数据融合新技术在识别变压器油中四种特征气体的研究

A NEW DATUM FUSION TECHNIQUE FOR PATTERN RECOGNITION OF FOUR CHARACTERISTIC GASES IN TRANSFORMER OIL

通过将气敏元件阵列技术和遗传神经网络相结合[1] ,来检测电力变压器油中的 4种微量故障特征气体 (H2 、C2 H4 、C2 H2 和CO)。实验结果表明 ,该技术的泛化能力较强 ,但识别精度在某些值处达不到实用的要求。例如 ,针对变压器油中故障特征气体的临界值在电力变压器早期故障诊断中的重要性 ,遗传神经网络数据融合技术需对混合气体临界值的识别精度作进一步提高。并在已有的融合技术基础上提出了一种新技术—分步分档识别法 ,可在大范围内保证识别的准确基金项目 :国家自然科学基金资助项目 (5 0 0 770 16 ) ;教育部博士点基金资助项目 (980 6 982 8)。ProjectSupportedbyNationalNaturalScienceFoundationofChina(5 0 0 770 16 ) .性 ,提高数据融合技术的实用性。该方法既可用于正常环境气氛 。

Through combination of gas sensitive element array and genetic neural network , four characteristic gases with slight amount in electric power transformer oil(H 2,C 2H 4, C 2H 2 and CO) are tested. Experiment result demonstrates that the technique could recognise the pattern accurately in large range, at the same time the recognition precision at some values couldn't satisfy the requirements of applications. Since the threshold density of various failures are very important to the early diagnosis of transformer failures, the recognition precision of the threshold density of the mixed gases should be increased greatly. So an improved method of staged sectioning recognition is proposed in this paper to recognise the mixed gases accurately and enhance the practicability of the datum fusion technique. The new method could recognise the gas pattern not only in normal range but also in failure condition.