| 基于改进电法装置和深度神经网络的填埋场渗滤液水位探测 |
| |
| 引用本文： | 夏广培,能昌信,侯晓姝,刘景财,杨枫,姚光远,徐亚.基于改进电法装置和深度神经网络的填埋场渗滤液水位探测[J].环境科学研究,2022,35(8):1945-1957. |
| |
| 作者姓名： | 夏广培 能昌信 侯晓姝 刘景财 杨枫 姚光远 徐亚 |
| |
| 作者单位： | 1.山东工商学院信息与电子工程学院，山东 烟台 264005 |
| |
| 基金项目： | 国家重点研发计划项目(No.2020YFC1806304，2018YFC1800902)；国家自然科学基金项目(No.51708529) |
| |
| 摘 要： | 渗滤液水位检测和管控是填埋场环境风险管控的关键之一，高密度电法装置具有无损、快速和分辨率高等优点，但填埋场铺设的HDPE防渗膜的高阻特性掩盖了渗滤液的低阻特性，反演结果存在探测精度差、探测范围有限等问题，无法有效识别渗滤液水位. 为此，本文设计了改进的川字型采集装置并配套提出了基于深度神经网络的非线性反演算法——EConvNet-C，通过构建高仿真模拟模型获得了代表性场景下的学习样本，并进行学习训练得到了渗滤液水位与观测数据的非线性映射关系. 对上述改进的电法装置和配套的反演算法的有效性进行了验证，并开展了实际填埋场地的案例研究. 结果表明:①基于川字型装置的EConvNet-C反演的电阻率存在明显的分层特征，在场景A中EConvNet-C反演算法的均方误差(MSE)均在0.00230以下，渗滤液水位探测精度均在90.0%以上. ②以“浅层滞水”形式非正常积存情况下，EConvNet-C探测精度略有下降，但MSE仍在0.00420以下，水位探测精度仍在80.0%以上. ③基于传统探测装置的深度神经网络耦合方法得到的渗滤液水位探测精度略低于基于川字型装置的EConvNet-C，而基于传统探测装置的线性反演耦合方法在HDPE膜高阻特性的影响下无法有效识别渗滤液水位. 因此，基于川字型装置的EConvNet-C在填埋场渗滤液水位探测领域具有较大潜力，可为后期开展垃圾填埋场的性能和风险评估提供指导.
|
| 关 键 词： | 电阻率 反演 饱和水位 人工智能 |
| 收稿时间： | 2021-10-31 |
Landfill Leachate Level Detection Based on Improved Electrical Device and Deep Neural Network |
| |
| Affiliation: | 1.School of information an Electronic Engineering, Shandong Technology and Business University, Yantai 264005, China2.State Key Laboratory of Environmental Criteria and Risk Assessment, Research institute of Solid Waste Management, Chinese Research Academy of Environmental Sciences, Beijing 100012, China3.Chinese Academy of Environmental Planning, Beijing 100012, China |
| |
| Abstract: | The identification and control of landfill leachate level are critical components of landfill environmental risk management. The traditional multi-electrode resistivity method device has the advantages of non-destructive, fast and high resolution. However, the high resistance characteristics of HDPE anti-seepage film laid in landfill mask the low resistance characteristics of the leachate. The inversion results of the Least Squares using the traditional linear inversion method have problems such as poor detection accuracy and limited detection range, and inability to accurately determine the leachate level. Therefore, an improved ‘Chuan’ device is proposed to collect data, and a nonlinear inversion algorithm based on depth neural network EConvNet-C is proposed. By constructing a high-simulation physical model, the learning samples of representative scenes are obtained, and the nonlinear mapping relationship between leachate level and observation data is obtained through training and learning. The effectiveness of the improved electrical method device and the matching inversion algorithm is verified, and a case study of an actual landfill site is carried out. The results show that: (1) The resistivity of EConvNet-C inversion based on ‘Chuan’ device has obvious layered characteristics. In scenario A, the MSE of EConvNet-C inversion algorithm is less than 0.002,30, and the detection accuracy of leachate water level is more than 90.0%. (2) In the case of abnormal accumulation in the form of ‘shallow stagnant water’, the detection accuracy of this method decreases slightly, but the MSE is still below 0.004,20 and the water level detection accuracy is still above 80.0%. (3) The detection accuracy of leachate water level based on the coupling method of traditional detection device and depth neural network is slightly lower than that of this method, while the method based on the coupling of traditional detection device and linear inversion can not effectively identify the leachate water level under the influence of the high resistance characteristics of HDPE membrane. Therefore, the EConvNet-C based on ‘chuan’ device has great potential in the field of landfill leachate level detection, which can provide guidance for the performance and risk assessment of landfill in the later stage. |
| |
| Keywords: | |
|
| 点击此处可从《环境科学研究》浏览原始摘要信息 |
|
点击此处可从《环境科学研究》下载全文 |
|
