Steam turbine rising rate and steam temperature cooperative control method based on allowable temperature difference feedback

Abstract

The invention discloses a turbine rising rate and steam temperature cooperative control method based on allowable temperature difference feedback, and belongs to the field of turbine operation control, and the method comprises the steps: collecting deep peak regulation operation data, and carrying out cleaning and normalization; generating thermal state prediction data by using the unsteady heat conduction analytical model; establishing a cyclic fatigue and creep damage superposition model to calculate an accumulated damage value; dynamically correcting an allowable temperature difference limit value envelope line; according to the stress sensitivity weight, control vectors are distributed between the steam superheat degree loop and the rising rate loop, and a collaborative decision instruction is generated; and feedforward compensation is carried out by using the extended state observer and a final control signal is issued. According to the method, the digital twinborn prediction and multivariable decoupling cooperation technology is adopted, the delay of the executing mechanism can be compensated in real time, the thermal stress track can be actively regulated and controlled, and the operation safety and the whole-life-cycle economical efficiency of the unit under the frequent variable working condition are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of steam turbine operation control, and in particular to a method for coordinated control of steam turbine acceleration rate and steam temperature based on allowable temperature difference feedback. Background Technology

[0002] As the core power equipment in thermal power plants, steam turbines play a crucial role in the deep peak shaving process of the power system. To ensure the safe operation of the unit under large load fluctuations, it is necessary to closely monitor and control the thermal stress and life loss of key high-temperature components such as rotors.

[0003] Among related technologies, Chinese invention patent CN120175433A discloses a steam turbine operation control method, intelligent agent, and electronic device. This technology obtains the current operating parameters of the high-temperature components of the steam turbine and inputs them into stress prediction models for the start-up and shutdown phases and load adjustment phases, respectively, to obtain the corresponding predicted stress parameters. It then calculates the predicted values ​​of low-cycle fatigue crack initiation loss and cumulative loss, and finally determines the target depth peak-shaving operation control parameters that meet stress and life constraints, thereby controlling the steam turbine.

[0004] Regarding the aforementioned technologies, the use of single-stage stress prediction models and traditional life loss assessment methods is insufficient to fully account for the dynamic changes in the safety boundary of the rotor under complex operating conditions due to the accumulation of historical service life. Furthermore, the lack of feedforward compensation for mechanical hysteresis and total system disturbance at the execution level results in room for improvement in control accuracy and response quality, which is detrimental to the economical operation of the unit throughout its entire life cycle. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a method for coordinated control of turbine acceleration rate and steam temperature based on allowable temperature difference feedback. This method employs real-time analysis of internal rotor thermal stress, dynamic assessment of accumulated damage to correct safety boundaries, and coordinated allocation of acceleration rate and steam temperature control vectors for feedforward compensation control. This enables precise, safe, and efficient control of thermal stress in the turbine during deep peak shaving.

[0006] To achieve the above objectives, this application adopts the following technical solution: Firstly, a method for coordinated control of turbine acceleration rate and steam temperature based on allowable temperature difference feedback is provided, including: real-time acquisition of turbine operating data during deep peak shaving; cleaning and normalizing the operating data using thermodynamic state equations to form a basic dataset supporting the calculation of thermal kinetic energy gradient, wherein the operating data includes main steam pressure, steam flow rate, regulating stage enthalpy, and multi-point temperature measurements distributed on the turbine rotor surface; inputting the basic dataset into a preset unsteady heat conduction analytical model, using the rotor's material thermal conductivity, specific heat capacity, and surface heat transfer coefficient determined by steam flow rate and pressure to analyze the equivalent stress field at unmeasurable points inside the rotor in real time, generating thermal state prediction data characterizing the real-time evolution trajectory of rotor thermal stress; retrieving start-stop frequency and load fluctuation amplitude from the rotor's historical operating records to establish cyclic fatigue damage caused by alternating thermal stress and creep damage accumulated over time under high-temperature conditions. The superimposed mathematical model calculates the current cumulative fatigue damage value of the rotor. Based on the cumulative fatigue damage value, the standard allowable temperature difference is nonlinearly contracted or expanded using a fracture mechanics correction function to establish the dynamic allowable temperature difference limit envelope at the current moment. The thermal state prediction data is compared with the dynamic allowable temperature difference limit envelope in real time to determine the stress safety margin. According to the stress sensitivity weight, control vectors are allocated between the steam superheat regulation loop and the rate of increase regulation loop to generate collaborative decision instructions. Based on the vector allocation results in the collaborative decision instructions, the corresponding steam temperature regulation target value and rate of increase regulation target value are calculated, and corresponding PID control instructions are generated respectively. The mechanical hysteresis of the actuator and the total system disturbance are captured by the expanded state observer. The PID control instructions are then fed forward to compensate and correct, generating the final control signal and sending it to the turbine speed control system and the steam desuperheating water actuator.

[0007] Based on the above technical solution, the turbine acceleration rate and steam temperature coordinated control method based on allowable temperature difference feedback provided in this application adopts the means of real-time analysis of internal rotor thermal stress, dynamic evaluation of cumulative damage to correct safety boundaries, and coordinated allocation of acceleration rate and steam temperature control vectors for feedforward compensation control, which can achieve accurate, safe and efficient control of thermal stress of the turbine during deep peak shaving.

[0008] In conjunction with the first aspect mentioned above, in one possible implementation, the real-time acquisition of the turbine's operating data during deep peak shaving, and the cleaning and normalization of the operating data through the thermodynamic equation of state, includes: unifying the time base of the acquired pressure, flow rate, enthalpy, and temperature at each measuring point; using interpolation algorithms to eliminate phase differences caused by inconsistent sampling frequencies of different physical quantity sensors, ensuring the synchronization of multi-dimensional operating data on the same time cross-section; establishing the state correlation law between various operating parameters using the thermodynamic equation of state, identifying and eliminating transient abrupt changes that deviate from thermodynamic logic, and reconstructing and filling missing or abnormal positions based on historical trends or adjacent valid data; combining the time change rate of rotor surface measuring point temperature, real-time pressure, and real-time flow rate to analyze and obtain nonlinear characteristic parameters reflecting the heat transfer intensity of the rotor surface; mapping the cleaned operating parameters to a unified per-unit interval to eliminate the influence of different physical dimensions on the calculation of the unsteady heat conduction analytical model, and outputting a standardized basic dataset.

[0009] In conjunction with the first aspect mentioned above, in one possible implementation, the basic dataset is input into a pre-defined unsteady heat conduction analytical model. The equivalent stress field at unmeasurable points inside the rotor is analyzed in real time using the rotor's material thermal conductivity, specific heat capacity, and surface heat transfer coefficient determined by steam flow rate and pressure. This includes: numerically simulating the transient temperature field of the rotor under different variable load conditions using a three-dimensional finite element model to obtain historical response samples of stress-prone points inside the rotor; using the basic dataset as input features and the equivalent stress at the corresponding time as output labels, pre-training the weight coefficients in the unsteady heat conduction analytical model by optimizing the mean square error loss function; inputting the currently collected basic dataset as an excitation signal into the pre-trained unsteady heat conduction analytical model in real time, and using the weight coefficients to perform convolution operations based on physical response features to map the thermal flow impact on the rotor surface; and outputting the equivalent thermal stress distribution and strain intensity of the pre-defined hazardous area inside the rotor in real time from the unsteady heat conduction analytical model to form thermal state prediction data.

[0010] In conjunction with the first aspect above, in one possible implementation, the step of establishing a mathematical model that superimposes cyclic fatigue damage caused by alternating thermal stress and creep damage accumulated over time under high-temperature conditions to calculate the current cumulative fatigue damage value of the rotor includes: matching a preset material life loss curve with the start-stop frequency and corresponding speed change characteristics to calculate the cyclic damage component generated by the alternating action of mechanical stress and thermal stress; analyzing the creep damage component reflecting the deterioration of the internal microstructure of the material based on the metal steady-state operating temperature corresponding to the load fluctuation amplitude and the cumulative high-temperature operating time in the historical operating archive; and using a linear loss accumulation criterion to weight and superimpose the cyclic damage component and the creep damage component to output the current cumulative fatigue damage value characterizing the current service state and material deterioration degree of the rotor.

[0011] In conjunction with the first aspect above, in one possible implementation, based on the accumulated fatigue damage value, the standard allowable temperature difference is nonlinearly contracted or expanded using a fracture mechanics correction function to establish the dynamic allowable temperature difference limit envelope at the current moment. This includes: establishing a mapping relationship between the rotor material strength and the degradation of the current accumulated fatigue damage value; calculating a resistance attenuation coefficient reflecting the current real-time load-bearing capacity of the rotor material; obtaining the standard allowable temperature difference curve of the rotor in its initial service state as the initial reference boundary for nonlinear adjustment; scaling the initial reference boundary using the resistance attenuation coefficient, and combining the rate of temperature change over time in the operating data to perform real-time compensation on the scaled boundary, thereby generating a dynamic allowable temperature difference limit envelope that dynamically changes with the rotor's service life and real-time operating conditions.

