A novel healthy state monitoring method for offshore booster station platforms is proposed to enhance the damage detection capabilities under complex operating conditions. Using a deep learning framework based on memory unit autoencoders, the method magnifies fault-relevant features via denoising and angular domain resampling high-frequency vibration data from offshore boosting stations. The model employs a deep convolutional neural network to learn historical data patterns, constructs a hidden state memory bank, and achieves sparse matching between sample encoded features and the memory bank. Finally, a Gaussian mixture probability model is employed to model the generated membership scores to assess the health status of the booster station. A case study of the offshore booster station in Rudong, Jiangsu, validates the approach, achieving an anomaly recall rate and accuracy of over 98%, outperforming other comparison algorithms.

To explore the ignition and stable combustion performance of ammonia fuel in simulated combustion chambers of gas turbines, ignition and combustion experiments were conducted on ammonia gas with different preheating temperatures and cracking degrees, and the ignition and combustion laws of ammonia fuel under certain experimental conditions were obtained. The results indicate that, stable combustion of ammonia requires a cracking degree of not less than 30% and an air preheater temperature of not less than 643 K. Within the temperature range of 743~943 K and combustion duration of 5~40 seconds in the air preheater, the internal temperature, tail temperature, and pressure of the combustion chamber generally increase with the preheater temperature and combustion duration. The NO emission volume fraction is significantly affected by the temperature of the preheater, it reaches the minimum (376 μL/L) at 673 K when the combustion efficiency is 96%. The zero dimensional simulation results show that, increasing pressure, ammonia cracking degree and temperature can help shorten the ignition delay time, and higher hydrogen content and slightly enriched combustion state can promote the increase of laminar flame velocity and optimize the combustion of ammonia.

Ammonium bicarbonate is a potential denitrification reducing agent that can efficiently produce ammonia gas through direct solid pyrolysis. The pyrolysis reaction of ammonium bicarbonate solid is numerically simulated, a pyrolysis ammonia production system suitable for coal-fired power plants is designed, and the economic feasibility of the ammonium bicarbonate pyrolysis ammonia production process is analyzed. The simulation results show that, the pyrolysis process of ammonium bicarbonate favors the atmosphere pressure and the conversion rate of pyrolysis rapidly increases when the reaction temperature is above 110 ℃. The pyrolysis system of ammonium bicarbonate for a 660 MW unit has been designed and calculated. An external heating pyrolysis reactor is adopted to realize the utilization of waste heat and stable solid feeding. Steam or flue gas from the coal-fired power plant is used as the heat source for pyrolysis. At 110 ℃, a conversion rate of 95% can be reached within 10 minutes for ammonium bicarbonate feed. Compared with the urea hydrolysis process, the equipment cost, land occupation and operating cost of the ammonium bicarbonate solid pyrolysis process all significantly reduce, showing good prospects for promotion and application.

At home and abroad, the locations suitable for developing concentrated solar power are mainly in desert areas. Dust in these environments may accumulate on the heat absorbing surfaces of the receiver in the solar power tower system, resulting in failure of the wall and coating of the pipe. To protect the heat absorbing walls, a coupled heat transfer model is developed for the sand-pipe, and the effects of several parameters on the wall temperature are investigated, such as the dust particle diameter, the contact areas between the dust and tube wall, and the concentrated solar energy flux density. The results show that, the influence of dust particles on the temperature of the heat-absorbing pipes is limited to a small area, but it will cause local high-temperature hot spots on the pipes. With a high concentrated solar energy flux density, a large dust particle diameter and a small contract area between the dust particles and the heat-absorbing pipes, both the temperature of the dust particle and the hot spot at the pipes will increase greatly. The temperature of the dust particles could exceed their melting point, forming calcium-magnesium-aluminum-silicate (CMAS) deposits, which means the receiver is at risk of CMAS corrosion. Meanwhile, the high-temperature hot spots on the heat-absorbing pipes will affect the local thermal stress distribution, exacerbating the damage to the receiver. Therefore, during actual operation, the cleanliness of the heat-absorbing pipe walls should be regularly inspected to avoid the accumulation of large-sized dust particles. The research results can provide technical guidance for the operation and maintenance of the receiver in the concentrated solar power system.

The structural characteristics of once-through steam generators and throttling assembly in the demonstration project of high-temperature gas cooled reactors were introduced, and the reasons and influencing factors of sediment blocking the throttling holes were analyzed. Moreover, the deposition law of corrosion products on the throttling holes of steam generators in high-temperature gas cooled reactors was studied by dynamic cyclic tests at high temperatures and high pressures, including the effects of different iron sources, iron mass fractions, flow rates, and pH values on throttling hole deposition. The results show that, free iron is the main precursor of throttling pore sediments. The phenomenon of throttling hole sediment increases with the iron mass fraction in water. With the increase of local flow velocity, the sedimentation rate of throttling hole increases at first and then decreases, and there exists a maximum deposition velocity range. As the pH value of the water increases from 9.1 to 9.7, the phenomenon of throttle hole deposition intensifies. It is found that appropriately reducing the pH value of feed water and optimizing throttle hole structure size (adjusting flow rate) are effective methods to inhibit the deposition and blockage of steam generator throttling components.

