is high in reliability.

Description

Technical field [0001] The invention relates to the technical field of wind power generation, in particular to an ultra-short-term prediction method of wind farm power based on feature selection and cyclic neural network. Background technique [0002] In the context of the sharp reduction of traditional fossil energy and the increasingly prominent environmental pollution problems, wind energy has become the most mature and fastest-growing renewable energy source. However, wind power is highly intermittent, random and uncertain. [0003] Ultra-short-term forecasting refers to forecasting the output power of the wind farm 0-4 hours in advance with a time resolution of not less than 15 minutes. The prediction results are mainly used for the optimization of frequency regulation and the adjustment of spinning reserve capacity of the power grid, as well as the optimal combination of online generating units and economic load dispatch. [0004] Super short-term wind farm power prediction m...

AI Technical Summary

Problems solved by technology

However, the prediction accuracy of the model with more input variables is not higher. On the contrary, too many inputs will make the prediction model more complicated, and problems such as over-learning and dimensionality disaster will occur.

In other words, some input variables may be useless and redundant to the output

Images