基于神经网络误差补偿的预测控制研究（毕业学术论文设计）.doc

基于神经网络误差补偿的预测控制 摘 要 非线性系统预测控制已成为一个有着广阔前景的研究热点。预测控制的多步预测和滚动优化有克服建模误差影响的能力，使控制系统具有一定的鲁棒性。但研究表明,预测控制对模型失配的鲁棒性是有限的，仍需要一个较准确的预测模型。 实际系统大都是难以准确建模的非线性系统，如何提高预测的精确性以增强控制系统鲁棒性是非线性系统预测控制研究的重要问题。建立在反馈校正原理上的预测误差补偿是一种有效的方法。 本文提出了一种基于对象定量和定性信息的组合预测控制方法，但其只是对一类特殊对象而言。 本文对模型失配误差用Bp 网络建模, 构成误差修正DMC 算法, 然而由于经训练的网络权值固定不变, 当工况发生较大变化时, 修正后的预测值仍有可能产生较大误差。 本文利用一种权值可以在线调整的动态Bp 网络对模型预测误差进行拟合并与预测模型一起构成动态组合预测器，在此基础上形成对模型误差具有动态补偿能力的预测控制算法。 由于该算法显著提高了预测精度，从而增强了预测控制算法的鲁棒性。 关键词：预测控制 神经网络 动态矩阵 误差补偿 Error compensation?based on neural network?predictive control Abstract Predictive control of nonlinear systems has become a promising research focus. Predictive control of multi-step prediction and rolling optimization has the ability to overcome the effects of modeling errors, the control system has a certain robustness. But studies show that mismatch the model predictive control robustness is limited, still need a more accurate prediction models.? The actual system are mostly difficult to accurately model nonlinear systems, how to improve forecast accuracy to enhance the robustness of control systems predictive control of nonlinear systems is an important issue.?Based on the principle of feedback on the forecast error correction is an effective method of compensation.? This paper presents an object-based combination of quantitative and qualitative information predictive control, but only in terms of a special class of objects.?In this paper, the model mismatch error with Bp network modeling, error correction form DMC algorithm, but because of trained network weights are fixed when the conditions changed greatly, the revised forecast is still possible to have a greater?error.?This use of a weight you can adjust the dynamic online Bp network model prediction error and prediction models to be combined together constitute the dynamic combination of predictors, on the basis of the formation of the model error has the ability to predict the dynamic compensation control algorithm.?Since the algorit