基于RBF神经网络的热工过程在线自适应建模算法研究

A STUDY ON RBF NEURAL NETWORK BASED ONLINE ALGORITHM MODELS ADAPTIVE TO THERMAL PROCESSES

传统的各种神经网络建模方法由于自身的局限性不能很好地应用于复杂的热工过程建模。该文提出了一种新型的基于RBF网络的热工过程在线自适应建模算法:近似相关性网络(ACN)建模和阶层补偿式网络结构(HCN)建模。文中与资源分配网络(RAN)进行了详细的算例比较,并进一步计算了实际的热工非线性模型。计算结果表明:该文提出的建模算法不仅能提高模型的输出精度,而且也可有效地减小网络的规模,较好地解决了神经网络超界空间的自适应构造问题,为热工过程全局非线性模型的建立提供了一个新的解决方法。

Due to their inherent imperfections, the traditional neural networks cannot be well used in complicated thermal process modeling. Therefore, this paper proposes a new adaptive RBF network algorithm composed of Approximate Correlation Network (ACN) and Hierarchies Compensatory Networks (HCN).In addition, the paper compares the new algorithm with the Resource-Allocating Network (RAN) by a typical calculation example in details, and finally applies the new algorithm for a nonlinear real thermal process modeling. The computational results show that the new algorithm not only improves precision of the outputs but also reduces the network size evidently, that leads the neural network adaptive algorithm to a new appropriate range beyond the boundary of the primary space. It also shows itself as an effective approach for whole nonlinear modeling for thermal processes by the proposed algorithm.