Intelligent monitoring and early warning control methods and systems for gas turbine power plants

Abstract

This invention relates to the field of gas turbine power plant technology, and particularly to an intelligent monitoring and early warning control method and system for gas turbine power plants. The method acquires multi-source data such as the combustion chamber exhaust temperature field, gas flow velocity distribution, and thermal channel vibration spectrum, performs multi-modal fusion and three-dimensional thermal flow field reconstruction to generate a fused thermal flow image and three-dimensional modeling data of the thermal channels. Based on the fused image, temperature dispersion gradient analysis is performed to achieve accurate identification and location of thermal anomaly contours. Combining the creep characteristics of the thermal channel materials, thermal stress field reconstruction is performed on the abnormal areas to predict the lifespan decay trend. Furthermore, the prediction results are used to conduct thermal channel failure induction analysis, generating failure risk channel data, and implementing dynamic heat load regulation accordingly to complete lifespan early warning and proactive control. This improves the early identification capability of thermal anomalies and the accuracy of lifespan management, realizing a shift from passive alarm to proactive prevention and control, and significantly enhancing the safety and intelligence level of gas turbine thermal channel operation.

Description

Technical Field

[0001] This invention relates to the field of gas turbine power plant technology, and in particular to an intelligent monitoring and early warning control method and system for gas turbine power plants. Background Technology

[0002] With the rapid development of modern gas turbine power plants towards higher efficiency and higher parameters, the combustion chamber and hot passage components of gas turbines operate under extreme conditions of high temperature, high pressure, and high-speed airflow for extended periods. The stability of their thermodynamic state directly affects the safety and reliability of the unit. The uniformity of the exhaust temperature field in the combustion chamber is a crucial indicator of combustion stability. Local combustion anomalies or uneven fuel distribution can easily lead to uneven exhaust temperature distribution, resulting in significant temperature dispersion. This can then cause problems such as localized overheating and thermal stress concentration in the hot passage structure, accelerating material creep, fatigue damage, and even crack initiation, seriously threatening the service life of the hot passage. Traditional monitoring methods often rely on threshold alarm mechanisms for discrete measurement points, lacking the dynamic sensing capability of the overall temperature field distribution characteristics, making it difficult to accurately identify and locate early thermal anomalies.

[0003] Existing monitoring systems typically analyze only single physical quantities (such as temperature and vibration) independently, failing to effectively integrate multi-source information such as gas flow rate, heat flow distribution, and structural response. This results in insufficient ability to accurately reconstruct the true thermodynamic state of thermal channels. Particularly in thermal channel lifespan assessment, empirical prediction methods based on cumulative operating time or simplified heat load models are commonly used, neglecting the coupling effects between local thermal gradient evolution, nonlinear material creep behavior, and structural dynamic response. This leads to low prediction accuracy and fails to meet the needs of refined operation and maintenance. Furthermore, after anomalies occur, existing systems lack a closed-loop control mechanism from thermal anomaly identification to lifespan prediction and proactive regulation, making it difficult to achieve true early warning and lifespan extension control.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide an intelligent monitoring and early warning control method and system for gas turbine power plants. It aims to solve the technical problems of existing gas turbine power plant monitoring systems, which are unable to integrate multi-source data to achieve accurate perception and anomaly identification of the thermal channel temperature field, and lack the ability to predict and dynamically regulate life based on the thermal stress-creep coupling mechanism, resulting in low accuracy of early warning of thermal anomalies and extensive life management.

[0006] To achieve the above objectives, the present invention provides an intelligent monitoring and early warning control method for gas turbine power plants, the method comprising: Acquire multi-source monitoring data from gas turbine power plants, including combustion chamber exhaust temperature field data, gas flow velocity distribution data, and thermal channel vibration spectrum data, and perform multi-modal data fusion and three-dimensional thermal flow field reconstruction of thermal channels to generate fused thermal flow images and three-dimensional modeling data of thermal channels; Temperature dispersion gradient analysis is performed on the fused heat flow image to generate temperature dispersion gradient data; thermal anomaly contour recognition and localization are performed on the fused heat flow image using the temperature dispersion gradient data to generate initial thermal anomaly localization data. Acquire creep characteristic data of thermal channel material; reconstruct thermal stress field of thermal channel region based on initial thermal anomaly location data to obtain thermal stress field; predict life decay trend based on thermal stress field and creep characteristic data of thermal channel material to generate life decay trend prediction data. Using lifetime decay trend prediction data, thermal channel failure induction analysis is performed on the 3D modeling data of the thermal channel to generate thermal channel failure risk channel data; the thermal channel failure risk channel data is used to dynamically regulate the thermal load of the corresponding thermal channel area in order to perform thermal channel lifetime early warning control operations.

[0007] Optionally, the acquisition of multi-source monitoring data from the gas turbine power plant, including combustion chamber exhaust temperature field data, gas flow velocity distribution data, and thermal channel vibration spectrum data, and the subsequent multi-modal data fusion and three-dimensional thermal flow field reconstruction of the thermal channels to generate fused thermal flow images and three-dimensional thermal channel modeling data, includes: The combustion chamber exhaust temperature field data and thermal channel vibration spectrum data are collected synchronously by a distributed thermocouple array, and the gas flow velocity distribution data are obtained by a laser Doppler velocimeter. Radiation compensation and spatiotemporal registration are performed on the combustion chamber exhaust temperature field data to generate standard temperature field data. Turbulence correction and noise filtering are performed on the gas velocity distribution data to generate velocity gradient distribution data. Based on the vibration spectrum data of the thermal channel, the three-dimensional point cloud of the thermal channel is reconstructed to generate three-dimensional modeling data of the thermal channel. Standard temperature field data, velocity gradient distribution data, and thermal channel 3D modeling data are fused in a multi-scale spatiotemporal manner to generate a fused heat flow image.

[0008] Optionally, the step of reconstructing the three-dimensional point cloud of the thermal channel based on the thermal channel vibration spectrum data to generate three-dimensional modeling data of the thermal channel includes: Frequency domain equalization is performed on the vibration spectrum data of the hot channel to generate spectral equalization data; Structural mode segmentation is performed on the thermal channel vibration spectrum data based on spectrum equalization data to distinguish the main structure of the thermal channel and remove interfering modes, thereby generating thermal channel main structure segmentation data. Thermal deformation gradient estimation is performed on the segmented data of the main structure of the thermal channel to generate thermal deformation information data of the thermal channel surface; based on the thermal deformation information data of the thermal channel surface, surface heat flow reconstruction is performed on the thermal channel to generate surface heat flow reconstruction data of the thermal channel. Thermodynamic meshing was performed on the thermal flow reconstruction data of the thermal channel surface to obtain the three-dimensional modeling data of the thermal channel.

[0009] Optionally, the step of performing temperature dispersion gradient analysis based on the fused heat flow image to generate temperature dispersion gradient data; and using the temperature dispersion gradient data to perform thermal anomaly contour recognition and localization on the fused heat flow image to generate initial thermal anomaly localization data, includes: Temperature anomaly regions are extracted based on fused heat flow images, and the temperature anomaly regions are divided into core regions to obtain temperature anomaly core region images and temperature anomaly edge region images. Temperature morphology and gradient analysis are performed on the image of the core region of the temperature anomaly to generate temperature feature data of the core region; the directional consistency and heat flux density concentration of the image of the edge region of the temperature anomaly are calculated to obtain the temperature feature data of the edge region. 3D heat flow projection is performed on the fused heat flow image using temperature feature data from the core region and the edge region, and the heat flow value and thermal gradient change after projection are extracted; the dispersion gradient of the temperature anomaly region is calculated based on the heat flow value and thermal gradient change to generate temperature dispersion gradient data. Obtain historical thermal channel design parameters; perform cross-segment structural thermal feature matching of historical thermal channel design parameters based on temperature dispersion gradient data to generate thermal anomaly matching data; perform coordinate transformation and positioning of the thermal channel using thermal anomaly matching data to generate initial thermal anomaly positioning data.

[0010] Optionally, the step of performing cross-segment structural thermal characteristic matching of historical thermal channel design parameters based on temperature dispersion gradient data includes: Extract the segmental structure from the historical hot passage design parameters to generate hot passage segment-level profile data; Perform a thermodynamic projection transformation on the temperature dispersion gradient data to generate dispersion structure mapping data; By performing inter-segment trajectory overlap analysis on dispersion structure mapping data and thermal channel segment-level profile data, dispersion segment coupling map data is generated. Common thermal features are identified in the dispersion segment coupled spectral data to generate thermal anomaly consistency data; Based on the thermal anomaly consistency data, thermal path consensus reconstruction is performed on the dispersion segment coupling spectrum data to generate thermal anomaly matching data.

[0011] Optionally, the steps of acquiring creep characteristic data of the thermal channel material; reconstructing the thermal stress field of the thermal channel region based on the initial location data of the thermal anomaly to obtain the thermal stress field; and predicting the lifetime decay trend based on the creep characteristic data of the thermal channel material using the thermal stress field to generate lifetime decay trend prediction data include: Acquire creep property data of thermal channel materials, deconstruct the creep constitutive features of the thermal channel material creep property data, and generate material creep modulus field data; The initial location data of the thermal anomaly is meshed into a thermal channel region to generate finite element discrete mesh data of the thermal channel; the thermal stress field is solved based on the material creep modulus field data of the thermal channel finite element discrete mesh data to generate the thermal stress field. The thermal gradient direction of the thermal stress field is extracted, and nonlinear coupling prediction is performed on the creep characteristics data of the thermal channel material to generate lifetime decay trend prediction data.

[0012] Optionally, the step of extracting the thermal gradient direction of the thermal stress field and performing nonlinear coupling prediction on the creep characteristic data of the thermal channel material to generate lifetime degradation trend prediction data includes: Multi-scale thermal tensor differentiation processing is performed on the thermal stress field data to generate thermal channel principal thermal gradient tensor data; thermal field stability analysis is performed on the thermal channel principal thermal gradient tensor data to generate thermally stable directional domain data. Dynamic creep partitioning is performed on the creep characteristic data of thermal channel materials to generate layered creep mapping data; Nonlinear interactive fusion of thermal stability directional domain data and hierarchical creep mapping data is performed to generate lifetime decay probability field data. The lifetime decay probability field data is subjected to time entropy value evolution to generate lifetime decay trend prediction data.

[0013] Optionally, the step of using lifetime decay trend prediction data to perform thermal channel failure induction analysis on the three-dimensional modeling data of the thermal channel to generate thermal channel failure risk channel data; and using the thermal channel failure risk channel data to dynamically regulate the thermal load of the corresponding thermal channel area to execute thermal channel lifetime early warning control operations, includes: Thermal-structure coupling mapping is performed on the three-dimensional modeling data of thermal channels using lifetime decay trend prediction data to generate thermal channel failure path prediction data. The path topology of the hot aisle failure path prediction data is analyzed, and based on the results of the path topology analysis, inter-segment failure nodes are extracted from the hot aisle failure path prediction data to generate failure path node map data. Thermal load simulation is performed on the failure path node map data to generate path thermal load penetration data; dynamic thermal load regulation analysis is performed on the initial location data of thermal anomalies based on the path thermal load penetration data to generate dynamic regulation data of thermal channels. Based on the dynamic control data of the thermal channel, the initial location data of the thermal anomaly is fed back with thermal response data to obtain thermal channel control feedback data, so as to perform thermal channel lifespan early warning control operations.

[0014] Optionally, the step of performing thermal load simulation on the failure path node map data to generate path thermal load penetration data; and performing dynamic thermal load regulation analysis on the initial location data of thermal anomalies based on the path thermal load penetration data to generate dynamic regulation data of thermal channels, including: Multiphysics thermal response modeling is performed on the failure path node map data to generate local node thermal coupling model data; time-series thermal disturbance simulation is performed on the local node thermal coupling model data to generate thermal load response sequence data. Three-dimensional thermal field inversion calculations are performed on the heat load response sequence data to generate path heat load infiltration data; Non-uniform heat-sensitive region clustering is performed on the path heat load infiltration data to generate high-risk heat penetration cluster data; location heat gradient cross-analysis is performed on the high-risk heat penetration cluster data and the initial location data of thermal anomalies to generate dynamic heat load configuration data. A global thermal control simulation of the thermal channel is performed on the dynamic heat load configuration data to generate dynamic control data for the thermal channel.

