研究扩展线性化神经状态空间的数学理论基础，基于该模型的面向实时控制且保证收敛的快速训练算法，保证闭环系统稳定的控制器设计方法，考虑噪声和未知扰动时的扩展线性化神经状态空间鲁棒控制器设计，基于以上理论成果的温度控制和水轮机发电控制的实时控制应用研究，形成实用化软件。成果对神经网络控制的理论和实际应用有重要促进。

In this project, modeling and control of nonlinear systems are studied based on neural state space, which is neural networks combined with state space. In system modeling, an extended linearized neural state space model and pseudolinear neural networks model for multi-input and multi-output systems are proposed respectively. The structure of extended linearized neural state space model is similiar to the canonical controllable form. The parameters in this model are the output of neural networks. Extended Kalman Filter is adopted as learning algorithm. The performance of modeling for several typical nonlinear systems are good. Under the guarantee of stability of closed-loop system, one-step-ahead predictive control and pole assignment algorithm are proposed respectively based on neural state space model. A two-tier real-time computer system based on MATLAB environment is developed. Various control algorithms such as neural networks adaptive PID, single-neuon intelligent control, and predictive control algorithm are applied to real-time control experiments. The performance is satisfactory. This research provides new ideas and methods for the nonlinear systems modeling and control. The development of real-time control platform based on MATLAB is meaningful for the application of theoretical results. Moreover, l∞ predictive control based on neurofuzzy networks and B-spline neural networks based output PDFs control, suboptimal mean control, and robust predictive mean control for stochastic systems are studied respectively.