[0012] In conjunction with the first aspect above, in one possible implementation, the step of comparing the predicted thermal state data with the dynamic allowable temperature difference limit envelope in real time to determine the stress safety margin, and allocating control vectors between the steam superheat regulation loop and the rate of increase regulation loop according to the stress sensitivity weight, and generating collaborative decision instructions includes: calculating the deviation between the peak thermal stress in the predicted thermal state data at the current moment and the corresponding boundary value of the dynamic allowable temperature difference limit envelope to obtain stress safety margin data reflecting the real-time thermal safety space of the rotor; establishing sensitivity matrices of the influence of steam superheat and rate of increase on rotor surface heat transfer and central hole stress, respectively, and analyzing the real-time contribution weight of each regulation variable to stress change; preset collaborative regulation priority, and when the stress safety margin is greater than the preset safety threshold, preferentially performing micro-regulation through the steam superheat regulation loop; when the predicted thermal state data reaches the boundary of the dynamic allowable temperature difference limit envelope, increasing the intervention ratio of the rate of increase regulation loop according to the contribution weight, thereby realizing the dynamic allocation of control vectors and outputting collaborative decision instructions.

[0013] In conjunction with the first aspect above, in one possible implementation, the step of using an extended state observer to capture the mechanical hysteresis of the actuator and the total system disturbance, performing feedforward compensation correction on the PID control command, generating a final control signal, and sending it to the turbine speed control system and the steam desuperheating water actuator includes: using the extended state observer to obtain the actual feedback opening of the turbine speed control system and the steam desuperheating water actuator in real time, comparing it with the PID control command, and analyzing to obtain the internal disturbance component of the actuator composed of mechanical friction and response delay; defining the external heat load impact caused by the fluctuation of the main steam parameters as the external disturbance component, and constructing a total system disturbance model in combination with the internal disturbance component; generating a reverse compensation vector based on the total system disturbance model, and performing feedforward superposition on the output pulses corresponding to the steam temperature regulation target value and the rate of increase regulation target value to offset the influence of mechanical hysteresis on the thermal stress control accuracy; and synchronously sending the corrected final control signal to the speed control valve and the desuperheating water regulating valve in the form of an analog signal or a digital bus to achieve real-time closed-loop control of the thermal stress trajectory.

[0014] In conjunction with the first aspect above, in one possible implementation, the method further includes: acquiring in real time the actual evolution data of thermal stress of the steam turbine after executing the final control signal, comparing it with the thermal state prediction data, and calculating the prediction residual reflecting the prediction accuracy of the model; based on the prediction residual, using an incremental learning algorithm to adjust the weight coefficients in the unsteady heat conduction analytical model online to reduce the prediction deviation caused by equipment performance degradation; and dynamically correcting the stress sensitivity weight according to the evolution trend of the prediction residual to achieve adaptive optimization of the allocation ratio between the steam superheat regulation loop and the rate of rise regulation loop.

[0015] In conjunction with the first aspect above, in one possible implementation, the calculation of the prediction residual reflecting the model's prediction accuracy includes: simultaneously acquiring the predicted temperature values ​​of the corresponding rotor surface measuring points output by the unsteady heat conduction analytical model during the calculation of internal thermal stress; comparing the predicted temperature values ​​with the actual temperature values ​​of the rotor surface measuring points collected by sensors to obtain the prediction residual characterizing the model's response deviation to thermal boundary conditions; when the fluctuation amplitude of the prediction residual or its moving average exceeds a preset accuracy threshold, using an online gradient descent algorithm to calculate the update increment of the weight coefficients according to the direction of the prediction residual, and iteratively correcting the unsteady heat conduction analytical model.

[0016] Secondly, a turbine acceleration rate and steam temperature coordinated control system based on allowable temperature difference feedback is provided, including: an operation data acquisition and preprocessing module, which collects the turbine's operation data in real time during deep peak shaving, cleans and normalizes the operation data using the thermodynamic state equation to form a basic dataset supporting the calculation of thermal kinetic energy gradient. The operation data includes main steam pressure, steam flow rate, regulating stage enthalpy, and temperatures at multiple measuring points distributed on the turbine rotor surface; a thermal state prediction module, which inputs the basic dataset into a preset unsteady heat conduction analytical model, and uses the rotor's material thermal conductivity, specific heat capacity, and surface heat transfer coefficient determined by steam flow rate and pressure to analyze the equivalent stress field at unmeasurable points inside the rotor in real time, generating thermal state prediction data characterizing the real-time evolution trajectory of rotor thermal stress; and a cumulative fatigue damage calculation module, which retrieves the start-stop frequency and load fluctuation amplitude from the rotor's historical operation records to establish a mathematical model superimposed with cyclic fatigue damage caused by alternating thermal stress and creep damage accumulated over time under high-temperature conditions. The current cumulative fatigue damage value of the rotor is calculated; the dynamic allowable temperature difference limit envelope establishment module, based on the cumulative fatigue damage value, uses a fracture mechanics correction function to nonlinearly contract or expand the standard allowable temperature difference to establish the dynamic allowable temperature difference limit envelope at the current moment; the collaborative decision command generation module compares the thermal state prediction data with the dynamic allowable temperature difference limit envelope in real time to determine the stress safety margin, and allocates control vectors between the steam superheat regulation loop and the rate of increase regulation loop according to the stress sensitivity weight, generating collaborative decision commands; the control command generation module calculates the corresponding steam temperature regulation target value and rate of increase regulation target value according to the vector allocation results in the collaborative decision commands, and generates corresponding PID control commands respectively; the dynamic compensation and command issuance module uses an expanded state observer to capture the mechanical hysteresis of the actuator and the total system disturbance, performs feedforward compensation correction on the PID control commands, generates the final control signal, and issues it to the turbine speed regulation system and the steam desuperheating water actuator.

[0017] Compared with the prior art, the present invention has the following advantages: This invention achieves adaptive control strategies based on equipment age and health status by quantitatively assessing historical rotor damage and dynamically adjusting safety boundaries accordingly. Compared to traditional methods using fixed safety margins, this invention can release more peak-shaving potential in the early stages of the unit or when its health is good, while automatically tightening operating constraints after equipment aging or accumulated damage. This achieves the optimal balance between safety and economy throughout the entire life cycle, effectively extending the service life of critical hot-end components such as rotors.

[0018] This invention establishes a collaborative control mechanism between two key regulation loops: the rate of increase and steam temperature. It dynamically allocates the control vector through stress sensitivity weighting. This multi-variable collaborative strategy overcomes the limitations of single-loop control, enabling flexible combination of regulation methods based on current operating conditions and safety margins. It allows for fine-tuning using the rapid adjustment of steam temperature, while also providing strong intervention by adjusting the rate of increase when necessary. This improves the dynamic response quality and load tracking accuracy of the unit during deep peak shaving.

[0019] This invention achieves real-time, high-precision online prediction of thermal stress at unmeasurable points inside the rotor by introducing an analytical model of unsteady heat conduction. This method transforms the complex heat transfer and stress coupling problem into an efficient data-driven model, overcoming the shortcomings of traditional finite element analysis, which involves huge computational loads and cannot be used for real-time control. This allows the control system to directly perform closed-loop feedback control based on the actual stress state of core components, improving operational safety during variable load processes.

[0020] This invention employs an extended state observer for feedforward compensation in the control execution stage, which can proactively identify and counteract mechanical hysteresis of the actuator and total system disturbances such as external load fluctuations. This design improves the execution accuracy of control commands and the system's anti-interference capability, ensuring that upper-level collaborative decisions can be transmitted to field equipment without distortion, and guaranteeing that the thermal stress control system maintains high stability and robustness under complex and variable operating conditions.

[0021] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. The technical features, technical solutions, and beneficial effects described in this embodiment can also be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A structural architecture diagram of a turbine acceleration rate and steam temperature coordinated control system based on allowable temperature difference feedback provided in an embodiment of this application; Figure 2 A flowchart illustrating the method for coordinated control of turbine acceleration rate and steam temperature based on allowable temperature difference feedback, provided in an embodiment of this application. Figure 3 This is a comparative schematic diagram of the dynamic allowable temperature difference limit envelope provided in the embodiments of this application.