[0015] Furthermore, to achieve the above objectives, the present invention also provides an intelligent monitoring and early warning control system for gas turbine power plants, the system comprising: The multimodal fusion module is used to acquire multi-source monitoring data from gas turbine power plants, including combustion chamber exhaust temperature field data, gas flow velocity distribution data, and thermal channel vibration spectrum data. It also performs multimodal data fusion and thermal channel three-dimensional thermal flow field reconstruction to generate fused thermal flow images and thermal channel three-dimensional modeling data. The anomaly localization module is used to perform temperature dispersion gradient analysis based on the fused heat flow image to generate temperature dispersion gradient data; and to perform thermal anomaly contour recognition and localization on the fused heat flow image using the temperature dispersion gradient data to generate initial thermal anomaly localization data. The lifetime prediction module is used to acquire creep characteristic data of thermal channel materials; reconstruct the thermal stress field of the thermal channel region based on the initial location data of thermal anomalies to obtain the thermal stress field; and predict the lifetime decay trend based on the creep characteristic data of thermal channel materials using the thermal stress field to generate lifetime decay trend prediction data. The dynamic control module is used to perform thermal channel failure induction analysis on the three-dimensional modeling data of the thermal channel using the life decay trend prediction data, and generate thermal channel failure risk channel data; and to perform dynamic heat load control on the thermal channel area based on the thermal channel failure risk channel data, so as to execute thermal channel life early warning control operations.

[0016] This invention provides an intelligent monitoring and early warning control method for gas turbine power plants. The method integrates multi-source monitoring data, including combustion chamber exhaust temperature field, gas velocity distribution, and thermal channel vibration spectrum, to achieve high-precision reconstruction and visualization of the three-dimensional thermal flow field of the thermal channel, overcoming the limitations of traditional single-point monitoring methods in terms of spatial coverage and physical correlation. Based on the fused thermal flow image, temperature dispersion gradient analysis and thermal anomaly contour recognition can accurately locate early-stage local overheating areas, improving the sensitivity and accuracy of thermal anomaly detection. Combining thermal anomaly location with thermal stress field reconstruction and integrating material creep characteristic data enables nonlinear prediction of lifespan decay trends, significantly improving the scientific rigor and reliability of thermal channel remaining lifespan assessment. Furthermore, failure excitation analysis generates risk channel data and drives dynamic heat load regulation, forming a closed-loop control mechanism of perception, diagnosis, prediction, and regulation, realizing a shift from passive alarm to proactive early warning and lifespan extension control. This method effectively improves the safety, stability, and intelligent operation and maintenance level of the thermal channel in gas turbine power plants, demonstrating promising engineering application prospects and widespread value. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating an embodiment of the intelligent monitoring and early warning control method for gas turbine power plants according to the present invention. Figure 2 This is a schematic diagram illustrating the specific process of generating lifespan decline trend prediction data in an embodiment of the intelligent monitoring and early warning control method for gas turbine power plants according to the present invention. Figure 3 This is a structural block diagram of an embodiment of the intelligent monitoring and early warning control system for gas turbine power plants according to the present invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] Reference Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the intelligent monitoring and early warning control method for gas turbine power plants according to the present invention.

[0021] In one embodiment, the intelligent monitoring and early warning control method for gas turbine power plants includes: Step S100: Acquire multi-source monitoring data of the gas turbine power plant, including combustion chamber exhaust temperature field data, gas flow velocity distribution data and thermal channel vibration spectrum data, and perform multi-modal data fusion and thermal channel three-dimensional thermal flow field reconstruction to generate fused thermal flow image and thermal channel three-dimensional modeling data.

[0022] The multi-source monitoring data from the gas turbine power plant can be a collection of real-time operational monitoring information from different physical dimensions of the gas turbine power plant. This data can provide raw input for hot channel status sensing and support multi-physics coupling analysis. Combustion chamber exhaust temperature field data can be measurement data reflecting the gas temperature distribution at various locations on the combustion chamber outlet section. This data can be used to characterize combustion uniformity and localized combustion anomalies. Gas flow velocity distribution data can describe the spatial distribution characteristics of gas flow velocity at the hot channel inlet or interior. This data can be used to correct for thermal convection effects and improve the accuracy of thermal flow field reconstruction. In an exemplary embodiment, the gas flow velocity distribution data can be obtained through a Pitot tube array, a laser Doppler velocimeter, or pressure gradient inversion. Hot channel vibration spectrum data can be the frequency domain representation of the vibration signals generated by the hot channel structure during operation after Fourier transform. This data can be used to indirectly reflect thermal stress fluctuations and structural dynamic response characteristics. Furthermore, the hot channel vibration spectrum data can be obtained by collecting time-domain vibration signals from an accelerometer and then performing spectrum analysis.

[0023] Multimodal data fusion can be a process of jointly modeling heterogeneous physical quantities (temperature, velocity, vibration) based on spatiotemporal alignment. It can be used to eliminate blind spots in single-modal observations and enhance the physical consistency of reconstructed thermal flow fields. Three-dimensional thermal flow field reconstruction of thermal channels can be based on multi-source monitoring data to construct a numerical model of heat flux density and temperature distribution within the three-dimensional space of the thermal channel. This can be used to convert discrete point measurements into continuous field representations. Fusion heat flow images can be two-dimensional visualizations of heat flow distribution that integrate multi-source information. They can be used as input for temperature dispersion gradient analysis and thermal anomaly profile recognition. Three-dimensional modeling data of thermal channels can be a three-dimensional digital model containing the thermal channel geometry, material partitioning, and coordinate mapping relationships. This can be used to provide a spatial positioning reference for thermal stress field reconstruction and failure initiation analysis. In an exemplary embodiment, three-dimensional modeling data of thermal channels can be used in conjunction with lifetime decay trend prediction data for thermal channel failure initiation analysis.

[0024] Multimodal data fusion and 3D thermal flow field reconstruction of hot channels can involve spatiotemporal registration of multi-source data, physical correlation modeling, and solving the heat conduction-convection coupling equations. Furthermore, this operation can be achieved by using a data-driven neural network fusion model combined with physical constraints for field reconstruction, or by embedding measured boundary conditions into the finite element method for inverse thermal flow field solving, thereby generating a physically consistent continuous 3D thermal flow distribution. Generating the fused thermal flow image and 3D thermal channel modeling data can be achieved by projecting the reconstructed 3D thermal flow field into a 2D image and binding it to the geometric model coordinate system. Furthermore, this operation can provide both visual and structured representations.

[0025] Step S200: Perform temperature dispersion gradient analysis based on the fused heat flow image to generate temperature dispersion gradient data. Use the temperature dispersion gradient data to identify and locate the thermal anomaly contour in the fused heat flow image, generating initial thermal anomaly localization data.

[0026] Temperature dispersion gradient analysis can be a process of spatial differentiation of the local temperature change rate in the fused heat flow image, which can be used to quantify the spatial evolution characteristics of temperature field inhomogeneity. Temperature dispersion gradient data can be a numerical matrix describing the magnitude and direction of temperature gradients in each region of the fused heat flow image, which can be used as a criterion for thermal anomaly contour recognition, highlighting potential overheating boundaries. Thermal anomaly contour recognition and localization can be an image processing operation based on temperature gradient feature extraction and delineation of local overheating region boundaries, which can be used to achieve precise spatial localization of early thermal anomalies. Initial thermal anomaly localization data can be a set of spatial coordinates and contour information identifying suspected overheating areas in the thermal channel, which can be used to guide local area focusing in subsequent thermal stress field reconstruction.

[0027] Temperature dispersion gradient analysis based on fused heat flow images can be achieved by performing spatial gradient operators (such as Sobel and Laplacian) on image pixels. Furthermore, this operation can be implemented by preprocessing with anisotropic diffusion filtering to calculate the gradient and suppress noise, or by using a deep learning edge detection network to replace traditional operators for gradient feature extraction, thereby highlighting regions of abrupt temperature changes and enhancing anomaly sensitivity. Generating temperature dispersion gradient data can be achieved by outputting vector field data consisting of gradient magnitude and direction. Furthermore, this operation can provide quantitative criteria for contour recognition.

[0028] Thermal anomaly contour identification and localization using temperature dispersion gradient data in fused heat flux images can be achieved by applying closed contour extraction algorithms to determine anomaly boundaries in gradient-significant regions. For example, this operation can be accomplished by iteratively fitting anomaly region boundaries using an active contour model (Snake), or by automatically generating contours using connected component analysis combined with gradient thresholding, thus achieving sub-measurement point-level anomaly region localization. Generating initial thermal anomaly localization data can involve converting the identified contours into region identifiers in a three-dimensional coordinate system. Furthermore, this operation can provide structured anomaly location information.

[0029] Step S300: Obtain creep characteristic data of the thermal channel material. Reconstruct the thermal stress field of the corresponding thermal channel region based on the initial thermal anomaly location data to obtain the thermal stress field. Based on the thermal stress field, predict the lifetime degradation trend of the thermal channel material creep characteristic data to generate lifetime degradation trend prediction data.

[0030] The creep characteristics data of the thermal channel material can describe the time-dependent plastic deformation of the high-temperature alloy used in the thermal channel under different temperature-stress combinations, and can be used to provide a material constitutive relation basis for predicting the life decay trend. The creep characteristics data of the thermal channel material can be obtained through accelerated creep tests in the laboratory or material databases. Thermal stress field reconstruction can be a numerical process of calculating the spatial distribution of thermal stress inside the thermal channel structure based on the location and temperature distribution of local thermal anomalies, and can be used to quantify the mechanical load state caused by local overheating. The thermal stress field can be the distribution state of the internal stress of the thermal channel structure due to the temperature gradient in three-dimensional space, and can be used as a key input variable for the life prediction model, reflecting the thermo-mechanical coupling effect.

[0031] Lifetime decay trend prediction can combine thermal stress field and material creep characteristics to deduce the nonlinear process of the remaining life of the thermal channel evolving over time, providing dynamic, coupled life assessment results. The prediction can take thermal stress field and thermal channel material creep characteristic data as inputs and output lifetime decay trend prediction data. This data can be time-series data characterizing the remaining life decay rate and failure probability in different regions of the thermal channel, supporting failure risk quantification and control decisions. Obtaining thermal channel material creep characteristic data can be done by reading the creep curve of the corresponding alloy from a material performance database or experimental report. Reconstructing the thermal stress field of the corresponding thermal channel region based on the initial location data of the thermal anomaly can be achieved by applying measured temperature boundary conditions within a defined region and solving the thermoelastic equations. For example, this operation can be achieved by using sub-model technology to refine the global model locally before performing thermal stress calculations, or by combining dynamic load correction of static thermal stress results with vibration spectrum inversion, thereby obtaining a local stress state matching the actual anomaly location.

[0032] Obtaining the thermal stress field can be achieved by outputting the stress tensor data of each node or element, which quantifies the level of thermo-mechanical coupling load. Predicting the lifetime decay trend based on the creep characteristics of the thermal channel material using the thermal stress field can be done by substituting the thermal stress field into a creep damage accumulation model (such as Larson-Miller or Orr-Sherby-Dorn) for time integration. In a specific embodiment, this operation can be accelerated by using a machine learning surrogate model to accelerate the nonlinear creep integration process, or by introducing Monte Carlo simulation to consider material parameter uncertainties for probabilistic lifetime prediction, thereby enabling dynamic lifetime assessment considering multi-field coupling. Generating lifetime decay trend prediction data can be achieved by outputting the curves of the remaining lifetime or damage index of each region over time, which provides quantifiable lifetime status indicators.

[0033] Step S400: Using lifetime decay trend prediction data, perform thermal channel failure induction analysis on the three-dimensional modeling data of the thermal channel to generate thermal channel failure risk channel data. Dynamic heat load regulation is then implemented in the corresponding thermal channel area using the thermal channel failure risk channel data to perform thermal channel lifetime early warning control operations.

