基于神经网络模型的海南变电站接地网Q235钢腐蚀率预测

运用MATLAB软件在土壤腐蚀等级评价指标上随机生成了2 000组训练样本和200组测试样本来增强网络的鲁棒性(抗变换性)和样本识别准确性,找出了适合BP和RBF神经网络模型的结构参数,构建出了性能和稳定性都较好的BP和RBF神经网络模型。用现场采集的海南省变电站土壤腐蚀相关数据分别对已建并训练的BP和RBF神经网络模型进行检验,并用这两种模型对变电站接地网普遍使用的Q235钢的腐蚀速率进行了预测。结果表明:两种模型预测的准确率均在95%以上;BP神经网络模型在结构和运算方面比RBF神经网络模型好,但需要设定的参数多、较繁琐,而RBF神经网络模型只需设定Spread值,较简单,且RBF神经网络模型在训练精度和泛化能力方面均优于BP神经网络模型。

Corrosion Rate Prediction of Q235 Steel in Hainan Substation Grounding Grid Based on Neural Network Models

Using MATLAB software,2000 training samples and 200 test samples were randomly generated among soil corrosion grade evaluation indexes in order to enhance the robustness and accuracy of sample identification,find out proper structural parameters for BP and RBF network models with good performance and stability.The BP and RBF network models were tested using the data of soil erosion in the substation of Hainan province after building and training.The corrosion rate of Q235 steel widely used in substation grounding grid was predicted by these two models.The results show that the accuracy of these two models was more than 95％.BP neural network model is better than RBF neural network model in structure and operation,but it needs to set more parameters and is more cumbersome.On the contrary,the RBF neural network model is more simple and only needs to set the Spread value.Meanwhile,the training accuracy and generalization ability of RBF neural network model are better than those of BP neural network model.