【作者】 田园；

【导师】 杨承恩；

【作者基本信息】 大连海事大学 ， 控制理论与控制工程， 2004， 硕士

【摘要】 本文对自适应神经模糊推理系统(ANFIS)在非线性船舶舵阻横摇中的应用进行了仿真研究。ANFIS设计方法是一种将模糊逻辑系统(FLS)和人工神经网络系统(ANN)相结合，利用两者各自的优点所形成的混合智能系统。用神经网络的连接结构表述模糊逻辑系统，通过ANN的学习功能使这些FLS规则中出现的许多参数的优化和自适应得以实现。 自适应神经模糊系统最大的特点就是，该系统是基于数据的建模方法。自适应神经模糊系统中的模糊隶属度函数及模糊规则是通过对大量已知数据的学习得到的，而不是基于经验或是直觉给定的，所以尤其适用于缺乏专家经验知识的一类复杂的工业控制过程问题。本设计主要通过下面的工作来完成的： 首先在M文件中编写出船舶的非线性模型，并通过龙格库塔法将其求解出来，将海浪干扰编写成函数，调入到船舶模型中，将得到的整个船舶模型先设计出PID控制器，取出ANFIS所要的训练数据，为后面的ANFIS训练作准备，同时可以得出舵阻摇PID控制的仿真曲线，以及减摇效果在30％左右。 然后利用MATLAB中的ANFIS工具箱，训练初始模糊推理系统，即在训练之前先设计一个sugeno型的模糊推理系统，作为初始模糊推理系统，用训练数据加以训练，可以看出隶属函数通过训练发生了变化，将训练后的模糊推理系统保存，调入到所要控制的船舶模型中，得出舵阻摇的控制仿真曲线，其减摇效果在25％左右。为了加深对ANFIS的理解，同时设计了基于BP网络的PID控制器，进一步理解了神经网络的自学习能力，以及神经网络所具有的任意非线性表达能力，通过神经网络的自学习来实现具有最佳组合的PID控制。更多还原

【Abstract】 Adaptive neural-fuzzy inference (ANFIS) and its application on the nonlinear ship rudder damping system are studied in this thesis. The ANFIS design method is a blend intelligent system which combines the Fuzzy Logic System(FLS) and the Annual Neural Network (ANN) and uses their’s strongpoints. It expresses the FLS by the connection structure of ANN , meanwhile realize the parameter optimization of the FLS rules by the self-learning function of ANN.The biggest point of ANFIS is that it is a modeling method based on data. Through a great deal of datas not experience or instinct can get the Fuzzy membership and rules of ANFIS, so ANFIS especially fits the complex industry control processes without specialist knowledge. This design mainly finish through the below work:First the non-linear ship model is programmed in M file and uncoiled by Runge-Kutta. The wave disturbance function is programmed to be used to ship model. In order to get the training data of ANFIS,PID control is adopted to the ship model which prepare for the ANFIS training, meanwhile get the simulation result. The roll reduction can reach about 30%.Second the ANFIS toolbox is used in matlab, training the preliminary Fuzzy Logic System. Before the training a sugeno Fuzzy system is designed as the preliminary FLS which use the training data to train. It can be seen that the Fuzzy membership have changed. The FLS is saved after training and used to the ship model. The roll reduction is about 25%.In order to understand the ANFIS further , a PID control based on BP network is designed. The scale coefficient of PID is adjust by the self-learning of network, resolve the abuse which it need test repeatly to get the three PID parameter. It can be proved there is good effect.更多还原