Research On LVQ Neural Network Sucker Rod Pump Fault Diagnostic Based On Rough Sets

In the oilfield at present, dynamometer cards collected from the field for fault diagnosisare manually checked, which is time-consuming and inefficient. This essay points out adiagnostic method for LVQ Neural Network based on rough sets. Specific content as follows:First, existing dynamometer cards of pumping unit were analyzed and three kinds ofcommon fault samples were searched. The characteristic value was extracted, whichconcluded15features such as the7invariant moments of image Hu, the perimeter of thecurve, and the area.Second, the rough sets and LVQ Neural Network were combined to solve the pooranti-interference and the disadvantage of unable to reduce the complicated information. Therough sets was used as leading end data processor to build decision table for conditionattribute consisting of15features. The continuous condition attribute of the decision tablewere discretized with the use of the self-organization characteristic mapping SOM. The roughsets was used to cancel the repetitive or inconsistent data and thus build a reduced decisiontable. K-ford was used to work out the optimum number of the nerve cells of LVQ NeuralNetwork, and thus the LVQ Neural Network was built. Some parts of the reduced decisiontable were used to train LVQ Neural Network, and the other parts were used to test it.Third, the method referred in this essay was contrastively analyzed with the BP NeuralNetwork based on rough sets. The result shows that the LVQ Neural Network based on roughsets were well adapted to practical diagnosing for dynamometer cards and the diagnosticaccuracy was superior to that of BP Neural Network based on rough sets.