基于多目标LSSVM回归的火电厂烟气含氧量软测量

Soft-sensing Method Based on Multiobjective LSSVM for Oxygen Content in Flue Gas of a Coal-fired Power Plant

摘要:

为了提高基于LSSVM的软测量模型的可信度,提出将平均拟合误差、平均预测误差与最大预测误差作为模型参数优化的3个目标,并根据两个预测误差目标之间的差值来设置模型参数选择的偏好；基于某电厂600 MW超临界机组运行时采集的数据,对烟气含氧量进行的建模仿真结果表明:根据偏好选择LSSVM的正则化参数γ与核函数宽度σ可以同时兼顾模型的拟合能力与预测能力,并确保模型的最大预测误差小于一定的上限,从而能够提高模型的可信度；在此基础上,对γ与σ值变化的仿真试验进一步验证了综合考虑上述3个目标来进行模型参数优化选择的合理性。

Abstract:

In order to improve the credibility of LSSVM based soft-sensing model, mean value of the fitting error, mean value of the prediction error and the biggest prediction error are suggested as the objectives for model parameters optimization. The preference for the best parameters selection is set according to the difference between the biggest prediction error and the mean value of prediction error. Simulation results of oxygen content in flue gas of a 600 MW supercritical unit in a coal-fired power plant verify the rationality of the preference. By selecting the parameter γ and σ based on the preference, the fitting ability and the predicting ability are considered simultaneously, and the biggest prediction error will not exceed the upper bound, thus the reliability of the model is enhanced. On the basis of this, the further simulation of changing the parameter proves that considering the whole three objectives for model parameters selection is reasonable.