A method and system for intelligent scheduling of operating equipment in thermal power plant units

Abstract

This invention relates to the field of power system technology, and in particular to an intelligent scheduling method and system for the operation of thermal power plant units. The method includes the following steps: collecting multi-dimensional historical operating data of the thermal power plant units to obtain a panoramic historical operating dataset; constructing a digital twin model of the thermal power plant units based on the panoramic historical operating dataset to obtain a digital twin model; performing nonlinear dynamic system analysis on the panoramic historical operating dataset to obtain dynamic characteristic data of unit performance indicators; predicting the equipment performance degradation trend of the thermal power plant units based on the digital twin model and the dynamic characteristic data of unit performance indicators to obtain equipment performance degradation trend data; and performing smart grid load forecasting for the thermal power plant to obtain grid load demand forecasting data. This invention can adapt to rapid changes in grid load and real-time changes in unit operating status.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to an intelligent scheduling method and system for the operation of thermal power plant units. Background Technology

[0002] In the field of power system technology, especially in the operation and management of thermal power plants, traditional dispatching methods often rely on experience-based judgment and simple mathematical models. These methods have certain limitations when dealing with complex unit operation problems. With the continuous expansion of thermal power plant scale and the diversification of operating conditions, traditional dispatching methods are no longer sufficient to meet the requirements of high efficiency, economy, and environmental protection. First, existing dispatching systems often lack in-depth analysis and utilization of historical unit operating data, resulting in the inability to accurately predict unit performance degradation trends and changes in thermal efficiency. This directly affects the accuracy and foresight of dispatching.

[0003] Secondly, the models and algorithms commonly used by thermal power plants for grid load forecasting are likely to fail to fully capture the spatiotemporal characteristics and multi-scale features of grid load, thus affecting the accuracy and reliability of load forecasting. Inaccurate load forecasting leads to unreasonable load allocation among generating units, which in turn affects the operating efficiency and fuel consumption of the units.

[0004] Furthermore, existing intelligent fuel management systems for thermal power plants are inadequate in real-time monitoring, making it difficult to accurately grasp dynamic data on fuel inventory. This prevents the full utilization of fuel inventory information for optimization during unit operation scheduling, leading to low fuel efficiency and increased operating costs.

[0005] Finally, existing multi-objective optimization scheduling methods do not fully consider multi-dimensional factors such as unit performance degradation, thermal efficiency changes, and fuel inventory, resulting in scheduling schemes that cannot maximize energy utilization and minimize costs while ensuring safe unit operation. Summary of the Invention

[0006] Based on this, it is necessary for the present invention to provide an intelligent scheduling method and scheduling system for the operation equipment of thermal power plant units, so as to solve at least one of the above-mentioned technical problems.

[0007] To achieve the above objectives, a method for intelligent scheduling of operating equipment in thermal power plant units includes the following steps:

[0008] Step S1: Collect multi-dimensional historical operation data of the thermal power plant units to obtain a panoramic historical operation dataset of the units; construct a digital twin model of the thermal power plant units based on the panoramic historical operation dataset of the units to obtain the digital twin model of the units.

[0009] Step S2: Perform nonlinear dynamic system analysis on the unit's panoramic historical operation dataset to obtain dynamic characteristic data of unit performance indicators; based on the unit's digital twin model, predict the equipment performance degradation trend of the thermal power plant unit according to the dynamic characteristic data of the unit's performance indicators to obtain equipment performance degradation trend data of the unit.

[0010] Step S3: Perform smart grid load forecasting for the thermal power plant to obtain grid load demand forecasting data; optimize load allocation for the thermal power plant units based on the grid load demand forecasting data using the unit digital twin model, and estimate fuel consumption and combustion time to obtain unit fuel consumption forecasting data and unit fuel combustion time forecasting data.

[0011] Step S4: Perform thermodynamic efficiency analysis on the thermal power plant units based on the unit's panoramic historical operation dataset to obtain the current thermoelectric conversion efficiency data of the units; predict the thermal efficiency degradation gradient of the thermal power plant units based on the unit equipment performance degradation trend data and the current thermoelectric conversion efficiency data of the units to obtain thermal efficiency degradation gradient prediction data.

[0012] Step S5: Conduct real-time monitoring of the thermal power plant using the intelligent fuel management system to obtain dynamic data on the plant's fuel inventory; based on the thermal efficiency degradation gradient prediction data, perform multi-objective optimization scheduling of the thermal power plant units according to the predicted data on unit fuel consumption, predicted data on unit fuel combustion time, and dynamic data on the plant's fuel inventory to obtain a unit operation scheduling scheme.

[0013] This invention utilizes a digital twin model of the generating unit to accurately predict performance degradation trends and thermal efficiency changes by fully leveraging historical operating data. This data-driven approach overcomes the limitations of traditional experience-based judgment, making dispatching decisions more scientific and precise. By capturing the spatiotemporal characteristics and multi-scale features of grid load, the accuracy and reliability of load forecasting can be significantly improved. Accurate load forecasting provides a solid foundation for unit load allocation, contributing to more rational load distribution. Real-time monitoring through the intelligent fuel management system allows thermal power plants to accurately grasp the dynamic changes in fuel inventory. This function enhances the utilization of fuel inventory information during dispatching, helping to optimize fuel efficiency and reduce operating costs. Through dynamic characteristic analysis of unit performance indicators and prediction of equipment performance degradation trends, operating parameters can be adjusted in a timely manner to avoid efficiency reduction due to performance degradation, thereby improving the overall operating efficiency of the unit. Accurate prediction of thermal efficiency degradation and optimized dispatching help reduce unnecessary energy waste and pollutant emissions, meeting current requirements for environmental protection and sustainable development. In summary, this invention can adapt to rapid changes in grid load and real-time fluctuations in unit operating status, providing a flexible scheduling strategy to cope with various complex and changing operating conditions. Through automated data analysis and model prediction, it reduces manual intervention, improves the intelligence and automation level of thermal power plant operation management, and lowers the probability of human error.

[0014] Preferably, the present invention also provides an intelligent scheduling system for operating equipment of thermal power plant units, used to execute the intelligent scheduling method for operating equipment of thermal power plant units as described above, the intelligent scheduling system for operating equipment of thermal power plant units includes:

[0015] The unit modeling module is used to collect multi-dimensional historical operation data of thermal power plant units to obtain a panoramic historical operation dataset of the units; based on the panoramic historical operation dataset of the units, a digital twin model of the thermal power plant units is constructed to obtain the digital twin model of the units.

[0016] The performance analysis and prediction module is used to perform nonlinear dynamic system analysis on the panoramic historical operation dataset of the unit to obtain dynamic characteristic data of the unit's performance indicators; based on the unit's digital twin model, the module predicts the equipment performance degradation trend of the thermal power plant unit according to the dynamic characteristic data of the unit's performance indicators, and obtains the unit's equipment performance degradation trend data.

[0017] The load forecasting module is used to perform smart grid load forecasting for thermal power plants and obtain grid load demand forecasting data. Based on the unit digital twin model, the module optimizes the load allocation of thermal power plant units according to the grid load demand forecasting data, and estimates fuel consumption and combustion time to obtain unit fuel consumption forecasting data and unit fuel combustion time forecasting data.

[0018] The equipment degradation prediction module is used to perform thermodynamic efficiency analysis on thermal power plant units based on the unit's panoramic historical operation dataset to obtain the current thermoelectric conversion efficiency data of the unit; and to perform thermal efficiency degradation gradient prediction on thermal power plant units based on the unit's equipment performance degradation trend data and the current thermoelectric conversion efficiency data of the unit to obtain thermal efficiency degradation gradient prediction data.

[0019] The optimization scheduling module is used to monitor the thermal power plant in real time through the intelligent fuel management system and obtain dynamic data on the fuel inventory of the thermal power plant. Based on the thermal efficiency degradation gradient prediction data, the module performs multi-objective optimization scheduling of the thermal power plant units according to the predicted data of unit fuel consumption, the predicted data of unit fuel combustion time, and the dynamic data of the thermal power plant fuel inventory, and obtains the unit operation scheduling scheme.

[0020] In this invention, a digital twin model constructed using deep learning technology enables high-precision simulation and prediction of unit operating status, significantly enhancing the accuracy and response speed of dispatching. Simultaneously, multi-dimensional historical operating data acquisition and nonlinear dynamic system analysis provide a scientific basis for real-time monitoring and prediction of unit performance, making preventative maintenance and operational adjustments more precise. Furthermore, the application of intelligent algorithms in grid load forecasting ensures the reliability of load allocation and power generation planning. The system also improves energy utilization efficiency and reduces operating costs through real-time fuel inventory monitoring and multi-objective optimization algorithms. The integration of these technologies not only enhances the system's adaptability and flexibility but also reduces human error. Attached Figure Description

[0021] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0022] Figure 1 A flowchart illustrating the steps of an embodiment of an intelligent scheduling method for operating equipment of thermal power plant units is shown.

[0023] Figure 2 A detailed flowchart of step S4 of one embodiment is shown.

[0024] Figure 3 A detailed flowchart of step S5 of one embodiment is shown. Detailed Implementation

[0025] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0026] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0027] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0028] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides an intelligent scheduling method for operating equipment of thermal power plant units, comprising the following steps:

[0029] Step S1: Collect multi-dimensional historical operation data of the thermal power plant units to obtain a panoramic historical operation dataset of the units; construct a digital twin model of the thermal power plant units based on the panoramic historical operation dataset of the units to obtain the digital twin model of the units.

