基于多值编码混合遗传算法的变压器故障诊断

Power Transformer Fault Diagnonis with Hybrid Genetic Algorithm Based on Multi-Encoding

根据网络的结构和参数的特性采用多值编码方式构造染色体结构,同时确定BP神经网络的结构,优化网络参数。在遗传算法中嵌入一个梯度下降算子,使得混合算法既有较快的收敛性,又能以较大概率得到全局极值。基于油中气体分析法的变压器故障诊断仿真结果表明,该算法有较快的收敛速度,较高的逼近精度。

A hybrid genetic algorithm based on multi-encoding method for the optimization of neural networks is put forward. This method can be used to optimize the structure and the parameters of ANN in the same training process. Through embedding a gradient descend operator into the generic algorithm, a hybrid algorithm is achieved with fast convergence and great probability for global optimization. Simulation results of power transformer fault diagnosis by dissolved gas-in-oil analysis show that it can both meet the precision request and enhance the generalization ability.