【摘要】 针对一类多输入多输出非线性被控对象,利用前向神经网络逼近原系统的逆系统,将其作为控制器,采用预测滚动优化性能指标训练该神经网络逆控制器,以克服干扰和不确定性影响,实现对多变量非线性对象的解耦控制。对某微型锅炉对象进行了控制算法仿真,结果表明,所提出的控制方法能够克服模型误差的影响,实现稳定解耦控制,且易于实现。在仿真过程中通过实验方法建立该锅炉对象的神经网络预测模型,并注意采用泛化方法采集训练样本数据和训练神经网络,以提高神经网络模型的泛化能力。更多还原

【Abstract】 A multi-layer forward neural network acted as the inverse controller,which was trained with predictive optimization algorithm to compensate for disturbances and uncertain plant nonlinearities.The control scheme can steadily decouple nonlinear-time-varying systems over a wide operating range without priori knowledge of the exact mathematic model.And computation complexities of the classical predictive optimal methods can be rejected.A control system with such a control method is designed for a mini-boiler.Simulation results illustrate the effectiveness of this technique.Generalization measures,including ensuring the quantity and quality of samples during experiment period and training the neural network with Bayes Method,are adopted to improve the model’s generalization ability on modeling the mini-boiler with a neural network.更多还原