[0030] Step S2: Perform nonlinear dynamic system analysis on the unit's panoramic historical operation dataset to obtain dynamic characteristic data of unit performance indicators; based on the unit's digital twin model, predict the equipment performance degradation trend of the thermal power plant unit according to the dynamic characteristic data of the unit's performance indicators to obtain equipment performance degradation trend data of the unit.

[0031] Step S3: Perform smart grid load forecasting for the thermal power plant to obtain grid load demand forecasting data; optimize load allocation for the thermal power plant units based on the grid load demand forecasting data using the unit digital twin model, and estimate fuel consumption and combustion time to obtain unit fuel consumption forecasting data and unit fuel combustion time forecasting data.

[0032] Step S4: Perform thermodynamic efficiency analysis on the thermal power plant units based on the unit's panoramic historical operation dataset to obtain the current thermoelectric conversion efficiency data of the units; predict the thermal efficiency degradation gradient of the thermal power plant units based on the unit equipment performance degradation trend data and the current thermoelectric conversion efficiency data of the units to obtain thermal efficiency degradation gradient prediction data.

[0033] Step S5: Conduct real-time monitoring of the thermal power plant using the intelligent fuel management system to obtain dynamic data on the plant's fuel inventory; based on the thermal efficiency degradation gradient prediction data, perform multi-objective optimization scheduling of the thermal power plant units according to the predicted data on unit fuel consumption, predicted data on unit fuel combustion time, and dynamic data on the plant's fuel inventory to obtain a unit operation scheduling scheme.

[0034] This invention utilizes a digital twin model of the generating unit to accurately predict performance degradation trends and thermal efficiency changes by fully leveraging historical operating data. This data-driven approach overcomes the limitations of traditional experience-based judgment, making dispatching decisions more scientific and precise. By capturing the spatiotemporal characteristics and multi-scale features of grid load, the accuracy and reliability of load forecasting can be significantly improved. Accurate load forecasting provides a solid foundation for unit load allocation, contributing to more rational load distribution. Real-time monitoring through the intelligent fuel management system allows thermal power plants to accurately grasp the dynamic changes in fuel inventory. This function enhances the utilization of fuel inventory information during dispatching, helping to optimize fuel efficiency and reduce operating costs. Through dynamic characteristic analysis of unit performance indicators and prediction of equipment performance degradation trends, operating parameters can be adjusted in a timely manner to avoid efficiency reduction due to performance degradation, thereby improving the overall operating efficiency of the unit. Accurate prediction of thermal efficiency degradation and optimized dispatching help reduce unnecessary energy waste and pollutant emissions, meeting current requirements for environmental protection and sustainable development. In summary, this invention can adapt to rapid changes in grid load and real-time fluctuations in unit operating status, providing a flexible scheduling strategy to cope with various complex and changing operating conditions. Through automated data analysis and model prediction, it reduces manual intervention, improves the intelligence and automation level of thermal power plant operation management, and lowers the probability of human error.

[0035] In this embodiment, multi-dimensional historical operating data of thermal power plant units can be collected in real time through a data acquisition system (such as a SCADA system). This data includes operating parameters of the boiler, steam turbine, generator, and condenser, such as temperature, pressure, and flow rate, thereby constructing a panoramic historical operating dataset of the unit. Next, using this data, a digital twin model of the unit is constructed using a deep learning framework (such as TensorFlow or PyTorch). This model can simulate the unit's operating state and performance with high accuracy. Subsequently, nonlinear dynamic system analysis is performed on the panoramic historical operating dataset of the unit, using a system identification toolbox such as MATLAB to identify and extract dynamic feature data of the unit's performance indicators. Further, this dynamic feature data is input into the unit's digital twin model, and forward extrapolation and multi-scale degradation trend analysis methods are used to predict equipment performance degradation trends, obtaining unit equipment performance degradation trend data. Simultaneously, smart grid load forecasting technologies, such as the ARIMA model based on time series analysis or the LSTM model based on deep learning, are used to predict grid load demand, obtaining grid load demand forecast data. Combining these predicted data and the unit's digital twin model, load allocation is optimized using optimization algorithms (such as linear programming or genetic algorithms), while simultaneously estimating fuel consumption and combustion time to obtain predicted fuel consumption and combustion time data. Furthermore, based on the unit's comprehensive historical operation dataset, thermodynamic efficiency analysis is performed using software such as Aspen Plus to obtain current thermoelectric conversion efficiency data. Combining unit equipment performance degradation trend data and current thermoelectric conversion efficiency data, a thermal efficiency degradation prediction model is constructed to predict thermal efficiency degradation gradients, yielding predicted thermal efficiency degradation gradient data. Finally, the power plant's fuel inventory status is monitored in real-time through an intelligent fuel management system to obtain dynamic fuel inventory data. Using this data, along with the previously obtained predicted fuel consumption, combustion time, and thermal efficiency degradation gradient data, a multi-objective optimization algorithm (such as Multi-Objective Particle Swarm Optimization, MOPSO) is applied for comprehensive consideration to formulate a unit operation scheduling plan. For a detailed example of this embodiment, please refer to the specific implementation details of steps S1-S5.

[0036] Preferably, step S1 includes the following steps:

[0037] Step S11: Obtain the unit equipment topology list;

[0038] Specifically, a list of unit equipment topology can be obtained by interacting with the power plant's engineering documents, design drawings, or equipment management system. Detailed configuration information of the unit, including the connection methods and hierarchical relationships of each equipment component, can be retrieved using engineering information management software or database query tools. Furthermore, graphical user interfaces, such as configuration software, can be used to visually display the unit's topology.

[0039] Step S12: Collect multi-dimensional time-series data of the thermal power plant units according to the unit equipment topology list to obtain the unit panoramic historical operation dataset. The unit panoramic historical operation dataset includes at least boiler historical operation data, steam turbine historical operation data, generator historical operation data and condenser historical operation data.

[0040] Specifically, the types and frequencies of data to be collected can be determined, such as operating parameters of boilers, steam turbines, generators, and condensers, including temperature, pressure, and flow rate. These parameters are then monitored and recorded in real time using a Data Acquisition System (DAS) or SCADA system. The required time-series data can be extracted from different data sources by writing custom scripts or using automated data acquisition tools. This data is then transferred to a central database or data lake for storage and initial processing. Ultimately, a comprehensive historical operational dataset of the unit is obtained.

[0041] Step S13: Perform deep neural network modeling on the historical operating data of the boiler, steam turbine, generator and condenser respectively to obtain the learning model set of the unit subsystem. The learning model set of the unit subsystem includes the boiler learning model, steam turbine learning model, generator learning model and condenser learning model.

[0042] Specifically, the historical operational data of various devices can be preprocessed using Python's Pandas library, including data cleaning, standardization, and splitting into training and testing sets. Next, a suitable deep learning framework, such as TensorFlow or PyTorch, is chosen to build a deep neural network model for each device type. For example, a Long Short-Term Memory (LSTM) network can be used to process time-series data, or a Convolutional Neural Network (CNN) can be used to extract local features. Through training, each subsystem learning model can capture the key operational characteristics and patterns of its corresponding device. Ultimately, a set of subsystem learning models is obtained, including boiler learning models, steam turbine learning models, generator learning models, and condenser learning models.

[0043] Step S14: Integrate the boiler learning model, steam turbine learning model, generator learning model and condenser learning model into a graph neural network to obtain the overall graph neural network model of the unit;

[0044] Specifically, the boiler learning model, steam turbine learning model, generator learning model, and condenser learning model can be integrated into a single graph neural network to form a comprehensive graph neural network model of the entire unit. First, a graph representation learning library, such as the Deep Graph Library (DGL), is used to construct the graph structure of the unit, where nodes represent equipment components and edges represent interactions between components. Then, the output of each subsystem's learning model is used as the feature input to the graph nodes, and graph neural network algorithms (Graph Convolutional Network (GCN) or Graph Attention Network (GAT)) are applied to learn patterns on the graph structure. In this way, the graph neural network can comprehensively consider the relationships and interactions between the various subsystems, generating a comprehensive graph neural network model of the entire unit.

[0045] Step S15: Perform spatiotemporal feature extraction and abnormal pattern recognition on the unit's panoramic historical operation dataset to obtain unit operation feature data and unit abnormal pattern data;

[0046] Specifically, Python's SciPy library can be used to analyze the temporal characteristics of the historical operational dataset of the generating units, such as periodicity, trends, and seasonality. Then, machine learning algorithms, such as Isolation Forest or Autoencoder, are used to identify anomalous patterns in the data. These algorithms can reveal outliers or behaviors that do not conform to normal patterns. Simultaneously, deep learning techniques, such as Recurrent Neural Networks (RNNs) or Long Short-Term Memory Networks (LSTMs), are used to extract spatiotemporal features from the data. Through these methods, both generating unit operational characteristic data and generating unit anomalous pattern data can be obtained.

[0047] Step S16: Use the overall unit diagram neural network model to perform transfer learning on the unit operation characteristic data and the unit abnormal mode data, and perform fine-tuning and optimization to obtain the unit digital twin model.

[0048] For a detailed implementation process of this embodiment, please refer to the sub-step of step S16.

