Thermal insulation protection system of steam turbine

Abstract

The invention discloses a steam turbine heat insulation protection system, and relates to the technical field of steam turbine data monitoring, the protection system comprises a thermal field sensing and data acquisition module, a thermodynamic modeling and risk assessment module, a dynamic cooling regulation and control and execution module and an adaptive learning and strategy optimization module which operate in sequence; wherein the thermal field sensing and data acquisition module is used for acquiring a data parameter set of target equipment in real time; according to the technical key points, by operating the protection system, on one hand, historical operation data are learned and subjected to model iteration, a cooling control strategy is continuously optimized, and the adaptability of the system to complex working conditions is improved; and on the other hand, an LSTM neural network and a multi-objective optimization algorithm are adopted, and a digital twinborn simulation verification mechanism is combined, so that the cooling energy consumption is reduced while the safety is ensured, the technical problems that a traditional protection system is solidified in strategy and cannot be adaptively adjusted along with equipment aging and environment change are solved, and the optimization performance of autonomous regulation and control is realized.

Description

Technical Field

[0001] This invention relates to the field of steam turbine data monitoring technology, specifically a steam turbine thermal insulation protection system. Background Technology

[0002] Steam turbine data monitoring mainly involves installing sensors at key locations to collect parameters such as temperature, pressure, vibration, and speed during operation. Existing technologies typically use thermocouples or resistance temperature detectors to measure metal wall temperature, pressure transmitters to monitor steam pressure, and vibration probes to detect bearing vibration amplitude and phase. The signals are transmitted to DCS or TSI systems via field instruments for display and over-limit alarms. Meanwhile, the monitoring data is mostly recorded as point values. Some systems have trend curve functions, but the sampling frequency is low and the spatial coverage is limited. Cooling control is mostly based on fixed programs or manual intervention, and a closed-loop system linking perception, evaluation, and control has not been formed, making it difficult to cope with thermal shock problems under complex and variable operating conditions.

[0003] Existing turbine thermal protection systems largely rely on a few temperature measurement points and fixed cooling programs, making it difficult to fully grasp the true thermal state of components. For example, during a cold start-up of a 300MW unit in a power plant, monitoring was only conducted by two thermocouples on the outer flange wall, failing to detect the rapid temperature rise on the inner wall, which led to micro-cracks on the split surface. Similarly, some units use constant airflow cooling, but the fans continue to operate at full power during sudden load drops, causing a reverse increase in cylinder temperature difference and triggering thermal deformation. Due to the lack of comprehensive temperature distribution data, thermal stress calculations are often based on empirical formulas, failing to dynamically respond to changes in operating conditions and resulting in delayed early warnings. Furthermore, risk assessments often rely on single temperature or stress thresholds to trigger alarms, lacking a tiered evaluation mechanism, leading to false alarms or missed alarms. Cooling adjustments depend on operator experience, lacking closed-loop feedback and resulting in untimely adjustments. More importantly, existing or designed protection systems lack learning capabilities, often causing the same problems to recur. For instance, frequent thermal shocks occur during winter equipment startup, yet the control strategy cannot be automatically corrected, leaving the overall turbine protection system configuration in a relatively rigid control state. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: A steam turbine thermal insulation protection system, comprising: The thermal field sensing and data acquisition module collects the data parameter set of the target device in real time; The thermodynamic modeling and risk assessment module receives a set of data parameters, constructs a transient heat conduction model of the target equipment, discretizes each component in the target equipment into three-dimensional mesh elements, determines the boundary conditions and calculates the temperature gradient, thermal stress tensor and equivalent stress of each three-dimensional mesh element, compares them with the corresponding set thresholds to determine the risk level of different target components, and outputs a risk level map and trend curves of target parameters in the data parameter set. The dynamic cooling control and execution module receives and adjusts the cooling system and execution system based on the risk level map; The adaptive learning and strategy optimization module performs structured cleaning of stored historical operating data during the data preprocessing stage. During model training, a dual-channel LSTM network architecture is used: the main channel inputs the operating condition feature vector to predict the maximum temperature rise and equivalent stress peak value of each component of the target equipment; the auxiliary channel introduces time-lag variables to capture thermal inertia effects. In the strategy generation stage, backpropagation is performed based on the trained model to obtain a candidate strategy set. Before strategy deployment, offline simulation verification is performed: the candidate strategy set is imported into the digital twin model, and after simulation, if the simulation results show a risk level reduction of ≥1, the corresponding candidate strategy is written to the database and marked as effective.

