椭球定界算法在混合建模中的应用研究

Research of Ellipsoid Bounded Algorithm in Hybrid Modeling

并行结构混合建模主要由机理模型与误差补偿模型组成.一般地,误差补偿模型不宜过于复杂,且模型应具有校正功能,以免精度随时间不断下降.针对这个问题,本文选择单层神经网络作为误差补偿模型,并将椭球定界算法应用于单层神经网络的参数更新,不仅能够保证建模误差稳定有界,同时能够提高网络的收敛速度.将提出的方法应用于氧化铝生产过程,改进了原有的苛性碱和氧化铝组分浓度软测量方法.实验研究结果表明,椭球定界算法的应用提高了模型的精度和网络的收敛速度.除此之外,在存在噪声干扰下,改进的方法比原有方法更稳定,进一步证明了方法的有效性和优越性.

A parallel structure hybrid model is mainly composed of a mechanism model and an error compensation model. In general, the error compensation model should be simple and have correction functions to prevent the accuracy from varying with time. Aiming at this problem, a single layer neural network with an ellipsoid bounded algorithm is proposed, and it is taken to enhance the convergence speed and ensure the boundedness of modeling error. The proposed method is applied to an alumina production process to improve the soft sensing method of caustic soda and alumina component concentration. The experimental results show that the ellipsoid bound algorithm can improve the model accuracy and the speed of network convergence. In addition, the improved method is more stable than the original method the presence of noise.