Self-organization Membrane Computing Based Attribute Weights Allocation Method And Its Application

Case-Based Reasoning(case-based reasoning, CBR) feature attribute weight distribution is important, whether attribute weights is reasonable or not will directly affect the quality of their problem-solving. Therefore, the research of attribute weight allocation method has been widespread concern. In addition, subjective weighting method given attribute weight currently used mainly rely rights researcher experience and subjective judgment, with a great uncertainty. And objective methods such as genetic algorithms, neural networks and so on, there are also some flaws. These flaws cause weight distribution is difficult to accurately reflect the importance of property. The new branch of natural computing- membrane computing(membrane computing, MC) optimization method weight distribution has opened up a new way. However, since the structure of a single, MC algorithm is difficult to determine the number of basic membrane, and the algorithm time complexity higher, resulting in their lack of ability to learn, therefore, this article from MC and self-organizing principle, the optimization of membrane structure, basic rules and the number of basic membranes were designed and systematic studied, the main contents are as follows:(1) For the allocation of CBR attribute weight, the membrane structure with a simpler single-layer cell is designed. Then, selection, crossover, evolutionary rule are designed to evolve weights objects in basic membrane; at the same time, a two-way communication rule will be used to keep the weight object with largest fitness appeared in the process of evolution and make it participate in the loop iteration, and ensure the diversity and evolution of performance of weights objects.(2) For the number of basic membrane is difficult to determine, building a MC which the basic membrane can build self-organization according to the principle of self-organization. In this method, a portion of the source case will be used to train the performance of MC, when the number of basic membrane is not at the same time. Thus, the weight object with largest fitness is obtained. Then, determine a reasonable number of basic membranes according to MC performance evaluation function. It plays a MC parallel computing capabilities while ensuring the computational efficiency.(3) The stop condition has been improved for low operating efficiency of MC. Saving the weight object with largest fitness and its fitness in each basic membrane, if there is no improvement in the results for two generations, mean that the objects set in this basic membrane had been saturated, or fall into local minimum point. Then, the basic membrane will stop calculating, can reduce the convergence time.(4) Combining with the examples of regression based on data, an experiment platform based on MATLAB-GUI is developed. The contrast of performance is studied in three cases, which is defferet parameter settings, the presence of basic membrane builded self-organization in MC and whether with or without optimizing stop condition. In addition, based on the typical problem of regression forecasts and dissolved oxygen concentration prediction in the sewage treatment process, the proposed method is compared with other weight allocation methods by the contrast experiments. The experimental results show the effectiveness of this method.