[0005] Furthermore, the data parameter set includes at least: temperature field distribution, heat flux density, ambient temperature and humidity, and cooling medium parameters; wherein, the cooling medium parameters include at least: the flow rate, pressure, temperature, and velocity of the cooling steam.

[0006] Furthermore, the transient heat transfer model is established based on the finite difference method and solved using an implicit difference scheme with a set time step; the boundary conditions are determined based on the measured temperature and the convective heat transfer coefficient.

[0007] Further, the comparison process with the corresponding set thresholds is as follows: when the equivalent stress does not exceed 70% of the material's allowable stress threshold, and the temperature gradient is ≤50℃ / m, and the thermal stress change rate is ≤2MPa / min, it is judged as a low-risk level; when the equivalent stress exceeds 70% but does not reach 85% of the material's allowable stress threshold, or the temperature gradient is within (50, 100), or the thermal stress change rate is within (2, 4), it is judged as a medium-risk level; when the equivalent stress exceeds 85% of the material's allowable stress threshold for three consecutive cycles T, or the temperature gradient is within [100, 150), or the thermal stress change rate is within [4, 6), it is judged as a high-risk level, and a thermal fatigue damage accumulation algorithm is introduced, based on the Miner linear accumulation method to evaluate the remaining life of the target component; when the equivalent stress reaches or exceeds the material's allowable stress threshold, or the temperature gradient is ≥150℃ / m, or the thermal stress change rate is ≥6MPa / min, it is judged as an emergency risk level.

[0008] Furthermore, the cooling system includes at least: a steam cooling circuit, a forced air cooling device, and an adjustable airflow structure; the actuation system includes at least: a PID controller group, an electric regulating valve group, a variable frequency fan, and a shape memory alloy actuator.

[0009] Furthermore, the adjustment of the cooling system and the execution system is based on the following: receiving the trend curve of the target temperature and the measured value, calculating the absolute value of the deviation between the two, recording it as the deviation amount, and outputting a control signal: when the deviation amount ≥ M, the rapid cooling mode is activated, the main cooling valve is fully opened and the dual fans are started; when the deviation amount ∈ (1 / 3M, M), the proportional adjustment mode is activated, and the valve opening is controlled by pulse width modulation; when the deviation amount ≤ 1 / 3M, the maintenance mode is activated, and the air volume is adjusted according to the set standard amount.

[0010] Furthermore, during the data preprocessing stage, after the stored historical operating data is structured and cleaned, the following steps are also included: removing outliers and filling in short-term missing data using cubic spline interpolation; classifying and archiving historical operating data according to operating condition type, with each type of historical operating data meeting the minimum sample size requirement; among which, the operating condition types include at least: cold start, hot start, coasting stop, and variable load; the minimum sample size is 20.

[0011] Furthermore, during the model training phase, the operating condition feature vector input to the main channel includes at least: the main steam temperature change rate, the load increase / decrease rate, the ambient temperature and humidity, the initial state of the cooling medium, and the cumulative number of start-stop cycles. An early stop mechanism is implemented during training, terminating training when the validation set loss fails to decrease for five consecutive rounds. After model training is completed, a triple verification mechanism is executed. This triple verification mechanism is based on the following steps: backtracking test: using recent 7 days of data not used in training to verify the prediction error, requiring a root mean square error ≤ 8%; trend consistency test: verifying whether the stress change direction output by the model matches the measured trend; boundary condition simulation: inputting pre-set extreme conditions to test the model's robustness.

