Convolutional Long and Short-term Memory Network Prediction of Daily Water Intake from Waterworks Based on Multidimensional Input
Given that the prediction accuracy of traditional single-dimensional input convolutional neural network-long short term memory(CNN-LSTM)depends on the regularity of historical data,this paper constructs a prediction model based on convolutional long short term memory network with multi-dimensional input,in which Pearson correlation anal-ysis is used to identify the internal characteristics of the data or the correlation of external environmental factors,on the basis of which the model input is constructed and applied to the prediction of daily water intake of two waterworks.Com-pared with the traditional unidimensional input prediction model,the results show that the proposed multidimensional in-put CNN-LSTM prediction leads to a reduction of 32%and 17%of the average absolute percentage error in the daily wa-ter intake of water plant A,and a reduction of 47%and 12%of the average absolute percentage error in water plant B,indicating that the multidimensional input based on the internal features of the data is a higher prediction model.The rest of the evaluation indicators showed similar changes;Increasing the amount of water withdrawn from the water plant helped to improve the prediction accuracy of the model.This model input analysis method can provide an effective exam-ple for improving the accuracy of prediction models.
water withdrawal predictionconvolutional neural networkcorrelation analysismultidimensional input
NETL
NSTL
万方数据
