中国电机工程学报
中國電機工程學報
중국전궤공정학보
ZHONGGUO DIANJI GONGCHENG XUEBAO
2012年
2期
39-44
,共6页
燃烧优化%最优操作变量%决策模型%模糊规则%数据挖掘
燃燒優化%最優操作變量%決策模型%模糊規則%數據挖掘
연소우화%최우조작변량%결책모형%모호규칙%수거알굴
combustion optimization%optimal manipulatedvariables%decision-model%fuzzy rules%data-mining
基于锅炉燃烧模型的非线性寻优和基于历史运行工况的数据挖掘是两种常见的锅炉燃烧优化技术,且各有利弊。前者可得到全局最优解,但算法复杂度较高;后者计算较为简易,但只能实现局部最优。结合两种方案的优点,提出基于离线非线性寻优所得最优知识库,采用模糊关联规则挖掘算法,建立最优操作变量(manipulated variables,MVs)决策模型,实现高效、稳定的锅炉燃烧优化。关联规则挖掘中,提出基于缸均值聚类的语言变量非均等模糊分割,以提高所得规则库的可信度;并基于改进的支持度和置信度概念实现规则库的精简。仿真结果表明,基于该文最优MVs决策模型的锅炉燃烧优化结果与全局寻优结果接近,且算法复杂度低、稳定性高,适合于在线实时优化与自适应更新。
基于鍋爐燃燒模型的非線性尋優和基于歷史運行工況的數據挖掘是兩種常見的鍋爐燃燒優化技術,且各有利弊。前者可得到全跼最優解,但算法複雜度較高;後者計算較為簡易,但隻能實現跼部最優。結閤兩種方案的優點,提齣基于離線非線性尋優所得最優知識庫,採用模糊關聯規則挖掘算法,建立最優操作變量(manipulated variables,MVs)決策模型,實現高效、穩定的鍋爐燃燒優化。關聯規則挖掘中,提齣基于缸均值聚類的語言變量非均等模糊分割,以提高所得規則庫的可信度;併基于改進的支持度和置信度概唸實現規則庫的精簡。倣真結果錶明,基于該文最優MVs決策模型的鍋爐燃燒優化結果與全跼尋優結果接近,且算法複雜度低、穩定性高,適閤于在線實時優化與自適應更新。
기우과로연소모형적비선성심우화기우역사운행공황적수거알굴시량충상견적과로연소우화기술,차각유리폐。전자가득도전국최우해,단산법복잡도교고;후자계산교위간역,단지능실현국부최우。결합량충방안적우점,제출기우리선비선성심우소득최우지식고,채용모호관련규칙알굴산법,건립최우조작변량(manipulated variables,MVs)결책모형,실현고효、은정적과로연소우화。관련규칙알굴중,제출기우항균치취류적어언변량비균등모호분할,이제고소득규칙고적가신도;병기우개진적지지도화치신도개념실현규칙고적정간。방진결과표명,기우해문최우MVs결책모형적과로연소우화결과여전국심우결과접근,차산법복잡도저、은정성고,괄합우재선실시우화여자괄응경신。
There are two typical boiler combustion optimization techniques. One is global searching with evolutionary algorithm based on the boiler combustion model and the other is employing data-mining technology to the historical operating data. Theoretically, the global optimum can be achieved with the first method; however, the relevant computation is very complex. The computation of the second technology is much simpler but only with local optimum. A new boiler combustion optimization proposal was presented by combining the different advantages of the existing two methods It was based on the optimal manipulated variables (MVs) decision-model, which was established by employing fuzzy-association-rule-mining method to the optimal knowledge base; and the optimal knowledge base was the results of global searching. In the process of mining association rules, fuzzy language variables were determined by unequal partition based on the k-mean clustering algorithm, so as to improve the confidence of the achieved fuzzy rules. Modified definitions of support and confidence were adopted for the rules reduction. The final numerical experiment indicates that the optimizing results based on the optimal MVs decision-model built as the algorithm in this paper is close to the global optimum, and moreover, the calculation of the new optimization technology is much less and more stable than the evolutionary algorithm; thus it is more suitable for online use and easy for timely updating when the system feature is time- varying.