Design of an Ultra-Short-Term Wind Power Forecasting Model Combined with CNN and LSTM Networks
Accurate short-term wind power forecasting has a significant impact to economic dispatch, which ensures the efficiency and smooth operation of power system. In this paper, a hybrid wind power forecasting method based on a typical algorithm of feedforward
- PDF / 246,500 Bytes
- 5 Pages / 439.37 x 666.142 pts Page_size
- 107 Downloads / 172 Views
[email protected] 2 State Grid Gansu Electric Power Company, Lanzhou 730030, China
1 Introduction With the advent of the new energy era, the low-carbon energy development model is accelerating the transformation, and the establishment of a green and diversified energy supply system is boosting. Wind power, as one of the clean, low-carbon, safe, and efficient new energy sources, has achieved remarkable promotion and application worldwide. The new energy represented by wind energy and solar energy is random, intermittent, and volatile. Accurate wind power forecasting is of great significance for determining reasonable dispatching plan and ensuring efficiency and economic operation of power grid. Therefore, the accuracy of wind power prediction becomes a key indicator of realtime adjustment of wind power in power system. Wind power prediction can adjust power distribution, optimize control strategy, and reduce operating cost [1]. There are four periods prediction of wind power that could be summarized according to different time ranges: ultra-short-term prediction, short-term prediction, medium-term prediction, and long-term prediction. Ultra-short-term wind forecasting is a prediction for wind power production in the next few minutes to a few hours, which can provide a basis for grid scheduling and power trading [2]. Currently, commonly used methods of wind power prediction are machine learning and deep learning. Machine learning method refers to the modern wind power prediction method which combines neural connection model, fuzzy model, support vector machine, and other intelligent algorithms. Deep learning extracts low-level features through feature engineering, and then forms them in an abstract way, which aims to figure out the relationship between features and data. In this paper, a hybrid CNN–LSTM model including Convolutional Neural Network and Long Short-Term Memory is proposed, which applies the advantages of data feature extraction and transformation in deep learning theory to wind power prediction. It not only uses convolutional neural network to reduce data size and complexity and improve the learning efficiency and overall generalization of the model, LSTM is also used to further explore the information features provided by different data sources in wind farms, establish the nonlinear relationship between multivariable time series and wind power time series, and solve the problems of low accuracy and modeling difficulty caused by high uncertainty and noise interference of wind energy in traditional methods.
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 C. H. WU et al. (eds.), Recent Developments in Intelligent Computing, Communication and Devices, Advances in Intelligent Systems and Computing 1185, https://doi.org/10.1007/978-981-15-5887-0_20
142
Y. Zhang et al.
2 Related Works 2.1 Wind Power Generation Power The output power of the wind turbine [3] is Pw =
1 Cp Aρv3 2
(1)
In the formula, Cp is the performance coefficient
Data Loading...
