Search

Timely and accurate three-dimensional (3-D) NOx concentration distribution prediction is essential for achieving low-emission and efficient operation in power plants. This study proposed a theory-guided data-driven prediction method for the 3-D NOx concentration distribution prediction. Firstly, the method created a foundational dataset by fusing numerical simulation data from the computational fluid dynamics (CFD) with operational data from the distributed control system (DCS). Then, the data was classified into three load condition categories, and the center operating conditions for each category were computed separately. Subsequently, the K-means algorithm was employed to extract representative data to address the computational challenges associated with big data. Finally, a Theory-Guided Deep Neural Network model (TG-DNN) was established leveraging the principle of carbon element mass conservation and deep neural network. Experimental results demonstrate that the method effectively monitors the 3-D NOx concentration distribution, potentially facilitating efficient production processes. • Develop a framework for 3-D NOx distribution prediction online with high-precision. • Propose a Theory-Guided DNN merging conservation theory for interpretability. • Utilize the K-Means algorithm for resampling preserving data characteristics. • The proposed model outperformed other comparison models. [ABSTRACT FROM AUTHOR]

The development of an accurate nitrogen oxide (NO x) prediction model is difficult because of multiple parameters, strong coupling, and long delay time of selective catalytic reduction (SCR) systems. In this study, a modeling scheme based on combined feature selection, the JAYA optimization algorithm, and deep neural network (DNN) was developed. First, historical operating data preprocessing, including eliminating outlier points and normalization, was completed. Next, a combined feature selection algorithm based on classification and regression tree, random forest, extreme gradient boosting, and maximal information coefficient (MIC) was developed to select critical input variables. Subsequently, the delay time of the input variables was calculated based on the MIC and JAYA algorithm, and the modeling data were reconstructed. Finally, the real-time dynamic prediction of the SCR outlet NO x concentration was realized based on the DNN model. Experimental results based on operation data of 1000 MW ultra-supercritical boiler revealed that the prediction errors of the established models were less than 5%. Thus, could accurately predict the NO x emission concentration at the outlet of SCR system. • A combined feature selection algorithm based on CART, RF, XGBoost, and MIC is developed. • A JAYA-based method for calculating delay time between NOx emission and input variables is developed. • A real-time dynamic prediction model of the SCR outlet NOx concentration is developed. • Case study on NOx emissions at a 1000 MW coal-fired power plant. • The proposed model outperformed other comparable models. [ABSTRACT FROM AUTHOR]

Accurate and reliable wind direction prediction is important for improving wind power conversion efficiency and operation safety. In this paper, a two-phase deep learning model is proposed and constructed for high-performance short-term wind direction forecasting. In the first phase, a hybrid data processing strategy, including data reconstruction, outlier deletion, dimension reduction, and sequence decomposition, is proposed to extract the most meaningful information from practical data. Then, in the second phase, a robust echo state network is developed for wind direction forecasting. In addition, its hyper-parameters are optimized using an improved flower pollination algorithm (IFPA) to achieve high efficiency. Experiments conducted on data from real wind farms validate the proposed hybrid data processing method. Finally, comparisons with benchmark prediction models show that the proposed network achieves superior performance. • A two-phase short-term wind direction prediction model is proposed. • Hybrid data processing method is used to extract data's most meaningful information. • Improved echo state network (ESN) is developed in prediction modeling. • An improved flower pollination algorithm is proposed to optimize ESN's parameters. • The proposed model is validated efficient and effective in wind direction prediction. [ABSTRACT FROM AUTHOR]

Three-dimensional (3D) reconstructions of temperature distributions can be used to effectively design power plants and ensure production safety. Typically, 3D temperature reconstruction based on the flame image processing technology and finite element calculation of furnace combustion using computational fluid dynamics (CFD) simulation are performed to obtain the furnace temperature field. In this study, a novel online method that overcomes the defects of image detection devices was proposed for reconstructing the temperature field with improved evaluation accuracy. Numerical simulations were used to perform numerous calculations. In this method, deep neural network (DNN) models were used for reconstructing the 3D temperature distribution. The training set was derived from offline CFD simulations that were set for a specific boiler and a series of typical working conditions. Based on established DNN models, the online calculation of 3D temperature distribution was realized for current operating conditions. The result revealed that the furnace temperature field could be accurately reconstructed online in a 350-MW tangentially fired boiler. Compared with the numerical simulation results, the mean absolute percent error under the tilt angles of 0°, 10°, and −10° were 3.61 %, 4.25 %, and 4.52 %. The proposed integrated method was applied to actual boilers with average error 3.448 % and achieved feasible solutions within 20 s. • A novel method was used to reconstruct the furnace temperature field. • The boundary condition of the method was defined from actual operation data. • The measuring point, CFD simulation, and proposed reconstruction errors were compared. • Multiple hyperparameter configurations were tested and optimal model was found. • The combustion process was monitored using a three-dimensional visualization system. [ABSTRACT FROM AUTHOR]

