基于迟延估计与Kalman状态跟踪的热工过程动态数据驱动建模

Dynamic Data Driven Modeling for Thermal Processes Based on Delay Estimation and Kalman State Tracking

针对常规热工对象历史数据建模过程中稳态数据获取困难的问题,从数据选取、数据处理和方法设计3方面出发,提出一种基于迟延估计与Kalman状态跟踪的热工过程动态数据驱动建模方法。该方法选取动态历史数据作为建模数据,将数据末端输入值作为输入稳态分量,将纯迟延时间及输出的稳态分量作为寻优变量的2个维度,并依据各量的值对数据进行处理;应用Kalman滤波算法获取系统初始状态后,结合智能寻优算法对系统进行建模仿真,对某火电机组高温过热器惰性区进行建模。结果表明:该建模方法能够直接应用系统的动态历史数据进行模型辨识,且所建模型具有较好的模型精度,为动态历史数据建模提供了参考。

Aiming at the difficulty in obtaining steady-state data for modeling of conventional thermal processes using historical data, a dynamic data driven modeling method was proposed based on delay estimation and Kalman state tracking, in consideration of following three aspects, such as the modeling data selection, modeling data processing and modeling method design, etc. , which takes dynamic history data as the modeling data, end data input as the steady-state input component, as well as pure delay time and steady-state output component as the two dimensions of optimization variables. According to the values mentioned above, the data were processed. Meanwhile, Kalman filtering algorithm was used to obtain the initial state of a system, and combining the optimization algorithm, the system was then modeled, which was subsequently applied to an inert area object of a finishing superheater in a thermal power unit. Results show that the modeling method can directly use the dynamic history data to identify the model parameters, and the model established has high accuracy, which may serve as a reference for further study of dynamic data driven modeling.