Ana Paula Barbosa Sobral ana_sobral@vm.uff.br Fluminense Federal University, Rio das Ostras, RJ, Brazil. ABSTRACT Goal: To evaluate the performance of a set of forecasting methods in the prediction of future values on a dataset of time series collected from sensors installed in an industrial gas turbine. Design / Methodology / Approach: Forecasting methods tested include the use of multivariate and univariate neural networks (FNN and LSTM), exponential smoothing and ARIMA models. Results: Results show that the use of ARIMA models to forecast on the dataset is the best default method to apply, and is the only forecasting method that consistently beats a simple naïve no-change model. Limitation of the investigation: There was a focus on evaluating neural networks. This limited resources available to evaluate other forecasting methods. There is no guarantee that it would not be possible to find neural networks capable of yielding better forecasts than the ones achieved by the best performing methods in this research. Practical implications: The broadest possible implications of the results are that the best default method to forecast industrial machinery time series is the use of ARIMA models. Additionally, neural networks are not capable of beating methods well stablished within the forecasting community, namely ARIMA models. Originality / Value: To the best of the authors’ knowledge, there is a scarce amount of published evaluations of multiple forecasting methods on data from real machines. This knowledge is useful for the understanding of the best forecasting methods available for the estimation of machine’s RUL using sensor time series.

中文翻译：

---

工业机械传感器采样的多个时间序列的预测

Ana Paula Barbosa Sobral ana_sobral@vm.uff.br巴西巴西Rio das Ostras的Fluminense联邦大学。摘要目标：在从安装在工业燃气轮机中的传感器收集的时间序列数据集上，评估一组预测方法在预测未来值方面的性能。设计/方法论/方法：测试的预测方法包括使用多元和单变量神经网络（FNN和LSTM），指数平滑和ARIMA模型。结果：结果表明，使用ARIMA模型对数据集进行预测是应用的最佳默认方法，并且是唯一击败简单的朴素不变模型的唯一预测方法。研究的局限性：重点是评估神经网络。这种有限的资源可用于评估其他预测方法。不能保证不可能找到比通过本研究中性能最好的方法获得的预测更好的神经网络。实际意义：结果的最广泛含义是，预测工业机械时间序列的最佳默认方法是使用ARIMA模型。此外，神经网络无法击败在预测社区中已建立好的方法，即ARIMA模型。原创性/价值：据作者所知，对来自真实机器的数据的多种预测方法的已发表评估很少。该知识有助于理解最佳的预测方法，该方法可用于使用传感器时间序列估算机器的RUL。