Improved Particle Swarm Optimization Algorithm And Its Application To Predicting Petroleum Properties

In the field of petrochemical industry, as the market competition between enterprises is severe and they continuously pursue economic efficiency, it's very worthy to predict petroleum properties accurately based on near infrared data in production. Artificial neural network (ANN) can simulate the structure of human brain and intelligent behavior, and has the property of parallel process, self-organization, self-adaptation and so on. It is an important means of nonlinear modeling. At present, the ANN has been application in prediction for petroleum properties. But due to its inherent imperfections, the precision of prediction and generalization of traditional ANN for petroleum properties need to be improved.Particle Swarm Optimization (PSO) algorithm is an evolutionary computation technique based on swarm intelligence optimization algorithm, which was inspired by social behavior of bird flocking. Because of its strong ability to global search, less parameters and simplicity, its much attention has gained since it is proposed, and a lot of PSO algorithms are developed rapidly. Up till now, these algorithms have been successfully applied in many areas such as function optimization, neutral network training, model identification, fuzzy system control, etc.Aiming at the disadvantage of these algorithms that are easily trapped in the local optimization and convergence speed is slow in the evolution later, a particle swarm collaborative optimization algorithm based on velocity angle (V-PSCO) is proposed in this paper, the adjustment strategy of inertia weight is proposed based on cumulative distribution function of Cauchy distribution. The V-PSCO is used to resolve two function optimization, which are widely used to test of algorithm. Simulation results show that the performance of V-PSCO is effectively improved.In this paper the V-PSCO is introduced into neural network to optimize connecting weighs and threshold values of neural network, the principle and process of optimization is given. The ANN prediction model of petroleum properties based on V-PSCO is established. The experimental result shows that ANN prediction model based on V-PSCO is better than traditional BP algorithm and SPSO in prediction precision and generalization, and it is a effective method for predicting petroleum properties accurately.