Coal reservoir in Shanxi is influenced by sedimentary environment, diagenesis and structure. The reservoir is characterized by heterogeneity and nonlinearity distribution. An intelligence evaluation model, based on modified particle swarm optimization (MPSO) algorithm with Elman neural network (ENN), is constructed. The model is corresponding to reservoir physical property by the use of logging, core and test data, and it is employed to classify rock types of coal reservoir. The case results indicate that this model is feasible and effective in the rock classification, so it provides an effective method for identification of lithology and mineral rock classification.
The rock types of coal reservoir in core research area are classified by analyzing the characteristics of rock core and logging parameters. Lithologic recognition patterns are established by mathematic methods, based on the parameters of representational samples in various rocks and the basic relationship between lithology and logging parameters.
There are two main traditional model recognition methods: linear equation method and statistics classification method. The former requires highly reliable logging curves. The latter is based on probability statistics and can achieve accurate classification through anterior probability and posterior probability. However, anterior probability may be seldom known. For the extremely complexity of sedimentary environment and physicochemical condition practically, there are not obvious differences among the logging response curves of the different lithology formations (Li and Zhao, 2005). This makes it very difficult to describe the formation of heterogeneity and real properties by adopting the linear logging response equations and the empirical modes. Therefore, how to effectively improve the recognition accuracy of rock classification has been a significant research subject of geologic analyses of logging data.
Recently, the intelligent techniques, such as particle swarm optimization (PSO), genetic algorithm (GA) and ant colony algorithm (ACA), have developed rapidly. They have achieved important progresses in parameter optimization, fuzzy simulation and problem-solving of acceptableness observation data, etc (Kennedy et al., 2010).
