基于DEPSO-RBFNN的变压器表面温度预测模型

Prediction of Shell Temperature for Transformers Based on DEPSO-RBFNN

提出一种基于差异进化算法(DE)和粒子群优化算法(PSO)的新型混合进化算法DEPSO,以及基于DEPSO的径向基函数神经网络(RBFNN)模型,并应用于预测SF6气体绝缘变压器表面温度。该模型用DEPSO算法训练RBFNN隐层中心的数量和位置,并采用递推最小二乘法确定网络输出层的权值。对某变电站SF6气体绝缘变压器的表面温度预测结果表明:与BP网络、基于进化规划(EP)、PSO的RBFNN相比,这种建模方法具有更高的预测精度。

A novel radial basis function neural network （RBFNN） model based on a hybrid learning algorithm differential evolution and particle swarm optimization （DEPSO） is proposed in this paper to predict the shell temperature for SF6-insulated transformers. The DEPSO automatically adjusts the number and positions of hidden layer RBF centers o The weights of output layer are decided by the recursive least squares algorithm. The proposed DEPSO-RBFNN model is trained and tested based on the field data collected from a SF6-insulated transformer. The test results reveal that the DEPSO- RBFNN possesses far superior forecast precision than BP neural network （BPNN）, EP-RBFNN and PSO- RBFNN.