Power prediction method of small hydropower cluster based on convolutional neural network technology

Abstract

The invention provides a small hydropower cluster power prediction method based on convolutional neural network technology. The steps include: preprocessing of rainfall and runoff data; using random oversampling and SMOTE technology to balance the data set; building a convolutional neural network suitable for runoff prediction; converting the runoff prediction results into corresponding small hydropower output. Compared with traditional neural network technology, the method proposed in this patent adopts convolutional neural network input, takes planar space precipitation information as input, and has more abundant input data dimensions. The method of combining random oversampling and SMOTE is used to randomly generate moderate rain samples, and the invention effectively solves the problem of imbalance between light rain and moderate rain samples in the sample set. The convolutional neural network can effectively establish the mapping relationship between precipitation information and runoff information in planar space, avoiding the problem of difficult parameter setting in runoff forecasting model forecasting methods, and is suitable for online small hydropower output forecasting by power grid dispatching departments.

Description

technical field [0001] The invention relates to a small hydropower cluster power prediction method based on convolutional neural network technology. Background technique [0002] The day-ahead forecasting of power grid load is the basis for ensuring the safe and stable operation of the power system. Small hydropower is an uncontrollable power source, which is generally regarded as a "negative" load by grid dispatchers. Some mountainous areas in the southeastern coastal provinces of my country are rich in water resources, and the installed capacity of local run-of-the-river small hydropower accounts for a large proportion. These small hydropower stations under the management of prefecture-level power supply units, when there is a lot of incoming water, the hydropower stations will run at full capacity, and the discarded water will be released downstream; rather than being used. Therefore, the run-of-river hydropower station is basically in a state of generating water when ...

AI Technical Summary

Problems solved by technology

[0004] Based on the above deficiencies, the object of the present invention is to provide a small hydropower cluster power prediction method based on convolutional neural network technology, which can improve the accuracy of the predicted output of the small hydropower plan and reduce the deviation between the planned output of the small hydropower and the actual output. It is applicable to the conditions of light rain and moderate rain in the target area, but not suitable for the output power prediction under the condition of heavy rain, so as to effectively solve the problem of unbalance between light rain and moderate rain samples in the sample concentration, and avoid the runoff prediction model-based prediction method. Difficult to set parameters

Images