[0012] Furthermore, during the strategy generation stage, the basis for back-optimization based on the trained model is as follows: with minimizing the maximum thermal stress and minimizing the cooling energy consumption as the dual objective functions, a non-dominated sorting genetic algorithm is used to search for the optimal combination of cooling parameters and generate control strategies that meet the constraints; among which, the constraints include at least: the temperature rise rate does not exceed the material's allowable limit, the cooling response time is ≤30s, and there is no secondary thermal shock; the control strategies are filtered by a pre-built expert rule base to form a candidate strategy set.

[0013] Furthermore, in the stage before strategy deployment, the simulation process is as follows: simulate the operation process for the next 24 hours; write the corresponding candidate strategy records into the database as update records, and store the update records in the audit log.

[0014] This invention provides a steam turbine thermal insulation protection system, which has the following beneficial effects: This solution enables high-precision, high-frequency, and multi-dimensional real-time monitoring of the temperature field, heat flux density, and cooling medium parameters of key components of the steam turbine, improving the ability to acquire thermal state information. At the same time, it effectively overcomes the problems of insufficient spatial coverage and response lag in traditional single-point temperature measurement, solves the technical defects of traditional thermal insulation protection systems that lead to misjudgment or omission due to incomplete data acquisition, and realizes full-domain dynamic perception of the thermal field distribution of the steam turbine. This scheme employs a transient heat conduction model based on the finite difference method, combined with measured boundary conditions for dynamic solution, enabling refined prediction of thermal stress and deformation of turbine components. It introduces a four-level risk assessment and spatial continuity judgment logic, which not only evaluates based on single indicators such as stress and temperature gradient, but also analyzes through multi-parameter coupling and regional correlation, solving the technical bottleneck of traditional methods in accurately identifying local thermal stress concentrations and predicting their evolution trend, and realizing the transformation from passive response to proactive early warning. This solution automatically generates and executes cooling adjustment commands based on the risk level map. Through the coordinated action of PID control group, variable frequency fan and shape memory alloy flow guide structure, it realizes on-demand allocation and spatial orientation guidance of cooling resources. At the same time, its cooling regulation adopts closed-loop feedback to ensure that the regulation process is stable and controllable, realizing the dynamic and precise cooling intervention, which can solve the related problems of over- or under-cooling in traditional fixed cooling mode to a certain extent. This solution learns from historical operating data and iterates the model to continuously optimize the cooling control strategy, thereby improving the system's adaptability to complex operating conditions. On the other hand, it adopts LSTM neural networks and multi-objective optimization algorithms, combined with a digital twin simulation verification mechanism, to reduce cooling energy consumption while ensuring safety. This solves the technical problem of traditional protection system strategies being fixed and unable to adapt to equipment aging and environmental changes, and achieves optimized performance through autonomous regulation. Attached Figure Description

[0015] Figure 1 This is a simplified schematic diagram of the operation process of a steam turbine thermal insulation protection system according to the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1:

