Abstract: The environment of hazardous chemicals leakage field is complex and changeable. The identification of leakage source is affected by multiple factors such as wind speed, light intensity and noise. At present, intelligent robots are mostly used for detection and target identification by smell or vision. However, the reliability of single type sensor identification results is poor. This paper proposes a recognition method based on improved Dempster-Shafer（D-S） evidence theory to simulate the human multi-sensory recognition mechanism and comprehensively deal with the olfactory, visual and auditory recognition reliability. By introducing reference evidence to obtain the global information of the evidence source, the evidence conflict problem in the classical D-S algorithm is solved, and an example is verified in MATLAB. The results show that the recognition reliability of olfactory, visual and auditory sensors is improved by 42.5%, 51.2% and 38.4% respectively; for conflict data, it overcomes the defect of fusion failure of classical D-S algorithm and Yager algorithm. Compared with Murphy algorithm, the recognition reliability is increased by 33%, which can realize high reliability identification of leakage source, and the reliability of fusion decision is good.

Key words: leak source identification, humanoid multi-perception, information fusion, D-S evidence theory, evidence conflict

摘要： 危化品泄漏场域环境复杂多变，泄漏源的识别受风速、光照强度、噪声等多重因素影响，当前多以智能机器人探测，利用嗅觉或视觉进行目标识别，但单类传感器识别结果可靠性差，提出一种改进Dempster-Shafer（D-S）证据理论的识别方法，模拟人类多感官识别机制，综合处理嗅觉、视觉和听觉识别信度；通过引入参考证据获取证据源全局信息，解决了经典D-S算法中存在的证据冲突问题，并在MATLAB中进行了算例验证。结果表明，相比嗅觉、视觉和听觉单类传感器识别信度分别提高了42.5%、51.2%和38.4%；对于冲突数据，克服了经典D-S算法、Yager算法融合失效的缺陷，相比Murphy算法识别信度提高了33%，可实现泄漏源的高信度识别，融合决策可靠性好。

关键词: 泄漏源识别, 仿人多感知, 信息融合, D-S证据理论, 证据冲突