基于金属氧化物气敏传感器的空气绝缘电力设备放电分解特征组分检测方法研究

Discharge faults in air insulated power equipment happen frequently with severe consequences, and seriously threaten the security of the power system. The characteristic components of discharge decomposition can directly reflect the state of discharge faults and thus urgently need a reliable detection method. Metal oxide gas sensors are widely applied in the field of gas detection, but with difficulty in the selection of gas-sensing materials, high working temperature and poor identification of mixed gases. In order to detect the characteristic components of discharge decomposition with metal oxide gas sensors, the research will be carried out as follows. (1) A method , combining first principle calculation and in situ measurement, is proposed to select suitable metal oxide materials for detecting the characteristic components of discharge decomposition; (2) microstructure regulation and modification with graphene-like materials are adopted to reduce the optimum working temperature of metal oxide and improve the sensing properties; (3) a virtual sensor array is constructed with dynamic heating mode, and an intelligent pattern recognition method based on deep convolution neural network is designed to achieve reliable detection of the characteristic components of discharge decomposition. This project will provide a reliable method to detect the characteristic components of discharge decomposition in air insulated power equipment, and lay a foundation both in theory and method for online monitoring of discharge faults.

空气绝缘电力设备放电故障频发且危害大，严重威胁电力系统安全。放电分解特征组分可以直接反映设备放电故障情况，因此亟需一套可靠的检测空气绝缘设备放电分解特征组分的方法。金属氧化物气敏传感器广泛应用于气体检测领域，但仍存在以下问题：气敏材料筛选困难、工作温度高和混合气体识别能力差。为了应用金属氧化物气敏传感器开展放电分解物检测，本项目拟从以下几个方面开展研究：（1）采用第一性原理计算和原位测量相结合的方法，筛选可用于放电分解特征组分检测的金属氧化物气敏材料；（2）通过微观结构调控和类石墨烯材料修饰，降低金属氧化物的最佳工作温度，提高其气敏性能；（3）采用动态加热模式构筑虚拟传感器阵列，设计深度卷积神经网络智能模式识别方法，实现对放电分解特征组分的可靠检测。基于以上研究，本项目将提出一套可靠的空气绝缘电力设备放电分解特征组分检测方法，为放电故障在线监测奠定理论和方法基础。

空气绝缘电力设备放电故障频发且危害大，严重威胁电力系统安全。放电分解特征组分可以直接反映设备放电故障情况，亟需一套可靠的检测空气绝缘设备放电分解特征组分的方法。首先，本项目构建了特定晶面暴露的碲烯、磷化锗、以及二硫化钼二维材料模型，结合第一性原理计算与原位实验，分析揭示了气敏材料与不同空气放电特征分解物的作用机制。之后，合成了碲烯与Cu2Se二维材料，实现了对空气放电特征分解物NO2和CO的高灵敏度检测，阐明了分级结构调控对CNTs-MoS2传感器性能的优化作用，提出了基于多指标综合评估的气敏材料筛选方法。然后，采用脉冲温度调制的方式构建出虚拟传感器阵列，扩容传感器阵列的物理规模，改善了传感器的线性度。最后，采用深度神经网络提取可迁移的特征，构建了基于自注意力机制的气体组分浓度识别模型，实现了对不同类型、不同功率下的空气放电故障检测，证明了本项目的研究成果有着巨大的应用潜力。