[0018] Please see Figure 1This embodiment provides a steam turbine thermal insulation protection system. This system achieves dynamic thermal protection for high-temperature components of the steam turbine through multi-level collaborative operation modules. The operation of the entire protection system improves the operational safety and energy efficiency stability of the equipment, namely the steam turbine and its related components. The following provides a detailed description of each module in the protection system: I. Thermal Field Sensing and Data Acquisition Module: This module is used to collect data parameter sets from the target device in real time; In this embodiment, the target equipment refers to a steam turbine. The data parameter set includes at least: temperature field distribution, heat flux density, ambient temperature and humidity, and cooling medium parameters. The cooling medium parameters include the flow rate, pressure, temperature, and velocity of cooling steam or air, used to characterize the supply status and heat exchange capacity of the cooling medium, ensuring that the thermal insulation protection system dynamically adjusts based on actual cooling performance. Simultaneously, the various data included in the data parameter set are collected from the surface and interior of key components of the target equipment, and the data parameter set is obtained from a thermal field sensing unit, which includes a distributed fiber optic temperature sensor, an infrared thermal imager array, an embedded thermocouple group, and cooling medium flow and temperature sensors. Specifically, the distributed fiber optic temperature sensor is arranged circumferentially and axially along the outer wall of the cylinder, with a sampling frequency of no less than 10Hz and a spatial resolution of 0.1m, used to acquire continuous temperature field data. The infrared thermal imager array is installed inside the protective cover, scanning the outer surface of the steam turbine every 30 seconds to generate... Two-dimensional thermal maps; embedded thermocouple groups are embedded in easily overheated areas such as cylinder flanges and rotor ends to provide high-precision point temperature data; cooling medium flow and temperature sensors are installed in the cooling medium flow pipelines of turbine components to obtain cooling medium parameters; the thermal field sensing unit transmits data to the central processing unit via industrial Ethernet; when the module is running, it first starts a self-test program to verify the communication status and zero-point drift of each sensor; then it enters continuous monitoring mode, packaging the raw data into standard data frames and adding timestamps and location codes; the data acquisition cycle is set to once every 3 seconds, and when high-frequency events, such as start-up, shutdown, or sudden load changes occur, it automatically switches to a 1-second acquisition cycle; Example: During the cold start of a 300MW subcritical turbine, the module detected that the temperature in the front flange area of ​​the high-pressure cylinder rose from 85℃ to 387℃ within 40 minutes after the start-up, with a heating rate as high as 7.5℃ / min, while the cooling steam flow decreased by 12%, thus achieving data acquisition of the target equipment, namely the turbine.

[0019] By adopting the above technical solutions, high-precision, high-frequency, and multi-dimensional real-time monitoring of the temperature field, heat flux density, and cooling medium parameters of key components of the steam turbine is achieved, improving the ability to acquire thermal state information. At the same time, the use of distributed fiber optic temperature measurement and infrared thermal imager array for collaborative sensing, combined with embedded thermocouple group fixed-point calibration, effectively overcomes the problems of insufficient spatial coverage and response lag in traditional single-point temperature measurement. It also solves the technical defects of traditional thermal insulation protection systems that lead to misjudgment or omission due to incomplete data acquisition, realizing full-domain dynamic perception of the steam turbine thermal field distribution, and providing an effective data foundation and basis for subsequent modeling and control.

[0020] II. Thermodynamic Modeling and Risk Assessment Module: Receive the data parameter set, construct the transient heat conduction model of the target device, and predict thermal stress and deformation; The transient heat conduction model is based on the finite difference method, discretizing each component within the target device into three-dimensional mesh elements (e.g., cylinders, rotors, etc., are discretized into three-dimensional network elements). Boundary conditions are determined based on measured temperature and convective heat transfer coefficient. The transient heat conduction model is solved using an implicit difference scheme, with a set time step, for example, 2s in this embodiment, to ensure numerical stability. The temperature gradient, thermal stress tensor, and equivalent stress of each three-dimensional mesh element node are calculated and compared with corresponding set thresholds to determine the risk level of different target components. A risk level map and trend curves of target parameters in the data parameter set are output. The target parameter can be selected as temperature, i.e., the trend curve of the target temperature; if heat flux density is selected, it corresponds to the trend curve of the target heat flux density. When the equivalent stress in a certain area does not exceed 70% of the material's allowable stress threshold, and the temperature gradient is ≤50℃ / m, and the thermal stress change rate is ≤2MPa / min, it is judged as a low-risk level. Here, 70%, 50℃ / m, and 2MPa / min are the corresponding set thresholds. This state indicates that the component is in thermal equilibrium or a slow temperature change process, with no significant thermal stress concentration and sufficient structural safety margin. When the equivalent stress exceeds 70% but does not reach 85% of the material's allowable stress threshold, or the temperature gradient is within (50, 100), the unit is ℃ / m, or the thermal stress... If the rate of change is within (2, 4), in MPa / min, it is classified as medium risk. Among these, 70% but less than 85% of the allowable stress threshold of the material, as well as the interval values ​​corresponding to the temperature gradient and the rate of change of thermal stress, are all set thresholds. Details regarding these set thresholds will not be elaborated upon here or thereafter. This state indicates that the component has experienced a relatively rapid thermal shock, with localized mild stress concentration, and its development needs to be monitored. When the equivalent stress exceeds 85% of the allowable stress threshold of the material for three consecutive cycles T, or the temperature gradient is within [100, 150), or the rate of change of thermal stress is within [4, 6), it is classified as medium risk. Classified as high-risk, a thermal fatigue damage accumulation algorithm is introduced, based on the Miner linear accumulation method, to assess the remaining life of the target component. This state corresponds to significant thermal unevenness conditions, such as rapid start-up and shutdown or sudden load changes, posing a risk of inducing plastic deformation or microcrack propagation, requiring immediate active cooling intervention. When the equivalent stress reaches or exceeds the material's allowable stress threshold, or the temperature gradient is ≥150℃ / m, or the thermal stress change rate is ≥6MPa / min, it is classified as an emergency-risk level, indicating that the component has entered the yield critical state or experienced local failure, requiring the triggering of the highest priority protection actions, including automatic... The system can trigger load reduction, forced cooling, or a shutdown warning. It should be noted that the above-mentioned levels are determined based on the principle that any condition being met will trigger the response. This is combined with spatial continuity assessment. If five or more adjacent grid cells simultaneously reach the same level, the overall response priority is increased by one level to prevent the spread of local hotspots. Example: In addition to determining the high-pressure cylinder front flange as a high-risk level, this module also identifies eight consecutive units in its surrounding shaft seal section with a temperature gradient of 110℃ / m. Therefore, the entire area is upgraded to an emergency risk level, and the command packet priority is adjusted, requiring forced cooling to be implemented within 3 minutes.