[0049] This invention achieves a deep understanding and accurate prediction of the performance of thermal power plant units through meticulous data acquisition and deep neural network modeling. The technological advancement lies in its ability to comprehensively integrate historical operating data from boilers, steam turbines, generators, and condensers, constructing a multi-dimensional, panoramic dataset of unit operation. By leveraging the integration advantages of graph neural networks, it not only simulates the mutual influence of various components within the unit but also enhances the predictive ability for overall performance. Through the extraction of spatiotemporal features and anomaly pattern recognition, the system's ability to detect potential operational problems early is further enhanced, improving the accuracy of fault prediction. The combination of transfer learning and fine-tuning optimization ensures the model's rapid adaptability to new operating conditions and the effectiveness of scheduling strategies. Ultimately, the establishment of the digital twin model provides a highly realistic and reliable virtual environment for intelligent scheduling, thereby significantly improving the scientific nature of decision-making and the accuracy of scheduling.

[0050] Preferably, step S16 includes the following steps:

[0051] Step S161: Perform feature dimensionality reduction and extraction on the unit operation feature data and unit anomaly mode data to obtain unit operation feature dimensionality reduction data and unit anomaly mode dimensionality reduction data;

[0052] Specifically, the Python Pandas library can be used to conduct preliminary exploratory analysis of the unit operation characteristic data and unit anomaly pattern data to understand the basic structure and distribution of the data. Next, feature dimensionality reduction techniques, such as Principal Component Analysis (PCA) or Linear Discriminant Analysis (LDA), are applied using the Scikit-learn library to reduce the dimensionality of the data. These methods reduce the dimensionality of the feature space by identifying the principal components in the data while preserving the variability of the original data. Furthermore, feature selection algorithms, such as Recursive Feature Elimination (RFE) or Selective Random Forest, are used to further filter out the features most influential on unit operation and anomaly pattern identification. Finally, dimensionality-reduced data of unit operation characteristics and unit anomaly patterns are obtained, which still reflect the key operating characteristics and anomaly patterns of the unit.

[0053] Step S162: Perform hierarchical freezing preprocessing on the overall neural network model of the unit to obtain the preprocessed network model of the unit;

[0054] Specifically, a graph neural network library (Deep Graph Library (DGL) or PyTorchGeometric) can be used to initialize the overall graph neural network model for the unit. Then, a layer-by-layer freezing technique is employed, fixing some weights of the graph neural network layer by layer. This is typically achieved by setting parameter update rules for each network layer. Layer-by-layer freezing allows the network to converge quickly in the early stages of training, while fine-tuning all layers in later training to improve the model's generalization ability. This method yields a pre-processed network model for the unit that can more effectively capture complex patterns in the data during subsequent training.

[0055] Step S163: Construct a target domain dataset from the dimensionality reduction data of unit operation characteristics and the dimensionality reduction data of unit anomaly modes to obtain a balanced target domain dataset for the unit.

[0056] Specifically, data augmentation techniques, such as the SMOTE algorithm, can be used to balance the data distribution of different categories in the dimensionality-reduced data of unit operation characteristics and the dimensionality-reduced data of unit anomaly patterns, especially when there is limited data on anomaly patterns. Next, data segmentation techniques are used to divide the dataset into training, validation, and test sets, ensuring that the model can be effectively evaluated on unseen data. Finally, a balanced and diverse balanced dataset of the unit target domain is constructed.

[0057] Step S164: Perform domain adaptive layer integration on the preprocessed network model of the unit to obtain the domain adaptive network model of the unit;

[0058] Specifically, domain-adaptive layers can be defined, which learn the mapping relationship between the source and target domains. This can be implemented using deep learning frameworks such as TensorFlow or PyTorch. These layers are then integrated into the preprocessing network model of the generator set, enabling the model to adapt to different operating conditions and environments. Transfer learning techniques, particularly domain-adaptive methods such as Domain Adversarial Neural Networks (DANNs) or Domain-Adversarial Training (DAT), are used to train the ensemble model, ultimately yielding the generator set domain-adaptive network model. In this way, the generator set domain-adaptive network model can reduce the distributional differences between the source and target domains, improving the model's generalization ability in new environments.

[0059] Step S165: Use the unit domain adaptive network model to perform adversarial transfer learning on the unit target domain balanced dataset to obtain the unit transfer optimization network model;

[0060] Specifically, a balanced dataset of the crew target domain can be prepared and initially trained using a crew domain adaptive network model. Then, an adversarial method, such as the Wasserstein domain adversarial method, is employed to further train the crew domain adaptive network model. In this method, two networks—a domain discriminator and a feature extractor—compete with each other. The domain discriminator attempts to distinguish between data from the source and target domains, while the feature extractor attempts to minimize the accuracy of the domain discriminator, thereby learning more generalizable feature representations. Through this adversarial training, a crew migration optimization network model can be obtained.

[0061] Step S166: Perform incremental fine-tuning optimization on the unit migration optimization network model to obtain the unit fine-tuning optimization network model;

[0062] Specifically, a small, balanced dataset of the unit target domain can be used to fine-tune certain layers of the unit migration optimization network model, typically a few layers near the output layer, to adapt to specific operating conditions. Then, the number and depth of the fine-tuned layers are gradually increased while the learning rate is decreased to ensure the model does not overfit to specific noise in the target domain. A validation set is used to monitor the fine-tuning process to ensure continuous improvement in model performance. This gradual fine-tuning yields the unit migration optimization network model.

[0063] Step S167: Perform knowledge distillation and integration on the unit fine-tuning and optimization network model to obtain the unit digital twin model.

[0064] In this embodiment, knowledge distillation is applied to transfer the ability to fine-tune and optimize a network model for the generator set to a simplified student network, with the aim of creating a highly efficient digital twin model of the generator set. The initial phase involves selecting a teacher model that performs exceptionally well on historical data. Subsequently, a student network with a small number of parameters is constructed and initialized using a Python deep learning framework (TensorFlow or PyTorch). During training, the student network combines a traditional loss function with distillation loss, the latter measuring the difference between the student and teacher network outputs through Kullback-Leibler divergence or mean squared error. This allows the student network to learn high-level feature representations from the teacher network without accessing all the training data. During training, regularization measures such as Dropout and weight decay are implemented to prevent overfitting. The knowledge-distilled student network is transformed into a compact and efficient digital twin model of the generator set, capable of accurately simulating generator set operation and suitable for critical tasks such as real-time monitoring, fault diagnosis, and performance prediction.

[0065] This invention optimizes the representativeness and predictability of data through feature reduction and extraction techniques while reducing computational resource consumption. Layered freezing preprocessing and domain-adaptive layer integration enhance the model's generalization and adaptability, ensuring its stability and reliability under different operating environments. The application of adversarial transfer learning further improves the model's ability to quickly learn from new data. Progressive fine-tuning ensures the model maintains its original performance level while absorbing new data. The application of knowledge distillation technology enables efficient transfer of knowledge from complex networks, improving the model's compactness and practicality. Ultimately, the establishment of the digital twin model provides a highly realistic and reliable virtual environment for intelligent scheduling and decision-making, improving the model's practicality and flexibility, enabling it to better cope with the changing needs of industrial applications, thereby promoting the in-depth development of intelligent decision-making.

[0066] Preferably, step S2 includes the following steps:

[0067] Step S21: Perform time series segmentation and noise filtering on the unit's panoramic historical operation dataset to obtain the unit's preprocessed time series operation dataset;

[0068] Specifically, a data acquisition system, such as a SCADA system, can be used to collect a comprehensive historical operational dataset of the generator set. This data includes, but is not limited to, time-series records of key parameters such as temperature, pressure, and flow. The Pandas library in Python can be used to process the data, dividing it into fixed-length sequences by setting time windows. Next, filtering algorithms, such as moving average filtering or Kalman filtering, are applied to remove noise. These algorithms can remove high-frequency noise components from the data while retaining the main trends and periodic information. The final result is a preprocessed time-series operational dataset of the generator set.

[0069] Step S22: Extract nonlinear dynamic features from the preprocessed time-series operation dataset of the unit to obtain the nonlinear dynamic feature data of the unit;

[0070] Specifically, the NumPy library in Python can be used to clean the preprocessed time-series dataset of the unit, including removing outliers and filling in missing values. Then, based on physical laws and engineering principles, system dynamic equations describing the interactions of various components of the unit are constructed. These equations originate from the principles of energy and momentum conservation and reflect the complex nonlinear characteristics of the unit's operation. For complex systems, empirical models or machine learning methods can be combined to derive these equations when necessary. After obtaining the dynamic equations, the odeint function in the SciPy library is used to numerically integrate the equations, simulating the temporal evolution of the unit's state, and the tqdm library is used to monitor the computation progress. Based on the numerical integration results, the Lyapunov exponent is calculated to quantitatively assess the system's degree of chaos and sensitivity to initial conditions. Furthermore, Takens' theorem is used to reconstruct the phase space, selecting appropriate delay times and embedding dimensions to recover the system's multidimensional state from the one-dimensional time series and constructing attractors in the phase space to reveal the system's long-term behavioral patterns. Finally, the attractors are visualized using the matplotlib and seaborn libraries, displaying the distribution and trajectories of data points in the phase space through scatter plots and phase diagrams, intuitively identifying the system's potential nonlinear dynamic behavior. Through these steps, nonlinear dynamic characteristic data, including Lyapunov exponents, attractor geometry, and system periodic solutions, were successfully extracted from the dataset.