[0034] Among these, hot channel failure initiation analysis can be a simulation process based on lifetime decay trend prediction data, simulating failure triggering conditions in different regions within a 3D model. This can be used to identify high-risk failure paths and critical weak areas. Hot channel failure risk channel data can be spatial data identifying areas with high failure probability and their risk levels within hot channels. This can be used to provide target area guidance for dynamic heat load control. Dynamic heat load control can be an operation that adjusts fuel distribution or cooling strategies in real time based on hot channel failure risk channel data. This can be used to proactively reduce heat load in high-risk areas and delay damage accumulation. Hot channel lifetime early warning control operations can be a set of closed-loop intervention measures implemented based on risk assessment results. This can be used to achieve a complete operation and maintenance closed loop from monitoring to control.

[0035] Using lifetime decay trend prediction data to perform thermal channel failure induction analysis on 3D modeling data of thermal channels can involve marking areas with lifetimes below a threshold in the 3D model and simulating their expansion paths. Generating thermal channel failure risk channel data can output the spatial topology and risk level labels of high-risk areas. Furthermore, this operation can provide precise targets for regulation. Dynamic heat load regulation of the corresponding thermal channel area using thermal channel failure risk channel data can involve adjusting the fuel flow rate or cooling air distribution ratio of the corresponding burner. In a specific embodiment, this operation can achieve localized heat power reduction by adjusting the fuel nozzle opening or optimizing the cooling orifice flow distribution of the guide vanes to enhance cooling in high-risk areas, thereby proactively balancing the heat load and suppressing damage development. Performing thermal channel lifetime early warning control operations can involve issuing regulation commands to the gas turbine control system and recording intervention logs. Furthermore, this operation can complete a closed loop of perception-diagnosis-prediction-regulation.

[0036] For example, in the scenario of thermal channel health management during high-load operation of a gas turbine, the intelligent monitoring and early warning control method for gas turbine power plants in this embodiment can be as follows: During the full-load operation of the gas turbine, the system simultaneously collects exhaust temperature field, gas flow velocity, and thermal channel vibration data; a three-dimensional thermal flow field is reconstructed through multimodal fusion and a fused thermal flow image is generated; after temperature dispersion gradient analysis, the image identifies a weak overheating profile at the leading edge of a blade of the third-stage moving blade; based on this, the thermal stress field of the region is reconstructed, and combined with the creep characteristics of the nickel-based alloy in this area, it is predicted that its remaining life will decline at an accelerating rate; failure initiation analysis shows that a through-crack channel may form in this area; the system then issues a fuel flow fine-tuning command to the corresponding burner to reduce the local heat load and increase the proportion of cooling airflow to the blade, thereby effectively delaying the material damage process and avoiding unplanned shutdowns.

[0037] In one embodiment, multi-source monitoring data from a gas turbine power plant is acquired, including combustion chamber exhaust temperature field data, gas flow velocity distribution data, and thermal channel vibration spectrum data. Multimodal data fusion and three-dimensional thermal flow field reconstruction of the thermal channels are then performed to generate fused thermal flow images and three-dimensional thermal channel modeling data, including: The combustion chamber exhaust temperature field data and thermal channel vibration spectrum data are collected synchronously by a distributed thermocouple array, and the gas flow velocity distribution data are obtained by a laser Doppler velocimeter. Radiation compensation and spatiotemporal registration are performed on the combustion chamber exhaust temperature field data to generate standard temperature field data. Turbulence correction and noise filtering are performed on the gas velocity distribution data to generate velocity gradient distribution data. Based on the vibration spectrum data of the thermal channel, the three-dimensional point cloud of the thermal channel is reconstructed to generate three-dimensional modeling data of the thermal channel. Standard temperature field data, velocity gradient distribution data, and thermal channel 3D modeling data are fused in a multi-scale spatiotemporal manner to generate a fused heat flow image.

[0038] Among them, a distributed thermocouple array can be a synchronous temperature measurement network composed of multiple thermocouples densely arranged on the surface of the gas turbine exhaust section or hot channel. It can be used to provide raw combustion chamber exhaust temperature field data with high spatial density and high temporal synchronization, and can also integrate vibration sensors to achieve co-position measurement. A laser Doppler velocimeter is an optical instrument that measures gas flow velocity non-contactly based on the Doppler frequency shift principle. It can be used to obtain high-precision, high-response velocity distribution data without disturbing the high-speed gas flow field. Synchronously acquiring combustion chamber exhaust temperature field data and hot channel vibration spectrum data through a distributed thermocouple array can be achieved by using thermocouples and accelerometers in the same sensor network to acquire temperature and vibration signals in parallel under the same sampling clock. Using a laser Doppler velocimeter to acquire gas flow velocity distribution data can be achieved by deploying non-contact optical velocimeters at key sections of the hot channel to scan and acquire multi-point velocity vectors. Radiation compensation is a process of physical model correction for temperature measurement deviations caused by high-temperature radiation from thermocouples. It can be used to eliminate false high-temperature readings caused by radiative heat transfer and improve the accuracy of temperature field data. In one specific embodiment, radiation compensation can be applied to the original combustion chamber exhaust temperature field data to output a corrected temperature value for spatiotemporal registration.

[0039] Spatiotemporal registration is the process of unifying temperature data from different locations and sampling times to the same time reference and spatial coordinate system. It ensures the temporal and geometric consistency of multi-point temperature data, supporting subsequent field reconstruction. Radiation compensation and spatiotemporal registration of combustion chamber exhaust temperature field data can be achieved by applying a radiation heat transfer model to correct temperature measurement deviations and aligning the timestamps and spatial coordinates of each measurement point through interpolation or resampling. Furthermore, this operation can be achieved through radiation error inversion compensation based on the Stefan-Boltzmann law or by using a dynamic time warping (DTW) algorithm to time-align asynchronous sampling sequences, thereby generating physically consistent and spatiotemporally aligned standard temperature field data.

[0040] Standard temperature field data can be combustion chamber exhaust temperature distribution data after radiation error correction and temporal-spatial alignment. It can serve as a reliable temperature input for multi-scale spatiotemporal fusion, ensuring the physical accuracy of the reconstructed thermal flow field. Turbulence correction involves separating the average flow and pulsating components from the original velocity signal and reconstructing the steady-state velocity distribution. This can suppress the interference of high-frequency turbulence disturbances on thermal convection modeling and extract effective heat transport characteristics. Noise filtering uses signal processing techniques to remove random noise and instrument interference from velocity measurements, improving the signal-to-noise ratio of velocity data and enhancing the stability of gradient calculations. Turbulence correction and noise filtering of the gas velocity distribution data can involve first extracting the average velocity through low-pass filtering or Reynolds decomposition, and then applying a denoising algorithm to purify the signal. Velocity gradient distribution data, after turbulence correction and filtering, can be vector field data reflecting the spatial rate of change of gas velocity. This can characterize the intensity of local convective heat transfer and accurately express the convection term in thermal-fluid coupling modeling.

[0041] Generating standard temperature field data can be achieved by outputting a temperature matrix or field function after double correction, providing high-quality temperature input for fusion. Generating velocity gradient distribution data can be achieved by performing spatial differentiation on the processed velocity field to output a velocity gradient tensor field, quantifying the potential of local flow shear and convective heat transfer. Reconstructing the 3D point cloud of the thermal channel can be a process of inverting the surface geometry of the structure based on the thermal channel vibration spectrum and generating a discrete spatial point set. This can be used to achieve dynamic geometric modeling of the thermal channel under actual operating conditions, replacing static CAD models. Reconstructing the 3D point cloud of the thermal channel based on its vibration spectrum data can be achieved by utilizing the mapping relationship between structural vibration modes and geometric deformation to invert the coordinates of points on the thermal channel surface under operating conditions. Generating 3D modeling data for the thermal channel can be achieved by fitting the reconstructed point cloud data into a 3D mesh model with material partitioning, providing structural boundary conditions that match the current operating state.

[0042] Multi-scale spatiotemporal fusion of standard temperature field data, velocity gradient distribution data, and 3D modeling data of thermal channels can be performed within a unified coordinate system, weighted according to physical scales (e.g., macroscopic temperature field, microscopic gradient field) and time scale. Furthermore, this operation can be achieved by employing a multi-resolution wavelet fusion framework to integrate features at different scales or by constructing a graph neural network to fuse structure, temperature, and velocity data in a node-edge manner, thereby generating a fused heat flow image with multi-physics semantic consistency. Generating the fused heat flow image can be achieved by outputting the fusion result as a pseudo-color 2D image, preserving spatial location and heat flow intensity information. This operation provides high-fidelity visualization input for subsequent anomaly identification.

[0043] In one embodiment, the thermal channel is reconstructed into a three-dimensional point cloud based on the thermal channel vibration spectrum data, generating three-dimensional modeling data for the thermal channel, including: Frequency domain equalization is performed on the vibration spectrum data of the hot channel to generate spectral equalization data; Frequency domain equalization can be a signal processing operation that corrects the amplitude of the thermal channel vibration spectrum in the frequency dimension to eliminate the influence of sensor response deviation or propagation path attenuation. It can be used to improve the relative reliability of each modal component in the spectrum, providing balanced input for subsequent modal identification. In an exemplary embodiment, frequency domain equalization can be achieved by inverse filtering compensation based on the system frequency response function under reference white noise excitation. Furthermore, frequency domain equalization can also employ an adaptive spectrum shaping algorithm to adjust the gain in low signal-to-noise ratio frequency bands. Performing frequency domain equalization on the thermal channel vibration spectrum data can involve applying a frequency response compensation function to correct the amplitude of each frequency point in the original spectrum. The equalized spectrum data can be the thermal channel vibration spectrum after frequency domain equalization, where the amplitude of each frequency component has been corrected to a physically consistent level. This data can be used as a reliable input for structural modal segmentation, avoiding modal misjudgment caused by uneven frequency response.

[0044] Structural mode segmentation is performed on the thermal channel vibration spectrum data based on spectrum equalization data to distinguish the main structure of the thermal channel and remove interfering modes, thereby generating thermal channel main structure segmentation data. Structural modal segmentation can be a process of identifying and separating the inherent vibration modes of the main load-bearing structure of the thermal channel from the equalized vibration spectrum, and eliminating non-structurally related interference components. It can be used to focus on effective modes that truly reflect the dynamic characteristics of the thermal channel itself, excluding external interference such as support components and airflow excitation. In a specific embodiment, structural modal segmentation can include, but is not limited to, one or more of the following: modal clustering based on modal confidence criteria (MAC), matching and screening using a finite element simulation modal library, and extraction of dominant structural modes based on sparse representation. Structural modal segmentation can take spectral equalization data as input and output thermal channel main structure segmentation data. Performing structural modal segmentation on the thermal channel vibration spectrum data based on spectral equalization data, distinguishing the main structure of the thermal channel and eliminating interference modes, can be achieved by extracting the natural frequencies and mode shapes corresponding to the thermal channel body through modal recognition algorithms, and filtering out non-structurally related components. The thermal channel main structure segmentation data can be a dataset identifying the set of vibration modes and their spatial mode shapes corresponding to the main structure of the thermal channel. This can be used to limit the analysis scope of subsequent thermal deformation estimation, ensuring that deformation inversion is based only on the actual structural response.

[0045] Thermal deformation gradient estimation is performed on the segmented data of the main structure of the thermal channel to generate thermal deformation information data of the thermal channel surface. Thermal deformation gradient estimation can be a process of calculating the spatial rate of change of local deformation on the surface of the thermal channel caused by the temperature gradient based on the vibration modal data of the main structure. It can be used to map the dynamic vibration response to static thermally induced deformation characteristics, indirectly reflecting the local thermal load distribution. For example, thermal deformation gradient estimation can be achieved by using the modal superposition method combined with thermoelastic theory to invert the surface displacement gradient field. In an exemplary embodiment, thermal deformation gradient estimation can also be achieved end-to-end by learning the modal-thermal deformation mapping relationship through a deep neural network. Thermal deformation gradient estimation can take the segmented data of the main structure of the thermal channel as input and output the thermal deformation information data of the thermal channel surface. Thermal deformation gradient estimation of the segmented data of the main structure of the thermal channel can be performed by using thermal-structural coupling theory or a data-driven model to map the modal response to the surface deformation gradient. The thermal deformation information data of the thermal channel surface can be vector field data describing the local deformation and its spatial gradient caused by uneven thermal expansion on the outer surface of the thermal channel, which can be used as geometric and thermal load coupling constraints for surface heat flow reconstruction.

