【作者】 刘东；

【导师】 冯全源；

【作者基本信息】 西南交通大学 ， 通信与信息系统， 2013， 博士

【摘要】 粒子群优化(PSO)算法是一种基于种群的随机优化方法。与传统的优化方法相比,PSO算法具有结构简单、参数较少、易于实现以及全局寻优能力强等优点。然而,PSO算法的理论基础还不完善,存在早熟收敛、易陷入局部极值等问题,并且在应用于工程实际问题时存在很多值得改进和提高之处。本文从PSO控制参数策略和实现框架上提出了几种不同的改进算法,并将改进的算法用于PID控制器参数整定以及阵列天线方向图综合。粒子群优化算法是基于种群的群智能算法,种群的收敛特性直接关系到算法的寻优性能。通过分析粒子群优化算法的收敛特性,明确了控制参数策略以及算法平衡搜索能力与算法性能的关联,并进一步对粒子群优化算法进行了改进：(1)提出了基于聚集度反馈控制的粒子群优化算法,使得种群的聚集程度可控。这样在种群陷入局部最优情况下,算法可以重新使种群按一种可控的方式重新发散开,从而改善算法的全局寻优能力。(2)搭建了一个继承学习的算法框架,将单个PSO优化进程纳入到这个框架中。在新的算法框架中,多个并行的PSO进程构成一个循环。上一个循环中各PSO进程结果中较好的一部分,以及随机生成一部分粒子位置,共同构成一个完整的种群位置来作为下一个循环各PSO进程的初始化种群位置。新算法在很大程度上改善了随机性对优化结果的影响,在多维复杂优化问题上具有良好的性能,而且具备很大的灵活性,可以很容易地将此算法框架应用到其他智能优化算法中。在控制系统中,PID参数整定是一个经典的研究方向。如果将PID控制器的3个参数看做是待优化变量,用控制器响应评价函数作为适应值函数,就可以使用优化算法来处理PID参数整定。本文针对这个问题进行了相应研究：(1)建立了磁浮列车控制器参数的PSO优化模型,采用改进的PSO算法对磁浮控制器PID参数进行了优化,仿真结果和实验结果表明基于PSO算法的PID参数优化模型具有较好的可行性和适用性。(2)在实际应用中,有时候无法得到被控对象的准确模型,也无从进行有效的PID参数整定,只能靠经验对PID控制器参数进行整定。为此,将PSO算法用于摊铺机控制系统的模型辨识,并在辨识基础上对摊铺机控制器参数进行了整定,取得了很好的效果。阵列天线方向图综合是智能算法的一个重要应用领域,本文也针对这个问题进行了研究,主要包括连续变量的方向图综合以及离散变量的方向图综合,具体内容如下：(1)针对多目标方向图优化,提出了分阶段适应值函数策略。由于各个指标分阶段提高,使得各个阶段更容易平衡各个优化目标,不会导致某个优化目标已经满足收敛条件而另外的优化目标还差距很远。这样在解空间中,局部极值区域的深度更浅,有利于种群跳出局部极值点,实现全局收敛。在实际方向图优化中,可以分两次或者更多阶段将方向图指标逐步提高至设计指标。(2)将停滞检测PSO算法和基于聚集度反馈控制PSO算法用于均匀间距直线阵的低旁瓣方向图综合中。实验表明两种改进算法能有效生成多零陷,抑制旁瓣电平值。(3)将继承学习粒子群优化算法(ILPSO)用于不等距线阵天线低旁瓣方向图综合,仿真结果表明ILPSO可以使用较少的优化次数取得与国内外最新文献相当或更好的结果。(4)针对阵列天线方向图综合中的离散优化问题,提出了基于实数PSO算法和粒子位置取整相结合的优化策略,有效地处理了4bit数字移相器阵列的方向图综合、稀疏直线阵列的方向图综合和不等距稀疏阵列方向图综合。仿真结果表明此策略可以有效地将实数PSO算法用于方向图综合中的离散优化问题,优化结果优于已有二进制粒子群优化算法和其他智能优化算法。为了有效处理带稀疏比例约束的大型稀疏平面阵列方向图综合,在前面算法基础上进一步加入了概率调整策略,结果优于其他离散粒子群优化算法。最后,对本文的工作进行了总结,并提出了进一步研究工作的方向。更多还原

