Pumping well fault diagnosis method based on improved fish school algorithm

Abstract

The invention provides a pumping well fault diagnosis method based on improved fish school algorithm, and belongs to the field of oil pumping well fault diagnosis. The method comprises the following steps: carrying out pretreatment on a well indicator diagram of a rod pumped well by using a known fault type, extracting characteristics by using a deep belief network method from the pre-processed indicator diagram, classifying the characteristics extracted by the deep belief network method by using a classification function of a support vector machine, calculating a classification function valueof a known fault type, optimizing the parameters of the classification function of the support the vector machine by using the improved fish school algorithm, carrying out classification calculationon the characteristics extracted by the depth belief network according to the support vector machine, obtaining the classification function value of the indicator diagram of the diagnosis fault type to be diagnosed for each type of fault, the wherein diagnosis fault type to be diagnosed is classified into the diagnosis type which has the maximum classification function value, According to the invention, the method avoids the uncertainty caused by manual weight value and preset bias, the accuracy of classifying the characteristics extracted by the depth belief network is improved, and the convergence speed is improved.

Description

technical field [0001] The invention belongs to the field of fault diagnosis of pumping wells, and in particular relates to a fault diagnosis method of pumping wells based on an improved fish swarm algorithm. Background technique [0002] The dynamometer diagram of the rod pump pumping unit well is a concentrated expression of the working status of the rod pump pumping system. The diagnostic dynamometer diagram is the quickest way to judge the working condition of the rod pumping system. Traditional dynamometer diagnosis mainly relies on manual observation, but manual observation is easily affected by various subjective factors. Therefore, there will inevitably be various interference factors in the diagnosis results, and the accuracy and stability of the analysis results cannot be guaranteed. In addition, the efficiency of manual analysis is very low, which is contrary to the high efficiency requirements of oil production. [0003] With the continuous upgrading of techno...

AI Technical Summary

Problems solved by technology

In the selection of the classifier, the traditional neural network as a classifier is easy to fall into local optimum, and the quality of the training samples has a great impact on the neural network, and the neural network model generally adopts the principle of empirical risk minimization. poor generalization ability

Images