In modern power systems, unit coordinated control faces complex dynamic characteristics and is influenced by faults and external disturbances. To address this challenge, a fault-tolerant control scheme designed for unit coordinated control systems under fault conditions is proposed. Firstly, the transfer function of the unit system is derived using mechanism analysis, and a mathematical model incorporating actuator faults is developed. This model enables the analysis of transient response, stability, and dynamic performance of the system. Secondly, by integrating adaptive techniques with H_∞ control theory, an adaptive fault-tolerant guaranteed-cost tracking control method is designed. This method can automatically compensate for degraded signals when faults occur in the system while further enhancing the robustness and fault tolerance of the system through performance indicator optimization. It satisfies the combined requirements for tracking accuracy and dynamic response. Finally, a simulation is conducted using a 300 MW unit coordinated control system as an example. The results demonstrate that, compared with the conventional fault-tolerant control methods, the proposed approach exhibits superior dynamic tracking performance, disturbance rejection capability, and fault recovery ability. This study provides a reliable control solution for ensuring the safe and stable operation of unit systems in power systems.

The effectiveness of a data-driven model relies on the completeness of its training samples. For operating conditions beyond the scope of the training samples, the model’s generalization ability is compromised. Therefore, to develop a condenser model that can adapt to the wide load variation of the unit, it is essential for the training samples to involve a diverse range of power generation loads and ambient temperatures. However, achieving this complete dataset is difficult for newly-commissioned units because of their short operation time. To address these challenges, a method for characterizing condensing units using multi-fidelity data and transfer learning is proposed, even with incomplete data. In this method, a pre-trained model is firstly built based on the comprehensive operational dataset collected from a similar unit. On the basis of the pre-trained model, additional linear and nonlinear calibration networks are introduced. The calibration networks are updated through the incomplete data of newly constructed units, enabling the transfer of the pre-training model to the feature space that is adapted to the incomplete dataset. The effectiveness of this method is validated through the condenser of a 1 000 MW supercritical unit. The results indicate that, even with limited training samples, the method accurately predicts parameters such as condenser pressure and circulating water outlet temperature, with an average R² of 0.95, significantly outperforming the conventional data-driven model based on a single data set, of which the average R² is only 0.81.

To address the challenges of low diagnostic accuracy and poor interpretability for minority fault classes caused by imbalanced data distribution in coal mill pulverizing systems of coal-fired power plants, a fault diagnosis method integrating SMOTE data enhancement, Dirichlet prior smoothing, and Bayesian networks is proposed. The SMOTE technology expands the feature space of minority fault samples to alleviate data scarcity, while Dirichlet prior smoothing optimizes conditional probability estimation in Bayesian networks, resolving zero-probability issues caused by insufficient samples. A hierarchical Bayesian network architecture is constructed by incorporating domain knowledge and data-driven structure learning, enabling a dual-mode diagnosis strategy that combines rapid fault node inference with indirect attribute node analysis. The experimental results based on real industrial data demonstrate that the proposed method achieves high diagnostic accuracy and interpretability under imbalanced data scenarios. The solution provides real-time performance, precision, and transparency for coal mill fault diagnosis, offering significant engineering value.

To construct a prediction model for carbon emission from coal-fired power plants and address the problem of general lack of real-time elemental analysis for coal entering the furnace of coal-fired units, according to the in-furnace coal quality information of a million kilowatt unit in 2023, the low calorific value, volatile matter, and sulfur content were used as the basis for coal quality classification, K-means++ algorithm was used for clustering analysis, and correlation analysis was used to screen the input parameters of the carbon emission prediction model. The BP neural network suffered Bayesian optimization was used to construct carbon emission prediction models for each cluster data after clustering, and the models were tested for working conditions such as load increase and decrease. The results show that, the accuracy of the coal quality clustering model in predicting carbon emissions increases significantly. Compared with the non clustered model, the optimal cases of average root mean square error and average relative error reduce by about 53.4% and 49.2%, respectively. Especially under variable load conditions, the predicted results are more in line with the actual values. This indicates that the proposed method can not only effectively predict the carbon emissions of coal-fired power plants, but also maintain high accuracy in the case of complex and variable coal quality.

A hybrid prediction model combining enhanced grey wolf optimization algorithm (EGWO) and long short-term memory (LSTM) neural network is proposed to address the problem of low accuracy in predicting the mass concentration of NO_x at the outlet of selective catalytic reduction (SCR) denitrification reactors using conventional mechanism modeling methods. Firstly, based on principal component analysis (PCA), the raw data is processed and filtered to achieve dimensionality reduction of input variables. Then, the EGWO is used to optimize the hyperparameters of LSTM. Finally, the input variables are used as inputs for the EGWO-LSTM model to predict the mass concentration of NO_x at the outlet. Taking a 1 000 MW ultra supercritical thermal power unit in China as an example, simulation results show that the proposed model performs the best in error control, with root mean square error reduces by 50.36% compared to the conventional LSTM model, and by 76.14% compared to the BP model, and the mean absolute percentage error of the model is only 1.01%. The EGWO has fewer iterations and higher convergence accuracy compared to the GWO when converging to the optimal solution.