【Abstract】 Particle Swarm Optimization (PSO) algorithm is a typical swarm intelligence optimization algorithm. Comparing with traditional optimization methods, PSO algorithm has many advantages, such as simple structure, less parameter, easy to implement and strong global optimization capability. However, the theoretical basis of PSO is still far from mature. PSO has problems of premature convergence and easy to fall into local optimum trap, there is still a lot of room for improvement when applying PSO to engineering practice.The paper proposes several distinct improved algorithms, including the parameters control strategy improvement and new implementation framework. These improved PSO versions are appled into PID tunning and array antenna pattern synthesis.PSO is an intelligent optimization technique based on swarm intelligence, its swarm convergence directly relations with the optimization performance. After analyzing the convergence performance of PSO, the paper improves PSO in the aspects of parameter control strategy improvement and enhancing the balanced search capacity of algorithm:(1) The paper proposes an improved PSO on the basis of convergence control, so the swarm convergence becomes controllable. When swarm is converged in a local optimum trap, the algorithm makes swarm diverge again.In this way, the global search ability of the algorithm can be improved a lot.(2) The paper establishes an algorithm framework based on inheritance learning strategy, and single PSO process is incorporates into this framework as a basic unit. Under the new framework, several parallel PSO processes construe one cycle. The better part of the results of last cycle and the randomly generated results composed the whole initial swarm positions of next cycle. New algorithm has greatly reduced the impact of randomness, and has good performance in multidimensional optimization problems. The improved algorithm also has great flexibility, and the algorithm frame is easy to apply in other intelligent optimization algorithms.In the control system, PID parameter tunning is a classical research direction. If3parameters of PID controller are deemed as promising optimization variants, and the controller’s response is used as fitness value, then optimization algorithm can be applied to handle PID parameter tunning. To solve this problem, the paper makes relevant studies:(1) The paper establishes PSO optimization model for Maglev train controller PID tunning and the improved PSO algorithm is used. The simulation and experimental results show that PID tunning by PSO algorithm has good feasibility and applicability.(2) In practical application, sometimes it is impossible to obtain accurate model for the control objects, so it is hard to tune PID parameter and the feasible way is to use the experience tuning method. To improve this situation, PSO algorithm is applied in model identification of paver’s control system. The identification paver model is used to PID tunning and the experimental results show that this method is very effective.Array antenna pattern synthesis is a key application field of intelligent algorithm. The paper also studies this issue, mainly consisting of pattern synthesis with continuous variants and discrete variants. The details are as below:(1) As for multi-objective pattern optimization, the paper proposes the strategy of stage fitness function. For new strategy, different indicators increase by stages, and it is much easier to balance each optimization objective in different stage. Compared with fixed fitness function strategy, new strategy can avoid a certain indicators being satisfied convergence conditions while the others far from the convergence conditions. As a result, local optimum trap’s depth becomes shallower in the solution space, which is helpful for the PSO swarm to escape the local optimum trap and can help PSO swarm realize global convergence. In pattern optimization applications, the fitness function can be divided into two or more stages.(2) The paper applies stagnation detecting PSO algorithm and PSO algorithm based on convergence control into uniform linear array’s low sidelobe pattern synthesis. The simulation results show that these two improved algorithms can effectively applicate in multi-null and low sidelobe pattern synthesis.(3) The paper applies ILPSO into unequal spaced linear array’s low sidelobe pattern synthesis. Simulation results show that ILPSO obtains results equal to or better than the results of the latest domestic and foreign researches by using less computation.(4) In order to solve the discrete optimization problems in antenna array pattern synthesis, the paper proposes an optimization strategy based on the combination of PSO algorithm and particle position rounding. Therefore, it effectively handles the pattern synthesis of4bit digital phase shifter array, the pattern synthesis of thinned linear array, and the pattern synthesis of non-uniform thinning linear array. Simulation results show that this strategy can effectively apply real number PSO algorithm into the discrete optimization in pattern synthesis, and the optimization results are better than existed binary PSO and other intelligent optimization algorithms. In order to handle the pattern synthesis of large thinned plannar array with thinned element proportion constraint, the probability adjustment strategy is added, and the optimization results are superior to results of other discrete particle swarm optimization algorithms.In the end, the paper makes final conclusion and proposes further research directions.更多还原