[0021] Specifically, when the thermodynamic modeling and risk assessment module is running, the pre-stored geometric and material parameter library can be called first to initialize the model; then the boundary conditions are updated every 2 seconds to resolve the full field distribution; finally, the risk level map and the trend curves of the key parameters selected in the data parameter set are output; the risk level map corresponds to the low risk level, medium risk level, high risk level and emergency risk level given above.

[0022] By adopting the above technical solution, on the one hand, it uses a transient heat conduction model based on the finite difference method and combines it with measured boundary conditions for dynamic solution, thereby realizing the refined prediction of thermal stress and deformation of turbine components; on the other hand, it introduces a four-level risk assessment and spatial continuity judgment logic, which not only evaluates based on single indicators such as stress and temperature gradient, but also analyzes through multi-parameter coupling and regional correlation, avoiding the insufficient adaptability of traditional static threshold judgment under complex working conditions, solving the technical bottleneck of traditional methods in accurately identifying local thermal stress concentration and predicting its evolution trend, and realizing the transformation from passive response to active early warning.

[0023] III. Dynamic Cooling Regulation and Execution Module: Receive and dynamically adjust the pre-installed cooling system and execution system according to the risk level map; wherein, the cooling system includes at least: a steam cooling circuit, a forced air cooling device and an adjustable flow guide structure; the execution system includes at least: a PID controller group, an electric regulating valve group, a variable frequency fan and a shape memory alloy actuator; The PID controller group receives the trend curve of the target temperature and the measured value, calculates the absolute value of the deviation between the two, records it as the deviation, and outputs a control signal: when the deviation is ≥ M, the rapid cooling mode is activated, the main cooling valve is fully opened and the dual fans are started; when the deviation is ∈ (1 / 3M, M), the proportional regulation mode is activated, and the valve opening is controlled by pulse width modulation; when the deviation is ≤ 1 / 3M, the maintenance mode is activated, and the air volume is adjusted only according to the set standard amount. The value of M is a positive integer greater than 0. In this embodiment, the value of M is 15, so 1 / 3M = 5. Specifically, the airflow guiding structure is made of nickel-titanium alloy sheets, which can automatically change the direction of cooling airflow upon thermal deformation, prioritizing coverage of high-risk areas. During module operation, the status of the actuator is first verified to confirm that the valve's movable range and fan speed feedback are normal. Then, the control strategy is loaded, and the adjustment is initiated according to the command package parameters. The execution effect is fed back every 10 seconds, forming a closed loop. For example, in response to the high-pressure cylinder risk given in the previous example, this module instructs the main cooling valve opening to increase from 40% to 75%, starts the backup fan to increase the airflow to 12000m³ / h, and at the same time, the guide plate deforms due to local high temperature, directing the airflow to the bottom of the flange. After 5 minutes, the temperature in this area is measured to drop to 362℃, the stress drops to 102MPa, and the risk level is reduced to medium.