[0046] Based on the thermal deformation information data of the thermal channel surface, surface heat flow reconstruction is performed on the thermal channel to generate thermal channel surface heat flow reconstruction data; Surface heat flux reconstruction can be a numerical process of reconstructing the surface heat flux density distribution of a thermal channel based on the actual thermal deformation surface geometry. This can be used to achieve a synergistic expression of heat flux and geometric deformation, improving the realism of thermal field modeling. Surface heat flux reconstruction can generate surface heat flux reconstruction data for a thermal channel by taking thermal deformation information data of the thermal channel surface as input. Reconstructing surface heat flux for the corresponding thermal channel based on the thermal deformation information data can involve solving the inverse heat conduction problem on the deformed surface or fusing temperature field data to reconstruct the heat flux density. Furthermore, this operation can be achieved through numerical inversion or data fusion strategies, thereby enabling synergistic modeling of heat flux and actual geometric morphology. The surface heat flux reconstruction data for the thermal channel can be continuous heat flux density distribution data defined on a dynamically deformed surface, which can be used to provide a continuous field input with geometric-thermal coupling semantics for thermodynamic meshing.

[0047] Thermodynamic meshing was performed on the thermal flow reconstruction data of the thermal channel surface to obtain the three-dimensional modeling data of the thermal channel.

[0048] Thermodynamic meshing can be the process of discretizing continuous surface heat flow reconstruction data into a finite element mesh model with material properties, thermal boundary conditions, and mechanical degrees of freedom. This can be used to generate structured 3D modeling data that can be directly used for thermal stress calculation and life prediction. In one specific embodiment, thermodynamic meshing can be achieved by generating a high-conformity mesh using curvature-adaptive triangulation. For example, thermodynamic meshing can also employ multiphysics conformal meshing technology to simultaneously meet the requirements of heat conduction and structural mechanics solutions. Thermodynamic meshing of the thermal channel surface heat flow reconstruction data can involve discretizing the continuous surface into a finite element mesh with material partitions, thermal boundaries, and mechanical nodes. Further, this operation can be implemented as described above, thereby generating structured 3D modeling data that can be directly used for multiphysics simulation. The obtained 3D modeling data of the thermal channel can be output as a 3D thermodynamic model file in a standard format (such as HDF5 or VTK), which can support subsequent thermal stress field reconstruction and life prediction.

[0049] For example, in the scenario of monitoring the creep deformation of the hot channel after long-term operation of a gas turbine, the intelligent monitoring and early warning control method for gas turbine power plants in this embodiment can be as follows: After the gas turbine has been running continuously for 5000 hours, the system collects the vibration spectrum of the hot channel; firstly, the spectrum is frequency domain equalized to correct the sensor sensitivity attenuation caused by high temperature; then, the dominant bending mode of the second-stage nozzle ring is identified through structural mode segmentation to eliminate broadband interference caused by cooling airflow pulsation; based on the dominant mode data, the system estimates that there is a significant thermal deformation gradient on the inner arc surface of the nozzle, indicating insufficient local cooling; accordingly, the surface heat flux distribution in this area is reconstructed, showing that the heat flux density is concentrated at the top; finally, a three-dimensional model containing the actual creep deformation geometry is generated through thermodynamic meshing to accurately calculate the thermal stress concentration area and guide the subsequent maintenance priority ranking.

[0050] In one embodiment, temperature dispersion gradient analysis is performed based on the fused heat flow image to generate temperature dispersion gradient data; thermal anomaly contour recognition and localization are performed on the fused heat flow image using the temperature dispersion gradient data to generate initial thermal anomaly localization data, including: Temperature anomaly regions are extracted based on fused heat flow images, and the temperature anomaly regions are divided into core regions to obtain temperature anomaly core region images and temperature anomaly edge region images. Temperature morphology and gradient analysis are performed on the image of the core region of the temperature anomaly to generate temperature feature data of the core region; the directional consistency and heat flux density concentration of the image of the edge region of the temperature anomaly are calculated to obtain the temperature feature data of the edge region. 3D heat flow projection is performed on the fused heat flow image using temperature feature data from the core region and the edge region, and the heat flow value and thermal gradient change after projection are extracted; the dispersion gradient of the temperature anomaly region is calculated based on the heat flow value and thermal gradient change to generate temperature dispersion gradient data. Obtain historical thermal channel design parameters; perform cross-segment structural thermal feature matching of historical thermal channel design parameters based on temperature dispersion gradient data to generate thermal anomaly matching data; perform coordinate transformation and positioning of the thermal channel using thermal anomaly matching data to generate initial thermal anomaly positioning data.

[0051] In this context, an abnormal temperature region can be a set of connected pixels in the fused heat flow image whose temperature significantly deviates from the normal distribution. This set can be used as input for region delineation, identifying the potential range of thermal anomalies. Extracting an abnormal temperature region from the fused heat flow image can be achieved by using adaptive Otsu thresholding combined with morphological closing operations, or by using a Gaussian mixture model to model the temperature distribution and then extracting low-probability tail regions. This can achieve the technical effect of initially delineating the range of potential thermal anomalies.

[0052] Region core segmentation is an image processing operation that divides a temperature anomaly region into core and edge sub-regions based on differences in thermodynamic characteristics. It can be used to distinguish between the high-gradient core region and the diffusion transition edge region within the anomaly, supporting differential feature extraction. Region core segmentation of the temperature anomaly region yields images of the temperature anomaly core region and temperature anomaly edge regions. This can be achieved by dividing the inner and outer regions based on the temperature extreme point as the center, according to the gradient decay ratio or distance threshold. This operation can be implemented using the watershed algorithm with local maxima as seeds for region growth, or by automatically separating the core and edge through the nested relationship of temperature contour lines, thus achieving the technical effect of structural decoupling of the anomaly region. The temperature anomaly core region image can be a sub-image representing the concentrated area of ​​local maximum temperature after region core segmentation, used to analyze superheat intensity, morphological stability, and internal gradient structure. The temperature anomaly edge region image can be an image of the temperature transition zone surrounding the core region, reflecting the characteristics of the heat diffusion boundary, and can be used to assess the consistency of heat flow direction and the degree of energy accumulation.

[0053] Temperature morphology and gradient analysis can be a joint analysis process of geometric morphology description and internal temperature spatial derivative calculation of a core region image. This can be used to quantify the degree of heat concentration and structural stability of the core region. Performing temperature morphology and gradient analysis on images of temperature anomaly core regions generates core region temperature characteristic data, which can include calculating the geometric moments, skewness, kurtosis, and internal gradient statistics of the temperature distribution. Furthermore, this analysis can be achieved by using Zernike moments to describe the symmetry of the core region's temperature morphology, or by analyzing the type of internal gradient structure (such as ridges and valleys) using Hessian matrix eigenvalues, thereby achieving the technical effect of quantifying the heat concentration characteristics and stability of the core region.

[0054] Core region temperature characteristic data can be structured data containing parameters such as core region temperature extrema, shape moments, and internal gradient magnitudes, which can be used to characterize the intensity and evolution trend of local overheating. Directional consistency refers to the degree of alignment of the temperature gradient vector in the local neighborhood of the edge region, which can be used to reflect the orderliness of heat flow diffusion; high consistency indicates directional heat conduction paths. Heat flux density concentration refers to the statistical concentration of the heat flux vector magnitude per unit area in the edge region, which can be used to measure whether heat energy accumulation or dissipation channels form in the edge region.

[0055] The temperature characteristic data of the edge region can be a thermodynamic feature vector composed of directional consistency and heat flux density concentration, which can be used to describe the interaction characteristics between the thermal anomaly boundary and the surrounding medium. In a specific embodiment, the directional consistency and heat flux density concentration of the temperature anomaly edge region image are calculated to obtain the edge region temperature characteristic data. This can be achieved by calculating the directional entropy and magnitude variance of the edge gradient vector field. Furthermore, this calculation can be achieved by using structural tensor analysis to analyze the consistency of the principal directions of the local gradient, or by calculating the concentration exponent of the heat flux vector magnitude through kernel density estimation, thereby achieving the technical effect of characterizing the behavior of the thermal diffusion boundary.

[0056] 3D heat flow projection is a mapping process that reconstructs the heat flow vector field on a 3D geometric model of a thermal channel based on core and edge feature data. It can be used to upscale 2D image features to 3D physical space, enhancing the realism of the reconstructed heat flow field. In this embodiment, 3D heat flow projection is performed on the fused heat flow image using core and edge region temperature feature data. This can be achieved by using 2D feature parameters as constraints to invert the 3D heat flow vector field and project it onto the geometric model. For example, this projection can be achieved by constructing a shallow neural network to map 2D features to a 3D initial heat flow field, followed by optimization using physical equations, or by extrapolating the heat flow direction from the core outwards based on ray tracing and superimposing edge constraints. This achieves the technical effect of improving the physical realism of the reconstructed heat flow field.

[0057] The heat flux value can be the scalar magnitude of the heat flux density at each spatial location after 3D heat flux projection, and can be used as a fundamental physical quantity for dispersity gradient calculation. The thermal gradient change can be the spatial rate of change of the heat flux value along different directions in 3D space, and can be used to reflect the local thermal stress driving potential. In this embodiment, extracting the projected heat flux value and thermal gradient change can be achieved by sampling the scalar heat flux and gradient tensor of key points from the 3D heat flux projection result, thereby achieving the technical effect of obtaining high-dimensional thermodynamic variables for dispersity calculation.

[0058] The dispersion gradient of the temperature anomaly region is calculated based on the heat flux value and the change in the thermal gradient, generating temperature dispersion gradient data. This can be achieved by normalizing the heat flux gradient in 3D space and calculating its spatial coefficient of variation. Furthermore, this calculation can be achieved by using the residual of the anisotropic diffusion equation as a dispersion index, or by using the graphical Laplace operator to measure the smoothness of the heat flow field to infer the degree of dispersion, thereby achieving a more physically accurate representation of dispersion. Historical thermal channel design parameters can be data on thermal channel geometry, material zoning, and expected heat load distribution defined in the original design phase of the gas turbine, which can be used as a benchmark reference for matching the thermal characteristics of cross-section structures. In an exemplary embodiment, historical thermal channel design parameters can be obtained by reading original thermal design data from the gas turbine digital twin model or engineering database, thereby achieving the technical effect of establishing a benchmark reference system for anomaly judgment.

[0059] Cross-segment structural thermal feature matching can be achieved by comparing the current temperature dispersion gradient data with the multi-segment thermal response patterns in historical design parameters. This can be used to identify whether anomalies deviate from the normal thermal behavior under design conditions. In this embodiment, cross-segment structural thermal feature matching is performed on historical thermal channel design parameters based on temperature dispersion gradient data to generate thermal anomaly matching data. This can be achieved by matching the current dispersion pattern with the expected thermal features of each segment (such as the burner outlet segment, the first-stage nozzle segment, etc.) under design conditions. For example, this matching can be achieved by using Dynamic Time Warping (DTW) to match unaligned thermal feature sequences, or by constructing a thermal feature fingerprint database and retrieving the closest design segment using cosine similarity. This can achieve the technical effect of identifying the structural segment to which the anomaly belongs and the type of deviation. Thermal anomaly matching data can be a structured result representing the matching degree and deviation type between the current anomaly pattern and historical design thermal features. This data can be used to provide semantic-level anomaly classification basis for coordinate transformation and positioning.

