【作者】 段群杰；

【导师】 张铭钧；

【作者基本信息】 哈尔滨工程大学 ， 机械设计及理论， 2007， 博士

【摘要】 21世纪是人类研究、开发及和平利用海洋的世纪,随着人类对海洋开发利用的不断增加,能够探测水下环境并且自主完成作业任务的水下机器人受到国内外研究机构的广泛重视。作为在复杂海洋环境下工作的载体,自主性及安全性是水下机器人的重要特征,其中自主能力的含义是具有和外部环境进行交互的能力,这种交互能力的一个重要方面就是具有自主运动规划与实时控制的能力,能够在复杂的海洋环境下完成复杂的使命。智能控制技术是保障其自主性和安全性的重要基础和核心技术。水下机器人智能控制的内涵包括自主规划、控制与状态监控。研究水下机器人自主任务规划、智能运动控制、传感器信息融合以及自主监控技术,对于提高水下机器人的智能化水平和加快实用化进程具有重要的理论研究意义和实用价值。水下机器人运动规划一般以全局规划为指导,利用在线得到的局部环境信息,避开出现的未知的障碍物的过程。由于水下机器人工作在未知海洋环境下,实际上难以得到完全准确的全局环境信息,且水下机器人具有非线性、强耦合性和时变性的特点,因此强调环境知识完备性的全局规划方法并不适合于水下机器人系统。本文从水下机器人的自适应性和实时性出发,不注重环境知识的完备性,并解决全局规划的离线性和实时局部规划不能对路径进行优化的缺陷,提出基于全局规划目标、实时局部避障行为并结合滚动规划的方法,使用模糊控制策略来展开未知环境下水下机器人实时路径规划方法的研究。通过机器人上传感器系统的配置方式和对障碍物的探测方法,根据每个采样时刻规划窗口内机器人上传感器对周围环境的感知,对水下机器人工作的周围环境进行了类别划分,按照每种类别情况专家的操纵经验建立了模糊规则库,设计了基于模糊控制的实时滚动路径规划器。针对实时局部路径规划问题,提出了基于拟人行为的实时滚动路径规划算法,进行了静动态障碍物环境下的计算机仿真与模拟实验研究,实验结果验证了所提出的实时滚动路径规划算法的可行性。针对凹形障碍物区域内水下机器人的行走问题,提出了实虚目标转换方法,仿真和实验结果表明该方法对基于实时传感器信息的水下机器人在凹形障碍物区域内行走时的振荡和徘徊问题的有效性和可行性。研究了动态障碍物环境中的水下机器人路径规划问题。给出了对动态障碍物的处理方法—采用大范围内的随机运动和采样周期内的匀速运动相结合,用极大似然估计法对动态障碍物的位置不确定性进行估计。针对动态障碍物的避碰问题提出了主动预测避碰方法,并在动态障碍物环境下进行了路径规划的计算机仿真与实验研究。仿真与实验结果验证了对动态障碍物避碰的主动预测避碰方法的有效性。研究了基于模糊神经网络的模糊控制规则的化简及相关参数的学习方法。以模糊自适应学习控制网络(FALCON)为基础,采用分步式学习方法对网络结构和参数进行调整,用模糊C均值聚类方法对输入输出变量的隶属函数初始化,通过最大权值矩阵方法进行了模糊规则的提取和化简,使用遗传算法对隶属函数的参数进行了调整。提出根据输入、输出数据的相关特性,基于密度的减法聚类法和自适应神经模糊网络设计了一种水下机器人自适应神经模糊运动规划器,采用混合参数学习算法对模糊系统的初始结构和参数进行了优化。并将自适应神经模糊规划器与模糊规划器在相同仿真条件下的规划效果进行了比较,结果表明自适应神经模糊规划器的规划效果好于模糊规划器。更多还原

【Abstract】 In the 21st century, the ocean will be investigated, developed and utilized peacefully. With the development of the sea, more and more international and domestic researchers apply themselves to the development of autonomous underwater vehicle (AUV) which can explore the underwater circumstance and accomplish missions. As a vehicle which works in complicated oceanic environments, automation and safety are the main characters of AUV. And the mean of autonomous ability refers that underwater vehicle is of the interactive ability with external environment. The important aspect of interactive ability is of ability of autonomous movement plan and the real-time control, and can complete the complex mission under the complex ocean environment. Intelligent control is the key technology to keep AUV autonomous and safe. Intelligent control includes autonomous mission planning, control and status monitoring. It’s important to research autonomous mission planning, motion control and status monitoring of AUV for improving AUV’s intelligence and widening the application.The motion planning of AUV is mostly under the direction of global planning, making use of the online local information to avoid the unknown obstacles. However, because of the complicated and varied circumstances, the AUV not only has the characters of essential nonlinearity, tight coupling and frequent changeability, but also has been affected by many undetermined factors. Starting from the adaptation and real-time of AUV, ignoring the integrality of the environment knowledge, and solving the disadvantage which the off-linearity of overall planning and real-time partial planning can’t optimize the path. A real-time path planning method of AUV based on the combine of behavior and rolling planning is presented. The method of real-time path planning for underwater vehicle in the uncertain environment is researched by fuzzy control strategy. The real-time path planning method is researched for underwater vehicle in the static uncertain environment. Through the allocation of the AUV’s sensor system and the detection of the obstacles, depending on the apperceiving of the ambience at every sampling time, the compartmentalization of the kinds of ambience is established. And the real-time rolling path planning instrument basing on fuzzy control is designed, and built the fuzzy rule library based on expert experiences.The anthropopathic real-time rolling path planning algorithm is presented aiming at the local path planning. Computer simulation and experiment are performed in the environment existing dynamic and static obstacles. And experiment results validated the feasibility of the real-time rolling path planning algorithm.When the AUV is walking under the environment that existing concave obstacles, the method of switching between the real target and virtual target is presented. The computer simulation and experiment results show that the method is effective and feasible for vibration and hover problem when AUV walking in concave obstacles district based on the real-time sensor information.AUV path planning method in the working circumstances with dynamic obstacles is researched. The processing method of dynamic obstacles—random movement in wide extension and constant speed movement in sampling period combine together. The uncertain location of the dynamic obstacles is estimated by maximum likelihood estimation method. The initiative obstacles avoidance method is presented aiming on collision avoidance problem of the dynamic obstacles. And the computer simulation and experiment research of path planning are carried out under the circumstances with dynamic obstacles. The results of simulation and experiment improved the initiative obstacles avoidance method is feasible and effective.Predigestion of the fuzzy control rule using fuzzy neural network and study method of correlation parameter is researched. Based on the fuzzy adaptive learning control network (FALCON), using distributed learning method to adjust the network architecture and parameter, using the fuzzy C average value cluster method to initiate the membership function of the input and the output variable, extracting and simplifying the fuzzy rules by the maximum value method based on the weight matrix, using the genetic algorithm to adjust the parameter of the membership function. The computer simulation results confirm the feasibility of the research method.According to correlation property of input and output datum, the adapting nerve fuzzy movement planning instrument of AUV is designed based on subtractive-clustering algorithm of density. The initial structure and parameters of fuzzy system are optimized by mixing parameters learning method. The simulation results of adapting nerve fuzzy movement planning instrument and fuzzy planning instrument of the same condition are compared and analyzed. The results show that adapting nerve fuzzy movement planning instrument is better than fuzzy planning instrument.更多还原