[0024] By adopting the above technical solution, cooling adjustment commands are automatically generated and executed based on the risk level map. Through the coordinated action of PID control group, variable frequency fan and shape memory alloy flow guide structure, cooling resources are allocated on demand and guided spatially. At the same time, its cooling regulation adopts closed-loop feedback to ensure that the regulation process is stable and controllable. It solves the problems of over- or under-cooling, high energy consumption or easy to cause secondary thermal shock in traditional fixed cooling mode. It realizes dynamic and precise cooling intervention and reflects the high responsiveness design of the system at the execution level.

[0025] IV. Adaptive Learning and Policy Optimization Module: In the data preprocessing stage, the historical operating data stored in the first three modules is structured and cleaned to remove outliers, and cubic spline interpolation is used to fill in short-term missing data. The historical operating data is categorized and archived according to operating condition type, with each category meeting the minimum sample size requirement. It should be noted that historical operating data refers to all types of data generated by the first three modules, including data parameter sets, risk level maps, and parameter changes during cooling system regulation, etc., which will not be elaborated upon here. Outliers are usually removed due to sensor malfunctions or communication interruptions; for example, temperature fluctuations exceeding ±50℃ / s are considered outliers. Cubic spline interpolation is used to ensure the integrity of the training set. When categorizing and archiving historical operating data by operating condition type, the operating condition types must include at least: cold start, hot start, coasting stop, and variable load. The minimum sample size is 20, which satisfies 20 complete operating cycles to ensure statistical representativeness. During the model training phase, a dual-channel LSTM network architecture is adopted: the main channel inputs the operating condition feature vector, which includes at least the main steam temperature change rate, load increase / decrease rate, ambient temperature and humidity, initial state of the cooling medium, and cumulative start / stop count, predicting the maximum temperature rise and equivalent stress peak value of each component of the target equipment; the auxiliary channel introduces time lag variables to capture the thermal inertia effect; an early stop mechanism is set during training, terminating training when the validation set loss does not decrease for 5 consecutive rounds to prevent overfitting; after the model training is completed, a triple verification mechanism is executed; the triple verification mechanism is based on: backtracking test: using recent 7 days of data not involved in training to verify the prediction error, requiring the root mean square error ≤8%; trend consistency test: verifying whether the stress change direction output by the model matches the measured trend; boundary condition simulation: inputting pre-set extreme conditions to test the robustness of the model; in this embodiment, the corresponding extreme conditions can be achieved by -20℃ cold start + rapid load increase, and the specific extreme conditions can be set according to actual needs; In the strategy generation phase, back-optimization is performed based on the trained model: with the dual objective functions of minimizing maximum thermal stress and minimizing cooling energy consumption, a non-dominated sorting genetic algorithm is used to search for the optimal combination of cooling parameters; for example, valve opening curves and fan speed sequences; generating control strategies that meet the constraints; the constraints include: temperature rise rate not exceeding the material's allowable limit, cooling response time ≤30s, and no secondary thermal shock; the control strategies are filtered by a pre-built expert rule base to form a candidate strategy set; the content of the expert rule base includes: prohibiting the adjustment of cooling airflow in the rotor's critical speed range, etc., which are only examples here and will not be elaborated further. Before strategy deployment, offline simulation verification is performed: the candidate strategy set is imported into a high-fidelity digital twin model to simulate the typical operation process in the next 24 hours. When the simulation results show that the risk level has decreased by ≥1, the corresponding candidate strategy is written into the database and marked as effective. Specifically, all update records are stored in the audit log to support traceability review.