[0024] Figure 4 This is a stress trajectory and acceleration rate intervention response diagram under the cooperative control mode provided in the embodiments of this application. Detailed Implementation

[0025] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0026] The turbine acceleration rate and steam temperature coordinated control method based on allowable temperature difference feedback provided in this application can be applied to, for example... Figure 1 In the turbine acceleration rate and steam temperature coordinated control system 100 based on allowable temperature difference feedback shown, as Figure 1 As shown, the system includes: The data acquisition and preprocessing module collects the operating data of the steam turbine in real time during the deep peak shaving process. The operating data is cleaned and normalized by the thermodynamic state equation to form a basic dataset that supports the calculation of the thermal kinetic energy gradient. The operating data includes the main steam pressure, steam flow rate, regulating stage enthalpy value, and temperature at multiple measuring points distributed on the surface of the steam turbine rotor. The thermal state prediction module inputs the basic dataset into a preset unsteady heat conduction analytical model. It uses the rotor's material thermal conductivity, specific heat capacity, and surface heat transfer coefficient determined by steam flow rate and pressure to analyze the equivalent stress field at unmeasurable points inside the rotor in real time, generating thermal state prediction data that characterizes the real-time evolution trajectory of the rotor's thermal stress. The cumulative fatigue damage calculation module retrieves the start-stop frequency and load fluctuation amplitude from the rotor's historical operating records, establishes a mathematical model that superimposes cyclic fatigue damage caused by alternating thermal stress and creep damage accumulated over time under high-temperature conditions, and calculates the current cumulative fatigue damage value of the rotor. The dynamic allowable temperature difference limit envelope establishment module, based on the accumulated fatigue damage value, uses a fracture mechanics correction function to nonlinearly contract or expand the standard allowable temperature difference, and establishes the dynamic allowable temperature difference limit envelope at the current moment. The collaborative decision-making instruction generation module compares the thermal state prediction data with the envelope of the dynamic allowable temperature difference limit in real time to determine the stress safety margin, and allocates control vectors between the steam superheat regulation loop and the rate of rise regulation loop according to the stress sensitivity weight to generate collaborative decision-making instructions. The control command generation module calculates the corresponding steam temperature regulation target value and rise rate regulation target value based on the vector allocation result in the collaborative decision command, and generates the corresponding PID control commands respectively. The dynamic compensation and command issuance module uses an extended state observer to capture the mechanical hysteresis of the actuator and the total system disturbance, performs feedforward compensation correction on the PID control command, generates the final control signal, and issues it to the turbine speed regulation system and the steam desuperheating water actuator.

[0027] like Figure 2 As shown, this application provides a method for coordinated control of turbine acceleration rate and steam temperature based on allowable temperature difference feedback, including: Real-time acquisition of turbine operation data during deep peak shaving process; cleaning and normalization of the operation data through thermodynamic state equation to form a basic dataset supporting the calculation of thermal kinetic energy gradient; wherein the operation data includes main steam pressure, steam flow rate, regulating stage enthalpy value and temperature at multiple measuring points distributed on the surface of turbine rotor. The basic dataset is input into a preset unsteady heat conduction analytical model. The equivalent stress field of unmeasurable points inside the rotor is analyzed in real time using the rotor's material thermal conductivity, specific heat capacity, and surface heat transfer coefficient determined by steam flow rate and pressure. This generates thermal state prediction data characterizing the real-time evolution trajectory of rotor thermal stress. By retrieving the start-stop frequency and load fluctuation amplitude from the rotor's historical operating records, a mathematical model is established that combines the cyclic fatigue damage caused by alternating thermal stress with the creep damage accumulated over time under high-temperature conditions, and the current cumulative fatigue damage value of the rotor is calculated. Based on the accumulated fatigue damage value, the standard allowable temperature difference is nonlinearly contracted or expanded using the fracture mechanics correction function to establish the dynamic allowable temperature difference limit envelope at the current moment. The thermal state prediction data is compared with the envelope of the dynamic allowable temperature difference limit in real time to determine the stress safety margin. Based on the stress sensitivity weight, control vectors are allocated between the steam superheat regulation loop and the rate of rise regulation loop to generate collaborative decision instructions. Based on the vector allocation results in the collaborative decision-making instructions, the corresponding steam temperature regulation target value and rise rate regulation target value are calculated, and corresponding PID control instructions are generated respectively. The mechanical hysteresis of the actuator and the total system disturbance are captured by the extended state observer. The PID control command is then fed forward to compensate and correct the PID control command, generating the final control signal and sending it to the turbine speed control system and the steam desuperheating water actuator.

[0028] It should be noted that, firstly, a standardized basic dataset reflecting the rotor's thermal boundary conditions is constructed through in-depth processing of real-time operating parameters. Then, an unsteady heat conduction analytical model is used to map these measurable surface parameters in real time into unmeasurable equivalent strain fields at key locations inside the rotor, thereby achieving online transparent monitoring of core thermal stress. This method incorporates the service life dimension, quantifying the rotor's historical service damage by establishing a fatigue and creep damage superposition model, and dynamically correcting the standard safety boundary accordingly, generating a dynamic allowable temperature difference limit envelope that changes in real time with the equipment's health status. The core of the control decision is to continuously compare the predicted real-time thermal stress with this dynamic safety boundary, using the resulting stress safety margin as a feedback signal. Based on the sensitivity of each control variable to stress influence, intelligent allocation of control weights is performed between the steam temperature and rate of rise regulation loops, forming a collaborative decision. At the execution level, an extended state observer is used to identify and feedforward compensate for the non-ideal characteristics of the actuator and the total system disturbance online, ensuring that the collaborative decision commands can be executed accurately and without delay, achieving closed-loop precise tracking of the rotor's thermal stress trajectory.

[0029] In one possible implementation of the embodiments of this application, combined with Figure 2 The real-time acquisition of turbine operating data during deep peak shaving, and the cleaning and normalization of the operating data using the thermodynamic equation of state, includes: The time reference of the acquired pressure, flow rate, enthalpy and temperature at each measuring point is unified. Interpolation algorithm is used to eliminate the phase difference caused by the inconsistent sampling frequency of different physical quantity sensors, so as to ensure the synchronization of multi-dimensional operation data on the same time section. The thermodynamic equation of state is used to establish the state correlation law between various operating parameters, identify and eliminate transient abrupt data that deviates from thermodynamic logic, and reconstruct and fill missing or abnormal positions based on historical trends or adjacent valid data. By combining the time-varying rate of temperature change at rotor surface measuring points, real-time pressure, and real-time flow rate, nonlinear characteristic parameters reflecting the heat transfer intensity of the rotor surface are obtained analytically. The cleaned operating parameters are mapped to a unified per-unit interval to eliminate the influence of different physical dimensions on the calculation of the unsteady heat conduction analytical model and output a standardized basic dataset.

[0030] In some implementations, a time-synchronized, physically consistent, and format-standardized foundation dataset is established to provide a high-quality input source for subsequent accurate prediction of rotor thermal state. First, time-base alignment is performed on the raw data streams from different sensors. Since the sampling frequencies of physical quantities such as pressure, flow rate, and temperature differ (e.g., the main steam pressure sampling frequency can reach 1 Hz while the rotor metal temperature measurement point may be 0.1 Hz), linear or cubic spline interpolation algorithms are used to interpolate low-frequency sampled data points with a unified, high-frequency time base as the target. This generates a multi-dimensional data matrix that is completely synchronized across time sections, eliminating the phase difference introduced by asynchronous sampling. Artifacts and noise in the data are removed, and data loss due to sensor failure or communication interruption is repaired to ensure the physical authenticity of the dataset. The built-in International Association for the Properties of Water and Steam (IAPWS-IF97) thermodynamic equation of state is used as the physical constraint benchmark. For each synchronized data point, the theoretical regulating stage enthalpy is calculated using its main steam pressure and temperature values ​​through the equation of state. Subsequently, the theoretical enthalpy value is compared with the actual measured enthalpy value of the regulating stage by the sensor. If the deviation exceeds a preset engineering threshold, such as 5%, the data point is determined to be abnormal data deviating from thermodynamic logic and is removed. For data locations that are removed or are missing in the original data stream, a reconstruction and filling mechanism is initiated, using an interpolation filling algorithm based on 5 to 10 adjacent valid data points or an autoregressive model to ensure the integrity and continuity of the dataset. A composite characteristic parameter that directly characterizes the intensity of heat exchange on the rotor surface is extracted from the basic operating parameters. This parameter is the core physical quantity driving the change in the internal temperature field of the rotor. Using the temperature measured at the rotor surface after cleaning, real-time main steam pressure, and flow rate data, a nonlinear characteristic parameter is calculated. This parameter is equivalent to the convective heat transfer coefficient. Its calculation follows the following functional relationship: ; in, This represents the nonlinear characteristic parameter, namely the equivalent heat transfer intensity on the rotor surface; The derivative of the temperature at the measuring point on the rotor surface with respect to time is obtained through differential calculation, reflecting the instantaneous rate of thermal shock. Main steam flow rate; Main steam pressure, and This function uses ratio calculations to achieve dimensionless data processing based on the corresponding reference values. Typically, empirical formulas derived from experimental data or fluid dynamics simulations combine the thermophysical properties and flow state of steam with the rotor's thermal response rate to quantify the real-time intensity of heat exchange. The reference heat transfer coefficient under rated operating conditions is used as the static preset parameter of the system. and The index weights used to characterize the fluid flow state and thermal properties are typically taken from a range based on experience. as well as ; This is a thermal shock correction factor used to quantify the enhancing effect of transient temperature fluctuations on the boundary heat transfer layer. It eliminates numerical scale differences caused by varying dimensions of physical quantities, preventing large-scale parameters from disproportionately dominating the model weights when input into unsteady heat conduction analytical models. It applies to all processed operating parameters, including pressure, flow rate, enthalpy, temperature at each measuring point, and calculated nonlinear characteristic parameters. The parameter is then standardized. This process linearly maps the instantaneous value of each parameter to a uniform interval of 0 to 1. The mapping relationship is as follows: ; in, These are the normalized per-unit values; For real-time physical quantities of parameters; and These represent the maximum and minimum values ​​of the parameter within the turbine's design operating range. These boundary values ​​are pre-set as static configuration parameters of the system. The final output is a standardized basic dataset, in which all feature dimensions are dimensionless and can be directly used as input for subsequent models.

