基于LSTM-GA的过热蒸汽两级减温器协同预测控制  

Cooperative Optimal Predictive Control of Two-stage Superheated Steam Desuperheater Based on LSTM-GA

作　　者：刘明旭 夏飞 牟辰泽 王会永 益剑明 孙钦 LIU Mingxu;XIA Fei;MU Chenze;WANG Huiyong;YI Jianming;SUN Qin(Huaneng Xindian Power Generation Co.,Ltd.,Zibo 255414,Shandong,China)

机构地区：[1]华能辛店发电有限公司,山东淄博255414

出　　处：《工业锅炉》2024年第5期1-7,共7页Industrial Boilers

摘　　要：火力发电机组中,由于过热蒸汽减温控制系统存在高延迟、大惯性、参数多变等特点,使用PID控制器难以取得良好的控制效果,且传统的自动控制策略缺乏一二级减温器的互动手段。针对以上问题,提出了一种基于LSTM-GA的预测控制算法,使用长短期记忆网络(Long Short-Term Memory,LSTM)建立两级过热蒸汽减温系统的预测模型,遗传优化算法(Genetic Algorithm,GA)求解两级减温水最优控制量,实现两级减温器协同优化预测控制。根据某330 MW亚临界燃煤机组运行数据建立仿真模型,经计算分析,LSTM-GA预测控制器能够提前预测,做出判断,其快速性稳定性准确性均优于传统PID控制,并且实现了两级减温器的联动控制。验证结果表明了此算法的可行性、有效性,为优化过热蒸汽的减温控制提供了一种新方法。In the thermal power generation group,the high delay,large inertia and time-varying parameters of the superheated steam desuperheater control system make it difficult for a PID controller to achieve effective control.Additionally,the traditional automatic control strategy lacks interactive means for primary and secondary attemperators.To address these issues,a predictive control algorithm based on LSTM-GA was proposed.Long Short-Term Memory(LSTM) is used to establish a prediction model of the two-stage superheated steam desuperheating system,while Genetic Algorithm(GA) is employed to solve the optimal control quantity of the two-stage desuperheater and realize cooperative optimization predictive control.Based on operational data from a 330 MW sub-critical coal-fired unit,a simulation model is established.Through calculation and analysis,it is demonstrated that the LSTM-GA prediction controller can predict and make judgments in advance with superior speediness,stability,and accuracy compared to traditional PID control.The validation results demonstrate the feasibility and effectiveness of this algorithm,providing a new approach for optimizing overheated steam desuperheat control in practice.

关 键 词：火力发电机组 过热蒸汽温度 预测控制 LSTM神经网络 遗传优化算法 

分 类 号：TK223.7[动力工程及工程热物理—动力机械及工程]