Research On Multi-Criteria Decision-Making Model And Application For The Oilfield Development Area

The paper aiming at the problem of multi-criteria, multi-objective decision-making problems in the oil field development area, studying multi-criteria decision-making theory, models and learning algorithm in artificial intelligence and decision support field .Taking fuzzy logic, neural network theory into the multi-criteria decision-making rules, in order to study and apply the typical decision-making problems involved in the process of oil field development.Aiming at the typical decision-making problem of reservoir damage diagnoses, oil and water layers comprehensive discriminant, and the improvements program optimization of inefficient wells. Researching intelligent decision-making models and methods based on the combination of expert knowledge, fuzzy decision-making rules and neural network. Establish the regular fuzzy neural networks, fuzzy weighted reasoning network and fuzzy comprehensive evaluation model, as well as construction of the corresponding learning algorithm and taking each model into application.The application result shows the effectiveness of the proposed model and method.Aiming at multi-objective decision making problems of oil field development plan for the optimal, an improved BP neural network model and learning algorithm are presented in this paper, the feedback signal and the bias units are introduced into BP neural network. In order to introduce the empirical knowledge in the learning process, as well as improving the learning speed and the reliability of decision-making.Using fuzzy optimization theory model to replace the sigmoid function of neural network.Making the neural network activation function with a clearer, intuitive physical meaning.Taking into account the actual system variables have different roles and information transformation mechanism as well as the succession of system state in different stage, a cascade process neural network with some sub-networks is presented to establish the dynamic forecast model of actual system. At the same time, the theory of phase space reconstruction is applied to constructing sample set so as to make up for the lack of training sample data and improve the utilization of actual time series sample data. The information processing mechanisms and learning algorithm of prediction model are also proposed in this paper.Taking indicator forecasting of tertiary oil recovery process for example, the experimental result shows the effectiveness of the proposed model and method.