[0031] For example, data processing at a certain sampling moment under deep peak shaving conditions of a steam turbine. Time base alignment is performed. If the target synchronization time is 10:00:05, the sampling frequency of the main steam pressure sensor at this time... The physical quantity is directly measured as The sampling frequency of a certain rotor surface metal temperature sensor is... Measurements were taken at 00 seconds and 10 seconds respectively. and The synchronization temperature at 05 seconds was calculated using a linear interpolation algorithm. Physical constraint verification was performed using the IAPWS-IF97 thermodynamic equation of state, based on... and The theoretical enthalpy of the regulating stage was calculated. If the sensor measures the enthalpy value at this time The deviation between the two is approximately less than A preset threshold is used to determine if a data point is valid. Nonlinear feature parameters are then extracted. Assuming the current temperature derivative for Main steam flow for Substitute the empirical function The equivalent heat transfer intensity was calculated. for Finally, dimensional normalization was performed, using the main steam pressure. For example, let its designed operating range be... for , for Substitute into the formula Similarly, complete the calculations for flow rate, enthalpy, and... After mapping, the final output is a standardized basic dataset consisting of dimensionless values ​​such as 0.62.

[0032] In one possible implementation, combining Figure 2 The basic dataset is input into a preset unsteady heat conduction analytical model. Using the rotor's material thermal conductivity, specific heat capacity, and surface heat transfer coefficient determined by steam flow rate and pressure, the equivalent variable field at unmeasurable points inside the rotor is analyzed in real time, including: The transient temperature field of the rotor under different variable load conditions was numerically simulated using a three-dimensional finite element model to obtain historical response samples of stress critical points inside the rotor. Using the aforementioned basic dataset as input features and the equivalent stress at the corresponding time point as output labels, the weight coefficients in the unsteady heat conduction analytical model are pre-trained by optimizing the mean square error loss function. The currently collected basic dataset is used as an excitation signal and input into the pre-trained unsteady heat conduction analytical model in real time. The weight coefficients are used to perform convolution operation mapping on the heat flow impact on the rotor surface based on physical response features. The equivalent thermal stress distribution and strain intensity of the preset dangerous area inside the rotor are output in real time by the unsteady heat conduction analytical model, forming thermal state prediction data.

[0033] In some implementations, a proxy model is constructed and trained offline to quickly infer thermal stress at critical internal locations from easily measurable rotor surface and steam parameters, replacing the computationally intensive and unsuitable traditional finite element analysis. Before deployment, a digital model with geometry and material properties identical to the actual turbine rotor is established using 3D finite element software. This model precisely defines the nonlinear relationship between the rotor's thermal conductivity and specific heat capacity with temperature. Based on historical operating data, multiple scenarios covering typical deep peak-shaving conditions are selected, such as the rapid ramp-up process from 30% load to 90% load, and the transient temperature field of the rotor under these conditions is numerically simulated. Through simulation, a large amount of high-resolution temperature and equivalent stress response data of critical internal stress points in the rotor are obtained throughout the load variation process, forming a series of historical response samples including boundary conditions and internal responses. The complex thermo-mechanical coupling physical laws inherent in the finite element simulation are solidified into an unsteady heat conduction analytical model through machine learning. This analytical model is essentially a deep neural network structure, such as a temporal convolutional network or a long short-term memory network. Using the corresponding basic dataset from the simulation process as the input features of the model, and the equivalent stress values ​​of internal hazard points at the same moment calculated in the simulation as the output labels, a supervised learning training process is built and executed. The mean squared error loss function is used to quantify the difference between the model's predicted values ​​and the true label values. ; in, This represents the value of the mean squared error loss function; This represents the total number of training samples; This represents the equivalent stress value predicted by the unsteady heat conduction analytical model under the current weighting coefficients. This provides the true label for the equivalent stress at the corresponding time point calculated by finite element simulation. During training, optimization algorithms such as Adam are used to iteratively adjust the weight coefficients of the network within the model through backpropagation until the loss function is optimized. The model converges to a stable value below the preset engineering accuracy. After this process, the model is pre-trained, and its weight coefficients accurately reflect the nonlinear mapping relationship from surface thermal load to internal stress. Using the pre-trained unsteady heat conduction analytical model, high-frequency predictions of the rotor's internal thermal state are performed online in real time. During the actual operation of the turbine, the cleaned and normalized basic dataset serves as a continuous excitation signal, inputting into the model at millisecond-level frequencies. Within the model, the fixed weight coefficients perform a series of convolution operations based on physical response characteristics on the input time-series data. This operation is equivalent to a highly efficient filter bank, capable of quickly extracting dynamic features directly related to thermal stress generation from complex input signals, such as the intensity and rate of thermal flow impact. The model's computational results are transformed into quantitative indicators that can be used for control decisions. After forward propagation calculations, the model outputs a data vector in real time, which accurately describes the current equivalent thermal stress distribution and strain intensity of each critical region inside the rotor. This data stream containing multi-point stress prediction values ​​is defined as thermal state prediction data, which forms the direct basis for subsequent stress margin calculation and collaborative control decisions, thereby enabling real-time and accurate monitoring of the thermal state of unmeasurable points inside the rotor.

[0034] For example, during the offline pre-training process of an unsteady heat conduction analytical model, technicians used a three-dimensional finite element digital model to obtain... A set of samples regarding the thermal stress history response of the central hole in the regulating stage. Taking the sample data from one sampling time point as an example, the true label of the equivalent stress obtained from finite element calculation. for Meanwhile, the stress predicted by the deep neural network for The squared error term for this sample point is calculated as follows: The Adam optimization algorithm is used to iteratively adjust the weight coefficients of the internal network of the model through back-addition propagation, thereby reducing the mean squared error loss function value of the total samples. Reduce to Thus satisfying the requirement of being lower than The convergence accuracy requirements are met, and model pre-training is completed. During the unit's online operation phase, when the normalized main steam pressure is input in real time... Steam flow rate After receiving the excitation signal from the basic dataset, the model performs convolution operations based on physical response features using fixed weight coefficients, directly outputting a data vector describing the state of the dangerous area inside the rotor, such as the predicted equivalent stress of the current center hole. Strain intensity is .

[0035] In one possible implementation, combining Figure 2The mathematical model established by superimposing cyclic fatigue damage caused by alternating thermal stress and creep damage accumulated over time under high-temperature conditions calculates the current cumulative fatigue damage value of the rotor, including: Based on the start-stop frequency and the corresponding speed change characteristics, a preset material life loss curve is matched to calculate the cyclic damage component caused by the alternating action of mechanical stress and thermal stress. Based on the steady-state operating temperature of the metal corresponding to the load fluctuation amplitude, and combined with the cumulative high-temperature operating time in the historical operating records, the creep damage component reflecting the deterioration of the internal microstructure of the material is analyzed and obtained. Using a linear loss accumulation criterion, the cyclic damage component and the creep damage component are weighted and superimposed to output the current cumulative fatigue damage value, which characterizes the current service state and material degradation degree of the rotor.