[0071] Step S23: Perform multivariate coupling analysis on the preprocessed time-series operation dataset of the unit to obtain the coupling characteristic data of the unit parameters;

[0072] Specifically, the preprocessed time-series dataset for the unit can be loaded using Python's Pandas library for initial data exploration, including calculating the correlation coefficient matrix to assess the linear relationships between parameters. Subsequently, Granger causality analysis is performed using the Statsmodels library to determine the causal relationships between parameters. Information theory tools, such as mutual information measurements, are used to assess the strength of nonlinear coupling between variables. Furthermore, principal component analysis (PCA) or partial least squares regression (PLSR) methods from Python's Scikit-learn library are used to further explore the intrinsic connections between variables. Through these analyses, the coupling characteristics of the unit parameters can be extracted.

[0073] Step S24: Perform state-space reconstruction on the unit parameter coupling characteristic data to obtain the unit state-space reconstruction data;

[0074] Specifically, Takens' delayed embedding theorem can be applied to transform one-dimensional time series data (i.e., unit parameter coupled feature data) into a high-dimensional state-space representation, facilitating the capture of the system's dynamic characteristics. The specific operations involve determining appropriate time delays and embedding dimensions by scanning combinations of delay time pairs and embedding dimension pairs to find the parameters that best reconstruct the state space, ultimately obtaining the reconstructed unit state-space data. In Python, the Scikit-learn library can be used to assist in performing delayed embedding and dimensional expansion of the data.

[0075] Step S25: Perform dynamic pattern recognition on the unit state space reconstruction data to obtain the unit dynamic pattern feature data;

[0076] Specifically, time series analysis and pattern recognition techniques can be used to discover recurring behaviors or periodic features in unit state-space reconstructed data. This includes using clustering algorithms from Python's Scikit-learn library, such as K-means or hierarchical clustering, to identify different dynamic behavior groups within the reconstructed state-space data. Fourier transforms or wavelet transforms are employed to analyze the spectral characteristics of the time series, and time delay plots and phase plots are used to visualize the system's dynamic behavior. Through these methods, dynamic pattern characteristics of the unit can be extracted from the reconstructed state-space data, revealing the unit's operational patterns and potential abnormal behaviors under different operating conditions.

[0077] Step S26: Perform system stability analysis on the dynamic mode characteristic data of the unit to obtain the unit stability index data;

[0078] Specifically, the detailed process of system stability analysis begins with a thorough review of the unit's dynamic mode characteristic data under varying operating conditions. This data is imported, cleaned, and formatted using Python's Pandas library. Subsequently, mathematical modeling of the system dynamics is performed using the SciPy library, constructing a Jacobian matrix containing the first-order partial derivatives of the system state variables to reveal the interactions between them. After constructing the Jacobian matrix, its eigenvalues ​​are calculated, where all eigenvalues ​​with negative real parts indicate the linear stability of the system at the current operating point. Further nonlinear stability analysis, such as the Lyapunov method, is employed to demonstrate the decay trend of system energy over time by constructing or finding suitable Lyapunov functions. Simultaneously, time series analysis methods, such as the ARIMA model, are used to analyze the statistical characteristics of the data and identify periodicity and potential instability factors. These comprehensive analytical methods provide unit stability index data, including the maximum Lyapunov exponent, decay rate, and periodic behavior characteristics.

[0079] Step S27: Integrate the nonlinear dynamic characteristic data, the coupling characteristic data of the unit parameters, and the stability index data of the unit to obtain the dynamic characteristic data of the unit performance index;

[0080] Specifically, Python's Pandas library can be used to process and integrate nonlinear dynamic characteristic data of the generating unit, coupled characteristic data of generating unit parameters, and data on generating unit stability indices. Machine learning algorithms from the Scikit-learn library, such as random forests or support vector machines (SVM), can be used to evaluate the impact of different data sources on generating unit performance and to perform feature selection. Deep learning techniques, such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs), can be employed to learn the complex nonlinear relationships between the data. Through information fusion, dynamic characteristic data of generating unit performance indices are obtained, which comprehensively reflect the unit's operating performance and health status under different operating conditions.

[0081] Step S28: Based on the unit's digital twin model, predict the equipment performance degradation trend of the thermal power plant unit according to the dynamic characteristic data of the unit's performance indicators, and obtain the unit's equipment performance degradation trend data.

[0082] For a detailed implementation process of this embodiment, please refer to the sub-step of step S28.

[0083] This invention significantly improves the quality of the dataset through time series segmentation and noise filtering, laying a clear and accurate foundation for in-depth analysis. Furthermore, in-depth exploration of nonlinear dynamic characteristics enhances the understanding of the complex operating mechanisms of the unit, while multivariate coupling analysis provides a comprehensive perspective on the overall performance of the unit. The application of state-space reconstruction technology enables the system to gain deeper insights into unit behavior, and the accuracy of dynamic pattern recognition provides a reliable basis for performance analysis and fault diagnosis. In addition, real-time monitoring of system stability ensures timely response to the unit's stability status, while information fusion ensures the comprehensiveness and balance of performance indicators. Finally, the prediction of equipment performance degradation trends based on a digital twin model provides a forward-looking insight into the long-term operating status of the unit, offering a scientific basis for the formulation of maintenance and optimization strategies, and overall improving the intelligence and automation level of thermal power plant unit operation management.

[0084] Preferably, step S28 includes the following steps:

[0085] Step S281: Input the dynamic characteristic data of the unit performance indicators into the digital twin model of the unit for forward extrapolation to obtain the dynamic simulation data of the unit performance;

[0086] Specifically, dynamic characteristic data of unit performance indicators can be input into the unit's digital twin model. This model is a physics-based model or a data-driven machine learning model that can simulate the unit's performance under different conditions. Through forward propagation of the model, dynamic simulation data of unit performance can be obtained, which simulates the unit's performance changes over a future period.

[0087] Step S282: Perform multi-scale degradation trend analysis on the dynamic simulation data of unit performance to obtain multi-scale degradation characteristic data of the unit;

[0088] Specifically, a multi-scale degradation trend analysis is conducted. The aim of this analysis is to identify the degradation characteristics of unit performance at different time scales. First, time series analysis is performed on the dynamic simulation data of unit performance using Python's Pandas library, including sliding window calculation and time decomposition, to extract features at different time scales. Then, statistical methods or machine learning algorithms, such as random forests or gradient boosting trees (GBDT), are applied to identify key influencing factors of performance degradation. Furthermore, multi-scale analysis tools, such as wavelet transforms or multi-scale entropy methods, can be used to reveal the complexity of the data at different scales. Through these analyses, multi-scale degradation characteristic data of the unit can be obtained.

[0089] Step S283: Perform nonparametric regression prediction on the multi-scale degradation characteristic data of the unit to obtain preliminary degradation trend prediction data of the unit;

[0090] Specifically, appropriate nonparametric regression algorithms can be selected to model the data, such as Kernel Ridge Regression (RFR) or Extreme Gradient Boosting Tree (XGBoost), to predict the degradation trend of unit performance. The model is initialized and trained using the RFR implementation in the Scikit-learn library or the XGBoost library. During model training, multi-scale degradation feature data of the unit is used as input, and model parameters are optimized through cross-validation and parameter tuning to ensure that the model can capture the degradation trend in the data. After training, the model is used to perform regression analysis on historical data to obtain preliminary degradation trend prediction data for the unit.

[0091] Step S284: Perform hybrid model integration optimization on the preliminary degradation trend prediction data of the unit to obtain the comprehensive degradation trend prediction data of the unit;

[0092] Specifically, suitable ensemble learning algorithms can be selected, such as random forests, gradient boosting trees, or XGBoost. These algorithms combine the predictions of multiple models to reduce overfitting and variance of a single model. Ensemble models are built using the Scikit-learn library by setting different model parameters and ensemble strategies (such as bagging, boosting, or stacking). During the ensemble process, cross-validation is used to evaluate the performance of different ensemble configurations, and the optimal ensemble model is selected. After model training and optimization, the ensemble model is used to comprehensively analyze the preliminary degradation trend prediction data to obtain the comprehensive degradation trend prediction data for the unit. This data is obtained through the synergistic effect of multiple models and can more comprehensively reflect the degradation of unit performance.

[0093] Step S285: Quantify and evaluate the uncertainty of the unit's comprehensive degradation trend prediction data to obtain the unit equipment performance degradation trend data.

[0094] Specifically, statistical analysis methods, such as the calculation of confidence intervals or prediction intervals, can be used to quantify the uncertainty of the overall degradation trend prediction data for the unit. Next, sensitivity analysis is applied to identify key factors affecting the overall degradation trend prediction data and to assess the impact of different input parameters on the prediction output. Finally, through these analyses, the unit equipment performance degradation trend data is obtained, which includes not only the predicted values ​​but also uncertainty measures related to the prediction, such as standard deviation, confidence interval width, or the shape of the probability distribution.

