Research On Pump-jack Fault Diagnosis Method Based On Particle Swarm Optimization

Oil pump-jack is one of important equipment in oil field system, whose quantity is over seventy percents of the whole quantity of the equipment of oil extraction in oil field. Once the failure of pump-jack happened, no timely diagnosis, which will result in the waste of energy, even caused serious accident. So how to timely and accurate understanding of the rod pumping system in working condition and development of oil wells pumping and the intelligent fault diagnosis system is very important for watching a variety of state parameters of Oil well pump-jack in real-time, detecting faults and timely warning, to improve the efficiency of pumping wells, and reduce costs and increase mechanical oil production wells, difficulties. This paper adopted an improved particle swarm optimization algorithm (VCPSO) to train the weights and thresholds of neural network. Then the neural network trained was used to identify the faults of the failure of pump-jacks.In recent years, the artificial neural network that has a powerful learning and parallel processing capabilities which is exploit a new research approach to diagnose equipment fault, particle swarm optimization is a relatively new optimization algorithm have raised in recent years. As it depends on less empirical parameters, easy to control, we can use it to optimize the neural network connection weights and thresholds, not only can play the generalization ability of neural networks, but also can improve the convergence speed of neural networks.This paper particle swarm optimization and neural networks are combined, using the concept of a simple particle swarm optimization algorithm does not require the objective function gradient information to optimize the advantages of neural networks, to overcome the disadvantage of slow convergence of BP algorithm. And for the particle swarm algorithm is easy to fall into local optimum situation, put forward a dynamic changing the learning of multi-factor PSO. A new particle swarm optimization (VCPSO) base on unifying the study factor is proposed in this paper. To ensure that particles careful search in the neighborhood of its own in the earlier stage, prevent the particles fast convergence to a local optimal solution for having missed theirs own neighborhood that may exist in the global optimal solution. Particles were rapidly and accurately converge to the global optimal solution and improve algorithm convergence rapidity and accuracy in the later stage. Interfere with the speed of adding smaller items to ensure that the particles in the latter part of the search speed can’t dropped to zero and continue, in order to provide a local optimum particle jump out the possibility that changes in particle swarm algorithm to random search. The convergence of the proposed algorithm analysis and testing using standard function, verify the stability and convergence of the algorithm VCPSO accuracy.The conclusion is that the network based on VCPSO has better training performance, faster convergence rate and higher accuracy.