基于多源信息融合与知识发现的复杂工业过程工况识别方法研究

Intelligent recognition of the operational statuses and accurate prediction of the variation trend of the operational conditions in the complex industrial process play a key role in stabilizing the production indexes, as well as formulating the optimal production operation scheme to reduce the energy and resources consumption in the production process. In view of the problem that information monitoring method based on the traditional single and isolated sensor is difficult to effectively achieve the accurate recognition of the operation conditions in the complex industrial process, combining the mass process information obtained by various sensors in the industrial site with a large amount of on-site experience knowledge, this project introduces the idea of the multi-source information fusion and knowledge discovery into the operational statuses recognition of the mineral flotation process, which is the concrete industrial process focused in this project. The optimal configuration method of distributed visual sensors used for the visualization monitoring of the operational conditions of the overall circuit in the complex industrial process, the on-line processing of the multi-source heterogeneous information and the optimal feature selection method, the distributed multi-source heterogeneous information decision fusion method, the knowledge discovery method of the typical operation conditions as well as the pattern classification of the operational statuses with diverse classes of nonlinear uneven distribution and the unequal misjudgement cost will be investigate deeply so as to form the theory and methodology system aiming at intelligent recognition and accurate prediction the trend of the operation conditions in the real complex industrial processes. The research results of this project will greatly improve the automatic monitoring level of the complex industrial production processes, which can also provide the necessary theoretical basis and key technical support for the efficient, stable and optimal operation of the complex industrial production process.

复杂工业过程运行工况智能识别与工况变化趋势预测对于稳定工艺指标、制定最优生产操作方案以降低生产中的能源和资源消耗，具有关键性作用。基于单一、孤立传感器信息分别处理的监测方法难以有效实现复杂工业过程运行工况准确识别的问题，本项目以矿物浮选过程为具体对象，融合工业现场多种传感器获取的海量过程信息和大量现场经验知识，将多源信息融合和知识发现的思想引入到矿物浮选过程工况识别中。拟研究面向复杂工业过程全流程运行工况可视化监控的分布式视觉传感器最优配置方法、多源异构信息在线处理与最优特征选择方法、分布式多源异构信息决策融合方法、典型运行工况的知识发现和多类非线性不均匀分布与误判代价不均等的运行工况模式分类方法，形成面向复杂工业过程运行工况智能识别与趋势预报的理论和方法体系。研究成果将极大改进复杂工业过程的自动化监测水平，为我国复杂工业生产过程的高效稳定优化运行提供必须的理论依据和关键性的技术保障。

复杂工业过程运行工况状态智能识别与评价以及对工况变化趋势的准确预报，是及时准确掌握过程运行发展趋势、确保指标稳定以及过程优化运行的基础。本项目探讨了视觉传感器信息与传统过程工艺参量信息相融合的复杂工业过程工况状态智能感知与产品质量在线监测方法。在多源信息融合与特征提取中，深入分析了无前景、背景区分、由大量局部同质碎片随机堆积而成的复杂纹理结构构成的过程监测图像信息的空间结构特征描述方法，从理论上证明了该类复杂纹理模式的韦伯分布过程，提出了基于对称韦伯分布和t Location-Scale分布为基础的复杂纹理模式空间结构统计建模方法；基于视觉信息的统计分布特性，实现了视觉信息与传统过程参量的贝叶斯决策融合；同时，进一步研究了基于视觉图像统计分布特性与工艺过程参量相结合的复杂工业过程工况状态识别方法，特别探讨了基于稀疏多核最小二乘支持向量机的多类别非线性分布严重不均、工况状态误判代价不均等的复杂工业过程运行工况状态模式分类方法；并提出基于递归稀疏主成分分析的故障工况状态自动监测与诊断方法，以实现极端、故障工况的智能监测与预警。所提出的方法已在矿物浮选过程和某粮食加工过程进行了初步验证。结果表明，因所提方法拓宽了复杂工业过程运行状态信息智能化感知手段，有效获取了基于统计分布特性的多源信息的本质特征，并充分考虑了实际工业过程工况状态的不均衡分布特性。因而，能有效实现复杂工业过程工况状态的智能化感知，为保障复杂工业过程的安全稳定优化运行提供必须的理论依据和关键性的技术保障。