Wind power ultra-short-term rolling prediction method based on WT-TCN

Abstract

The invention relates to a wind power ultra-short-term rolling prediction method based on WT-TCN, and the method is characterized in that the method comprises the steps: combining historical wind power data, searching the internal relation of historical wind power in time by using wavelet transform, and respectively training and predicting a high-frequency component and a low-frequency component of the historical wind power by means of a PReLU-based activation function time sequence convolutional network; mining the historical wind power data to the maximum extent, obtaining a prediction result through wavelet reconstruction, saving a model, jointly employing historical data and latest data as training data through rolling prediction, performing fine adjustment on internal parameters of the model, and updating a data change rule in a rolling mode, so the instability problem caused by wind power grid connection can be effectively solved; meanwhile, the income of the wind power plant can be improved when the wind power plant participates in day-ahead scheduling.

Description

technical field [0001] The invention relates to the technical field of wind power forecasting, in particular to a method for super-short-term rolling forecasting of wind power based on WT-TCN. Background technique [0002] With the increasing installed capacity of wind power, wind power has brought huge challenges to the safe operation of the power grid due to its randomness and volatility. The uncertainty of large-scale wind power access to the power grid, accurate wind power prediction It is of great significance to promote wind power consumption. [0003] At present, wind power prediction mainly includes methods such as probability statistics and deep learning. Probability and statistics models are represented by regression models. Due to the use of NWP for prediction, errors are artificially introduced, which reduces the accuracy of the model; deep learning models are represented by recurrent neural networks, which can directly use the relationship between meteorologica...

AI Technical Summary

Problems solved by technology

Probability and statistics models are represented by regression models. Due to the use of NWP for prediction, errors are artificially introduced, which reduces the accuracy of the model; deep learning models are represented by recurrent neural networks, which can directly use the relationship between meteorological data and historical wind power data. The deep network model built by the relationship has higher fitting characteristics; at the same time, a large number of prediction methods introduce signal processing methods such as empirical mode decomposition, which reduces the mean square error of prediction; but the prediction accuracy of current prediction methods is generally above 7, and the stability is poor

Images