[0036] In some implementations, the fatigue damage to materials caused by the alternating mechanical and thermal stresses resulting from turbine start-up and shutdown and significant load changes is quantified. First, complete operating records are retrieved from the rotor's historical operating archive, and the frequency of start-up and shutdown events is automatically identified and counted using a pattern recognition algorithm. For each start-up and shutdown cycle, the speed change characteristics during that process are extracted and correlated with the maximum thermal stress amplitude calculated by an unsteady heat conduction analytical model for the corresponding time period. Then, this stress amplitude is used as input to query the strain-life loss curve (Coffin-Manson curve) of the rotor material pre-stored in the database. This curve describes the number of cycles the material can withstand within a specific strain range. The damage increment caused by a single start-up and shutdown cycle is calculated. The damage increments of all historical start-up and shutdown cycles are summed to obtain the total cyclic damage component caused by start-up and shutdown operations. The microstructural degradation, i.e., creep damage, that occurs in the material under high-temperature environments due to long-term stress is accurately assessed. The load fluctuation amplitudes in the rotor's historical operating archive are analyzed and mapped to the corresponding steady-state operating temperature range of the metal. For example, when the unit operates at 80% to 100% of its rated load, the metal temperature of critical rotor components typically remains between 520°C and 540°C. Using a preset creep-sensitive temperature as a threshold, all high-temperature operating periods are selected and precisely accumulated to obtain the cumulative high-temperature operating time. Combined with the steady-state stress level during this period, mature creep life prediction models such as the Larson-Miller parameter method are used to analytically calculate the fraction of life consumed by the material due to creep effects under the current cumulative high-temperature operating time; this is the creep damage component. Integrating the damage from these two different mechanisms yields a comprehensive index that fully characterizes the rotor's current service status and material degradation. Using the internationally recognized linear loss accumulation criterion, the calculated cyclic damage component and creep damage component are directly linearly superimposed. This criterion assumes that the two damage mechanisms are independent and that the cumulative effect can be simply added. The mathematical model for superposition is expressed as follows: ; in, The final output is the current cumulative fatigue damage value, which is a dimensionless parameter ranging from 0 to 1. When the value approaches 1, it indicates that the rotor's lifespan is nearing its end; The cyclic damage component represents the sum of the lifetime fractions consumed by all historical stress cycles. Its value is obtained by summing the actual number of cycles at each stress level divided by the allowable number of cycles at that stress level. The creep damage component represents the sum of the creep life fractions consumed under all high-temperature conditions. Its value is obtained by summing the actual operating time under each high-temperature condition divided by the creep fracture time under that condition. All terms in the formula are dimensionless damage fractions and can be directly added. The calculated current cumulative fatigue damage value... This will serve as a key input for dynamically adjusting the allowable temperature difference limit in the next stage.

[0037] For example, for a steam turbine rotor (made of 12CrMoV) during its service life, the frequency of start-stop events was identified and counted as 200 times from historical operation records. By extracting the speed change characteristics and associating them with the maximum thermal stress amplitude calculated by the analytical model, and querying the preset Coffin-Manson curve in the database, the allowable number of cycles at this stress level was found to be 4000. From this, the cyclic damage component was calculated. At the same time, the units were screened out in The cumulative operating time under the above high-temperature environment is 20,000 hours. Under the steady-state operating temperature and corresponding stress level of the metal, the allowable creep rupture time of the material was analytically determined to be 100,000 hours using the Larson-Miller parametric method, and the creep damage component was calculated. The linear loss accumulation criterion is adopted and substituted into the superposition mathematical model. The calculation is performed, and the current cumulative fatigue damage value is finally output. .

[0038] In one possible implementation, combining Figure 2 Based on the accumulated fatigue damage value, the standard allowable temperature difference is nonlinearly contracted or expanded using a fracture mechanics correction function to establish the dynamic allowable temperature difference limit envelope at the current moment, including: A mapping relationship is established between the rotor material strength and the degradation of the current cumulative fatigue damage value, and the resistance attenuation coefficient reflecting the current real-time load-bearing capacity of the rotor material is calculated. Obtain the standard allowable temperature difference curve of the rotor under initial service conditions as the initial reference boundary for nonlinear adjustment; The initial reference boundary is scaled proportionally using the resistance attenuation coefficient, and the scaling boundary is compensated in real time by combining the rate of temperature change over time in the operating data, thereby generating a dynamic allowable temperature difference limit envelope that changes dynamically with rotor service life and real-time operating conditions.

[0039] In some implementations, the performance degradation of rotor materials due to long-term service is quantified into engineering parameters that can be directly used to adjust safety boundaries. First, a degradation mapping relationship is established between the rotor material strength and its current cumulative fatigue damage value. This relationship is typically based on a material damage mechanics model, using a nonlinear function to calculate the current cumulative fatigue damage value. This is converted into a resistance attenuation coefficient that reflects the material's real-time load-bearing capacity. This coefficient is a dimensionless value between 0 and 1, where 1 represents a completely new state, and so on. It increases and then monotonically decreases. Its calculation model is expressed as: ; in, This is the resistance attenuation coefficient; The current cumulative fatigue damage value is input. The material degradation index is an empirical constant calibrated based on experimental data of the rotor material, typically ranging from 0.2 to 0.5, used to characterize the nonlinearity of the impact of damage accumulation on material strength. A theoretically ideal safe operating boundary, without any damage correction, is set as the benchmark for subsequent dynamic adjustments. The standard allowable temperature difference curve for this rotor model under initial service conditions, provided by the turbine manufacturer, is retrieved from the built-in equipment database. This curve is not a constant value but defines the maximum internal and external temperature difference allowed under different loads or metal temperatures to ensure that the stress does not exceed the design allowable value, forming the initial benchmark boundary for nonlinear adjustment. The historical damage of the material is combined with the current operational dynamics to generate dynamic constraints that reflect the rotor's true safety margin in real time. First, the calculated resistance attenuation coefficient is used... The standard allowable temperature difference curve is scaled proportionally to obtain a preliminary shrinkage temperature difference boundary that takes into account long-term service history. Subsequently, to address the instantaneous impact of thermal shock during rapid load changes, a dynamic compensation term based on real-time operating conditions is introduced. This compensation term uses the rate of temperature change over time at key rotor measuring points extracted from operating data—i.e., the rate of heating or cooling—to correct the scaled boundary. The more drastic the temperature change, the greater the additional thermal stress, thus imposing a larger shrinkage. The final dynamic allowable temperature difference limit envelope is calculated using the following formula: ; ; in, This represents the dynamic allowable temperature difference limit at the current moment; This is the baseline allowable temperature difference obtained from the standard curve; This represents the real-time rate of change of the rotor surface temperature. As a compensation function, the rate of temperature change is mapped to a temperature difference correction value. This function ensures that the safety boundary becomes more stringent during rapid load increases and decreases. The thermal sensitivity compensation coefficient is a constant preset based on the rotor geometry and the material's thermal diffusivity, typically taking a value within a certain range. Between these points, the envelope that dynamically changes with rotor age and real-time operating conditions provides precise and adaptive safety constraints for subsequent coordinated control decisions. For example... Figure 3The figure shows the shrinkage of the allowable temperature difference boundary with varying metal temperature under different service life damage levels. Compared to the initial baseline curve, the shrinkage is reduced by the resistance attenuation coefficient. The dynamic envelope after scaling and real-time thermal shock compensation is significantly tightened, providing a more targeted safety baseline for the control system.

[0040] For example, in a steam turbine rotor that is already in service, the calculated current cumulative fatigue damage value is received. Take the material degradation index Substituting into the degradation mapping model, the resistance attenuation coefficient is calculated. Subsequently, the initial reference boundary of this rotor model is retrieved from the database, assuming the standard allowable temperature difference under the current load conditions. for ;use The initial contraction boundary is obtained by scaling the reference boundary. Simultaneously, the real-time rate of change of rotor surface temperature was monitored. for Set the thermal sensitivity compensation coefficient for Through the compensation function The instantaneous thermal shock correction value was calculated. Finally, substitute into the formula Generate the dynamic allowable temperature difference limit for the current moment. .

