The amount of methane emission is crucial to the safety of coal mine. The paper proposes the Levenberg Marquardt (LM) algorithm (in the nonlinear least squares algorithms) that can reduce the training time of BP network. Genetic Algorithm (GA) is used to optimize weights in global search to prevent the inherent defects that neural network is liable to get stuck in local minimal points. Furthermore, neural network can prevent the defect of weak GA local search. Finally, BP-GA modal was trained and the sample data were precisely analyzed, which proves that this model features broad adaptability and precision.

Because the grounding grid corrosion rate has the property of nonlinearity and uncertainty, it is very difficult for us to predict precisely. The approach is proposed that ant colony clustering algorithm is combined with RBF neural network to predict the grounding grid corrosion rate, using ant colony clustering algorithm to get the center of hidden layer neurons. To find the best clustering result, local search is applied in ant colony algorithm. This model has good performance of strong local generalization abilities and satisfying accuracy. At last, it is proved with lots of experiments that the application is fairly effective.

In this paper, we proposed a training model to predict the corrosion rate for substation grounding grid based on the Similarity and Support Vector Regression (SSVR). In the proposed model, the effect of grounding grid corrosion rate was acted as a feature vector and processed by a dimensionless treatment. Then, the similarity between the feature vector of training terminal and index vector of actual site would be calculated. In the prediction of corrosion rate, the traditional Linear Average Method (LAM) to describe the nonlinear contribution has some fault defects. Therefore, we proposed the training model named SSVR. From the experimental results, the proposed SSVR can obtain better predicting performance than the traditional LAM.

Grounding grid is an important device which ensures the safe operation of substation. Its corrosion is influenced by various factors and the influence of these factors is varied. It is difficult to infer the relationship between the corrosion rate and its influencing factors, so this paper apply the grey clustering method of grey system theory in the grounding grid corrosion and gives an application instance and realizes it by MATLAB. The results demonstrate the practicality and effectiveness of the method mentioned in this paper. This method can provide reference for high reliability for related departments in the implementation of emergency situation due to grounding grid corrosion.

A forecasting model of the gradual optimization algorithm is established to predict substation grounding grip corrosion rate. In this model, according to the “Over Fitting” phenomenon in the neural network limited soil corrosion sample data are randomly combined and the training stops when the training error and validation error are equal. The model of smaller errors will be chosen as the optimal model. As shown in the simulation, the general performance and fitting accuracy from the forecasting model meet requirements.