Development of an optimized artificial neural network model for combined heat and power micro gas turbines
Development of an optimized artificial neural network model for combined heat and power micro gas turbines
复制标题
DOI：
10.1016/j.apenergy.2013.03.016
复制
发表时间：
2013-08
期刊：
影响因子：
11.2
通讯作者：
H. Nikpey;M. Assadi;P. Breuhaus
中科院分区：
文献类型：
--
作者：
H. Nikpey;M. Assadi;P. Breuhaus
微型燃气轮机被认为是昂贵的发电和输电的有效替代方案，特别是在偏远地区和热电联产（CHP）应用中。实现分布式热电联产需要易于使用的远程监控和诊断工具。本文报道了一种经过验证的人工神经网络（ANN）模型的开发，用于有效和适当地监测微型燃气轮机。本研究基于Turbec T100微型燃气轮机的实验数据。本研究中使用的燃气轮机试验台由一个改进的发动机和扩展的测点组成，为数据驱动建模提供了广泛的数据。与数学模型相比，人工神经网络不需要详细和精确的组件特征，这使得它们适用于许多情况下的监测。所建立的人工神经网络模型基于多层前馈网络和反向传播算法。人工神经网络模型的输入和输出之间的关系不是通过物理方程实现的，而是在训练过程中建立起来的。因此，在几个步骤中进行了系统的敏感性分析，以调查最初选择的输入和输出参数之间的依赖关系。根据灵敏度分析结果，在考虑预测精度的情况下，得出“优化”的输入参数集和最终输出。详细讨论了人工神经网络模型的建立过程和灵敏度分析。使用平均相对误差（MRE）来评价所开发的网络相对于训练中未使用的实验数据的预测精度。预测结果表明，最终优化的人工神经网络模型可以作为监测应用的准确基线模型。
Micro gas turbines are considered an efficient alternative to costly generation and transmission of electricity, especially in remote areas and in combined heat and power (CHP) applications. Tools for remote monitoring and diagnostics, which are easy to apply, would be needed for the realization of distributed CHP. This paper reports the development of a validated artificial neural network (ANN) model for efficient and appropriate monitoring of a micro gas turbine. This study is based on experimental data obtained from a Turbec T100 micro gas turbine. The gas turbine test rig used in this study consists of a modified engine with extended measurement points providing extensive data suitable for data-driven modeling. ANNs, in contrast to mathematical models, do not require detailed and exact component characteristics, making them applicable for monitoring in many cases. The developed ANN model was based on a multi-layer feed forward network with back-propagation algorithm. Relations between the inputs and outputs of an ANN model are not implemented by physical equations but are built up during the training process. Thus, a systematic sensitivity analysis, conducted in several steps, was performed to investigate the dependency between initially selected input and output parameters. Based on sensitivity analysis results, the “optimized” set of input parameters and final outputs were concluded, taking into account prediction accuracy. The procedure of the ANN model development and sensitivity analysis are discussed in detail. The mean relative error (MRE) was used to evaluate the prediction accuracy of the developed networks with respect to experimental data which were not used during the training. The prediction results showed that the final optimum ANN model can serve as an accurate baseline model for monitoring applications.