过热蒸汽温度控制是火力发电厂自动控制系统中主要的控制任务之一。过热蒸汽温度直接影响火电厂的安全经济运行。而目前在该过程控制中主要采用的控制策略是常规的PID算法，控制效果不理想，因此对其进行研究很有实际意义。 本论文分析了锅炉过热汽温系统的动静态特性，比较了基于PID算法的带导前微分双回路控制方案和串级控制方案的优点与不足；分析了传统预测控制在锅炉过热汽温系统中难以应用的原因，在此基础上提出了一种基于神经网络技术的预测控制算法，采用神经网络作为预测控制的预测模型，充分利用了神经网络对非线性映射很强的逼近能力和预测控制中滚动优化的实际最优控制能力。其中在在线校正环节采用了一种最大误差判断算法，减少了计算量。对神经网络预测控制器的主要参数进行了定量分析，确定了各个参数对系统控制性能的影响。最后，将神经网络预测控制算法用于锅炉过热汽温控制系统中，分别在喷水量扰动、烟气扰动以及负荷扰动下与常规PID控制的控制性能进行了定量的分析比较，通过Matlab仿真验证，取得了满意的效果。 仿真表明本文提出的神经网络预测控制算法在锅炉过热汽温控制系统中是切实可行的，如果能在火电厂工程实践中经过工程验证，将会广泛应用于此类过程控制中。

The superheated steam temperature control is one of the most important tasks in automatic control system in thermal power plants. The superheated steam temperature directly affects the safe and economic running of thermal power plants. While the traditional PID algorithm is mainly used in the process. And its effect is bad. So research of boiler superheated steam temperature control is very significance This article analyzes the static and dynamic characteristics of the boiler superheated steam temperature system. And compares the advantages and disadvantages of double-loop control with a differentiation and cascade control based on the PID . Analyzes the cause of that the traditional predictive control is difficult to apply in the boiler superheated steam temperature system. So proposed a neural network predictive control algorithm that use neural network as prediction model on the basis of the traditional predictive control. It takes full advantage of the capabilities of the neural network in nonlinear mapping approximation and the actual optimal control of receding-horizon in predictive control. And with an online real-time correction capability. In addition, this article using a maximum error judgment algorithm in the link of online correction to reduces the computation. Quantitative analyzes the main parameters of the neural network predictive controller and determine the effect from each parameter to controller. Finally, use the neural network predictive control algorithm into the boiler superheated steam temperature control system , and compare the control performance of the neural network predictive control and traditional PID in attemperating water disturbance, flue gas disturbance and load disturbance. Simulate in Matlab , and achieved satisfactory results Simulation shows that the neural network predictive control is feasible in the superheated steam temperature control system . If it can be tested and verify in engineering practice ,it will be widely used in this kind of process control.