[0026] By adopting the above technical solutions, historical operating data is learned and models are iterated to continuously optimize the cooling control strategy, thereby improving the system's adaptability to complex operating conditions. On the other hand, by using LSTM neural networks and multi-objective optimization algorithms, combined with a digital twin simulation verification mechanism, cooling energy consumption is reduced while ensuring safety. This solves the technical problem of traditional protection systems having fixed strategies and being unable to adapt to equipment aging and environmental changes, achieving optimized performance through autonomous control. In summary, the entire protection system constitutes a closed-loop technical design from perception to analysis, execution, and optimization. Data from each link is interconnected and functions are complementary, forming an inseparable overall technical system, thereby comprehensively improving the intelligence level and engineering practicality of turbine thermal insulation protection.

[0027] Example 2:

[0028] Based on Example 1, this embodiment also provides a method for heat insulation protection of a steam turbine, including the following specific steps: S1. Real-time acquisition of data parameter sets from the target equipment; S2. Receiving the data parameter set, constructing a transient heat conduction model of the target equipment, discretizing each component within the target equipment into three-dimensional mesh elements, determining boundary conditions, and calculating the temperature gradient, thermal stress tensor, and equivalent stress of each three-dimensional mesh element. These are then compared with corresponding set thresholds to determine the risk level of different target components, outputting a risk level map and trend curves of the target parameters in the data parameter set; S3. Receiving and adjusting the cooling and execution systems based on the risk level map; S4. In the data preprocessing stage, processing the stored historical operating data... The model undergoes structured cleaning. During training, a dual-channel LSTM network architecture is used: the main channel inputs the operating condition feature vector to predict the maximum temperature rise and equivalent stress peak of each component in the target equipment; the auxiliary channel introduces time-lag variables to capture thermal inertia effects. In the strategy generation stage, backpropagation is performed based on the trained model to obtain a candidate strategy set. Before strategy deployment, offline simulation verification is performed: the candidate strategy set is imported into the digital twin model, and after simulation, when the simulation results show a risk level reduction of ≥1, the corresponding candidate strategy is written into the database and marked as effective.

[0029] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0030] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0031] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A steam turbine thermal insulation protection system, characterized in that, include: The thermal field sensing and data acquisition module collects data parameter sets from the target device in real time. The thermodynamic modeling and risk assessment module receives a set of data parameters, constructs a transient heat conduction model of the target equipment, discretizes each component in the target equipment into three-dimensional mesh elements, determines the boundary conditions and calculates the temperature gradient, thermal stress tensor and equivalent stress of each three-dimensional mesh element, compares them with the corresponding set thresholds to determine the risk level of different target components, and outputs a risk level map and trend curves of target parameters in the data parameter set. The dynamic cooling control and execution module receives and adjusts the cooling system and execution system based on the risk level map; The adaptive learning and policy optimization module performs structured cleaning of the stored historical operating data during the data preprocessing stage; during the model training stage, it adopts a dual-channel LSTM network architecture: the main channel inputs the operating condition feature vector to predict the maximum temperature rise and equivalent stress peak value of each component of the target equipment; the auxiliary channel introduces time lag variables to capture the thermal inertia effect. In the strategy generation phase, back-optimization is performed based on the trained model to obtain a set of candidate strategies. In the phase before strategy deployment, offline simulation verification is performed: the set of candidate strategies is imported into the digital twin model, and after simulation, when the simulation results show that the risk level has decreased by ≥1, the corresponding candidate strategy is written into the database and marked as effective.

2. The turbine thermal insulation protection system according to claim 1, characterized in that: The data parameter set includes at least: temperature field distribution, heat flux density, ambient temperature and humidity, and cooling medium parameters; among which, the cooling medium parameters include at least: the flow rate, pressure, temperature, and velocity of the cooling steam.

3. The turbine thermal insulation protection system according to claim 1, characterized in that: The transient heat transfer model is established based on the finite difference method and solved using an implicit difference scheme with a set time step. The boundary conditions are determined based on the measured temperature and the convective heat transfer coefficient.

