Wind Electricity Power Prediction Based on CNN - LSTM Network Model

With wind power technology becoming more sophisticated, the scale of power generation capacity and grid-connected wind power facilities have been greatly expanded, and the power of wind power plants has increased, penetrating a series of problems here. System supplies power to the increasingly prominent, the stable operation of power system and seriously threatens. To better arrange dispatch, it is especially important to make the wind power prediction more precise. We propose a hybrid model wind power prediction method based on CNN (convolutional Neural Network) and LSTM (Long term Memory Network) for time series and nonlinear characteristics of power data. Firstly input large-volume historical wind data, relevant influence factor data and date information according to time-sliding window, in which a continuous feature map is built. The above data extracted and normalized feature vectors of the data through MULTI-layer convolution and pooling stacking through CNN. The obtained feature vectors are used as input data of the LSTM network in the form of time series, through the LSTM network for wind power prediction. By using the above method, the data set HDWF2 is selected for prediction experimental data information in this paper including cloud cover, dew point, humidity, ozone, precipitation intensity, pressure, temperature, UV index, visibility, wind bearing, gust, wind speed, and power. Experiments show that the method brought up in the paper can predict more precisely than BP network wind power prediction method and standard LSTM network wind power prediction method, and the R
²
value can reach 0.9026.