The Research On Corrosion Law And Prediction Model Of The Substation Grounding Grids

Safe operation of the substation is an important guarantee for safe and reliable operationof the power system, because most grounding grids conductor materials are mostly carbon,they are perennial buried in the ground, so corrosion make the grounding performancedegradation, seriously affect the safety of power grid operation. Therefore, exploring themethod of detection of corrosion status of grounding grid without excavation or power offhas important significance for grounding grid corrosion research. The paper use the corrosiondata in the soil environment of the substation grounding grid in recent years as researchobject, to study corrosion law and prediction model of the grounding grid. The main work isas follows:First, establish database of soil corrosion through field measurement; in order to find therelationship between the corrosion factors and corrosion rate, soil corrosion sample data isanalyzed; introduce the principle and implementation step of principal component analysismethod, using principal component analysis to preprocess soil corrosion data, eliminate thecorrelation between factors, reduce the dimension, then it is easier to establish model.Secondly, establish a non-parametric cluster classification model on the small sizeproblem of grounding grid corrosion data. Take Bootstrap generated bootstrap subset,combining KNN classification method and Adaboost method (non-parametric methods) tocreate multiple weak classifiers for all bootstrap and then cluster into a strong classifier.Experimental results show that the classification applies to the grounding grid corrosionclassification.Then, using BP neural networks, support vector machines and fuzzy analog method toestablish grounding grid corrosion rate prediction model. Aiming at neural networks”overfitting” phenomenon and combining with cross-validation method, a progressive optimizationalgorithm BP neural network is proposed, improved its generalization capability; propose a structure combined cluster analysis(K opt-means algorithm) and support vector regression,determine the optimal number of clusters, set the input samples into a type cluster havingmore similar feature, train and establish grounding grid corrosion factors SVR model; fromthe perspective of fuzzy mathematics to determine the membership functions and closenessdegree, to establish fuzzy analog prediction model of grounding grid corrosion rate. Andverify each of the three predictive models, comparative analysis showed that the threepredictive models have advantages and disadvantages, the predict result of fuzzy analogmodel is better than the previous two models.Finally, combine Visual Studio C#with MATLAB language, develop a ground gridcorrosion prediction application software, data importing, modeling and predicting thecorrosion rate and other functions are realized.