[0060] Coordinate transformation positioning can be the process of mapping the logical position in thermal anomaly matching data to the three-dimensional physical coordinates of the thermal channel, which can be used to achieve precise positioning from feature space to the physical structure. In a specific embodiment, coordinate transformation positioning is performed on the thermal channel using thermal anomaly matching data to generate initial thermal anomaly positioning data. This can be achieved by converting the logical segment identifier in the matching result into a physical coordinate range in the three-dimensional model. Furthermore, this positioning can be achieved by directly looking up the segment-coordinate mapping table in the design parameters, or by aligning the feature space coordinates to the physical geometric coordinate system through affine transformation, thereby achieving the technical effect of outputting anomaly location information with engineering operability.

[0061] For example, in the scenario of detecting weak thermal anomalies during the transition of gas turbine load, the intelligent monitoring and early warning control method for gas turbine power plants in this embodiment can be as follows: During the process of the gas turbine increasing from 50% load to 100%, the fused heat flow image shows a weak high-temperature patch downstream of a certain burner. The system first extracts the temperature anomaly area and divides it into a central high-temperature core area and an outer diffusion edge area through regional core division; morphological analysis of the core area reveals that it has a single-peak sharp shape and an extremely high internal gradient, while the edge area shows strong directional consistency but low heat flow density concentration, indicating that heat is rapidly diffusing along a specific channel; based on these two characteristics, 3D heat flow projection is performed to reconstruct the heat flow vector distribution of the anomaly in three-dimensional space; the dispersion gradient calculated accordingly is significantly higher than the historical parameters of the corresponding section (leading edge section of the first-stage guide vane) under the design conditions; cross-segment matching confirms that the pattern is inconsistent with the expected thermal response of the design section, and it is determined to be due to partial blockage of the fuel nozzle causing off-center burning; the system then locates the specific blade number and circumferential angle through coordinate transformation, generating initial thermal anomaly location data to provide a precise target for subsequent fuel adjustment.

[0062] In one embodiment, cross-segment structural thermal characteristic matching is performed on historical thermal channel design parameters based on temperature dispersion gradient data, including: Extract the segmental structure from the historical hot passage design parameters to generate hot passage segment-level profile data; Perform a thermodynamic projection transformation on the temperature dispersion gradient data to generate dispersion structure mapping data; By performing inter-segment trajectory overlap analysis on dispersion structure mapping data and thermal channel segment-level profile data, dispersion segment coupling map data is generated. Common thermal features are identified in the dispersion segment coupled spectral data to generate thermal anomaly consistency data; Based on the thermal anomaly consistency data, thermal path consensus reconstruction is performed on the dispersion segment coupling spectrum data to generate thermal anomaly matching data.

[0063] The segment structure extraction can be an operation of dividing discretized structural segments from historical hot channel design parameters according to thermodynamic function or geometric boundaries. This can be used to deconstruct continuous hot channels into segment-level units with defined thermal response characteristics, establishing a structural prior knowledge base. In this embodiment, segment structure extraction can divide the hot channel into several logical segments and extract their cross-sectional features based on functional zoning or thermal boundary conditions in the design drawings. For example, segment structure extraction can automatically divide thermodynamic segments based on the thermal flow line clustering results in CFD simulation, or manually define structural segment boundaries based on material boundaries, cooling hole layout, or geometric abrupt change points, thereby constructing a structural prior model with thermodynamic semantics. Furthermore, the hot channel segment-level cross-sectional data can be structured data obtained after segment structure extraction, characterizing the geometric contours and design heat load distribution of each hot channel segment. This data can be used as a benchmark template for inter-segment matching, providing structural semantic reference.

[0064] Thermodynamic projection transformation is a mathematical transformation process that maps temperature dispersion gradient data from the sensor coordinate system to a thermodynamic space aligned with the geometry of the thermal channel. It can be used to achieve spatial alignment between monitoring data and the physical structure of the equipment, improving comparability. In one specific embodiment, thermodynamic projection transformation can utilize the coordinate mapping relationship in the 3D modeling data of the thermal channel to project the dispersion gradient from the image plane to the surface of the solid structure. Furthermore, thermodynamic projection transformation can achieve field projection on an unstructured mesh through radial basis function interpolation, or back-project the 2D gradient onto a 3D surface and correct for viewpoint distortion through inverse ray tracing, thereby achieving spatial semantic alignment between monitoring data and the physical structure. The dispersion structure mapping data can be the temperature dispersion gradient distribution data expressed in the thermal channel structure coordinate system after thermodynamic projection transformation, which can be used as structural alignment input for coupled analysis with segment-level profiles.

[0065] Inter-segment trajectory overlap analysis can be a quantification process that calculates the degree of overlap and coupling strength between dispersion structure mapping data and each segment-level profile in spatial trajectory. It can be used to identify the thermal channel structure segment most likely associated with the current anomaly. In this embodiment, inter-segment trajectory overlap analysis can calculate the spatial intersection area and gradient direction similarity between high dispersion value regions and each segment profile. For example, inter-segment trajectory overlap analysis can use Hausdorff distance to measure the proximity of the anomaly trajectory to the segment boundary contour, or use a graph matching algorithm to calculate the topological overlap between the anomaly thermal streamline and the designed thermal streamline, thereby quantifying the association strength between the anomaly and each structural segment. Dispersion segment coupling map data can be a multi-dimensional association map characterizing the spatial coupling strength between each thermal channel segment and the current temperature dispersion anomaly. It can be used to visualize the anomaly-structure association relationship and support commonality discrimination.

[0066] Common thermal feature discrimination can be an analytical process of identifying cross-segment regions with similar thermal evolution patterns (such as gradient direction and amplitude trends) in a dispersion segment coupling spectrum. This can be used to filter out isolated noise points and enhance the robustness of real thermal anomaly signals. In one specific embodiment, common thermal feature discrimination can screen regions with similar gradient amplitude, direction, and temporal evolution trends across segments in the coupling spectrum. Furthermore, common thermal feature discrimination can employ spectral clustering to group multiple thermal feature vectors to identify common patterns, or utilize dynamic Bayesian networks to model the temporal dependencies of inter-segment thermal anomalies to determine consistency, thereby suppressing local noise and highlighting systemic thermal anomaly signals. Thermal anomaly consistency data can be a structured result identifying regions exhibiting coordinated thermal behavior in multiple structural segments and their degree of consistency, which can be used to provide high-confidence seed regions for thermal path reconstruction.

[0067] Thermal path consensus reconstruction can be a process of reconstructing continuous abnormal heat conduction paths across multiple segments within the three-dimensional structure of a thermal channel based on thermal anomaly consistency data. This can be used to generate anomaly propagation models that conform to the original heat flow design logic of the equipment. In this embodiment, thermal path consensus reconstruction can use consistent regions as nodes and connect them along the geometric path of the thermal channel to form continuous abnormal heat flow paths. For example, thermal path consensus reconstruction can connect highly consistent segments in the three-dimensional model based on a minimum energy path algorithm, or use graph neural networks to infer the optimal anomaly propagation path on the segment coupling graph, thereby outputting high-confidence anomaly matching results that conform to the physical laws of heat conduction.

[0068] Taking the identification of cross-thermal interference under multi-burner coordinated operation as an example, the intelligent monitoring and early warning control method for gas turbine power plants in this embodiment can be as follows: During gas turbine operation, the fused heat flow image shows that the high-temperature zone spans the downstream areas of two adjacent burners. The system first performs segmental structure extraction on historical design parameters to obtain profile data such as the first-stage nozzle section and transition section corresponding to each burner; the real-time temperature dispersion gradient data is mapped to the thermal channel surface through thermodynamic projection transformation; through inter-segment trajectory overlap analysis, it is found that the anomaly is highly coupled with the downstream sections of burners 5 and 6, and the coupling spectrum shows that there is a strong correlation between the two sections; the thermal feature commonality discrimination further confirms that the two abnormal regions have the same gradient direction and time growth trend, and is determined to be cross-thermal interference caused by uneven fuel distribution; based on this, the system performs thermal path consensus reconstruction, generates a consensus thermal anomaly path that runs through sections 5 and 6, and outputs thermal anomaly matching data, accurately locating the gas crossflow problem caused by the seal failure between the two burners, providing a decision basis for structural alignment for subsequent zone fuel regulation.

[0069] In one embodiment, reference Figure 2The process involves: acquiring creep characteristic data of the thermal channel material; reconstructing the thermal stress field of the thermal channel region based on the initial location data of the thermal anomaly; and predicting the lifetime degradation trend of the thermal channel material based on the thermal stress field, generating lifetime degradation trend prediction data, including: Step S301: Obtain creep characteristic data of thermal channel material, deconstruct the creep constitutive features of thermal channel material creep characteristic data, and generate material creep modulus field data; Step S302: Perform thermal channel region meshing on the initial thermal anomaly location data to generate thermal channel finite element discrete mesh data; solve the thermal stress field on the thermal channel finite element discrete mesh data based on the material creep modulus field data to generate the thermal stress field; Step S303: Extract the thermal gradient direction of the thermal stress field and perform nonlinear coupling prediction on the creep characteristics data of the thermal channel material to generate lifetime decay trend prediction data.

[0070] Creep constitutive feature decomposition can be the process of decomposing material creep characteristic data into a constitutive parameter field that can be embedded in a numerical model according to the temperature-stress dependence. This can be used to realize the spatial mapping of the material's nonlinear response characteristics and support field-specific mechanical simulation. In this embodiment, creep constitutive feature decomposition can take thermal channel material creep characteristic data as input and output material creep modulus field data as output. For example, creep constitutive feature decomposition can fit constitutive equations (such as Norton's law or Garofalo equations) based on material experimental data and map their parameters to spatial coordinates. Furthermore, creep constitutive feature decomposition can generate field data by using a parametric surrogate model (such as Gaussian process regression) to make discrete creep data continuous, or by inverting local creep performance based on material microstructure images and constructing a modulus field, thereby enabling the material's nonlinear behavior to have spatial resolution and supporting localized mechanical simulation.

[0071] Material creep modulus field data can describe the distribution field of creep modulus (such as time-temperature-stress-related stiffness) of the material in the thermal channel at different spatial locations. It can be used as a spatial variable of material properties in solving the thermal stress field, reflecting the local material degradation state. Thermal channel region mesh generation can be an operation of dividing the local thermal channel geometry into finite element units based on the initial location data of the thermal anomaly. This can be used to generate a computational mesh adapted to the shape and gradient characteristics of the anomaly region, improving numerical accuracy. In a specific embodiment, thermal channel region mesh generation can use the initial location data of the thermal anomaly as a regional constraint, outputting discrete finite element mesh data of the thermal channel. The discrete finite element mesh data of the thermal channel can be a structured or unstructured computational mesh consisting of nodes and elements covering the thermal anomaly region, providing a discretized spatial carrier for solving the thermal stress field. Solving the thermal stress field can be a numerical process of solving the thermoelastic or thermoelastic-plastic governing equations to obtain the stress distribution under given material property fields and boundary conditions. This can be used to generate a high-fidelity thermal stress field considering the non-uniform creep characteristics of the material. In this embodiment, the thermal stress field solution can take as input material creep modulus field data and thermal channel finite element discrete mesh data, and output the thermal stress field. The thermal gradient direction can be the principal direction of the temperature gradient vector in the thermal stress field, representing the dominant path of heat transfer and thermal expansion differences. It can be used as the main driving direction for material creep damage evolution, and for constructing direction-sensitive lifetime prediction models. In a specific embodiment, the thermal gradient direction can be extracted by performing spatial differentiation on the thermal stress field or the original temperature field. Nonlinear coupling prediction can be a process of jointly modeling the thermal gradient direction and material creep characteristics under a nonlinear damage framework to predict lifetime decay. It can be used to realize dynamic lifetime extrapolation under a multi-field interaction mechanism of thermo-mechanical-material. In this embodiment, nonlinear coupling prediction can fuse thermal gradient direction and thermal channel material creep characteristic data to generate lifetime decay trend prediction data. Furthermore, nonlinear coupling prediction can use the thermal gradient direction as a constraint on the damage evolution direction, combined with a creep constitutive model for time integration prediction; it can also construct a direction-dependent anisotropic creep damage model for lifetime integration, or use a deep neural network to learn the thermal gradient-creep-life mapping relationship to achieve rapid prediction, thereby realizing dynamic lifetime assessment considering heat flow directionality and material nonlinearity.

