摘要
Abstract
Taking a multi-heat source ring heat-ing network as an example,a simulation model is estab-lished,and a large amount of blockage condition data is generated through simulation to analyze the influence of blockage on heat sources and heating stations.Based on the neural network,a classification model is estab-lished to locate the blocked pipeline,and then a regres-sion model is established to predict the proportion of blockage.Blockage can be classified into two types ac-cording to location.One type occurs on pipelines con-necting heat sources,thermal stations,and heating net-work main lines.Only the pressure of this heat source or heating station is affected,and the pressure change rate of the heat source is much more significant.An-other type of blockage occurs on the main lines of the heating network,which affects the pressure of various heat sources and stations.The supply and return water networks where the blocked pipeline is located also has an impact.According to the relative position of the heat source and heating station in the supply and return water network and the blocked pipeline,the pressure of the heat source and heating station in the upstream of the blocked pipeline increases,while the pressure of the heat source and heating station in the downstream decreases.In addition,when the blockage occurs in the water supply network,only the water supply pressure is affected;when the blockage occurs in the return water network,only the return water pressure is affected.When neural network model is used to diagnose the blocked pipeline,the division of data set has an impor-tant impact on the diagnosis effect.With the decrease of the blockage proportion step,the accuracy of the blocked pipeline diagnosis increases.In addition,no matter how the data set is divided,the accuracy of the blocked pipeline diagnosis increases with the increase of the blocked proportion.Under 100％load condi-tions,the accuracy of the blocked pipeline prediction is 98.77％,and the R2 score for predicting the blocked proportion is 99.42％.When the load changes,the ac-curacy of the blocked pipeline prediction is more than 98％,and the R2 score for predicting the blocked pro-portion is more than 99％,which has a good prediction effect.关键词
供热管网/管道堵塞/故障诊断/神经网络Key words
heating pipeline network/pipeline blockage/fault diagnosis/neural network分类
建筑与水利
