航空发动机复杂磨损趋势的神经网络多变量预测模型

Artificial Neural Network Multi-Variable Forecasting Model of Aero-Engine Complex Wear Trend

针对非等间隔的受加油因素影响的光谱油样分析数据的建模预测问题,建立了BP神经网络的多变量预测模型,充分考虑了油样分析数据的非等间隔性及受加油因素影响的特点,同时,用遗传算法对网络参数进行了优化。最后,利用两组实际的航空发动机油样光谱分析数据对模型进行了验证。结果表明,所提出的神经网络多变量预测模型能有效解决实际的受多因素影响的油样分析数据建模问题,具有较强的工程实用价值和通用性。

Multi-variable predicting model by Back-Propagation Neural Networks （BPNN） was established, and it considered the complex characteristics of the oil analysis data fully such as unequal interval sampling and affected by adding oil factor. In addition, the effect of ANN＇s parameters on the predicting accuracy was also discussed, and Genetic Algorithm （GA） was used to optimize ANN＇s parameters. Finally, two time series and aero-engine spectrometric oil analysis data were used to verify this model. The results show that this model can solve effectively the prediction problem of oil analysis data, which is affected by multi-factors such as unequal interval sampling and adding oil factor. The new method has important engineering application value, and it is a common method of forecasting complex time series, which is affected by multi-factors.