[0072] Taking the refined life assessment of local overheated areas as an example, the intelligent monitoring and early warning control method for gas turbine power plants in this embodiment can be as follows: When the system identifies an early thermal anomaly at the root of a certain moving blade, it first performs constitutive deconstruction on the creep test data of the nickel-based alloy to generate a creep modulus field that varies with temperature; then, using the anomaly profile as the boundary, it generates a hexahedral finite element mesh with boundary layer refinement for the local area of ​​the blade; when solving the thermal stress field, the material properties of each element are assigned by the creep modulus field at the corresponding location, thereby obtaining the stress distribution considering material degradation; further, it extracts the principal direction of the thermal gradient in the region and finds that it is significant along the blade height direction; finally, it embeds the information of this direction into the anisotropic creep damage model to predict that the region will experience accelerated life decay in the next 2000 hours, rather than the 5000 hours estimated by the traditional method, providing a basis for accurately arranging maintenance windows.

[0073] In one embodiment, the thermal gradient direction of the thermal stress field is extracted, and nonlinear coupling prediction is performed on the creep characteristic data of the thermal channel material to generate lifetime degradation trend prediction data, including: Multi-scale thermal tensor differentiation processing is performed on the thermal stress field data to generate thermal channel principal thermal gradient tensor data; Multi-scale thermal tensor differential processing involves performing tensor differential operations on thermal stress field data at multiple spatial scales to extract gradient structure features. This can be used to reveal the non-uniformity and dominant conduction direction of the thermal field at different scales. For example, multi-scale thermal tensor differential processing can employ wavelet transform to achieve multi-scale decomposition and then calculate the gradient tensor separately, or utilize structural tensor methods to fuse multi-scale gradient information, thereby enhancing the ability to capture weak but persistent thermal anomaly directional features. The principal thermal gradient tensor data of the thermal channel can be a second-order tensor field characterizing the principal direction, amplitude, and anisotropy of the thermal gradient at various locations within the thermal channel. This can be used to provide a direction-sensitive description of the thermal flux structure for thermal field stability analysis.

[0074] Thermal stability analysis is performed on the principal thermal gradient tensor data of the thermal channel to generate thermally stable directional domain data. Thermal field stability analysis, based on the principal thermal gradient tensor, is an analytical method for assessing the long-term evolution trend of heat flow direction and its impact on structural damage accumulation. It can be used to identify stable heat conduction paths that are prone to continuous heat accumulation. For example, thermal field stability analysis can screen stable regions using the standard deviation of the direction cosine time series, or employ clustering algorithms to perform pattern recognition on gradient direction trajectories, thereby distinguishing between transient disturbances and structural thermal inhomogeneities and focusing on the true risk direction. Thermal stability direction domain data can identify regions in thermal channels where the thermal gradient direction remains stable over the long term and is prone to damage, along with their dominant direction sets. This can be used to focus on the orientation of high-risk thermodynamic effects and guide the directional constraints of lifetime prediction models.

[0075] Dynamic creep partitioning is performed on the creep characteristic data of thermal channel materials to generate layered creep mapping data; Dynamic creep partitioning can be an operation that divides the creep characteristics of a material into several response state intervals based on the local temperature-stress history. It can be used to reflect the differences in nonlinear behavior of materials under different thermo-mechanical coupling conditions. For example, dynamic creep partitioning can be based on K-means clustering to partition temperature-stress-strain rate samples into states, or a hidden Markov model can be used to identify creep stage transition points and divide intervals, thereby enabling the material model to adapt to the nonlinear response under different operating conditions. Hierarchical creep mapping data can be the mapping relationship data between the material's creep response and spatial location after stratifying it according to thermo-mechanical states. It can be used to achieve condition-adaptive expression of the material's nonlinear characteristics.

[0076] Nonlinear interactive fusion of thermal stability directional domain data and hierarchical creep mapping data is performed to generate lifetime decay probability field data. Nonlinear interactive fusion can be a process of coupling and modeling thermally stable directional domain data with hierarchical creep mapping data under physical mechanism constraints. This can be used to construct a thermal-material-directional ternary coupled damage probability field. For example, nonlinear interactive fusion can construct a graph neural network, with nodes as units and edge weights determined by directional consistency and material partition similarity. Alternatively, a Bayesian network can be used to fuse directional domain priors and creep observation data to generate a posterior probability field, thereby achieving physical coupling modeling of thermodynamic directionality and material nonlinear response. Lifetime decay probability field data can describe the probability distribution of lifetime decay in each region of a thermal channel under a given thermodynamic environment, and can be used to provide a lifetime state expression with uncertainty quantification capabilities.

[0077] The lifetime decay probability field data is subjected to time entropy value evolution to generate lifetime decay trend prediction data.

[0078] Among these methods, temporal entropy evolution can be used to model and deduce the uncertainty of the lifetime decay probability field over time based on information entropy theory. It can be used to transform a static probability field into a dynamic trend, reflecting the increase in information disorder during system degradation. For example, temporal entropy evolution can establish an entropy-time differential equation model to fit historical entropy increase trends for prediction, or utilize LSTM networks to learn the entropy evolution law of the probability field sequence, thereby quantifying the uncertainty evolution in lifetime prediction and improving the robustness of long-term predictions.

[0079] Taking the refined lifespan management of a high-parameter gas turbine under long-term variable load operation as an example, the intelligent monitoring and early warning control method for gas turbine power plants in this embodiment can be implemented after the gas turbine has experienced multiple start-ups, shutdowns, and load fluctuations. The system performs multi-scale thermal tensor differentiation on the thermal stress field and discovers a consistent radial principal thermal gradient across scales at the trailing edge of a certain stator blade. Thermal field stability analysis confirms that this direction remains stable during 80% of the operating period and is marked as a thermally stable direction domain. Simultaneously, based on the temperature-stress history experienced by this region, dynamic creep partitioning categorizes it into the "thermal cycle fatigue coupling mapping layer." The nonlinear interactive fusion module calculates that the failure probability of this region under continuous radial heat flow reaches 0.35, forming a local high-probability patch. The time entropy evolution model further predicts that if the current operating strategy is maintained, the entropy value of this region will exceed the critical threshold within 1500 hours, corresponding to an accelerated inflection point in lifespan decay. Based on this, the operation and maintenance system adjusts the burner mix ratio in advance to weaken the heat flow intensity in this direction, effectively delaying the entropy increase process.

[0080] In one embodiment, thermal channel failure induction analysis is performed on the three-dimensional modeling data of the thermal channel using lifetime decay trend prediction data to generate thermal channel failure risk channel data; dynamic heat load regulation is then performed on the corresponding thermal channel area using the thermal channel failure risk channel data to execute thermal channel lifetime early warning control operations, including: Thermal-structure coupling mapping is performed on the three-dimensional modeling data of thermal channels using lifetime decay trend prediction data to generate thermal channel failure path prediction data. The path topology of the hot aisle failure path prediction data is analyzed, and based on the results of the path topology analysis, inter-segment failure nodes are extracted from the hot aisle failure path prediction data to generate failure path node map data. Thermal load simulation is performed on the failure path node map data to generate path thermal load penetration data; dynamic thermal load regulation analysis is performed on the initial location data of thermal anomalies based on the path thermal load penetration data to generate dynamic regulation data of thermal channels. Based on the dynamic control data of the thermal channel, the initial location data of the thermal anomaly is fed back with thermal response data to obtain thermal channel control feedback data, so as to perform thermal channel lifespan early warning control operations.

[0081] The thermal-structural coupling mapping process involves associating lifetime degradation trend prediction data with the three-dimensional geometry and structural properties of thermal channels, establishing a spatial mapping relationship between material degradation state and structural load-bearing capacity. In this embodiment, the thermal-structural coupling mapping process uses lifetime degradation indices as spatial weights, mapping them onto the structural units of the three-dimensional model to establish a degradation-geometric correlation. Thermal channel failure path prediction data can be a set of spatial propagation paths characterizing the potential for continuous structural failure due to material creep degradation within the thermal channel, transforming abstract lifetime degradation into concrete potential crack or damage propagation trajectories.

[0082] By using lifetime degradation trend prediction data to perform thermal-structural coupling mapping on 3D thermal channel modeling data, lifetime degradation indices can be used as spatial weights and mapped onto the structural units of the 3D model to establish a degradation-geometric correlation. Furthermore, this operation can be achieved by employing finite element submodeling techniques to refine the mesh in high-degradation regions and assign degradation material parameters, or by embedding lifetime data into the feature vectors of 3D model nodes using graph neural networks, thereby achieving spatial alignment between the material degradation state and the structural load-bearing capacity. Generating thermal channel failure path prediction data can be based on the mapped degradation model, identifying the geometric paths most likely to lead to continuous damage, thus outputting a concrete failure evolution trajectory.

[0083] Path topology can be a graph-theoretic representation describing the nodes, branch structures, and connectivity in a failure path, and can be used to reveal critical links and hub regions in failure propagation. In an exemplary embodiment, path topology can serve as the analytical basis for inter-segment failure node extraction. Inter-segment failure node extraction can be an operation that identifies key transition nodes connecting different structural segments (such as leaf root-leaf body-leaf tip) in the failure path topology, and can be used to locate high-risk failure trigger points caused by structural geometric abrupt changes or load concentration.

[0084] Analyzing the path topology of hot aisle failure path prediction data can involve converting the failure paths into a graph structure and extracting topological features such as node degree and betweenness centrality. Further, this operation can be achieved by using community detection algorithms to identify functional sub-modules within the path, or by applying shortest path algorithms to identify the dominant failure propagation direction, thereby identifying critical propagation links and structural hubs. Based on the path topology analysis results, inter-segment failure node extraction can be performed on the hot aisle failure path prediction data. This can be done by marking high-risk nodes at the boundaries connecting different structural segments (such as the leaf root and leaf blade), thereby locating stress concentration failure points caused by geometrical abrupt changes. Failure path node graph data can be a dataset of key nodes and their components organized in a graph structure, which can be used to provide a discretized, networked analytical framework for thermal load simulation. Generating failure path node graph data can involve constructing a graph data structure containing node coordinates, connecting edges, and risk weights, thereby providing a computable failure network model.

[0085] Thermal load simulation can be a numerical calculation that simulates heat flow distribution and heat conduction processes on a failure path node map, used to quantify the penetration depth and intensity of heat energy along the failure path. Performing thermal load simulation on failure path node map data can involve applying thermal boundary conditions to the nodes and solving the heat conduction equation to simulate heat flow penetration. Furthermore, this operation can be simplified to graph edge weight calculation using a network thermal resistance model, or achieved by combining CFD results to perform local high-fidelity thermal simulations of key nodes, thereby quantifying the propagation characteristics of heat energy along the failure path.

[0086] Path heat load penetration data can describe the energy transfer efficiency and cumulative effect of heat flow at each node and connecting segment of the failure path, and can be used to reveal how local thermal anomalies accelerate damage to distant regions through structural paths. Generating path heat load penetration data can output the heat flux, temperature rise rate, and heat accumulation index of each node, thereby revealing the accelerated failure mechanism under thermal-structural coupling. Dynamic heat load regulation analysis can be a process of formulating differentiated heat load adjustment strategies by combining path heat load penetration characteristics and the original thermal anomaly location, and can be used to generate regulation instructions with path suppression intent. In an exemplary embodiment, dynamic heat load regulation analysis can fuse path heat load penetration data and initial thermal anomaly location data to output dynamic regulation data of thermal channels. Dynamic heat load regulation analysis based on path heat load penetration data and initial thermal anomaly location data can set the original thermal anomaly area corresponding to the high-permeability path as the priority regulation target and formulate differentiated fuel / cooling allocation strategies. Furthermore, this operation can be achieved by implementing fuel reduction on the upstream heat source of the high-permeability path to block heat input, or by enhancing local cooling airflow at key nodes of the path to break the heat conduction chain, thereby generating a precise regulation scheme with path suppression intent.