[0095] This invention generates accurate dynamic simulation data of unit performance through forward extrapolation using a digital twin model, laying a solid foundation for subsequent analysis. Multi-scale degradation trend analysis deepens the understanding of long-term changes in unit performance, while the flexible non-parametric regression prediction method, unrestricted by traditional models, improves the adaptability and accuracy of the predictions. Hybrid model integration optimization further enhances the overall performance of the predictions, ensuring the reliability and practicality of the results. Uncertainty quantification and assessment provide important evidence for risk management, enhancing operators' understanding of the prediction's credibility. In this invention, a forward-looking perspective on equipment performance management helps in planning maintenance strategies in advance, reducing downtime risks. Data-driven maintenance decision support improves the efficiency of resource allocation, while the significant improvement in the practicality and accuracy of the prediction model provides strong technical support for the long-term operation of thermal power plants. Furthermore, the comprehensive assessment considering uncertainty enhances the system's adaptability and robustness to data fluctuations and external disturbances.

[0096] Preferably, step S3 includes the following steps:

[0097] Step S31: Collect historical load data from thermal power plants to obtain historical power grid load time series data;

[0098] Specifically, time-series data of historical grid load can be exported from the monitoring system or database of the thermal power plant.

[0099] Step S32: Extract spatiotemporal features from historical power grid load time series data and perform load pattern recognition to obtain power grid load time feature data and power grid load spatial feature data;

[0100] Specifically, the Pandas library in Python can be used to perform time series analysis on historical power grid load data to identify periodic characteristics such as daily, weekly, and seasonal variations. Then, relevant algorithms from the Scikit-learn machine learning library, such as K-means or hierarchical clustering, are used to perform pattern recognition on the load data, classifying similar load patterns into different categories. Furthermore, time series decomposition methods are used to decompose the load data into trend, seasonal, and random components. Through these analyses, the temporal and spatial characteristics of power grid load are ultimately obtained.

[0101] Step S33: Perform feature fusion on the power grid load temporal feature data and the power grid load spatial feature data to obtain the power grid load multimodal feature data;

[0102] Specifically, time-series data of power grid load can be obtained from the historical operation records of thermal power plants. This data is automatically collected by the monitoring system at specific time intervals, recording the power grid load values ​​at different points in time. Using Python's Pandas library, this data is loaded and indexed with timestamps for chronological analysis. Subsequently, the time series is divided into fixed-length time windows for feature extraction. When comprehensively analyzing the temporal and spatial features of power grid load, the PCA method from the Scikit-learn library is used for dimensionality reduction to extract the most critical features for predicting power grid load. The RFE algorithm is then applied to further improve the accuracy of feature selection, identifying the features that contribute the most to the prediction. Finally, a comprehensive feature set is constructed by combining temporal and spatial features through a custom fusion strategy. This set comprehensively reflects the multi-dimensional characteristics of power grid load through the weighting, combination, or transformation of features, generating multimodal feature data of power grid load.

[0103] Step S34: Perform deep probabilistic load modeling on the multimodal characteristic data of the power grid load to obtain the power grid load probability distribution model;

[0104] Specifically, deep learning libraries in Python, such as TensorFlow or Keras, can be used to build a deep learning model, such as a Long Short-Term Memory (LSTM) network or a Convolutional Neural Network (CNN), to learn the temporal dynamics and spatial distribution of multimodal characteristics of power grid load. During training, historical load data is used to adjust model parameters, and cross-validation is used to evaluate the model's performance. After model training is complete, the model is used to make probabilistic predictions on the multimodal characteristics of power grid load, resulting in a power grid load probability distribution model. This model can predict the potential distribution and uncertainty of the load.

[0105] Step S35: Perform multi-model integration on the power grid load probability distribution model to obtain an adaptive probability load prediction model, and use the adaptive probability load prediction model to perform smart grid load prediction on thermal power plants to obtain power grid load demand prediction data.

[0106] Specifically, a range of different prediction models can be selected, each possessing unique algorithms and predictive capabilities. These models include traditional statistical models such as ARIMA, as well as advanced machine learning models such as Random Forest, Gradient Boosting Tree (GBDT), Support Vector Machine (SVM), and deep learning models such as Long Short-Term Memory (LSTM) networks. First, these basic models are implemented and trained using Python's Scikit-learn library. Each model is trained independently based on historical grid load data, and its parameters are tuned to optimize predictive performance. Subsequently, ensemble learning techniques, such as stacking or super learner frameworks, are employed to integrate the predictions from these models. In stacking, a meta-model (typically a simple model like logistic regression) can be trained to combine the predictions from the various basic models, serving as the final prediction output and resulting in an adaptive probabilistic load prediction model. This model is then used for smart grid load prediction of thermal power plants, yielding grid load demand prediction data that includes not only expected load values ​​but also probability distribution information.

[0107] Step S36: Based on the digital twin model of the unit, optimize the load allocation of the thermal power plant unit according to the grid load demand forecast data, and estimate the fuel consumption and combustion time to obtain the unit fuel consumption forecast data and the unit fuel combustion time forecast data.

[0108] For a detailed implementation process of this embodiment, please refer to the sub-step of step S36.

[0109] This invention establishes a detailed data foundation through comprehensive historical load data analysis, and then enhances the accuracy and resolution of load forecasting through precise spatiotemporal feature extraction technology. By applying multi-scale feature fusion technology, information from different time and spatial scales is integrated, significantly improving the ability to capture complex changes. The application of deep probabilistic load modeling further improves the accuracy and reliability of forecasts, while the flexibility of the adaptive probabilistic load forecasting model ensures the timeliness and adaptability of the forecast results. The multi-model integration method enhances the robustness of the system and reduces the limitations of a single model. Load allocation optimization based on a digital twin model achieves intelligent and refined unit operation scheduling, improving grid operating efficiency. Furthermore, the optimization algorithm's accurate estimation of fuel consumption and combustion time helps reduce operating costs and improve energy utilization efficiency. Overall, this invention, through rapid and accurate load forecasting and optimized scheduling, not only improves the response speed and flexibility of the power grid but also provides strong technical support for the efficient, economical, and environmentally friendly operation of the power grid.

[0110] Preferably, step S36 includes the following steps:

[0111] Step S361: Input the power grid load demand forecast data into the unit digital twin model to plan the multi-objective load allocation optimization algorithm and obtain the unit load allocation optimization algorithm;

[0112] Specifically, Python optimization libraries, such as SciPy.optimize, can be used to construct the optimization problem, define the objective function (e.g., minimizing fuel consumption and reducing emissions) and constraints (e.g., unit operating limitations and grid demand). Then, a suitable optimization algorithm, such as linear programming, mixed-integer programming, or particle swarm optimization (PSO), is selected to solve the optimization problem. Through algorithmic planning, a set of unit load allocation optimization algorithms is obtained.

[0113] Step S362: Use the unit load allocation optimization algorithm to generate load allocation schemes for thermal power plant units, obtain a set of candidate load allocation schemes, and perform rapid performance simulation of each candidate load allocation scheme through the unit digital twin model to obtain the unit load allocation simulation time series data.

[0114] Specifically, the unit load allocation optimization algorithm can be applied to actual unit operation data, including the performance parameters and technical specifications of each unit. The Python programming language is used, combined with the Pandas library for data processing, the Scikit-learn library for machine learning-assisted decision-making, and the SciPy library for optimization calculations. The unit load allocation optimization algorithm generates multiple candidate load allocation schemes that consider grid load demand and unit operating efficiency. Then, each candidate scheme is simulated using a digital twin model of the unit to quickly evaluate its performance. This step uses simulation software, such as MATLAB / Simulink or Python's SimPy library, to simulate the unit's operation under different load allocations. The simulation results provide time-series data of the unit load allocation simulation, reflecting the unit's performance under different schemes.

[0115] Step S363: Dynamically update the unit digital twin model based on the unit load distribution simulation time series data to obtain the unit operation dynamic constraint condition set;

[0116] Specifically, Python's Pandas library can be used to process and analyze the time-series data of unit load distribution simulation, identifying key performance indicators and potential operational constraints during the simulation process. Then, based on these indicators and constraints, the parameters and settings of the unit's digital twin model are adjusted to ensure the model reflects the dynamic changes the unit will encounter in actual operation. Model development tools, such as MATLAB or Python's deep learning libraries (e.g., TensorFlow or Keras), are used to update the unit's digital twin model to include the new constraints. The updated model will provide a more accurate set of dynamic constraints for unit operation.

[0117] Step S364: Use the preset unit performance evaluation index system to perform multi-dimensional performance evaluation on the unit load distribution simulation time series data to obtain the load distribution scheme performance evaluation matrix;

[0118] Specifically, a performance evaluation index system for the generating units can be defined, such as efficiency, stability, and environmental impact. Then, the Pandas library in Python is used to organize and analyze the time-series data of the unit load allocation simulation, calculating the value of each index. Next, multi-criteria decision analysis methods, such as weighted summation or the analytic hierarchy process (AHP), are employed to comprehensively score each index. This method yields a performance evaluation matrix for the load allocation schemes, which includes performance scores for different load allocation schemes across multiple dimensions.

[0119] Step S365: Select the optimal scheme based on the set of dynamic constraints for unit operation and the performance evaluation matrix of load allocation scheme to obtain the optimal load allocation scheme for the unit;

[0120] Specifically, optimization tools from Python's SciPy library can be used to process the dynamic constraint set of unit operation and the performance evaluation matrix of load allocation schemes. Then, an objective function is defined, such as minimizing cost or maximizing energy efficiency, while satisfying all operational constraints. Next, optimization algorithms, such as linear programming or genetic algorithms, are applied to search for the optimal load allocation scheme that satisfies all conditions. This method yields the optimal load allocation scheme for the unit.

