不确定过程和仿反馈调整机制的多源信息回转窑烧成状态智能认知模型研究

项目名称： 不确定过程和仿反馈调整机制的多源信息回转窑烧成状态智能认知模型研究

项目编号： No.61305029

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 李帷韬

作者单位： 合肥工业大学

项目金额： 26万元

中文摘要： 回转窑受限于旋转结构的特殊性和工艺设计的局限性，现有的测量手段难以精确检测窑内烧成状态，其烧成状态监控长期处于"人工看火"模式。由于烧成状态认知结果受主观因素的制约，致使熟料质量不稳定和人工劳动强度大，难以满足烧结过程监控和基于熟料质量闭环控制的需求。本项目在已取得研究基础上，研究基于多源信息的烧成状态智能认知模型。首先提取与烧成状态相关的滤波后的火焰图像感兴趣区域色彩和形态特征，融合过程数据特征构建完备的特征空间，引入监督学习机制选取最优滤波器参数以及火焰图像和过程数据特征。仿人设计具有具有不确定过程和仿反馈调整机制的烧成状态智能认知模型，从宏观到微观从全局到局部分层次认知烧成状态，并构建稳定趋势特征实现认知模型的稳定性监控，真正实现"机器看火"取代"人工看火"，为具有不确定过程和仿反馈调整机制的智能认知模型理论研究拓展思路,并为实现烧结过程监控和基于熟料质量闭环控制奠定应用基础理论。

中文关键词： 认知智能；仿反馈；广义认知误差；不确定信息；多层次变粒度

英文摘要： Rotary kilns have special continuous rotation structure and limited technological design. Hence, the existing measuring techniques are difficult to accurately measure the burning state inside the kiln. The burning state monitoring of rotary kiln sintering process still relies on "operators observing fire". Due to the burning state cognition results restricted by some personal elements, it will easily lead to low qualified rate of clinker and high operator's workload, and is difficult to meet the need of the rotary kiln sintering process monitoring and closed-loop control of the clinker quality. On the basis of the obtained research, this project will study the burning state intelligent cognition model via multi-source information. Firstly, for the filtered flame images, the color and configuration features of region of interests closed related to the burning state will be extracted, and then fuse with the process data features to form the complete feature space. The supervised learning mechanism is employed to select the optimal filter parameters and flame image and process data features. The burning state intelligent cognition model with uncertain process and feedback-simulated regulated mechanism will be discussed to hierarchically cognize the burning state from macroscopic to microcosmic and from global to lo

英文关键词： cognitive intelligence；feedback-simulated；generalized cognition error；uncertain information；multi-level and variable granularity