[0041] In one possible implementation, combining Figure 2 The step of comparing the predicted thermal state data with the envelope of the dynamic allowable temperature difference limit in real time to determine the stress safety margin, and allocating control vectors between the steam superheat regulation loop and the rate of rise regulation loop according to the stress sensitivity weight, and generating collaborative decision instructions includes: Calculate the deviation between the peak thermal stress in the thermal state prediction data at the current moment and the boundary value corresponding to the envelope of the dynamic allowable temperature difference limit to obtain stress safety margin data reflecting the real-time thermal safety space of the rotor. A sensitivity matrix was established to determine the effects of steam superheat and rise rate on rotor surface heat transfer and central hole stress, respectively, and the real-time contribution weights of each adjustment variable to stress changes were analyzed. The system prioritizes coordinated adjustment. When the stress safety margin is greater than the preset safety threshold, it prioritizes micro-adjustment through the steam superheat adjustment loop. When the thermal state prediction data reaches the boundary of the dynamic allowable temperature difference limit envelope, it increases the intervention ratio of the rate of rise adjustment loop according to the contribution weight, thereby realizing the dynamic allocation of the control vector and outputting coordinated decision instructions.

[0042] In some implementations, the distance between the current rotor thermal stress state and the dynamic safety boundary is quantified to provide a direct driving signal for subsequent control decisions. A comparison operation is performed at a fixed high frequency. The real-time output peak thermal stress is subtracted from the corresponding boundary value generated by the dynamic allowable temperature difference limit envelope establishment module. This boundary value is the allowable stress value obtained from the envelope based on the current load and metal temperature. The difference is defined as the stress safety margin data. When the stress peak is lower than the boundary value, the margin is positive, indicating a safe state; otherwise, it is negative, indicating an overstress risk. The relative effectiveness of the steam superheat and rise rate control variables on rotor thermal stress is determined so that control weights can be reasonably allocated when intervention is needed. An internal stress sensitivity matrix is ​​established, and the matrix is ​​calibrated by offline system identification or online recursive least squares method. The elements in the matrix represent the gains of unit steam superheat change and unit rise rate change on rotor surface heat transfer intensity and central hole stress, respectively. During real-time operation, this matrix is ​​consulted or calculated based on the current operating conditions to determine the real-time contribution weights of the steam superheat regulation loop and the rate of rise regulation loop to stress changes at the current moment. These two weight values ​​are denoted as... and All are dimensionless parameters, and their sum is 1, reflecting their relative dominance in thermal stress control. Based on the current stress safety margin, the system intelligently and dynamically allocates adjustment tasks between the two control loops, achieving coordinated adjustment that precisely controls stress while also considering load increase efficiency. An internally preset set of coordinated adjustment priority logic triggers different control strategies based on stress safety margin data. When the stress safety margin is greater than the preset safety threshold, it indicates sufficient system safety space, and the control system prioritizes rapid response, preferentially using the faster-responding steam superheat adjustment loop for minor adjustments to track the target load curve. When the stress value in the thermal state prediction data gradually approaches or even reaches the boundary of the dynamic allowable temperature-difference limit envelope, a significant overstress risk is identified, and the control strategy shifts to prioritizing safety. The system will adjust the control based on the real-time calculated contribution weights. and This significantly increases the intervention ratio in the rate of rise regulation loop. This means actively limiting the rate of rise, and even briefly reducing it when necessary, sacrificing some response speed for absolute thermal stress safety. The dynamic allocation process of the control vector ultimately generates a cooperative decision instruction containing two components. This instruction clarifies the direction and magnitude of adjustment for the steam temperature regulation target value and the rate of rise regulation target value in the next control cycle, thereby guiding the subsequent PID controller to make precise adjustments. For example... Figure 4As shown, the collaborative control response to the thermal stress trajectory during deep peak-shaving load variation was simulated. When the real-time thermal stress peak value approached the dynamic allowable boundary, the collaborative decision-making logic identified the insufficient safety margin and proactively adjusted the target value of the rate of increase from... Downgraded to The stress trajectory was successfully contained within the safety envelope.

[0043] For example, at a certain sampling moment during the deep peak shaving and load increase process of the steam turbine, the real-time peak value of the internal thermal stress of the rotor is obtained as follows: The corresponding boundary value at time is The stress safety margin is obtained by performing the subtraction operation. Since this margin accounts for a proportion of the dynamic limit, If the preset risk intervention threshold is reached, the control strategy is determined to shift from pursuing response efficiency to prioritizing thermal safety. Subsequently, the stress sensitivity matrix is ​​analyzed to obtain the real-time contribution weights of steam superheat and rise rate to stress changes under the current operating conditions. and Based on the logic of coordinated regulation priority, the intervention ratio of the rate of ascent regulation loop is significantly increased, generating a loop that includes shifting the rate of ascent target from... Actively restrict to And collaborative decision-making instructions for simultaneously fine-tuning the target value of steam temperature.

[0044] In one possible implementation, combining Figure 2 The method of using an extended state observer to capture the mechanical hysteresis of the actuator and the total system disturbance, performing feedforward compensation correction on the PID control command, generating the final control signal, and sending it to the turbine speed control system and the steam desuperheating water actuator includes: The actual feedback opening of the turbine speed control system and the steam desuperheating water actuator is obtained in real time using the extended state observer, and compared with the PID control command to obtain the internal disturbance component of the actuator composed of mechanical friction and response delay. The external heat load shock caused by fluctuations in main steam parameters is defined as the external disturbance component, and the total disturbance model of the system is constructed by combining the internal disturbance component. Based on the total disturbance model of the system, a reverse compensation vector is generated, and the output pulses corresponding to the steam temperature regulation target value and the rate of rise regulation target value are fed forward and superimposed to offset the influence of mechanical hysteresis on the thermal stress control accuracy. The corrected final control signal is synchronously sent to the speed regulating valve and the desuperheating water regulating valve in the form of analog or digital bus to realize real-time closed-loop control of the thermal stress trajectory.

[0045] In some implementations, the physical limitations of the control actuators themselves, such as mechanical friction, backlash, and response delay, are actively identified and quantified. These factors cause actual actions to lag behind control commands, affecting control accuracy. An extended state observer (ESO) is configured for both the turbine speed control system and the steam desuperheating water actuator. This observer is a dynamic system model that takes the control commands output by the PID controller as input and acquires the position feedback signals of the actuators in real time as observations. By comparing the difference between the command value and the actual feedback value, and combining this with the observer's internal state equations, a comprehensive internal disturbance component can be estimated online. This component is physically equivalent to the execution deviation caused by mechanical hysteresis. The impact of external environmental changes on the system, along with the internal disturbances of the actuators, is incorporated into the compensation model to form a comprehensive estimate of the total system disturbance. Unexpected fluctuations in parameters such as main steam pressure and temperature are defined as external disturbance components, which directly cause thermal load shocks to the turbine. By monitoring the rate of change of main steam parameters and combining this with their sensitivity to stress effects, the impact of this external disturbance on the control system is quantified. Subsequently, this external disturbance component is vector-superimposed with the internal disturbance component of the actuator extracted from the extended state observer to construct a total system disturbance model that reflects the sum of internal and external disturbances faced by the system in real time. Based on the accurate estimation of the total disturbance, a compensation signal is generated to offset its negative impact in advance, thereby improving the dynamic response performance and robustness of the control system. Based on the constructed total system disturbance model, a reverse compensation vector of equal magnitude and opposite direction is calculated in real time. This vector acts as a "prediction," predicting the magnitude of the deviation impact of the disturbance on the system in the next control cycle. This reverse compensation vector is directly superimposed in feedforward onto the PID output pulses corresponding to the steam temperature regulation target value and the rate of rise regulation target value calculated by the cooperative decision command. The calculated compensation control command is as follows: ; in, This is the final control signal issued; This refers to the output command of the original PID controller; This is the inverse compensation vector generated by the total disturbance model. Through this feedforward superposition, the control command already includes the amount to offset future disturbances before it reaches the actuator, thus effectively overcoming mechanical hysteresis and improving the tracking accuracy of thermal stress control. The optimized and corrected control decision is transmitted to the field actuators without delay, completing the entire closed-loop control circuit. The final control signal, after feedforward compensation correction, is converted into a standard 4-20 mA analog signal or transmitted synchronously to the servo mechanism of the speed regulating valve and the positioner of the desuperheating water regulating valve according to the field interface protocol. This ensures that the adjustment actions of the acceleration rate and steam temperature can be executed accurately and coordinatedly, achieving real-time, high-precision closed-loop control of the rotor thermal stress evolution trajectory.

[0046] For example, during the turbine load increase process, the control of the speed regulating valve monitors the original PID output command generated by the collaborative decision command in real time. for Valve opening degree. At this point, the Extended State Observer (ESO) dedicated to the speed control system compares the command value with the actual feedback valve opening degree to analyze the internal disturbance component composed of mechanical friction and response delay. Simultaneously, the external heat load impact, i.e., the external disturbance component, generated by the main steam pressure fluctuation was monitored. The vector superposition of the two constructs the total disturbance model of the system as follows: A reverse compensation vector with the opposite direction is generated based on the overall disturbance model. and according to the formula The final control signal is obtained by performing feedforward superposition calculation. The command is then converted into a corresponding digital bus signal and sent to the speed control valve servo mechanism, through pre-compensation. The execution deviation offset the impact of mechanical hysteresis on control accuracy.

