A Deep Learning Power Prediction Method Based on Multipoint NWP

Abstract

The invention discloses a depth learning power prediction method based on multi-point NWP. The method comprises steps that (1), data for power prediction is acquired in a designated region; (2), the data acquired in the step (1) is pre-processed to acquire a data set required by a training depth learning network; (3), each layer of the training depth learning network is trained layer by layer according to the data set acquired in the step (2) to acquire network parameters of each layer; (4), a neural depth network is initiated according to the network parameters of each layer acquired in the step (3), and fine tuning is carried out to acquire a final depth learning power prediction model; and (5), multi-point NWP data is inputted to the depth learning power prediction model acquired in the step (4) for prediction, and short-period power prediction results of any wind power set, any wind power field and any wind power field group in the designated region can be acquired through prediction.

Description

technical field [0001] The invention relates to the technical field of wind farms, in particular to a multi-point NWP-based deep learning power prediction method. Background technique [0002] The inherent volatility of wind power affects the safety, stability and economic operation of the power system, and is the main challenge for large-scale wind power grid integration. Wind power forecasting is one of the necessary means to solve this problem, and improving the accuracy of wind power forecasting is of great significance to the optimal operation of new energy power systems. [0003] The power forecasting model is a typical regression forecasting model, and its essence is the nonlinear regression function between forecasted wind conditions (ie numerical weather prediction, NWP) and wind power output power, which can be obtained through end-to-end learning. However, in the learning process, the regression function type, input method, output method, data preprocessing metho...

AI Technical Summary

Problems solved by technology

[0007] (1) Existing power prediction models usually only use a set of NWP as model input to establish a one-to-one mapping model between them and the output power of the wind farm (or wind turbine). This one-to-one modeling approach ignores the The connection between the positions of the field (that is, the temporal and spatial correlation of wind conditions), and the connection between the output of the unit are ignored, which greatly limits the improvement of the prediction accuracy;

[0008] (2) When modeling, only a single unit or a single wind farm is often considered, and all wind farms in the area are not considered, which greatly limits the scope of application of the model;

[0009] (3) The learning ability of previous models is difficult to meet the learning of large-scale data, and the value of data is often not fully exploited

Images