中文摘要：

针对模糊规则基难以自动在线设计问题,提出一种利用观测数据直接构造模糊规则基中心的新方法,并将其用于对运动过程的准确预测。该方法首先用较少的有限次状态和误差的观测数据,在不同时刻直接构造状态和误差的模糊规则基中心,然后,根据最大熵原理,求取当前状态的隶属度函数值。为使此方法适合于不同状态的运动跟踪,在确定最大熵原理的关键参数中,不使用经验方法,而利用粒子群算法在规定的进化代数内求取参数,在保证实时性要求的前提下,完成对参数的优化。为提高模糊推理的精确性,在模糊推理中设计加权因子,对下时刻的状态和预测误差进行预测,利用预测误差对预测状态进行修正。预测误差仅由有限次测量数据决定,消除了误差累积现象。分别对两类不同的运动进行对比跟踪实验,结果表明,此方法具有对先验知识的弱依赖性,较强的实时性和较高的预测精度。

英文摘要：

In order to overcome the difficulty in designing fuzzy rule bases automatically,a novel method using observed data to make the centers of fuzzy bases was proposed,and it was used to predict moving process well and truly.Above of all,only a few data of object states and error of limited measurement were used to construct fuzzy bases at different time directly in the method.Then,the membership value of current state was evaluated according to maximum entropy principle（MEP）.In order to make the method fit for different complex movements,Particle Swarm Optimization（PSO） algorithm was adopted to optimize the pivotal parameter of maximum entropy principle in setting generation on the condition of requirement of real time.Moreover,in order to improve the precise of prediction,weighted factor was designed in fuzzy inference processing,and the state and error in next time were calculated according to it,in addition,the state in next time was modified by prediction error.So,prediction error was only determined by limited times measured data,and accumulating error was eliminated.Two different complex movements were used in experiments comparing with method in the other reference,the results show that the method is independent of prior knowledge,and real-time performance and prediction precise are improved.