基于CS-GRU模型的短期负荷预测方法研究

Study on short-term load forecasting method based on CS-GRU model

针对门控循环单元(GRU)神经网络进行电力负荷预测时,其超参数选取困难等问题,提出一种布谷鸟搜索(CS)算法和GRU相结合的预测方法。研究发现,GRU的预测精度与超参数的设定有关,通过CS算法寻优GRU的超参数,包括迭代次数、学习率和隐含层节点数,节省了超参数选取时间,进一步提高了GRU的预测精度。最后,以河南某地区实例数据为例,在Python的TensorFlow框架下验证了预测方法的有效性。

Aiming at the problem of difficulty of choosing hyper-parameters for gated recurrent unit neural network in power load fore-casting, a method of prediction is proposed by combining cuckoo search(CS)algorithm with gated recurrent unit(GRU)neural network.The study found that the accuracy of power load prediction is related to the setting of hyper-parameters.Cuckoo search algorithm is used to optimize the super parameters of GRU,including iteration times, learning rate and number of hidden layer nodes, which saves the time of hyper-parameter selection and further improves the prediction accuracy of GRU.Finally, taking a county in Henan Province as an example, the effectiveness of the prediction method is confirmed in the tensorflow framework of python.