基于LSTM网络的综合射频模块温度的预测研究

Research onTemperature Prediction for Integrated RF Module Based on LSTM Network

摘要:

随着电子产品及集成电路的快速发展，其电子产品的故障预测研究引起了高度重视，但准确预测其使用寿命难度还是很大，目前针对电子产品主要采用状态监控和健康管理，从而实现状态的预测。以此为出发点，构建综合射频模块温度的状态预测模型，该预测模型首先将设备的时域特征数据转换为有监督的样本数据集，然后建立原始参数集、预测模型的训练集和测试集，接着建立LSTM深度学习网络结构，进行参数调整设置并运行模型，最后获得预测值和观测值的误差曲线。采用该方法在某典型任务场景中进行了应用验证，获得综合射频模块的温度预测的准确度为98.7%，达到了较好的预测效果和精度。

Abstract:

With the rapid development of electronic products and integrated circuits, the fault prediction research on electronics has attracted great attention. However, it is still very difficult to accurately predict electronics’ service life. At present, the main implementation is state monitoring and health management. A state prediction model for integrated RF module temperature based on long short-tern memory network（LSTM）neural network is constructed in this paper, Firstly, the time-domain data of a device is converted into a supervised sample dataset. Then the original parameter set, and training and test sets for prediction model are established. Next the LSTM deep learning network is constructed and run with parameters. Finally, the error curves of the predicted and observed values are obtained. The model is proved to achieve a good prediction effect and forecast precision with an accuracy of 98.7% when predicting the integrated RF module temperature in a typical task.