Abstract: Considering the characteristic of the quadruple-tanks, such as multiple variables, great time-delay, misalignment and coupling, the author applies the wavelet neural network generalized predictive control（WNNGPC）. Predictive model of the control system can be obtained through recognizing the system controlled object based on the good function approximation of the wavelet neural network. And with the combination of the good control performance of the generalized predictive control, the quadruple-tanks system can achieve stability control. During the reorganization of the system, the author applies the optimal BP neural network. This algorithm can correct weights and thresholds of the network quickly, and make the prediction output approaching the desired output smoothly. On solving the coupled problem of system, the author designs a fuzzy feed forward compensation decoupling by using the universal approximation of fuzzy control. Use the WNNGPC based on fuzzy compensation decoupling to do the experiments of the quadruple-tanks, and analyze the results of experiments. Through the experiments and analysis, it can be indicated that the control strategy achieves good control effect of the quadruple-tanks.

Key words: quadruple-tank, wavelet neural network, generalized predictive control, fuzzy compensation decoupling

摘要： 针对四容水箱系统的多变量、大时滞、非线性及强耦合等特性，采用了小波神经网络广义预测控制（WNNGPC）策略。利用小波神经网络良好的函数逼近能力，对系统被控对象进行辨识，得到控制系统的预测模型，再结合广义预测控制良好的控制性能，达到对四容水箱系统的稳定控制。在系统辨识的过程中，采用的是改进的BP学习算法，这一算法能够快速平稳地修正网络权值和阈值，使预测输出平滑地趋近期望输出。在解决系统的耦合问题上，利用了模糊控制的通用逼近性，设计了模糊前馈补偿解耦。基于模糊补偿解耦的WNNGPC对四容水箱进行控制实验，并对比分析实验结果。结果表明，这一控制策略对四容水箱进行控制时取得了较好的控制效果。

关键词: 四容水箱, 小波神经网络, 广义预测控制, 模糊补偿解耦