Improvements On CMAC Neural Network With Application In Reservoir Engineering

The lithology, physical property, electrical property and oiliness are the four characteristics in the study of reservoir. Among them, the porosity and permeability of reservoir, described by the physical characteristic, impact deeply on the development program of reservoir, study and research of reservoir, and dynamic management of the field. Commonly, the porosity and permeability can be deduced by the seismic data and logging interpretation. However, the two parameters are not easy to be measured and predicted and relationship between them is also difficult to be determined because of the type of cement. Analysis and prediction of the reservoir geological parameter, based on artificial intelligent technology, is one of the applications of logging interpretation and a new method of reservoir prediction, which can improve the reservoir description and the quantitative description of remaining oil, and increase the accuracy of reservoir forecast.In this paper, the main contents are outlines as follows:1. Fuzzy principal component analysis(FPCA) is designed to pre-process for reservoir data with strong anti-jamming capability. FPCA can not only save information better, but also reduce dimensions effectively to reduce the complexity of the analysis.2. According to the theory of the Cerebellar Model Articulation Controller (CMAC), a new CMAC model is proposed based on global dynamic information by batch learning. The model can retain the local generalization and avoid local minima. The learning rate is determined by the iterative errors in the learning process. Meanwhile, the concept of integral error in the automatic control strategy is introduced to enhance the performance of the new model. The model has a better approximation ability by weighting error dynamically. Through statistical learning from samples, the new model can improve the prediction accuracy of the network and has better convergence and a faster convergence rate. This conclusion can be obtained through both theoretical analysis and experimental proof.3. In the porosity prediction of reservoir, the global dynamic information CMAC neural network is applied to achieve a better result. This model provides a new approach for parameter measurement of the reservoir geology...