| 摘要: |
| 【目的】探究分类模型对灌溉渠道运行状况健康度检测效果。【方法】对陕西关中地区某灌区2014年10月―2018年10月各级渠道流量以及灌溉发生运行不良状况下异常终端报警信息等数据进行抽取和探索分析处理，提取渠道运行状况特征评价指标，构建灌区渠道运行状况健康识别模型。【结果】构建的LM神经网络模型与传统BP网络模型、CART决策树模型，对759个训练样本分类中综合分类准确率均高于98%；对156个测试样本，综合分类准确率LM神经网络模型相比BP网络模型同为96.2%，CART决策树模型为94.9%。对3个模型测试样本ROC曲线分析发现，LM神经网络模型中运行正常渠道分类准确度折线与运行不正常渠道分类准确度折线表现更优。【结论】LM神经网络模型是最优模型，可实际应用于渠道运行状态健康度识别检测，实际应用中对该灌区运行不正常渠道识别准确率为80.95%。 |
| 关键词: 数据挖掘;灌区渠道 神经网络;模型 |
| DOI： |
| 分类号:S |
| 基金项目:科技部国家重点研发计划课题(2017YFC0403203)和陕西水利科技计划项目(2014slkj-18); |
|
| Research on Health Status Detection of Irrigation Channel Operation Based on Data Mining |
|
zhaozhongsheng1,2, 许景辉1,2
|
|
1.Northwest A&2.F University
|
| Abstract: |
| 【Background】The irrigation channel project mainly includes main trunk, trunk, branch, bucket, agricultural canal and related water-retaining buildings. The health status of each channel is closely related to the utilization efficiency of water resources in the irrigation area. Traditional canal system leakage and other health status judgments are mainly carried out by manual inspection. This method is not only time-consuming and laborious, but also unable to determine the status of underwater structures. It also causes misjudgments or missed judgments due to different experience of patrolling personnel, resulting in water resources in the irrigation area. Serious waste. At present, the modern irrigation district has basically realized the automatic collection and recording of data such as channel flow, flow velocity, and water level changes. However, the irrigation area only counts the periodic water volume, and based on data mining technology and neural network analysis methods, there is little research on the detection of the health status of the channel water delivery in the irrigation area. Data mining refers to the extraction of potentially effective, understandable, and high-level processes from massive data based on specific business goals. If advanced technologies such as data mining can be used to reveal the water consumption rules of the canal system, discover and extract the health evaluation indicators of the canal system. It will play a positive role and have important significance in improving the efficiency of water resources utilization and the production and management of irrigation districts.【Objective】To explore the effect of classification model on the health status of irrigation channels【Method】Data extraction, exploration and analysis of data from various channels and abnormal terminal alarms of irrigation occurrence and operation in an irrigation district in Guanzhong, ShanXi Province from October 2014 to October 2018, and extraction of channel operation status evaluation indicators to construct channel operation status of irrigation districts health recognition model.【Result】The constructed LM neural network model, traditional BP network model and CART decision tree model classify 759 training samples. The comprehensive classification accuracy of the LM network model, traditional BP network model, and CART decision tree model is above 98%; In the normal channel classification, the results of the LM network model and the traditional BP network model are equal to 93.5%, which is higher than the 92.4% of the traditional CART decision tree model. In the classification of 156 test samples, the comprehensive classification accuracy rate of the LM network model and the traditional BP network model is 96.2%, which is higher than the traditional CART decision tree model of 94.9%; the normal channel classification 3 model is 100%; The accuracy rate of the LM network model and the traditional BP network model in the classification of abnormal channels is 8% higher than the traditional CART decision tree model. The analysis of the ROC curve of the test samples of the three models found that the LM neural network model performed better in the classification line of normal channel classification accuracy and that of abnormal channel classification.【Conclusion】The LM neural network model is the optimal model, which can be applied to the channel operation status detection and detection. In practice, the accuracy rate of the abnormal channel identification in the irrigation area is 80.95%. |
| Key words: data mining; irrigation district channel; neural networs;model |
