基于LSTM的舰船软件运行健康状态预测

Prediction of operational health status of ship software based on LSTM

针对舰船任务系统的复杂环境,综合考虑舰船软件自身性能、外在环境和运行工况等数据的影响,采用长短期记忆网络模型(LSTM)预测软件运行健康状态,并针对样本类别分布不均衡导致的预测效果不佳等问题,提出了一种加权焦点损失函数(WFL).实验结果表明:基于WFL与包含三个隐含层的LSTM模型(LSTM3-WFL)不仅比传统的机器学习算法能够更好地学习到特征在时间维度上的变化规律;而且相较于基于交叉熵损失函数的LSTM模型,该模型更容易学习到样本个数较少的类别信息,并最终在测试集上达到98.2%的准确率与0.947的宏平均F1-Socre值,在舰船软件运行健康状态的预测问题上有很高的应用价值.

In view of the complex environment of the ship mission system,the influence of the ship software’s self performane,external environment and operating conditions was considered.The long-short term memory(LSTM)model was used to predict the health status of the software.In order to solve the problem that the bad effect caused by the unbalanced distribution of sample categories,a weighted focal loss(WFL)loss function was proposed.The experimental results show that the variation of features in the time dimension can be better learn by the improved model named LSTM with three hidden layer based on WFL(LSTM3-WFL)than the traditional machine learning algorithm.The model is easier to learn the category information with less sample number rather than the LSTM model based on the cross entropy loss function,and achieves the accuracy of 98.2%and a macro-average F1-Socre of 0.947 on the test set.In general,it has a high application value in prediction of software operational health status.