Prediction of Dynamic Behavior of a Single Shaft Gas Turbine Using NARX Models

Gas turbines are internal combustion engines widely used in industry as main source of power for aircrafts, turbo-generators, turbo-pumps and turbo-compressors. Modelling these engines can help to improve their design and manufacturing processes, as well as to facilitate their operability and maintenance. These eventually lead to manufacturing of gas turbines with lower costs and higher efficiency at the same time. The models may also be employed to unfold nonlinear dynamics of these systems. The aim of this study is to predict the dynamic behavior of a single shaft gas turbine by using open-loop and closed-loop NARX models, which are subsets of artificial neural networks. To set up these models, datasets of significant variables of the gas turbine are used for training, test and validation processes. For this purpose, a comprehensive code is developed in MATLAB programming environment. In addition to the open-loop model, a closed-loop model is set up for multi-step prediction. The results of this study demonstrate the capability of the NARX models in reliable prediction of gas turbines’ dynamic behaviors over different operational ranges.

An Ensemble of Recurrent Neural Networks for Real Time Performance Modelling of Three-Spool Aero-Derivative Gas Turbine Engine

Gas turbine is a complex system operating in non-stationary operation conditions for which traditional model-based modeling approaches have poor generalization capabilities. To address this, an investigation of a novel data-driven neural networks based model approach for a three-spool aero-derivative gas turbine engine (ADGTE) for power generation during its loading and unloading conditions is reported in this paper. For this purpose, a non-linear autoregressive network with exogenous inputs (NARX) is used to develop this model in MATLAB environment using operational closed-loop data collected from Siemens (SGT-A65) ADGTE. Inspired by the way biological neural networks process information and by their structure which changes depending on their function, multiple-input single-output (MISO) NARX models with different configurations were used to represent each of the ADGTE output parameters with the same input parameters. Usage of a single neural network to represent each of the system output parameters may not be able to provide an accurate prediction for unseen data and as a consequence, provides poor generalization. To overcome this problem, an ensemble of MISO NARX models is used to represent each output parameter. The major challenge of the ensemble generation is to decide how to combine results produced by the ensemble's components. In this paper, a novel hybrid dynamic weighting method (HDWM) is proposed. The simulation results show improvement in accuracy and robustness by using the proposed modeling approach.

Evaluation of Kriging-NARX Modeling for Uncertainty Quantification of Nonlinear SDOF Systems with Degradation

Structural assessment for collapse is commonly approached by observing the failure or collapse of systems fully incorporating degradation. Challenges however exist in the performance indicator or damage measure due to compound impacts of uncertainties of external (seismic excitation) and internal (structural properties) characteristics with degradation behavior. To account for the impacts of uncertainties, the state-of-the-art kriging nonlinear autoregressive with exogenous (NARX) model is explored in this study to replicate the response of nonlinear single-degree-of-freedom systems. The generalized hysteretic Bouc-Wen model with internal uncertainties is selected to emulate the stiffness and strength degradation. A probabilistic stochastic ground motion model is introduced to represent the external uncertainties. The global terms of NARX model are selected by least-angle regression algorithm and the kriging model is utilized to surrogate uncertain parameters into corresponding NARX model coefficients. The predictions of kriging NARX models are further compared with that of the polynomial chaos nonlinear autoregressive with exogenous input form model as well as Monte Carlo simulation. The comparisons show that kriging NARX model presents an effective and efficient meta-model technique for uncertainty quantification of systems with degradation.

Groundwater Level Forecasting with Artificial Neural Networks: A Comparison of LSTM, CNN and NARX

It is now well established to use shallow artificial neural networks (ANN) to obtain accurate and reliable groundwater level forecasts, which are an important tool for sustainable groundwater management. However, we observe an increasing shift from conventional shallow ANNs to state-of-the-art deep learning (DL) techniques, but a direct comparison of the performance is often lacking. Although they have already clearly proven their suitability, especially shallow recurrent networks frequently seem to be excluded from the study design despite the euphoria about new DL techniques and its successes in various disciplines. Therefore, we aim to provide an overview on the predictive ability in terms of groundwater levels of shallow conventional recurrent ANN namely nonlinear autoregressive networks with exogenous inputs (NARX), and popular state-of-the-art DL-techniques such as long short-term memory (LSTM) and convolutional neural networks (CNN). We compare both the performance on sequence-to-value (seq2val) and sequence-to-sequence (seq2seq) forecasting on a 4-year period, while using only few, widely available and easy to measure meteorological input parameters, which makes our approach widely applicable. We observe that for seq2val forecasts NARX models on average perform best, however, CNNs are much faster and only slightly worse in terms of accuracy. For seq2seq forecasts, mostly NARX outperform both DL-models and even almost reach the speed of CNNs. However, NARX are the least robust against initialization effects, which nevertheless can be handled easily using ensemble forecasting. We showed that shallow neural networks, such as NARX, should not be neglected in comparison to DL-techniques; however, LSTMs and CNNs might perform substantially better with a larger data set, where DL really can demonstrate its strengths, which is rarely available in the groundwater domain though.

An Ensemble of Recurrent Neural Networks for Real Time Performance Modelling of Three-Spool Aero-Derivative Gas Turbine Engine

Gas turbine is a complex system operating in non-stationary operation conditions for which traditional model-based modelling approaches have poor generalization capabilities. To address this, an investigation of a novel data driven neural networks based model approach for a three-spool aero-derivative gas turbine engine (ADGTE) for power generation during its loading and unloading conditions is reported in this paper. For this purpose, a non-linear autoregressive network with exogenous inputs (NARX) is used to develop this model in MATLAB environment using operational closed-loop data collected from Siemens (SGT-A65) ADGTE. Inspired by the way biological neural networks process information and by their structure which changes depending on their function, multiple-input single-output (MISO) NARX models with different configurations were used to represent each of the ADGTE output parameters with the same input parameters. First, data preprocessing and estimation of the order of these MISO models were performed. Next, a computer program code was developed to perform a comparative study and to select the best NARX model configuration, which can represent the system dynamics. Usage of a single neural network to represent each of the system output parameters may not be able to provide an accurate prediction for unseen data and as a consequence, provides poor generalization. To overcome this problem, an ensemble of MISO NARX models is used to represent each output parameter. The major challenge of the ensemble generation is to decide how to combine results produced by the ensemble’s components. In this paper, a novel hybrid dynamic weighting method (HDWM) is proposed. The verification of this method was performed by comparing its performance with three of the most popular basic methods for ensemble integration: basic ensemble method (BEM), median rule and dynamic weighting method (DWM). Finally, the generated ensembles of MISO NARX models for each output parameter were evaluated using unseen data (testing data). The simulation results based on datasets consisting for experimental data as well as data provided by Siemens high fidelity thermodynamic transient simulation program show improvement in accuracy and robustness by using the proposed modelling approach.