关键词: 深度回声状态网络, 灰色关联度, 聚类, 坐标轮换法, 空调负荷预测

Abstract: In view of the problems such as too much randomness of input weights,the large number of intermediate states and the determination of key parameters by trial and error in the deep echo state network,this paper uses the grey correlation degree to calculate the correlation between the attributes to determine the input weights.Then,the clustering algorithm is used to simplify the intermediate states,and the coordinate rotation method is used to improve the algorithm by searching for the optimal depth network layers and the number of the reserve pools.Through the experiment of UCI standard data set,the improved algorithm in this paper improves the accuracy and speed of the prediction.Finally,the improved deep echo network is used to predict the air conditioning load of the cigarette factory.Through the internal and external conditions of the current moment,the low prediction efficiency caused by the periodic fluctuation of the load data is solved.The air conditioning load of the next moment is accurately predicted in time,and the operation strategy of the chiller is adjusted in advance to achieve the purpose of air conditioning energy saving.