基于WM与PSO的模糊分类器优化研究

发布时间：2018-10-18 08:43

【摘要】：从样本中提取规则进而进行构建模糊分类器是一种有效的建模方式。Wang-Mendel(WM)方法根据模糊数学理论方法从数据中直接提取模糊规则。WM方法具有简单、高效实用的特点。但是,处理过程中,该算法易提取出低效的模糊规则。因此对于WM方法提取后的模糊规则库需要进一步的优化整合处理。粒子群优化算法(Particle Swarm Optimization,PSO)是一种基于迭代的进化计算方法。PSO算法在模糊分类器领域的应用主要是对模糊知识库进行融合,从而将原有模糊规则库的结构整合成另一种拥有最好执行效率的组合。但是算法收敛速度过慢。本文采用基于PSO算法的优化算法智能单粒子优化算法(intelligence single particle optimizer,ISPO)来对规则库进行优化处理,ISPO算法相比PSO算法具有较快的算法收敛效果。通过对WM算法分析发现,虽然算法可以实现高效的规则提取,但是由于冲突机制中的设置使得规则缺少样本关联度,从而导致规则库的分类精度受到影响。为避免这一现象造成的影响,采用ISPO算法对规则库进行进一步的优化,在适应度函数中利用样本关联度对规则进行进一步的修改。从而加快算法的收敛速度。而且适应度函数与分类精度呈正相关关系,保证了规则库的高准确率。反向进行思考,ISPO算法有一大缺陷就是算法初始的种群为随机生成,这样降低算法的收敛速度,而本方法初始化的粒子是具有一定准确度的规则库,进而加快了寻优的速度。根据以上分析,本文提出一种基于WM与ISPO的模糊分类器WPFS算法对WM算法和ISPO算法更进一步分析发现,两种算法都拥有较高并行能力,因此,将算法进行并行化重构,设计出基于WM与ISPO的并行模糊分类器P-WPFS算法。为验证并行分类器模型有效性,将并行模型与MapReduce模型结合形成WPFS-MR模糊分类器模型,应用于大规模数据集分类问题。WPFS-MR模糊分类器模型较大提高了算法的处理效率,使得算法在可接受时间范围内给出分类结果。并且同时可以保证分类结果的准确度保持较高的精度。解决了面对大数据分类难题,模糊分类器效率低下的问题。
[Abstract]:Extracting rules from samples and constructing fuzzy classifier is an effective modeling method. Wang-Mendel (WM) method extracts fuzzy rules directly from data according to fuzzy mathematics theory. WM method is simple, efficient and practical. However, in the process of processing, the algorithm is easy to extract inefficient fuzzy rules. Therefore, the fuzzy rule base extracted by WM method needs further optimization and integration. Particle Swarm Optimization (Particle Swarm Optimization,PSO) is an iterative evolutionary algorithm. The application of PSO algorithm in fuzzy classifier is to fuse the fuzzy knowledge base. Thus, the structure of the original fuzzy rule base is integrated into another combination with the best execution efficiency. But the convergence rate of the algorithm is too slow. In this paper, the intelligent single particle optimization algorithm (intelligence single particle optimizer,ISPO) based on PSO algorithm is used to optimize the rule base. Compared with PSO algorithm, ISPO algorithm has faster convergence effect. Through the analysis of the WM algorithm, it is found that although the algorithm can achieve efficient rule extraction, the rules lack of sample correlation because of the conflict mechanism, which results in the impact of the classification accuracy of the rule base. In order to avoid the influence caused by this phenomenon, the rule base is further optimized by ISPO algorithm, and the rule is further modified by sample correlation degree in fitness function. In order to speed up the convergence of the algorithm. Furthermore, the fitness function is positively related to the classification accuracy, which ensures the high accuracy of the rule base. On the contrary, the ISPO algorithm has a big defect that the initial population of the algorithm is randomly generated, which reduces the convergence speed of the algorithm, while the particle initialized by this method is a rule base with certain accuracy, thus speeding up the speed of optimization. Based on the above analysis, a fuzzy classifier WPFS algorithm based on WM and ISPO is proposed to further analyze the WM algorithm and ISPO algorithm. It is found that both algorithms have high parallelism ability, so the algorithm is parallelized and reconstructed. A parallel fuzzy classifier P-WPFS algorithm based on WM and ISPO is designed. In order to verify the validity of the parallel classifier model, the parallel model and MapReduce model are combined to form the WPFS-MR fuzzy classifier model, which is applied to the large-scale data set classification problem. The WPFS-MR fuzzy classifier model greatly improves the processing efficiency of the algorithm. The classification results are given in the acceptable time range. At the same time, the accuracy of the classification results can be guaranteed to maintain a high accuracy. It solves the problem of low efficiency of fuzzy classifier in the face of big data classification problem.
【学位授予单位】：华侨大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【参考文献】