[0121] Step S366: Based on the digital twin model of the unit, construct the fuel consumption estimation model and the combustion time estimation model respectively to obtain the unit fuel consumption prediction model and the unit fuel combustion time inference model;

[0122] Specifically, data related to fuel consumption and combustion time, such as load size, unit operating status, and environmental conditions, can be collected. Then, statistical analysis methods or machine learning algorithms, such as regression analysis or random forests, are used to build predictive models. The models are trained using Python's Scikit-learn library, and cross-validation is used to evaluate their predictive capabilities. Through these steps, fuel consumption prediction models and combustion time inference models are obtained, capable of predicting fuel consumption and combustion time based on load allocation schemes.

[0123] Step S367: Input the optimal load allocation scheme of the unit into the unit fuel consumption prediction model and the unit fuel combustion time inference model respectively to predict fuel consumption and fuel combustion time, and obtain the unit fuel consumption prediction data and the unit fuel combustion time prediction data.

[0124] Specifically, the optimal load allocation scheme for the generating unit can be prepared using Python's Pandas library, ensuring that the data format matches the requirements of the unit's fuel consumption prediction model and combustion time inference model. Then, this data is input into the fuel consumption prediction model and combustion time inference model to obtain the prediction results. These results include predicted fuel consumption and combustion time data under the optimal load allocation scheme.

[0125] This invention utilizes a digital twin model combined with a multi-objective optimization algorithm to ensure optimal matching between grid demand and unit performance, achieving multi-objective optimized load allocation. Rapid performance simulation capabilities enable the system to promptly evaluate the temporal performance of different load allocation schemes, while dynamic model updates ensure that the scheduling scheme reflects changes in unit operating status in real time, enhancing the real-time performance and adaptability of the scheme. By applying a pre-defined performance evaluation index system, comprehensive evaluation can be performed across multiple performance dimensions, ensuring the selection of the optimal load allocation scheme, thereby improving the efficiency and economy of unit operation. Furthermore, the constructed fuel consumption prediction model and combustion time inference model provide accurate predictions based on the optimal load allocation scheme, offering data support for cost control and operational optimization. In summary, this invention enhances the generation capability of scheduling strategies, making scheduling decisions more scientific and rational, while improving overall energy utilization efficiency, reducing energy waste, and providing strong technical support for the efficient, economical, and environmentally friendly operation of thermal power plants.

[0126] Preferably, step S4 includes the following steps:

[0127] Step S41: Construct a thermodynamic model based on the unit's panoramic historical operation dataset to obtain the unit's thermodynamic parameterized model;

[0128] Specifically, key operating parameters, such as temperature, pressure, and flow rate, can be extracted from the unit's historical operational dataset. This data is preprocessed using Python's Pandas library, including cleaning, standardization, and outlier removal. Next, based on this data and thermodynamic principles, a thermodynamic model of the unit is constructed. This involves defining the equations of state, energy conservation, and mass conservation equations using the Engineering Equations Solver (EES) or MATLAB's Physical Modeling Toolbox. By parameterizing these equations, a parameterized thermodynamic model of the unit is obtained, capable of describing the unit's thermodynamic behavior under different operating conditions.

[0129] Step S42: Collect real-time operating data of the thermal power plant unit to obtain the real-time operating data of the unit, and use the thermodynamic parameterization model of the unit combined with the real-time operating data of the unit to perform heat balance calculation on the thermal power plant unit to obtain the current thermoelectric conversion efficiency data of the unit;

[0130] Specifically, monitoring systems for thermal power plant units, such as SCADA systems, can be used to collect real-time operating data. This data is transmitted to a data processing platform via a data acquisition interface. Then, combined with a thermodynamic parameterized model of the unit, heat balance calculations are performed using MATLAB or Python's SciPy library. These calculations include input and output analysis of energy and matter to obtain the unit's current thermoelectric conversion efficiency data.

[0131] Step S43: Perform data fusion between the unit equipment performance degradation trend data and the current thermoelectric conversion efficiency data of the unit to obtain the thermal efficiency degradation prediction feature matrix;

[0132] Specifically, the Pandas library in Python can be used to preprocess the data on the performance degradation trend of the unit equipment and the current thermoelectric conversion efficiency data of the unit, including synchronizing time series, handling missing values, and standardization. Then, Principal Component Analysis (PCA) from the Scikit-learn library is applied to extract key features and reduce data dimensionality, thus forming a comprehensive dataset. Next, machine learning algorithms, such as decision trees or random forests, are used to perform pattern recognition on the fused data to identify the key factors affecting thermal efficiency degradation. Finally, a feature matrix for predicting thermal efficiency degradation is constructed.

[0133] Step S44: Construct a thermal efficiency degradation prediction model based on the thermal efficiency degradation prediction feature matrix to obtain a hybrid thermal efficiency degradation prediction model;

[0134] Specifically, the Pandas library in Python can be used to conduct in-depth exploratory data analysis on the feature matrix for predicting thermal efficiency degradation, analyzing the correlation between features and thermal efficiency degradation. Appropriate machine learning algorithms, such as SVM, are selected to identify nonlinear patterns, or GBDT to capture interaction effects between features. Model instances are initialized using the Scikit-learn library and trained and validated using cross-validation. The feature selection process determines the most influential features through univariate methods or based on model importance scores. Key features are used in conjunction with labeled data to fine-tune model hyperparameters, such as the kernel function of SVM and the learning rate of GBDT, using grid search or random search techniques to find the optimal configuration. Finally, the optimized single or multiple models are combined to form a hybrid model for predicting thermal efficiency degradation.

[0135] Step S45: Use the thermal efficiency degradation prediction hybrid model combined with the current thermoelectric conversion efficiency data of the unit to perform multi-step prediction of the thermal efficiency degradation gradient, and obtain the thermal efficiency degradation gradient prediction data.

[0136] Specifically, the current thermoelectric conversion efficiency data can be prepared using the Pandas library and formatted into the form required for input to the hybrid model for predicting thermal efficiency degradation. This data is then fed into the hybrid model for prediction calculations. Next, the prediction results are used to analyze the trend and rate of thermal efficiency degradation, which can be achieved by calculating the differences between successive prediction steps. Finally, the predicted thermal efficiency degradation gradient data is obtained.

[0137] This invention ensures the accuracy and timeliness of thermoelectric conversion efficiency data through the effective utilization of real-time operational data. Data fusion enhances a comprehensive understanding of unit performance degradation characteristics, while a forward-looking model for predicting thermal efficiency degradation provides guidance for long-term unit performance changes. Precise performance degradation gradient prediction improves prediction accuracy, helping to optimize unit maintenance and operation strategies. Furthermore, the predicted thermal efficiency degradation gradient data enhances the scientific nature of operation scheduling, optimizes energy utilization efficiency, supports preventative maintenance, reduces unexpected downtime, and extends unit lifespan. This invention not only improves energy conversion efficiency but also promotes intelligent operation management, enhancing its efficiency and reliability.

[0138] Preferably, step S5 includes the following steps:

[0139] Step S51: Monitor the thermal power plant in real time using the intelligent fuel management system to obtain dynamic data on the plant's fuel inventory;

[0140] Specifically, this involves real-time monitoring of the intelligent fuel management system in thermal power plants. This requires integration with the plant's monitoring and data acquisition systems (such as SCADA systems) to automatically collect real-time fuel inventory data. Using the Python programming language, a data interface program is developed that periodically retrieves fuel inventory information from the SCADA system and stores this data in a database or data warehouse. By writing specific query and data processing scripts, the inflow, outflow, and storage of fuel can be monitored in real time, thus obtaining dynamic data on the power plant's fuel inventory.

[0141] Step S52: Perform data fusion and time alignment processing on the dynamic data of fuel inventory of thermal power plants, the predicted data of thermal efficiency degradation gradient, the predicted data of unit fuel consumption, the predicted data of unit fuel combustion time, and the dynamic data of fuel inventory of thermal power plants to obtain the integrated energy dynamic matrix of the unit.

[0142] Specifically, Python's Pandas library can be used to preprocess dynamic data on fuel inventory, thermal efficiency degradation gradient prediction, unit fuel consumption prediction, and unit fuel combustion time prediction data for thermal power plants. This preprocessing includes data cleaning, format standardization, and time synchronization. Then, data fusion techniques are used to integrate these data from different sources and with different time scales. This involves data interpolation, time series alignment, or window function techniques to ensure data consistency across the time dimension. Finally, a comprehensive unit energy dynamic matrix is ​​constructed, which fully reflects the energy use and inventory status of thermal power plants.

[0143] Step S53: Use a preset multi-objective optimization algorithm to update the constraints of the unit's integrated energy dynamic matrix in real time, and obtain the unit's optimization constraint parameter matrix;

[0144] Specifically, the objective of the optimization problem can be defined, such as minimizing fuel costs, maximizing energy efficiency, or reducing emissions. Then, a suitable multi-objective optimization algorithm, such as Multi-Objective Particle Swarm Optimization (MOPSO) or a genetic algorithm, is selected to handle this problem. These algorithms are implemented using Python's DEAP library or other optimization toolkits. During the optimization process, the unit's integrated energy dynamic matrix is ​​used as input, taking into account the unit's operating constraints and real-time changes in fuel inventory, and the optimization constraints are dynamically updated. In this way, the unit's optimization constraint parameter matrix can be obtained, which contains the optimal parameter settings for unit operation under current energy and operating conditions.