[Display omitted] • A 3D combustion temperature prediction method, GDNN, uses CFD simulations to achieve online reconstructions was proposed. • A temperature field correction method employs actual measurement points was proposed to correct the prediction results. • Optimal parameters for the proposed prediction and correction method were selected through a parameter analysis. • The proposed GDNN and correction method jointly yield the lowest error, demonstrating their real-world effectiveness. Predicting furnace temperature distribution is vital for coal-fired boiler safety. Existing methods, including finite element calculations and three-dimensional (3D) reconstruction still face limitations. A 3D combustion temperature field prediction method, GDNN, which utilizes offline-computed computational fluid dynamics (CFD) simulation results for online reconstructions of entire boiler, was proposed. GDNN method leverages the knowledge of the temperature field acquired by the base neural network model and Gaussian processes. Furthermore, a temperature field correction method is introduced, which employs intermediate variables of the GDNN model and measured values from temperature sensors to establish a correction model for the entire predicted temperature field. We compared GDNN's effectiveness with four well-established algorithms: Extreme Learning Machine (ELM), Least Absolute Shrinkage and Selection Operator (LASSO), Deep Neural Network(DNN), and Radial Basis Function (RBF) network, by also substituting these algorithms as the base model in our proposed method. The experimental results demonstrate that the proposed prediction method exhibits the highest performance, and the correction method effectively improves the overall results. The optimal parameters for predicting and correcting 3D furnace temperature field results were determined through experimental comparison, and the proposed method was applied to a 350 MW boiler, achieving an error of 2.41%, proving its real-world effectiveness. [ABSTRACT FROM AUTHOR]

Advancing methods for modeling combustion systems and optimizing their operations is beneficial to improve the combustion performance. This paper develops a deep data-driven framework for the optimization of combustion system operations. First, a deep belief network based method is developed to model both of the combustion efficiency and the NOx emission. Next, a multi-objective optimization model is developed by integrating the deep belief network based models, the considered operational constraints, and the control variable constraints. Two objectives, maximizing the combustion efficiency and minimizing the NOx emission, are considered in the optimization. Due to the nonlinearity and complexity of the optimization model, traditional exact solution methods are not applicable to solve it. A recently presented swarm intelligence method, the JAYA algorithm, is applied to obtain the optimal solutions of the developed optimization model. Advantages of using JAYA are proved by benchmarking against well-known computational intelligence methods. The feasibility and effectiveness of the developed framework for optimizing the combustion process using industrial data is validated by computational experiments. Results demonstrate the potential of further improving both of the combustion efficiency and NOx emission by optimizing control settings of the combustion system. • The multi-objective optimization of combustion system operations using deep data-driven models is studied. • A deep belief network based method is developed to model the combustion efficiency and NOx emission. • A data-driven multi-objective optimization model of combustion process operations is developed. • A JAYA algorithm is firstly applied to solve the developed optimization model. • Computational results show the potential of further improving the combustion performance based on the developed model. [ABSTRACT FROM AUTHOR]

An automatic encoder (AE) extreme learning machine (ELM)-AE-ELM model is proposed to predict the NOx emission concentration based on the combination of mutual information algorithm (MI), AE, and ELM. First, the importance of practical variables is computed by the MI algorithm, and the mechanism is analyzed to determine the variables related to the NOx emission concentration. Then, the time delay correlations between the selected variables and NOx emission concentration are further analyzed to reconstruct the modeling data. Subsequently, the AE is applied to extract hidden features within the input variables. Finally, an ELM algorithm establishes the relationship between the NOx emission concentration and deep features. The experimental results on practical data indicate that the proposed model shows promising performance compared to state-of-art models. • A MI-based method for calculating delay time between NOx emission and input variables is developed. • A new feature extraction method based on auto-encoder is developed. • An AE-ELM-based method for predicting NOx emissions is developed. • Case study on NOx emissions at a 1000 MW coal-fired power plant. • The proposed model outperformed other comparable models. [ABSTRACT FROM AUTHOR]

To schedule wind power efficiently and to mitigate the adverse effects caused by wind's intermittency and variability, an advanced wind power prediction model is proposed in this paper. This model is a combined model via switching different data-driven chaotic time series models. First, inputs of this model come from the reconstructed data based on the chaotic characteristics of wind power time series. Second, three different data mining algorithms are used to construct wind power prediction models individually. To obtain a regime for switching optimal models, a Markov chain is trained. Then, weights of different data-driven modes are calculated by the Markov chain switching regime, and used in the final combined model for wind power prediction. The industrial data from actual wind farms is studied. Results of the proposed model are compared with that of non-reconstructed input data, traditional data-driven models and two typical combined models. These results validate the superiority of proposed model on improving wind power prediction accuracy. • Wind power prediction is realized by chaotic time series and switching regime. • Chaotic wind power time series is analyzed and reconstructed as model inputs. • Different data-driven models are trained as prediction modes. • A regime constructed by Markov chain is proposed for modes switching. [ABSTRACT FROM AUTHOR]

Abstract Wind power ramps cause large-amplitude power fluctuation which harmfully affects the stability of power system's operation. As a new issue in wind power integration, the existing ramp forecasting methods still has some imperfection, e.g., harmonization on long-term trend and short-term precision. Therefore, an advanced method is proposed in this paper, mainly focus on improving the performance of wind power ramp prediction. This method utilizes wind power curve to build a primary model which can capture the trend of wind power variation. Then, prediction residual of the primary model is corrected by a MSAR (Markov-Switching-Auto-Regression) model which combining the advantages of AR models and Markov chain. Finally, an improved swinging door algorithm is applied to extract linear segments, and ramp definitions are used to detect ramp events. Actual wind farm data is used to test the proposed method. Comparison with traditional methods are presented, the numerical results validate that the proposed approach has improved performance not only on wind power prediction but also on ramp prediction. Highlights • An advanced approach is proposed to improve ramp prediction. • The approach contains wind power prediction, residual correction, ramp detection. • Primary models capture trend of wind power variation. • Details precision are corrected by MSAR correction prediction residual. [ABSTRACT FROM AUTHOR]