融合DID-Transformer与KAN的直流锅炉冷态启动智能控制策略OA
Intelligent Control Strategy for Cold Start-Up of Once-Through Boilers Based on the Fusion of DID-Transformer and KAN
火电机组直流锅炉冷态启动过程具有显著的大迟延、强非线性及多变量耦合特性,使得传统控制策略难以实现精准高效的自动控制.为此,文章提出一种融合深度迭代分解 Transformer(deep iterative decomposition Transformer,DID-Transformer)模型与柯尔莫戈洛夫-阿诺德网络(Kolmogorov-Arnold networks,KAN)的智能控制策略.首先,构建 DID-Transformer 模型,精准预测锅炉关键状态变量的未来一分钟趋势;然后,通过 KAN 拟合状态变量与控制量间的非线性映射关系,并根据预测值推演最优给煤指令,实现锅炉超前调节;最后,基于某 350 MW 机组实际运行数据进行仿真验证.仿真结果表明,相较于融合 Transformer 与多层感知器(multi-layer perceptron,MLP)的基准控制策略,所提策略将主汽温度变化率的闭环跟踪误差降低了 17.0%,同时将给煤指令的累计调节幅度减小了 76.4%,有效提升了直流锅炉冷态启动过程的控制品质与自动化水平.
The cold start-up process of once-through boilers used for thermal power units poses challenges to traditional strategies to achieve precise and efficient automatic control,due to large time delays,strong nonlinearity,and multivariable coupling.To address these challenges,an intelligent control strategy fusing a deep iterative decomposition Transformer(DID-Transformer)and a Kolmogorov-Arnold network(KAN)is proposed.First,the DID-Transformer is constructed to accurately predict the trends of key boiler state variables for the upcoming minute.Subsequently,the KAN is employed to fit the nonlinear mapping relationship between state variables and control inputs;based on the predicted values,an optimal coal feeding command is deduced to allow for anticipatory boiler regulation.Simulation verification was conducted based on real operational data from a 350 MW unit.Simulation results demonstrate that,compared to the baseline control strategy fusing Transformer and multi-layer perceptron(MLP),the proposed strategy reduces closed-loop tracking errors of the main steam temperature change rate by 17.0%and decreases the cumulative amplitude of coal-feeding command adjustments by 76.4%,effectively enhancing the control quality and automation level of the cold start-up process for once-through boilers.
鲁永明;郑智腾;谭祥帅;黄从智
华北电力大学 控制与计算机工程学院,北京 102206华北电力大学 控制与计算机工程学院,北京 102206西安热工研究院有限公司,陕西 西安 710054华北电力大学 控制与计算机工程学院,北京 102206||华北电力大学 新能源电力系统全国重点实验室,北京 102206
信息技术与安全科学
直流锅炉冷态启动智能控制Transformer模型柯尔莫戈洛夫-阿诺德网络
once-through boilercold start-upintelligent controlTransformerKolmogorov-Arnold network
《控制与信息技术》 2026 (2)
38-48,11
中央高校基本科研业务费专项资金资助(2025JC003)
评论