[0087] Dynamic control data for the thermal channel can be a structured instruction set containing the control target area, control intensity, and timing, which can be used to guide the gas turbine control system to implement precise thermal load intervention. Generating dynamic control data for the thermal channel can involve converting the control strategy into executable control parameters (such as nozzle opening and cooling valve position), thus providing an instruction set that can be issued to the actuators. Thermal response feedback can be a verification mechanism for re-monitoring and comparing changes in the thermal channel temperature field after control implementation, which can be used to evaluate the actual thermal suppression effect of the control measures. Furthermore, thermal response feedback can use dynamic control data for the thermal channel as input, combined with updated initial location data of thermal anomalies, to generate thermal channel control feedback data. Performing thermal response feedback on the initial location data of thermal anomalies based on dynamic control data can involve re-collecting temperature field data after control implementation and comparing changes in the thermal anomaly area before and after control. Furthermore, this operation can be achieved by using differential image analysis to quantify the temperature drop magnitude, or by updating the thermal anomaly contour recognition model through online learning to adapt to new operating conditions, thereby verifying the effectiveness of control and supporting strategy iteration.

[0088] Hot channel control feedback data can reflect the temperature change trend and failure risk evolution in the thermal anomaly area after control execution, and can be used to support continuous iteration of closed-loop optimization and early warning control operations. Obtaining hot channel control feedback data can output control effect evaluation indicators (such as temperature drop, gradient improvement rate, and risk level change), thus completing the closed-loop verification process. Executing hot channel lifespan early warning control operations can involve incorporating control feedback data into the operation and maintenance decision-making system to trigger early warnings or maintain the current strategy.

[0089] In one embodiment, thermal load simulation is performed on the failure path node map data to generate path thermal load penetration data; based on the path thermal load penetration data, dynamic thermal load regulation analysis is performed on the initial location data of thermal anomalies to generate dynamic regulation data of thermal channels, including: Multiphysics thermal response modeling is performed on the failure path node map data to generate local node thermal coupling model data; time-series thermal disturbance simulation is performed on the local node thermal coupling model data to generate thermal load response sequence data. Three-dimensional thermal field inversion calculations are performed on the heat load response sequence data to generate path heat load infiltration data; Non-uniform heat-sensitive region clustering is performed on the path heat load infiltration data to generate high-risk heat penetration cluster data; location heat gradient cross-analysis is performed on the high-risk heat penetration cluster data and the initial location data of thermal anomalies to generate dynamic heat load configuration data. A global thermal control simulation of the thermal channel is performed on the dynamic heat load configuration data to generate dynamic control data for the thermal channel.

[0090] Multiphysics thermal response modeling can be achieved by constructing a local physical model at each key node of the failure path node map, incorporating the coupling effects of heat conduction, fluid convection, and structural deformation. This model can transform abstract nodes into computational units with realistic thermo-mechanical-fluid response characteristics. In this embodiment, multiphysics thermal response modeling can take failure path node map data as input and output local node thermal coupling model data. The local node thermal coupling model data can be a set of parameterized models describing the thermal response characteristics of each failure path node under the action of multiphysics, which can be used to support dynamic thermal disturbance simulation and reflect the sensitivity of nodes to changes in thermal load. Multiphysics thermal response modeling of failure path node map data can be achieved by establishing a local model at each map node that couples the heat conduction equation, the Navier-Stokes equation, and the elasticity equation. Furthermore, this operation can be achieved by using reduced-order model (ROM) technology to compress the multiphysics equations to improve computational efficiency, or by embedding measured vibration spectrum data to correct structural damping terms to enhance the realism of dynamic response, thereby endowing the failure nodes with realistic physical response capabilities.

[0091] Generating local node thermal coupling model data can involve outputting the material properties, boundary conditions, coupling coefficients, and geometric simplification parameters of each node. This operation can create a simulateable node-level digital twin. Performing time-series thermal perturbation simulations on the local node thermal coupling model data can be achieved by applying typical gas turbine load variation curves (such as start-up, shutdown, and peak shaving) as thermal boundary conditions for transient solutions. Furthermore, this operation can be achieved by using random thermal perturbation sequences to simulate high-frequency thermal oscillations caused by combustion instability, or by introducing multi-condition combined perturbations (high load + low cooling flow) to test the node's limiting response, thereby capturing the node's thermal inertia and hysteresis characteristics under dynamic operating conditions.

[0092] Thermal load response sequence data can be a multidimensional time series recording the changes in physical quantities such as temperature and heat flux at each node under time-series thermal perturbation. It can be used as an observational basis for three-dimensional thermal field inversion, reflecting the time delay and cumulative effect of heat propagation. Generating thermal load response sequence data can involve recording the changes in temperature, heat flux density, and thermal strain at key points during the simulation process. This operation provides a time-dimensional observation of the heat propagation process. Three-dimensional thermal field inversion calculations can be performed on the thermal load response sequence data, using Tikhonov regularization or Bayesian inference methods to infer the continuous heat source distribution from the discrete response. Furthermore, this operation can be accelerated by combining a neural network surrogate model or by introducing prior thermal flow field constraints to improve inversion stability, thereby reconstructing the penetration path and intensity of thermal energy in three-dimensional space.

[0093] Three-dimensional thermal field inversion calculations can reconstruct the heat flow infiltration path and intensity distribution in three-dimensional space based on heat load response sequence data and through an inverse heat conduction algorithm. This can be used to map discrete node responses back to continuous space and generate a physically consistent heat infiltration field. In this embodiment, the three-dimensional thermal field inversion calculation can take heat load response sequence data as input and output path heat load infiltration data. Generating path heat load infiltration data can be achieved by outputting a heat flux vector field or an equivalent heat permeability scalar field on a three-dimensional mesh. This operation can reveal the heat flow conduction mechanism along the failure path.

[0094] Path heat load penetration data can be continuous field data, characterized by the intensity and direction of heat flow penetration in the failure path space, obtained through three-dimensional thermal field inversion calculations. This data can be used for subsequent high-risk area identification and control strategy generation. Non-uniform thermally sensitive region clustering identification of path heat load penetration data can be performed using DBSCAN or spectral clustering algorithms to automatically group high-thermal-sensitivity regions. Furthermore, this operation can be enhanced by combining the consistency of thermal gradient direction as a clustering feature to improve spatial coherence, or by introducing the difference in thermal time constant as a second clustering dimension to distinguish between fast and slow response zones. This allows for objective identification of high-risk heat penetration areas and avoids bias from empirical thresholds.

[0095] Non-uniform thermally sensitive region clustering identification involves unsupervised clustering of regions in path heat load penetration data with significantly higher thermal response sensitivity than their neighbors. This can be used to automatically identify high-risk thermal penetration clusters, avoiding the subjectivity of manually set thresholds. High-risk thermal penetration cluster data can be a set of spatially continuous regions with high thermal penetration tendency identified through clustering, along with their thermal sensitivity characteristics. This data can be used as the core target area for dynamic regulation, focusing intervention resources. Generating high-risk thermal penetration cluster data can output the geometric boundaries, center coordinates, thermal sensitivity index, and confidence level of the clustering results. This operation can provide a structured list of high-risk regions.

[0096] Locational thermal gradient cross-analysis of high-risk heat penetration cluster data and initial thermal anomaly location data can be performed by calculating the gradient vector dot product or correlation coefficient of the two datasets in the spatially overlapping region. Furthermore, this operation can be achieved by constructing a joint thermal gradient field and extracting saddle points or extreme points as control focuses, or by using image registration technology to align data from different sources and then performing pixel-level cross-analysis. This can identify areas where heat source-path synergistic aggravation occurs and improve the targeting of control measures. Locational thermal gradient cross-analysis can involve superimposing, comparing, and correlating the spatial location of high-risk heat penetration clusters with the local temperature gradient field in the initial thermal anomaly location data. This can be used to synthesize the original anomaly source and heat propagation path, and identify areas where gradient superposition amplification effects occur. In this embodiment, locational thermal gradient cross-analysis can take high-risk heat penetration cluster data and initial thermal anomaly location data as inputs and output dynamic heat load configuration data.

[0097] Dynamic heat load configuration data can be strategy data including fuel distribution adjustments, cooling airflow redistribution ratios, and control priorities for high-risk areas. This data can provide a physically interpretable and spatially precise basis for control commands. In an exemplary embodiment, dynamic heat load configuration data can be used to generate parameters such as fuel nozzle adjustment amounts and cooling orifice opening adjustment values ​​based on cross-analysis results, thus forming a physically executable intervention strategy. Generating dynamic heat load configuration data can involve generating parameters such as fuel nozzle adjustment amounts and cooling orifice opening adjustment values ​​based on cross-analysis results. This operation can form a physically executable intervention strategy.

[0098] Global thermal control simulation of the hot channel based on dynamic heat load configuration data can be performed by applying the configuration data to the whole-machine CFD-thermo-structure coupling model and evaluating the overall thermal response. Furthermore, this operation can be achieved by employing multi-fidelity simulation or introducing a control delay model to simulate the impact of actuator response lag on the effect, thereby ensuring the safety and effectiveness of the control strategy at the system level. Global thermal control simulation of the hot channel can simulate the overall temperature, stress, and flow field response after the implementation of dynamic heat load configuration in the whole-machine scale hot channel model. This can be used to verify the global feasibility and side effects of the control strategy and ensure that secondary thermal anomalies are not triggered. In this embodiment, global thermal control simulation of the hot channel can take dynamic heat load configuration data as input and output dynamic control data of the hot channel. Generating dynamic control data of the hot channel can output a globally verified final control instruction set, including the execution device, parameter values, and effective timing. This operation can provide closed-loop instructions that can be directly issued to the gas turbine control system.

[0099] Taking the risk prevention and control of thermal breakdown at the leading edge of the high-temperature blade as an example, the intelligent monitoring and early warning control method for gas turbine power plants in this embodiment can be as follows: The system identifies a certain area at the leading edge of the blade as a critical node in the failure path node map; a local coupled model including gas scouring, internal cooling and thermal barrier coating degradation is constructed through multiphysics modeling; time-series simulation shows that the temperature of this node rises sharply by 85K within 30 seconds after a load step; three-dimensional thermal field inversion reveals that heat rapidly penetrates from the leading edge to the blade tip; clustering identifies a high-risk thermal penetration cluster located in the middle of the leading edge; this area has a strong gradient superposition with the original thermal anomaly location data; cross-analysis suggests reducing the fuel by 4% and increasing the leading edge impact cooling flow for the corresponding burner in this area; global thermal control simulation confirms that this strategy will not cause adjacent blades to overcool or combustion oscillation; finally, dynamic control data is generated and issued for execution, successfully suppressing the thermal penetration trend.

[0100] In addition, refer to Figure 3 To achieve the above objectives, the present invention also provides an intelligent monitoring and early warning control system for gas turbine power plants, the system comprising: The multimodal fusion module 10 is used to acquire multi-source monitoring data of the gas turbine power plant, including combustion chamber exhaust temperature field data, gas flow velocity distribution data and thermal channel vibration spectrum data, and to perform multimodal data fusion and thermal channel three-dimensional thermal flow field reconstruction to generate fused thermal flow image and thermal channel three-dimensional modeling data. Anomaly localization module 20 is used to perform temperature dispersion gradient analysis based on fused heat flow image to generate temperature dispersion gradient data; and to perform thermal anomaly contour recognition and localization on fused heat flow image using temperature dispersion gradient data to generate initial thermal anomaly localization data. The lifetime prediction module 30 is used to acquire creep characteristic data of thermal channel material; reconstruct the thermal stress field of the thermal channel region based on the initial location data of thermal anomaly to obtain the thermal stress field; and predict the lifetime decay trend based on the creep characteristic data of thermal channel material using the thermal stress field to generate lifetime decay trend prediction data. The dynamic control module 40 is used to perform thermal channel failure induction analysis on the three-dimensional modeling data of the thermal channel using the life decay trend prediction data, and generate thermal channel failure risk channel data; and to perform dynamic heat load control on the thermal channel area through the thermal channel failure risk channel data to execute thermal channel life early warning control operations.

