基于ABC-LSTM的变压器表面振动基频幅值预测

Prediction of Fundamental Frequency Amplitude of Transformer Surface Vibration Based on ABC-LSTM

针对变压器表面振动基频幅值的预测问题,提出一种基于人工蜂群算法(Artificial Bee Colony,ABC)优化LSTM的混合预测模型(ABC-LSTM)。首先,在考虑变压器振动机理及其影响因素的基础上,选择运行电压、负载电流和油温作为输入、振动基频幅值作为输出构建LSTM模型;其次,采用ABC算法对LSTM的学习率、批处理数量、隐藏层神经元数目这三个直接决定其预测性能的超参数进行优化;最后,根据优化结果建立ABC-LSTM预测模型并通过某台变压器实测数据进行测试。结果表明,相比于LSTM和GRNN,文中所提模型的预测精度更高、稳定性更好。

In order to accurately predict the fundamental frequency amplitude of the transformer surface vibration,a hybrid prediction model combining artificial bee colony algorithm and LSTM is proposed in this paper.Firstly,consider the transformer vibration transmission mechanism and influencing factors,take the operating voltage,load current,oil temperature as inputs,and the amplitude of the fundamental frequency of vibration as the output to establish LSTM.Secondly,the three hyperparameters of LSTM's learning rate,batch size,and number of hidden layer neurons closely affect its prediction performance,and ABC is used to optimize them.Finally,establish a prediction model based on the results of ABC optimization,and use the measured data of the transformer for testing.The final experimental results confirm that ABC-LSTM has better prediction performance than LSTM and GRNN.