[0047] In one possible implementation, combining Figure 2 The method further includes: The actual thermal stress evolution data of the steam turbine after executing the final control signal is acquired in real time, and compared with the thermal state prediction data to calculate the prediction residual that reflects the accuracy of the model prediction. Based on the prediction residual, the weight coefficients in the unsteady heat conduction analytical model are adjusted online using an incremental learning algorithm to reduce the prediction bias caused by equipment performance degradation. Based on the evolution trend of the predicted residual, the stress sensitivity weight is dynamically corrected to achieve adaptive optimization of the allocation ratio between the steam superheat regulation loop and the rate of rise regulation loop.

[0048] Simultaneously acquire the predicted temperature values ​​of the corresponding rotor surface measuring points output by the unsteady heat conduction analytical model during the calculation of internal thermal stress; The predicted temperature value is compared with the actual temperature value of the rotor surface measuring point collected by the sensor to obtain the predicted residual characterizing the model's response deviation to thermal boundary conditions. When the fluctuation range of the predicted residual or its moving average exceeds a preset accuracy threshold, the online gradient descent algorithm is used to calculate the update increment of the weight coefficients according to the direction of the predicted residual, and the unsteady heat conduction analytical model is iteratively corrected.

[0049] In some implementations, a feedback mechanism is established to evaluate the prediction accuracy of the unsteady heat conduction analytical model in real time and quantify its deviation from the actual physical process. After issuing the final control signal, the unmeasurable internal thermal stress is not directly acquired as actual evolution data. Instead, the temperature values ​​predicted by the model simultaneously during the internal stress calculation at measurement points on the rotor surface are compared in real time with the actual temperature values ​​collected by sensors at these points. The difference between these two values ​​is calculated as the prediction residual, with the formula: ; in, To predict residual differences, the unit is Celsius; This is the predicted temperature at the corresponding measurement point location output by the unsteady heat conduction analytical model; This is the real-time temperature collected by the sensor at this measuring point. The residual... This directly reflects the model's prediction accuracy for the rotor surface thermal boundary conditions and serves as the primary basis for subsequent adaptive adjustments. Utilizing the prediction residuals, the unsteady heat conduction analytical model is continuously and online self-corrected to compensate for prediction biases caused by equipment performance degradation over time or model simplification. Incremental learning algorithms, such as online gradient descent, are employed to fine-tune the model's weight coefficients. When the moving average of the prediction residuals consistently exceeds a preset engineering threshold over a period of time, the online adjustment mechanism is triggered. Based on the residuals... Given the magnitude and direction of the weights, calculate the weight update amount that reduces the residual, and iteratively update the model weights. The update logic is expressed as follows: ; in, and These are the model weight coefficient matrices before and after the update, respectively. It is a preset, small learning rate, such as 0.001, used to ensure the stability of the adjustment process; It is the prediction residual For weighting coefficients The gradient indicates the steepest descent direction of the weight adjustment. Through this process, the model can gradually adapt to changes in the actual characteristics of the unit, maintaining high-precision prediction capabilities. Based on the error evolution law of the prediction model, the upstream control decision logic is optimized in reverse to achieve adaptive control strategy. The prediction residuals... Perform long-term trend analysis and correlation analysis with control commands executed during the same period. If frequent rate-of-rise adjustments are found, predict the residuals. The fluctuations in stress intensity increase significantly, while the residual remains stable during periods dominated by steam superheat regulation. This suggests that the sensitivity weight of the currently set rate of ascent to stress may be too low. Based on this analysis, the preset stress sensitivity weight is automatically and dynamically corrected. For example, the weight of the rate of ascent regulation loop is appropriately increased, while the weight of the steam superheat regulation loop is decreased. This correction, based on long-term data trends, achieves adaptive optimization of the allocation ratio between the steam superheat regulation loop and the rate of ascent regulation loop, enabling the collaborative control strategy to better match the actual dynamic response characteristics of the turbine.

[0050] For example, closed-loop correction is initiated in real time after the control command is executed. If the temperature at a certain measuring point on the rotor surface is synchronously predicted by the unsteady heat conduction analytical model... for The real-time temperature collected by the sensor for According to the formula The predicted residuals were calculated. for Because the residual exceeded If the preset threshold is exceeded and the moving average value continues to exceed the limit, the online adjustment mechanism is triggered. The current model weight coefficient matrix is ​​set. Corresponding element is Learning rate for If gradient The calculated value is Then, according to the update logic The new weights are calculated as follows This reduces deviations caused by equipment performance degradation. Simultaneously, long-term trend analysis revealed large residual fluctuations when the rate of increase regulation was frequent, indicating that the rate of increase sensitivity weight was too low. Therefore, the allocation ratio was dynamically corrected; for example, the stress sensitivity weight of the rate of increase regulation loop was adjusted from... Optimize to .

[0051] It should be noted that all equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A method for coordinated control of turbine acceleration rate and steam temperature based on allowable temperature difference feedback, characterized in that, The method includes: Real-time acquisition of turbine operation data during deep peak shaving process; cleaning and normalization of the operation data through thermodynamic state equation to form a basic dataset supporting the calculation of thermal kinetic energy gradient; wherein the operation data includes main steam pressure, steam flow rate, regulating stage enthalpy value and temperature at multiple measuring points distributed on the surface of turbine rotor. The basic dataset is input into a preset unsteady heat conduction analytical model. The equivalent stress field of unmeasurable points inside the rotor is analyzed in real time using the rotor's material thermal conductivity, specific heat capacity, and surface heat transfer coefficient determined by steam flow rate and pressure. This generates thermal state prediction data characterizing the real-time evolution trajectory of rotor thermal stress. By retrieving the start-stop frequency and load fluctuation amplitude from the rotor's historical operating records, a mathematical model is established that combines the cyclic fatigue damage caused by alternating thermal stress with the creep damage accumulated over time under high-temperature conditions, and the current cumulative fatigue damage value of the rotor is calculated. Based on the accumulated fatigue damage value, the standard allowable temperature difference is nonlinearly contracted or expanded using the fracture mechanics correction function to establish the dynamic allowable temperature difference limit envelope at the current moment. The thermal state prediction data is compared with the envelope of the dynamic allowable temperature difference limit in real time to determine the stress safety margin. Based on the stress sensitivity weight, control vectors are allocated between the steam superheat regulation loop and the rate of rise regulation loop to generate collaborative decision instructions. Based on the vector allocation results in the collaborative decision-making instructions, the corresponding steam temperature regulation target value and rise rate regulation target value are calculated, and corresponding PID control instructions are generated respectively. The mechanical hysteresis of the actuator and the total system disturbance are captured by the extended state observer. The PID control command is then fed forward to compensate and correct the PID control command, generating the final control signal and sending it to the turbine speed control system and the steam desuperheating water actuator.

2. The method for coordinated control of turbine acceleration rate and steam temperature based on allowable temperature difference feedback according to claim 1, characterized in that, The real-time acquisition of turbine operating data during deep peak shaving, and the cleaning and normalization of the operating data using the thermodynamic equation of state, includes: The time reference of the acquired pressure, flow rate, enthalpy and temperature at each measuring point is unified. Interpolation algorithm is used to eliminate the phase difference caused by the inconsistent sampling frequency of different physical quantity sensors, so as to ensure the synchronization of multi-dimensional operation data on the same time section. The thermodynamic equation of state is used to establish the state correlation law between various operating parameters, identify and eliminate transient abrupt data that deviates from thermodynamic logic, and reconstruct and fill missing or abnormal positions based on historical trends or adjacent valid data. By combining the time-varying rate of temperature change at rotor surface measuring points, real-time pressure, and real-time flow rate, nonlinear characteristic parameters reflecting the heat transfer intensity of the rotor surface are obtained analytically. The cleaned operating parameters are mapped to a unified per-unit interval to eliminate the influence of different physical dimensions on the calculation of the unsteady heat conduction analytical model and output a standardized basic dataset.

