基于LSTM-ECGRU的固体火箭发动机性能预测方法

随着数据挖掘技术、测量技术的不断发展，为了满足火箭发动机参数探索的需要，使用数据挖掘技术利用历史数据对发动机各种参数进行预测成为火箭发动机在数据探索方面新的发展方向。同时，火箭发动机的地面点火试验在向着尽可能还原真实运行环境的方向发展。基于以上情况，引入在地面点火试验中的环境因素与设计因素共同作为模型的输入变量，以此来补充环境因素对性能参数的影响。根据试验对象数据特性，使用长短期记忆(Long Short-Term Memory, LSTM)神经网络对性能进行初步预测。为了能够减少整体模型误差和引入环境因素带来的误差，提高模型预测精度和泛化能力，提出了基于误差修正分析和趋势判断的误差修正门控单元(Error Correction Gate Recurrent Unit, ECGRU)神经网络模型对初步预测结果进行误差修正。同时结合环境参数特点，设计规划ECGRU模型输入、输出参数的计算规则。基于历史试验数据完成对比试验，验证了新模型具有较高的预测精度和泛化能力。

Performance Prediction Method of Solid Rocket Motor Based on LSTM-ECGRU

With the continuous development of data mining technology and measurement technology,in order to meet the needs of rocket engine parameter exploration,data mining technology is used to predict various parameters of rocket engine with using historical data has become a new development direction in data exploration of rocket engine.At the same time,the ground ignition test of rocket engine is developing towards restoring the real operating environment as much as possible.Based on the above situation,environmental factors and design factors in ground ignition tests are introduced as input variables of the model,so as to supplement the influence of environmental factors on performance parameters.According to the data characteristics of experimental objects,the long short-term memory(LSTM) neural network is used to preliminarily predict the performance.In order to reduce the errors of the overall model and the errors caused by environmental factors,improve the prediction accuracy and generalization ability of the model,an error correction gate recurrent unit（ECGRU） neural network model based on error correction analysis and trend judgment is proposed to correct the initial prediction results.Combined with the characteristics of environmental parameters,the calculation rules for input and output parameters of ECGRU model are designed and planned.Based on the historical experimental data,the comparison experiment verifies that the new model has high prediction accuracy and generalization ability.