4. The turbine thermal insulation protection system according to claim 1, characterized in that: The process of comparing the equivalent stress with the corresponding set thresholds is as follows: when the equivalent stress does not exceed 70% of the allowable stress threshold of the material, and the temperature gradient is ≤50℃ / m, and the thermal stress change rate is ≤2MPa / min, it is judged as low risk level; when the equivalent stress exceeds 70% but does not reach 85% of the allowable stress threshold of the material, or the temperature gradient is within (50, 100), or the thermal stress change rate is within (2, 4), it is judged as medium risk level; when the equivalent stress exceeds 85% of the allowable stress threshold of the material for three consecutive cycles T, or the temperature gradient is within [100, 150), or the thermal stress change rate is within [4, 6), it is judged as high risk level, and a thermal fatigue damage accumulation algorithm is introduced, based on the Miner linear accumulation method to evaluate the remaining life of the target component; when the equivalent stress reaches or exceeds the allowable stress threshold of the material, or the temperature gradient is ≥150℃ / m, or the thermal stress change rate is ≥6MPa / min, it is judged as emergency risk level.

5. A steam turbine thermal insulation protection system according to claim 1, characterized in that: The cooling system includes at least: a steam cooling circuit, a forced air cooling device, and an adjustable airflow structure; the actuation system includes at least: a PID controller group, an electric regulating valve group, a variable frequency fan, and a shape memory alloy actuator.

6. A steam turbine thermal insulation protection system according to claim 1, characterized in that: The basis for adjusting the cooling system and the execution system is: receiving the trend curve of the target temperature and the measured value, calculating the absolute value of the deviation between the two, recording it as the deviation amount, and outputting a control signal: when the deviation amount is ≥ M, the rapid cooling mode is activated, the main cooling valve is fully opened and the dual fans are started; When the deviation is ∈ (1 / 3M, M), the proportional control mode is activated, and the valve opening is controlled by pulse width modulation; when the deviation is ≤1 / 3M, the maintenance mode is activated, and the air volume is adjusted according to the set standard.

7. A steam turbine thermal insulation protection system according to claim 1, characterized in that: During the data preprocessing stage, after the stored historical operating data is structured and cleaned, the following steps are also included: removing outliers and filling in short-term missing data using cubic spline interpolation; classifying and archiving historical operating data according to operating condition type, with each type of historical operating data meeting the minimum sample size requirement; among which, the operating condition types include at least: cold start, hot start, coasting stop, and variable load; the minimum sample size is 20.

8. A steam turbine thermal insulation protection system according to claim 1, characterized in that: During the model training phase, the main channel input feature vector includes at least: main steam temperature change rate, load increase / decrease rate, ambient temperature and humidity, initial state of cooling medium, and cumulative start / stop count. An early stop mechanism is implemented during training, terminating training when the validation set loss fails to decrease for five consecutive rounds. After model training, a triple validation mechanism is executed. This triple validation mechanism is based on the following steps: backtracking test: using recent 7 days of data not used in training to verify the prediction error, requiring a root mean square error ≤ 8%; trend consistency test: verifying whether the stress change direction output by the model matches the measured trend; and boundary condition simulation: inputting pre-set extreme conditions to test the model's robustness.

9. A steam turbine thermal insulation protection system according to claim 1, characterized in that: During the strategy generation phase, the basis for back-optimization based on the trained model is as follows: with minimizing the maximum thermal stress and minimizing the cooling energy consumption as the dual objective functions, a non-dominated sorting genetic algorithm is used to search for the optimal combination of cooling parameters and generate control strategies that meet the constraints. Among these constraints, at least the following are included: the temperature rise rate does not exceed the material's allowable limit, the cooling response time is ≤30s, and there is no secondary thermal shock. The control strategies are filtered through a pre-built expert rule base to form a set of candidate strategies.

10. A steam turbine thermal insulation protection system according to claim 1, characterized in that: In the phase before strategy deployment, the simulation process is as follows: simulate the operation process for the next 24 hours; write the corresponding candidate strategy records into the database as update records, and store the update records in the audit log.

Images