| This article has been：Browse 1402Times Download 2411Times |
scan it! |
|
|
| DOI：10.13522/j.cnki.ggps.2019278 |
|
| Using Data Mining to Diagnose Health Operation of Irrigation Channel Systems |
|
ZHAO Zhongsheng, XU Jinghui, WANG Lei, WANG Yichen
|
|
1.Northwest A & F University Key Laboratory of Agricultural Soil and Water Engineering in Arid Areas, Yangling 712100, China;2. College of Water Resources and Architectural Engineering, Northwest A&F University, Yangling 712100, China; 3.College of Mechanical and Electronic Engineering, Northwest A&F University, Yangling 712100, China
|
| Abstract: |
| 【Background】An irrigation channel system normally comprises a main channel, channels, branch channels, buckets and other water-retaining buildings. The health of the channel system is related to water utilization efficiency of its irrigation areas. Channel leakage and other health issues in the system were traditionally diagnosed by manual inspection, which is not only tedious and laborious but also unable to determine the health status underneath the structures. It could hence cause erroneous diagnoses or miss heath issues due to the difference in experience between inspectors. Given than most modern irrigation districts have been automatically operated and that data such as channel flow, flow velocity and fluctuation of water level are automaticity recorded, it is feasible to diagnose the channel system using these archived data.【Objective】The overarching objective of this paper is to present and test a data mining model to diagnose heath status of irrigation channel system. 【Method】 Archived data measured from October 2014 to October 2018 in an irrigation district at Guanzhong, Shanxi province were used to demonstrate the model development and validation. We first extracted and analyzed the data for different channels and the alarms reported during the operation of all channels in the irrigation district, from which we constructed the health diagnose system based on different neural network models and the CART decision-tree model.【Result】The training of the LM neural network model, the traditional BP network model and the CART decision-tree model based on 759 training samples showed that the accuracy of all three models was higher than 98%. For normal channel classification, the accuracy of the LM network model and the traditional BP network model was 93.5%, higher than the accuracy (92.4%) of the traditional CART decision tree model. For classifying 156 test samples, the overall accuracy of the LM network model and the traditional BP network model was 96.2%, higher than the accuracy (94.8) of the traditional CART decision-tree model. For classification of the normal channels, the accuracy of all three models reached 100%. The accuracy of the LM network model and the traditional BP network model was 8% higher than the traditional CART decision-tree model for classifying defect channels. Analysis of the ROC curves of the three models for the test samples showed that the LM neural network model worked better in the classification lines for both normal channels and the defect channels.【Conclusion】Of all three models tested, the LM neural network model worked best and can be used to detect operational status of the channel system in irrigation districts. Its accuracy in detecting defect channels was 80.95%. |
| Key words: data mining; irrigation district channels; neural networks; model |
|
|