[0145] Step S54: Perform energy efficiency optimization calculations on the unit optimization constraint parameter matrix to obtain the unit energy allocation efficiency matrix;

[0146] Specifically, the Pandas library in Python can be used to parse and preprocess the unit optimization constraint parameter matrix. Next, energy efficiency evaluation metrics are defined, such as power generation per unit of fuel or carbon emission intensity. Mathematical optimization methods, such as linear programming, dynamic programming, or nonlinear optimization algorithms, are employed using the `optimize` module in the SciPy library to minimize energy consumption or maximize energy output while satisfying unit operating constraints. During the optimization process, the interactions between different units and the overall demand of the power grid are considered to achieve global energy efficiency optimization. Finally, the unit energy allocation efficiency matrix is ​​obtained, which shows the efficiency of energy allocation for each unit under the current constraints.

[0147] Step S55: Generate a scheduling strategy from the unit energy allocation efficiency matrix to obtain the unit operation scheduling scheme.

[0148] Specifically, the unit energy allocation efficiency matrix can be analyzed to identify the most and least efficient units and their operating modes. Then, scheduling algorithms, such as heuristic algorithms, mixed-integer linear programming, or multi-objective optimization algorithms, are used to generate scheduling schemes that meet grid demand and unit operating efficiency. In Python, these algorithms can be implemented using the PuLP library to build scheduling models. The scheduling model considers unit start-up and shutdown states, load allocation, reserve requirements, and fuel consumption factors. Solving the model yields a unit operation scheduling scheme, which includes the operating parameters of each unit, such as load factor, start-up and shutdown times, and other operational commands.

[0149] This invention achieves precise data acquisition on the dynamics of fuel inventory in thermal power plants through the real-time monitoring function of an intelligent fuel management system, providing strong real-time data support for fuel management decisions. By integrating multi-source data on fuel inventory, thermal efficiency degradation, fuel consumption, and combustion time, a comprehensive dynamic energy matrix for the unit is constructed, significantly enhancing the comprehensiveness and depth of data analysis. By applying multi-objective optimization algorithms combined with real-time updated constraints, the energy use of the unit under varying conditions is optimized, improving operational flexibility and efficiency. Real-time calculation of the optimized constraint parameter matrix provides precise optimization guidance for unit operation, ensuring the rationality and efficiency of energy allocation. Optimized energy efficiency calculations further improve the economic and environmental benefits of energy use, reducing energy waste and increasing energy conversion efficiency. Intelligently generated scheduling strategies, based on the optimized energy allocation efficiency matrix, make unit operation scheduling more scientific, rational, and efficient. The application of real-time data processing and optimization algorithms significantly improves the response speed of operation scheduling, enabling thermal power plants to quickly adapt to changes in the power grid and market. This invention exhibits better adaptability and robustness, effectively addressing various operational challenges.

[0150] Preferably, the present invention also provides an intelligent scheduling system for operating equipment of thermal power plant units, used to execute the intelligent scheduling method for operating equipment of thermal power plant units as described above, the intelligent scheduling system for operating equipment of thermal power plant units includes:

[0151] The unit modeling module is used to collect multi-dimensional historical operation data of thermal power plant units to obtain a panoramic historical operation dataset of the units; based on the panoramic historical operation dataset of the units, a digital twin model of the thermal power plant units is constructed to obtain the digital twin model of the units.

[0152] The performance analysis and prediction module is used to perform nonlinear dynamic system analysis on the panoramic historical operation dataset of the unit to obtain dynamic characteristic data of the unit's performance indicators; based on the unit's digital twin model, the module predicts the equipment performance degradation trend of the thermal power plant unit according to the dynamic characteristic data of the unit's performance indicators, and obtains the unit's equipment performance degradation trend data.

[0153] The load forecasting module is used to perform smart grid load forecasting for thermal power plants and obtain grid load demand forecasting data. Based on the unit digital twin model, the module optimizes the load allocation of thermal power plant units according to the grid load demand forecasting data, and estimates fuel consumption and combustion time to obtain unit fuel consumption forecasting data and unit fuel combustion time forecasting data.

[0154] The equipment degradation prediction module is used to perform thermodynamic efficiency analysis on thermal power plant units based on the unit's panoramic historical operation dataset to obtain the current thermoelectric conversion efficiency data of the unit; and to perform thermal efficiency degradation gradient prediction on thermal power plant units based on the unit's equipment performance degradation trend data and the current thermoelectric conversion efficiency data of the unit to obtain thermal efficiency degradation gradient prediction data.

[0155] The optimization scheduling module is used to monitor the thermal power plant in real time through the intelligent fuel management system and obtain dynamic data on the fuel inventory of the thermal power plant. Based on the thermal efficiency degradation gradient prediction data, the module performs multi-objective optimization scheduling of the thermal power plant units according to the predicted data of unit fuel consumption, the predicted data of unit fuel combustion time, and the dynamic data of the thermal power plant fuel inventory, and obtains the unit operation scheduling scheme.

[0156] In this invention, a digital twin model constructed using deep learning technology enables high-precision simulation and prediction of unit operating status, significantly enhancing the accuracy and response speed of dispatching. Simultaneously, multi-dimensional historical operating data acquisition and nonlinear dynamic system analysis provide a scientific basis for real-time monitoring and prediction of unit performance, making preventative maintenance and operational adjustments more precise. Furthermore, the application of intelligent algorithms in grid load forecasting ensures the reliability of load allocation and power generation planning. The system also improves energy utilization efficiency and reduces operating costs through real-time fuel inventory monitoring and multi-objective optimization algorithms. The integration of these technologies not only enhances the system's adaptability and flexibility but also reduces human error.

[0157] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0158] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for intelligent scheduling of operating equipment in thermal power plant units, characterized in that, Includes the following steps: Step S1: Collect multi-dimensional historical operation data of the thermal power plant units to obtain a panoramic historical operation dataset of the units; construct a digital twin model of the thermal power plant units based on the panoramic historical operation dataset of the units to obtain the digital twin model of the units. Step S2: Perform nonlinear dynamic system analysis on the unit's panoramic historical operation dataset to obtain dynamic characteristic data of unit performance indicators; Based on the digital twin model of the unit, the dynamic characteristic data of the unit's performance indicators are used to predict the equipment performance degradation trend of the thermal power plant unit, and the unit equipment performance degradation trend data is obtained. Step S3: Perform smart grid load forecasting for the thermal power plant to obtain grid load demand forecasting data; optimize load allocation for the thermal power plant units based on the grid load demand forecasting data using the unit digital twin model, and estimate fuel consumption and combustion time to obtain unit fuel consumption forecasting data and unit fuel combustion time forecasting data. Step S4: Perform thermodynamic efficiency analysis on the thermal power plant units based on the unit's panoramic historical operation dataset to obtain the current thermoelectric conversion efficiency data of the units; predict the thermal efficiency degradation gradient of the thermal power plant units based on the unit equipment performance degradation trend data and the current thermoelectric conversion efficiency data of the units to obtain thermal efficiency degradation gradient prediction data. Step S5: Monitor the thermal power plant in real time using the intelligent fuel management system to obtain dynamic data on the plant's fuel inventory; Based on the thermal efficiency degradation gradient prediction data, the unit fuel consumption prediction data, the unit fuel combustion time prediction data, and the dynamic data of the thermal power plant fuel inventory are used to perform multi-objective optimization scheduling of thermal power plant units, and obtain the unit operation scheduling scheme.

2. The intelligent scheduling method for operating equipment of thermal power plant units according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain the unit equipment topology list; Step S12: Collect multi-dimensional time-series data of the thermal power plant units according to the unit equipment topology list to obtain the unit panoramic historical operation dataset. The unit panoramic historical operation dataset includes at least boiler historical operation data, steam turbine historical operation data, generator historical operation data and condenser historical operation data. Step S13: Perform deep neural network modeling on the historical operating data of the boiler, steam turbine, generator and condenser respectively to obtain the learning model set of the unit subsystem. The learning model set of the unit subsystem includes the boiler learning model, steam turbine learning model, generator learning model and condenser learning model. Step S14: Integrate the boiler learning model, steam turbine learning model, generator learning model and condenser learning model into a graph neural network to obtain the overall graph neural network model of the unit; Step S15: Perform spatiotemporal feature extraction and abnormal pattern recognition on the unit's panoramic historical operation dataset to obtain unit operation feature data and unit abnormal pattern data; Step S16: Use the overall unit diagram neural network model to perform transfer learning on the unit operation characteristic data and the unit abnormal mode data, and perform fine-tuning and optimization to obtain the unit digital twin model.

3. The intelligent scheduling method for operating equipment of thermal power plant units according to claim 2, characterized in that, Step S16 includes the following steps: Step S161: Perform feature dimensionality reduction and extraction on the unit operation feature data and unit anomaly mode data to obtain unit operation feature dimensionality reduction data and unit anomaly mode dimensionality reduction data; Step S162: Perform hierarchical freezing preprocessing on the overall neural network model of the unit to obtain the preprocessed network model of the unit; Step S163: Construct a target domain dataset from the dimensionality reduction data of unit operation characteristics and the dimensionality reduction data of unit anomaly modes to obtain a balanced target domain dataset for the unit. Step S164: Perform domain adaptive layer integration on the preprocessed network model of the unit to obtain the domain adaptive network model of the unit; Step S165: Use the unit domain adaptive network model to perform adversarial transfer learning on the unit target domain balanced dataset to obtain the unit transfer optimization network model; Step S166: Perform incremental fine-tuning optimization on the unit migration optimization network model to obtain the unit fine-tuning optimization network model; Step S167: Perform knowledge distillation and integration on the unit fine-tuning and optimization network model to obtain the unit digital twin model.

