A short-term wind power prediction method and system based on DWT and LSTM

A power forecasting and power technology, applied in forecasting, information technology support systems, instruments, etc., can solve the uncertainty, instability and volatility of wind energy, can not provide reliable data reference for grid scheduling and planning, can not ensure Safe and stable operation of power grids, etc., to avoid gradient disappearance and gradient explosion, improve prediction accuracy, and avoid obstacles

CN109376904AInactive Publication Date: 2019-02-22GUANGDONG POWER GRID CO LTD +1
8 Cites 12 Cited by

Patent Information

Authority / Receiving Office
CN · China
Current Assignee / Owner
Publication Date
2019-02-22
Estimated Expiration
Not applicable · inactive patent

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The present invention relates to the technical field of distributed renewable energy, more particularly, to a short-term wind power generation power prediction method and system based on DWT and LSTM,comprising: decomposing wind power raw data into low-frequency signals and high-frequency signals; standardizing the low frequency and high frequency signals using the z-score normalization method. so that the normalized data obeys normal distribution; dividing low frequency signal and the high frequency signal into a training set, a verification set and a test set according to the time sequence.The independent LSTM is used to train the training set and the verification set, and the test set of each signal is predicted to obtain the predicted value of each component, and the predicted valueof each component is summed up to obtain the predicted result. The invention can more fully mine the main information and the secondary information contained in the detailed information contained in the approximate signal in the data, can effectively avoid the problems of gradient disappearance and gradient explosion, and can better improve the prediction accuracy of the wind power generation.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The present invention relates to the technical field of distributed renewable energy, and more specifically, to a short-term wind power prediction method and system based on DWT and LSTM. Background technique

[0002] Global renewable energy has been showing rapid growth, among which wind power has grown exponentially, and its importance has become increasingly prominent. The increasingly serious environmental pollution and energy crisis have made countries all over the world pay more attention to wind power generation, and a lot of related research work has been carried out. The large-scale grid-connection of wind power can alleviate the problem of power supply shortage. However, due to the random uncertainty, non-stationarity, volatility and variability of wind energy to a certain extent, the greater the grid-connected power, the greater the impact on the grid. It will affect the stability and security of the power system.

[0003] Recently, releva...

Examples

Embodiment

[0035] like figure 1 Shown is the embodiment of the short-term wind power generation power prediction method based on DWT and LSTM of the present invention, comprises the following steps:

[0036] S10. Collecting raw wind power generation data, and using DWT to decompose the raw wind power generation data into low-frequency signals and high-frequency signals;

[0037] S20. Standardize the low-frequency signal and high-frequency signal in step S10 by using the z-score standardization method, and make the normalized data obey the normal distribution;

[0038] S30. Dividing the standardized low-frequency signal and high-frequency signal in step S20 into a training set, a verification set, and a test set according to time order;

[0039] S40. Using an independent LSTM to train the training set and verification set of each signal, and predicting the test set of each signal to obtain the predicted value of each component, and summing the predicted values ​​of each component to obta...