[0101] Other embodiments or specific implementations of the intelligent monitoring and early warning control system for gas turbine power plants described in this invention can be referred to the above-mentioned method embodiments, and will not be repeated here.

[0102] Furthermore, to achieve the above objectives, the present invention also provides an intelligent monitoring and early warning control device for gas turbine power plants. The device includes: a memory, a processor, and an intelligent monitoring and early warning control program for gas turbine power plants stored in the memory and executable on the processor. The intelligent monitoring and early warning control program for gas turbine power plants is configured to implement the steps of the intelligent monitoring and early warning control method for gas turbine power plants as described in any of the above descriptions.

[0103] In addition, to achieve the above objectives, the present invention also provides a medium storing an intelligent monitoring and early warning control program for a gas turbine power plant, wherein the intelligent monitoring and early warning control program for a gas turbine power plant, when executed by a processor, implements the steps of the intelligent monitoring and early warning control method for a gas turbine power plant as described above.

[0104] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A smart monitoring and early warning control method for gas turbine power plants, characterized in that, The method includes: Acquire multi-source monitoring data from gas turbine power plants, including combustion chamber exhaust temperature field data, gas flow velocity distribution data, and thermal channel vibration spectrum data, and perform multi-modal data fusion and three-dimensional thermal flow field reconstruction of thermal channels to generate fused thermal flow images and three-dimensional modeling data of thermal channels; Temperature dispersion gradient analysis is performed on the fused heat flow image to generate temperature dispersion gradient data; thermal anomaly contour recognition and localization are performed on the fused heat flow image using the temperature dispersion gradient data to generate initial thermal anomaly localization data. Acquire creep characteristic data of thermal channel material; reconstruct thermal stress field of thermal channel region based on initial thermal anomaly location data to obtain thermal stress field; predict life decay trend based on thermal stress field and creep characteristic data of thermal channel material to generate life decay trend prediction data. Using lifetime decay trend prediction data, thermal channel failure induction analysis is performed on the 3D modeling data of the thermal channel to generate thermal channel failure risk channel data; the thermal channel failure risk channel data is used to dynamically regulate the thermal load of the corresponding thermal channel area in order to perform thermal channel lifetime early warning control operations.

2. The intelligent monitoring and early warning control method for gas turbine power plants as described in claim 1, characterized in that, The acquisition of multi-source monitoring data from the gas turbine power plant includes combustion chamber exhaust temperature field data, gas flow velocity distribution data, and thermal channel vibration spectrum data. Multi-modal data fusion and three-dimensional thermal flow field reconstruction of the thermal channels are then performed to generate fused thermal flow images and three-dimensional thermal channel modeling data, including: The combustion chamber exhaust temperature field data and thermal channel vibration spectrum data are collected synchronously by a distributed thermocouple array, and the gas flow velocity distribution data are obtained by a laser Doppler velocimeter. Radiation compensation and spatiotemporal registration are performed on the combustion chamber exhaust temperature field data to generate standard temperature field data. Turbulence correction and noise filtering are performed on the gas velocity distribution data to generate velocity gradient distribution data. Based on the vibration spectrum data of the thermal channel, the three-dimensional point cloud of the thermal channel is reconstructed to generate three-dimensional modeling data of the thermal channel. Standard temperature field data, velocity gradient distribution data, and thermal channel 3D modeling data are fused in a multi-scale spatiotemporal manner to generate a fused heat flow image.

3. The intelligent monitoring and early warning control method for gas turbine power plants as described in claim 2, characterized in that, The process of reconstructing the three-dimensional point cloud of the thermal channel based on the vibration spectrum data of the thermal channel to generate three-dimensional modeling data of the thermal channel includes: Frequency domain equalization is performed on the vibration spectrum data of the hot channel to generate spectral equalization data; Structural mode segmentation is performed on the thermal channel vibration spectrum data based on spectrum equalization data to distinguish the main structure of the thermal channel and remove interfering modes, thereby generating thermal channel main structure segmentation data. Thermal deformation gradient estimation is performed on the segmented data of the main structure of the thermal channel to generate thermal deformation information data of the thermal channel surface; based on the thermal deformation information data of the thermal channel surface, surface heat flow reconstruction is performed on the thermal channel to generate surface heat flow reconstruction data of the thermal channel. Thermodynamic meshing was performed on the thermal flow reconstruction data of the thermal channel surface to obtain the three-dimensional modeling data of the thermal channel.

4. The intelligent monitoring and early warning control method for gas turbine power plants as described in claim 1, characterized in that, The temperature dispersion gradient analysis is performed based on the fused heat flow image to generate temperature dispersion gradient data; Thermal anomaly contour recognition and localization are performed on the fused heat flow image using temperature dispersion gradient data, generating initial thermal anomaly localization data, including: Temperature anomaly regions are extracted based on fused heat flow images, and the temperature anomaly regions are divided into core regions to obtain temperature anomaly core region images and temperature anomaly edge region images. Temperature morphology and gradient analysis are performed on the image of the core region of the temperature anomaly to generate temperature feature data of the core region; the directional consistency and heat flux density concentration of the image of the edge region of the temperature anomaly are calculated to obtain the temperature feature data of the edge region. 3D heat flow projection is performed on the fused heat flow image using temperature feature data from the core region and the edge region, and the heat flow value and thermal gradient change after projection are extracted; the dispersion gradient of the temperature anomaly region is calculated based on the heat flow value and thermal gradient change to generate temperature dispersion gradient data. Obtain historical thermal channel design parameters; perform cross-segment structural thermal feature matching of historical thermal channel design parameters based on temperature dispersion gradient data to generate thermal anomaly matching data; perform coordinate transformation and positioning of the thermal channel using thermal anomaly matching data to generate initial thermal anomaly positioning data.

5. The intelligent monitoring and early warning control method for gas turbine power plants as described in claim 4, characterized in that, The method of matching cross-segment structural thermal characteristics of historical thermal channel design parameters based on temperature dispersion gradient data includes: Extract the segmental structure from the historical hot passage design parameters to generate hot passage segment-level profile data; Perform a thermodynamic projection transformation on the temperature dispersion gradient data to generate dispersion structure mapping data; By performing inter-segment trajectory overlap analysis on dispersion structure mapping data and thermal channel segment-level profile data, dispersion segment coupling map data is generated. Common thermal features are identified in the dispersion segment coupled spectral data to generate thermal anomaly consistency data; Based on the thermal anomaly consistency data, thermal path consensus reconstruction is performed on the dispersion segment coupling spectrum data to generate thermal anomaly matching data.

6. The intelligent monitoring and early warning control method for gas turbine power plants as described in claim 1, characterized in that, The process involves acquiring creep characteristic data of the thermal channel material; reconstructing the thermal stress field in the thermal channel region based on the initial location data of the thermal anomaly; and obtaining the thermal stress field. Based on the thermal stress field, the creep characteristics data of the thermal channel material are used to predict the lifetime degradation trend, generating lifetime degradation trend prediction data, including: Acquire creep property data of thermal channel materials, deconstruct the creep constitutive features of the thermal channel material creep property data, and generate material creep modulus field data; The initial location data of the thermal anomaly is meshed into a thermal channel region to generate finite element discrete mesh data of the thermal channel; the thermal stress field is solved based on the material creep modulus field data of the thermal channel finite element discrete mesh data to generate the thermal stress field. The thermal gradient direction of the thermal stress field is extracted, and nonlinear coupling prediction is performed on the creep characteristics data of the thermal channel material to generate lifetime decay trend prediction data.

7. The intelligent monitoring and early warning control method for gas turbine power plants as described in claim 6, characterized in that, The extraction of the thermal gradient direction of the thermal stress field and the nonlinear coupling prediction of the creep characteristics data of the thermal channel material to generate lifetime degradation trend prediction data include: Multi-scale thermal tensor differentiation processing is performed on the thermal stress field data to generate thermal channel principal thermal gradient tensor data; thermal field stability analysis is performed on the thermal channel principal thermal gradient tensor data to generate thermally stable directional domain data. Dynamic creep partitioning is performed on the creep characteristic data of thermal channel materials to generate layered creep mapping data; Nonlinear interactive fusion of thermal stability directional domain data and hierarchical creep mapping data is performed to generate lifetime decay probability field data. The lifetime decay probability field data is subjected to time entropy value evolution to generate lifetime decay trend prediction data.

8. The intelligent monitoring and early warning control method for gas turbine power plants as described in claim 1, characterized in that, The method utilizes lifetime decay trend prediction data to perform thermal channel failure induction analysis on thermal channel three-dimensional modeling data, generating thermal channel failure risk channel data. Dynamic heat load regulation of the hot aisle area is performed based on hot aisle failure risk data to execute hot aisle life early warning control operations, including: Thermal-structure coupling mapping is performed on the three-dimensional modeling data of thermal channels using lifetime decay trend prediction data to generate thermal channel failure path prediction data. The path topology of the hot aisle failure path prediction data is analyzed, and based on the results of the path topology analysis, inter-segment failure nodes are extracted from the hot aisle failure path prediction data to generate failure path node map data. Thermal load simulation is performed on the failure path node map data to generate path thermal load penetration data; dynamic thermal load regulation analysis is performed on the initial location data of thermal anomalies based on the path thermal load penetration data to generate dynamic regulation data of thermal channels. Based on the dynamic control data of the thermal channel, the initial location data of the thermal anomaly is fed back with thermal response data to obtain thermal channel control feedback data, so as to perform thermal channel lifespan early warning control operations.

9. The intelligent monitoring and early warning control method for gas turbine power plants as described in claim 8, characterized in that, The thermal load simulation is performed on the failure path node map data to generate path thermal load penetration data. Dynamic heat load regulation analysis is performed on the initial location data of thermal anomalies based on path heat load infiltration data to generate dynamic regulation data of thermal channels, including: Multiphysics thermal response modeling is performed on the failure path node map data to generate local node thermal coupling model data; time-series thermal disturbance simulation is performed on the local node thermal coupling model data to generate thermal load response sequence data. Three-dimensional thermal field inversion calculations are performed on the heat load response sequence data to generate path heat load infiltration data; Non-uniform heat-sensitive region clustering is performed on the path heat load infiltration data to generate high-risk heat penetration cluster data; location heat gradient cross-analysis is performed on the high-risk heat penetration cluster data and the initial location data of thermal anomalies to generate dynamic heat load configuration data. A global thermal control simulation of the thermal channel is performed on the dynamic heat load configuration data to generate dynamic control data for the thermal channel.

10. An intelligent monitoring and early warning control system for gas turbine power plants, characterized in that, The system includes: The multimodal fusion module is used to acquire multi-source monitoring data from gas turbine power plants, including combustion chamber exhaust temperature field data, gas flow velocity distribution data, and thermal channel vibration spectrum data. It also performs multimodal data fusion and thermal channel three-dimensional thermal flow field reconstruction to generate fused thermal flow images and thermal channel three-dimensional modeling data. The anomaly localization module is used to perform temperature dispersion gradient analysis based on the fused heat flow image to generate temperature dispersion gradient data; and to perform thermal anomaly contour recognition and localization on the fused heat flow image using the temperature dispersion gradient data to generate initial thermal anomaly localization data. The lifetime prediction module is used to acquire creep characteristic data of thermal channel materials; reconstruct the thermal stress field of the thermal channel region based on the initial location data of thermal anomalies to obtain the thermal stress field; and predict the lifetime decay trend based on the creep characteristic data of thermal channel materials using the thermal stress field to generate lifetime decay trend prediction data. The dynamic control module is used to perform thermal channel failure induction analysis on the three-dimensional modeling data of the thermal channel using the life decay trend prediction data, and generate thermal channel failure risk channel data; and to perform dynamic heat load control on the thermal channel area based on the thermal channel failure risk channel data, so as to execute thermal channel life early warning control operations.

Images