HAILFINDER: Bayesian Prediction of Severe Weather

研究人员计划开发HAILFINDER，这是一个基于知识的系统，用于预测科罗拉多州东北部的恶劣天气。直到最近，大多数基于知识的系统都是基于生产规则、框架或各种非标准逻辑的。第四种方法现在正在出现，它使用层次贝叶斯推理和其他相关技术来进行诊断，得出结论，回答问题，并产生预测。HAILFINDER将是第一个为自然科学领域设计的此类系统。其目的是将规范设计的基于知识的系统的性能与基于规则的系统的性能以及专家的性能进行比较。国家海洋和大气管理局（NOAA）将收集和提供广泛的天气资料，并为它将于1992年夏季进行的预报恶劣天气的系统之间的竞争提供专家。HAILFINDER的研究团队由四名调查员组成，他们体现了各种技能，非常适合满足这项任务的需要。该项目有可能利用特定的天气应用产生具有重大科学意义的结果。即使只是部分成功，它也有望为概率专家系统作为实用工具的使用提供更坚实的基础。

The investigators plan to develop HAILFINDER, a knowledge-based system for forecasting severe weather in Northeastern Colorado. Until recently most knowledge-based systems have been based on production rules, frames, or various non-standard logics. A fourth approach is now emerging which uses hierarchical Bayesian inference and other related techniques to perform diagnoses, draw conclusions, answer questions, and produce forecasts. HAILFINDER will be the first such system designed for a natural science domain. Its purpose will be to compare the performance of normatively designed knowledge-based systems with that of rule-based systems and with expert performance. The National Oceanic and Atmospheric Administration (NOAA) will collect and provide extensive weather information and supply the experts for a competition among systems for forecasting severe weather that it will conduct in the summer of 1992. HAILFINDER's research team is comprised of four investigators who embody a variety of skills that are well suited to meeting the needs of this task. The project has the potential of using the specific weather application to yield results of substantial scientific significance. Even if only partially successful, it promises to provide a more solid footing for the use of probabilistic expert systems as practical tools.