4. The intelligent scheduling method for operating equipment of thermal power plant units according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Perform time series segmentation and noise filtering on the unit's panoramic historical operation dataset to obtain the unit's preprocessed time series operation dataset; Step S22: Extract nonlinear dynamic features from the preprocessed time-series operation dataset of the unit to obtain the nonlinear dynamic feature data of the unit; Step S23: Perform multivariate coupling analysis on the preprocessed time-series operation dataset of the unit to obtain the coupling characteristic data of the unit parameters; Step S24: Perform state-space reconstruction on the unit parameter coupling characteristic data to obtain the unit state-space reconstruction data; Step S25: Perform dynamic pattern recognition on the unit state space reconstruction data to obtain the unit dynamic pattern feature data; Step S26: Perform system stability analysis on the dynamic mode characteristic data of the unit to obtain the unit stability index data; Step S27: Integrate the nonlinear dynamic characteristic data, the coupling characteristic data of the unit parameters, and the stability index data of the unit to obtain the dynamic characteristic data of the unit performance index; Step S28: Based on the unit's digital twin model, predict the equipment performance degradation trend of the thermal power plant unit according to the dynamic characteristic data of the unit's performance indicators, and obtain the unit's equipment performance degradation trend data.

5. The intelligent scheduling method for operating equipment of thermal power plant units according to claim 4, characterized in that, Step S28 includes the following steps: Step S281: Input the dynamic characteristic data of the unit performance indicators into the digital twin model of the unit for forward extrapolation to obtain the dynamic simulation data of the unit performance; Step S282: Perform multi-scale degradation trend analysis on the dynamic simulation data of unit performance to obtain multi-scale degradation characteristic data of the unit; Step S283: Perform nonparametric regression prediction on the multi-scale degradation characteristic data of the unit to obtain preliminary degradation trend prediction data of the unit; Step S284: Perform hybrid model integration optimization on the preliminary degradation trend prediction data of the unit to obtain the comprehensive degradation trend prediction data of the unit; Step S285: Quantify and evaluate the uncertainty of the unit's comprehensive degradation trend prediction data to obtain the unit equipment performance degradation trend data.

6. The intelligent scheduling method for operating equipment of thermal power plant units according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Collect historical load data from thermal power plants to obtain historical power grid load time series data; Step S32: Extract spatiotemporal features from historical power grid load time series data and perform load pattern recognition to obtain power grid load time feature data and power grid load spatial feature data; Step S33: Perform feature fusion on the power grid load temporal feature data and the power grid load spatial feature data to obtain the power grid load multimodal feature data; Step S34: Perform deep probabilistic load modeling on the multimodal characteristic data of the power grid load to obtain the power grid load probability distribution model; Step S35: Perform multi-model integration on the power grid load probability distribution model to obtain an adaptive probability load prediction model, and use the adaptive probability load prediction model to perform smart grid load prediction on thermal power plants to obtain power grid load demand prediction data. Step S36: Based on the digital twin model of the unit, optimize the load allocation of the thermal power plant unit according to the grid load demand forecast data, and estimate the fuel consumption and combustion time to obtain the unit fuel consumption forecast data and the unit fuel combustion time forecast data.

7. The intelligent scheduling method for operating equipment of thermal power plant units according to claim 6, characterized in that, Step S36 includes the following steps: Step S361: Input the power grid load demand forecast data into the unit digital twin model to plan the multi-objective load allocation optimization algorithm and obtain the unit load allocation optimization algorithm; Step S362: Use the unit load allocation optimization algorithm to generate load allocation schemes for thermal power plant units, obtain a set of candidate load allocation schemes, and perform rapid performance simulation of each candidate load allocation scheme through the unit digital twin model to obtain the unit load allocation simulation time series data. Step S363: Dynamically update the unit digital twin model based on the unit load distribution simulation time series data to obtain the unit operation dynamic constraint condition set; Step S364: Use the preset unit performance evaluation index system to perform multi-dimensional performance evaluation on the unit load distribution simulation time series data to obtain the load distribution scheme performance evaluation matrix; Step S365: Select the optimal scheme based on the set of dynamic constraints for unit operation and the performance evaluation matrix of load allocation scheme to obtain the optimal load allocation scheme for the unit; Step S366: Based on the digital twin model of the unit, construct the fuel consumption estimation model and the combustion time estimation model respectively to obtain the unit fuel consumption prediction model and the unit fuel combustion time inference model; Step S367: Input the optimal load allocation scheme of the unit into the unit fuel consumption prediction model and the unit fuel combustion time inference model respectively to predict fuel consumption and fuel combustion time, and obtain the unit fuel consumption prediction data and the unit fuel combustion time prediction data.

8. The intelligent scheduling method for operating equipment of thermal power plant units according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Construct a thermodynamic model based on the unit's panoramic historical operation dataset to obtain the unit's thermodynamic parameterized model; Step S42: Collect real-time operating data of the thermal power plant unit to obtain the real-time operating data of the unit, and use the thermodynamic parameterization model of the unit combined with the real-time operating data of the unit to perform heat balance calculation on the thermal power plant unit to obtain the current thermoelectric conversion efficiency data of the unit; Step S43: Perform data fusion between the unit equipment performance degradation trend data and the current thermoelectric conversion efficiency data of the unit to obtain the thermal efficiency degradation prediction feature matrix; Step S44: Construct a thermal efficiency degradation prediction model based on the thermal efficiency degradation prediction feature matrix to obtain a hybrid thermal efficiency degradation prediction model; Step S45: Use the thermal efficiency degradation prediction hybrid model combined with the current thermoelectric conversion efficiency data of the unit to perform multi-step prediction of the thermal efficiency degradation gradient, and obtain the thermal efficiency degradation gradient prediction data.

9. The intelligent scheduling method for operating equipment of thermal power plant units according to claim 1, characterized in that, Step S5 includes the following steps: Step S51: Monitor the thermal power plant in real time using the intelligent fuel management system to obtain dynamic data on the plant's fuel inventory; Step S52: Perform data fusion and time alignment processing on the dynamic data of fuel inventory of thermal power plants, the predicted data of thermal efficiency degradation gradient, the predicted data of unit fuel consumption, the predicted data of unit fuel combustion time, and the dynamic data of fuel inventory of thermal power plants to obtain the integrated energy dynamic matrix of the unit. Step S53: Use a preset multi-objective optimization algorithm to update the constraints of the unit's integrated energy dynamic matrix in real time, and obtain the unit's optimization constraint parameter matrix; Step S54: Perform energy efficiency optimization calculations on the unit optimization constraint parameter matrix to obtain the unit energy allocation efficiency matrix; Step S55: Generate a scheduling strategy from the unit energy allocation efficiency matrix to obtain the unit operation scheduling scheme.

10. An intelligent scheduling system for operating equipment of thermal power plant units, characterized in that, For executing the intelligent scheduling method for operating equipment of thermal power plant units as described in claim 1, the intelligent scheduling system for operating equipment of thermal power plant units includes: The unit modeling module is used to collect multi-dimensional historical operation data of thermal power plant units to obtain a panoramic historical operation dataset of the units; based on the panoramic historical operation dataset of the units, a digital twin model of the thermal power plant units is constructed to obtain the digital twin model of the units. The performance analysis and prediction module is used to perform nonlinear dynamic system analysis on the panoramic historical operation dataset of the unit to obtain dynamic characteristic data of the unit's performance indicators; based on the unit's digital twin model, the module predicts the equipment performance degradation trend of the thermal power plant unit according to the dynamic characteristic data of the unit's performance indicators, and obtains the unit's equipment performance degradation trend data. The load forecasting module is used to perform smart grid load forecasting for thermal power plants and obtain grid load demand forecasting data. Based on the unit digital twin model, the module optimizes the load allocation of thermal power plant units according to the grid load demand forecasting data, and estimates fuel consumption and combustion time to obtain unit fuel consumption forecasting data and unit fuel combustion time forecasting data. The equipment degradation prediction module is used to perform thermodynamic efficiency analysis on thermal power plant units based on the unit's panoramic historical operation dataset to obtain the current thermoelectric conversion efficiency data of the unit; and to perform thermal efficiency degradation gradient prediction on thermal power plant units based on the unit's equipment performance degradation trend data and the current thermoelectric conversion efficiency data of the unit to obtain thermal efficiency degradation gradient prediction data. The optimization scheduling module is used to monitor the thermal power plant in real time through the intelligent fuel management system and obtain dynamic data on the fuel inventory of the thermal power plant. Based on the thermal efficiency degradation gradient prediction data, the module performs multi-objective optimization scheduling of the thermal power plant units according to the predicted data of unit fuel consumption, the predicted data of unit fuel combustion time, and the dynamic data of the thermal power plant fuel inventory, and obtains the unit operation scheduling scheme.

Images