3. The method for coordinated control of turbine acceleration rate and steam temperature based on allowable temperature difference feedback according to claim 1, characterized in that, The basic dataset is input into a preset unsteady heat conduction analytical model. Using the rotor's material thermal conductivity, specific heat capacity, and surface heat transfer coefficient determined by steam flow rate and pressure, the equivalent strain field at unmeasurable points inside the rotor is analyzed in real time, including: The transient temperature field of the rotor under different variable load conditions was numerically simulated using a three-dimensional finite element model to obtain historical response samples of stress critical points inside the rotor. Using the aforementioned basic dataset as input features and the equivalent stress at the corresponding time point as output labels, the weight coefficients in the unsteady heat conduction analytical model are pre-trained by optimizing the mean square error loss function. The currently collected basic dataset is used as an excitation signal and input into the pre-trained unsteady heat conduction analytical model in real time. The weight coefficients are used to perform convolution operation mapping on the heat flow impact on the rotor surface based on physical response features. The equivalent thermal stress distribution and strain intensity of the preset dangerous area inside the rotor are output in real time by the unsteady heat conduction analytical model, forming thermal state prediction data.

4. The method for coordinated control of turbine acceleration rate and steam temperature based on allowable temperature difference feedback according to claim 1, characterized in that, The mathematical model established by superimposing cyclic fatigue damage caused by alternating thermal stress and creep damage accumulated over time under high-temperature conditions calculates the current cumulative fatigue damage value of the rotor, including: Based on the start-stop frequency and the corresponding speed change characteristics, a preset material life loss curve is matched to calculate the cyclic damage component caused by the alternating action of mechanical stress and thermal stress. Based on the steady-state operating temperature of the metal corresponding to the load fluctuation amplitude, and combined with the cumulative high-temperature operating time in the historical operating records, the creep damage component reflecting the deterioration of the internal microstructure of the material is analyzed and obtained. Using a linear loss accumulation criterion, the cyclic damage component and the creep damage component are weighted and superimposed to output the current cumulative fatigue damage value, which characterizes the current service state and material degradation degree of the rotor.

5. The method for coordinated control of turbine acceleration rate and steam temperature based on allowable temperature difference feedback according to claim 1, characterized in that, Based on the accumulated fatigue damage value, the standard allowable temperature difference is nonlinearly contracted or expanded using a fracture mechanics correction function to establish the dynamic allowable temperature difference limit envelope at the current moment, including: A mapping relationship is established between the rotor material strength and the degradation of the current cumulative fatigue damage value, and the resistance attenuation coefficient reflecting the current real-time load-bearing capacity of the rotor material is calculated. Obtain the standard allowable temperature difference curve of the rotor under initial service conditions as the initial reference boundary for nonlinear adjustment; The initial reference boundary is scaled proportionally using the resistance attenuation coefficient, and the scaling boundary is compensated in real time by combining the rate of temperature change over time in the operating data, thereby generating a dynamic allowable temperature difference limit envelope that changes dynamically with rotor service life and real-time operating conditions.

6. The method for coordinated control of turbine acceleration rate and steam temperature based on allowable temperature difference feedback according to claim 1, characterized in that, The step of comparing the predicted thermal state data with the envelope of the dynamic allowable temperature difference limit in real time to determine the stress safety margin, and allocating control vectors between the steam superheat regulation loop and the rate of rise regulation loop according to the stress sensitivity weight, and generating collaborative decision instructions includes: Calculate the deviation between the peak thermal stress in the thermal state prediction data at the current moment and the boundary value corresponding to the envelope of the dynamic allowable temperature difference limit to obtain stress safety margin data reflecting the real-time thermal safety space of the rotor. A sensitivity matrix was established to determine the effects of steam superheat and rise rate on rotor surface heat transfer and central hole stress, respectively, and the real-time contribution weights of each adjustment variable to stress changes were analyzed. The system prioritizes coordinated adjustment. When the stress safety margin is greater than the preset safety threshold, it prioritizes micro-adjustment through the steam superheat adjustment loop. When the thermal state prediction data reaches the boundary of the dynamic allowable temperature difference limit envelope, it increases the intervention ratio of the rate of rise adjustment loop according to the contribution weight, thereby realizing the dynamic allocation of the control vector and outputting coordinated decision instructions.

7. The method for coordinated control of turbine acceleration rate and steam temperature based on allowable temperature difference feedback according to claim 1, characterized in that, The process of using an extended state observer to capture the mechanical hysteresis of the actuator and the total system disturbance, performing feedforward compensation correction on the PID control command, generating the final control signal, and sending it to the turbine speed control system and the steam desuperheating water actuator includes: The actual feedback opening of the turbine speed control system and the steam desuperheating water actuator is obtained in real time using the extended state observer, and compared with the PID control command to obtain the internal disturbance component of the actuator composed of mechanical friction and response delay. The external heat load shock caused by fluctuations in main steam parameters is defined as the external disturbance component, and the total disturbance model of the system is constructed by combining the internal disturbance component. Based on the total disturbance model of the system, a reverse compensation vector is generated, and the output pulses corresponding to the steam temperature regulation target value and the rate of rise regulation target value are fed forward and superimposed to offset the influence of mechanical hysteresis on the thermal stress control accuracy. The corrected final control signal is synchronously sent to the speed regulating valve and the desuperheating water regulating valve in the form of analog or digital bus to realize real-time closed-loop control of the thermal stress trajectory.

8. The method for coordinated control of turbine acceleration rate and steam temperature based on allowable temperature difference feedback according to claim 1, characterized in that, The method further includes: The actual thermal stress evolution data of the steam turbine after executing the final control signal is acquired in real time, and compared with the thermal state prediction data to calculate the prediction residual that reflects the accuracy of the model prediction. Based on the prediction residual, the weight coefficients in the unsteady heat conduction analytical model are adjusted online using an incremental learning algorithm to reduce the prediction bias caused by equipment performance degradation. Based on the evolution trend of the predicted residual, the stress sensitivity weight is dynamically corrected to achieve adaptive optimization of the allocation ratio between the steam superheat regulation loop and the rate of rise regulation loop.

9. The method for coordinated control of turbine acceleration rate and steam temperature based on allowable temperature difference feedback according to claim 8, characterized in that, The calculated prediction residuals, which reflect the accuracy of the model's predictions, include: Simultaneously acquire the predicted temperature values ​​of the corresponding rotor surface measuring points output by the unsteady heat conduction analytical model during the calculation of internal thermal stress; The predicted temperature value is compared with the actual temperature value of the rotor surface measuring point collected by the sensor to obtain the predicted residual characterizing the model's response deviation to thermal boundary conditions. When the fluctuation range of the predicted residual or its moving average exceeds a preset accuracy threshold, the online gradient descent algorithm is used to calculate the update increment of the weight coefficients according to the direction of the predicted residual, and the unsteady heat conduction analytical model is iteratively corrected.

10. A turbine acceleration rate and steam temperature coordinated control system based on allowable temperature difference feedback, characterized in that, The system is used in the turbine acceleration rate and steam temperature coordinated control method based on allowable temperature difference feedback as described in any one of claims 1-9, the system comprising: The data acquisition and preprocessing module collects the operating data of the steam turbine in real time during the deep peak shaving process. The operating data is cleaned and normalized by the thermodynamic state equation to form a basic dataset that supports the calculation of the thermal kinetic energy gradient. The operating data includes the main steam pressure, steam flow rate, regulating stage enthalpy value, and temperature at multiple measuring points distributed on the surface of the steam turbine rotor. The thermal state prediction module inputs the basic dataset into a preset unsteady heat conduction analytical model. It uses the rotor's material thermal conductivity, specific heat capacity, and surface heat transfer coefficient determined by steam flow rate and pressure to analyze the equivalent stress field at unmeasurable points inside the rotor in real time, generating thermal state prediction data that characterizes the real-time evolution trajectory of the rotor's thermal stress. The cumulative fatigue damage calculation module retrieves the start-stop frequency and load fluctuation amplitude from the rotor's historical operating records, establishes a mathematical model that superimposes cyclic fatigue damage caused by alternating thermal stress and creep damage accumulated over time under high-temperature conditions, and calculates the current cumulative fatigue damage value of the rotor. The dynamic allowable temperature difference limit envelope establishment module, based on the accumulated fatigue damage value, uses a fracture mechanics correction function to nonlinearly contract or expand the standard allowable temperature difference, and establishes the dynamic allowable temperature difference limit envelope at the current moment. The collaborative decision-making instruction generation module compares the thermal state prediction data with the envelope of the dynamic allowable temperature difference limit in real time to determine the stress safety margin, and allocates control vectors between the steam superheat regulation loop and the rate of rise regulation loop according to the stress sensitivity weight to generate collaborative decision-making instructions. The control command generation module calculates the corresponding steam temperature regulation target value and rise rate regulation target value based on the vector allocation result in the collaborative decision command, and generates the corresponding PID control commands respectively. The dynamic compensation and command issuance module uses an extended state observer to capture the mechanical hysteresis of the actuator and the total system disturbance, performs feedforward compensation correction on the PID control command, generates the final control signal, and issues it to the turbine speed regulation system and the steam desuperheating water actuator.

Images