基于长短期记忆网络汽轮机振动幅值预测

Prediction of turbine vibration amplitude based on long short-term memory network

火电机组的主轴振动幅值具有非线性,非平稳,时序相关,且与当前历史状态密不可分的特点,而实际火电厂所提取数据往往呈现无规则,长时间,数据量庞大的特点。提出了由麻雀搜索算法(SSA)进行优化的的长短期记忆网络(LSTM)相结合构建深度学习预测模型,对汽轮机主轴的振动幅值进行更高精度的预测模拟。相较于非时序神经网络模型和无优化时序神经网络模型预测性能大大提高。

The main axis vibration value of thermal power unit is non-linear,non-stable,sequentially related,and is inseparable from the current historical state.The data extracted by actual thermal power plants often show irregularities.For a long time,the data volume is huge.A long-term memory network(LSTM)that is optimized by the sparrow search algorithm(SSA)is proposed to build a deep learning predictive model,and the vibration amplitude of the spherical spindle of the steam turbine is made of higher accuracy and simulation.Compared with non-time-order neural network models and no optimized timing neural network model prediction performance greatly improved.