Molten salt energy storage turbine power generation system monitoring method and device, computer equipment, readable storage medium and program product
By adopting a monitoring method for molten salt energy storage turbine power generation system based on spatiotemporal attention network and near-end policy optimization controller, the problem of operational instability is solved, and the system achieves advanced perception and safety verification of load changes, thereby improving the system's operating efficiency and safety.
Patent Information
- Application Number
- CN202511773213.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-04-14
AI Technical Summary
Traditional molten salt energy storage turbine power generation systems are unstable in operation and have difficulty responding accurately to load changes, resulting in low operating efficiency and safety risks.
A load forecasting model based on a spatiotemporal attention network and a near-end policy optimization controller are adopted. The load forecasting and control policy optimization are carried out in combination with a simulation model. The system safety and efficiency are ensured through simulation and safety verification.
It enables proactive sensing of load changes, improves the dynamic matching accuracy and operational safety of the system, avoids efficiency losses and equipment damage caused by improper control, and significantly improves overall operating efficiency.
Smart Images

Figure CN121863674A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of molten salt energy storage turbine power generation technology, and in particular to a monitoring method, device, computer equipment, computer-readable storage medium and computer program product for a molten salt energy storage turbine power generation system. Background Technology
[0002] Molten salt energy storage turbine power generation systems are attracting increasing attention from the global energy industry as a key technology for the efficient utilization of renewable energy and peak shaving of the power grid. This system effectively solves the problems of high volatility and difficulty in direct grid connection of renewable energy by storing intermittent solar energy or surplus electricity in the form of high-temperature molten salt and then outputting stable power through turbine generator sets.
[0003] Traditional technologies mainly rely on empirical curves and fixed-parameter PID (proportional-integral-derivative) control strategies, using real-time monitoring of system parameter changes for feedback adjustment.
[0004] However, in the actual operation of traditional technologies, the overall operation of molten salt energy storage turbine power generation systems is still often unstable, which makes it difficult for molten salt energy storage turbine power generation systems to accurately respond to the actual needs of load changes, resulting in low operating efficiency of molten salt energy storage turbine power generation systems. Summary of the Invention
[0005] Therefore, it is necessary to provide a monitoring method, device, computer equipment, computer-readable storage medium, and computer program product for molten salt energy storage turbine power generation systems that can improve operational efficiency in response to the above-mentioned technical problems.
[0006] Firstly, this application provides a monitoring method for a molten salt energy storage turbine power generation system, the method comprising:
[0007] The process involves acquiring a simulation model, current operating data, and current load data for the molten salt energy storage turbine power generation system. Using a pre-set load forecasting model, the system predicts the load demand for the molten salt energy storage turbine power generation system over a pre-set future time period based on the current operating and load data. The corresponding control setpoint is then queried. This pre-set load forecasting model is obtained by training a spatiotemporal attention network based on historical operating and load data of the molten salt energy storage turbine power generation system. A near-end strategy optimization controller generates a target power output control strategy based on the load demand forecast and control setpoint, enabling the molten salt energy storage turbine power generation system to achieve a pre-set optimization objective. The simulation model is used to simulate the operating state of the molten salt energy storage turbine power generation system according to the target power output control strategy, obtaining simulation data. This simulation data is then subjected to safety verification based on pre-set unit operating constraints. If the safety verification passes, the molten salt energy storage turbine power generation system is configured according to the target power output control strategy and control setpoint. If the safety verification fails, the target power output control strategy is adjusted back until the safety verification passes.
[0008] Secondly, this application also provides a monitoring device for a molten salt energy storage turbine power generation system, comprising:
[0009] The system comprises four modules: an acquisition module, a load forecasting module, and a strategy generation module. The acquisition module obtains the simulation model, current operating data, and current load data of the molten salt energy storage turbine power generation system. The load forecasting module uses a pre-set load forecasting model to predict the load demand of the molten salt energy storage turbine power generation system over a preset future time period based on the current operating data and current load data. The module obtains the load demand forecasting result and queries the corresponding control setpoint. The pre-set load forecasting model is obtained by training a spatiotemporal attention network based on historical operating data and historical load data of the molten salt energy storage turbine power generation system. The strategy generation module optimizes the controller through a near-end strategy based on the load demand forecasting result and the control setpoint. The system comprises: a setpoint generation module to generate a target power output control strategy that enables the molten salt energy storage turbine power generation system to achieve a preset optimization objective; a simulation module to simulate the operating state of the molten salt energy storage turbine power generation system based on the target power output control strategy using a simulation model, obtaining simulation data of the operating state, and performing safety verification on the simulation data of the operating state according to preset unit operating constraints; a configuration module to configure the molten salt energy storage turbine power generation system according to the target power output control strategy and control setpoint if the safety verification passes; and a feedback adjustment module to adjust the target power output control strategy if the safety verification fails, until the safety verification passes.
[0010] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:
[0011] The process involves acquiring a simulation model, current operating data, and current load data for the molten salt energy storage turbine power generation system. Using a pre-set load forecasting model, the system predicts the load demand for the molten salt energy storage turbine power generation system over a pre-set future time period based on the current operating and load data. The corresponding control setpoint is then queried. This pre-set load forecasting model is obtained by training a spatiotemporal attention network based on historical operating and load data of the molten salt energy storage turbine power generation system. A near-end strategy optimization controller generates a target power output control strategy based on the load demand forecast and control setpoint, enabling the molten salt energy storage turbine power generation system to achieve a pre-set optimization objective. The simulation model is used to simulate the operating state of the molten salt energy storage turbine power generation system according to the target power output control strategy, obtaining simulation data. This simulation data is then subjected to safety verification based on pre-set unit operating constraints. If the safety verification passes, the molten salt energy storage turbine power generation system is configured according to the target power output control strategy and control setpoint. If the safety verification fails, the target power output control strategy is adjusted back until the safety verification passes.
[0012] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:
[0013] The process involves acquiring a simulation model, current operating data, and current load data for the molten salt energy storage turbine power generation system. Using a pre-set load forecasting model, the system predicts the load demand for the molten salt energy storage turbine power generation system over a pre-set future time period based on the current operating and load data. The corresponding control setpoint is then queried. This pre-set load forecasting model is obtained by training a spatiotemporal attention network based on historical operating and load data of the molten salt energy storage turbine power generation system. A near-end strategy optimization controller generates a target power output control strategy based on the load demand forecast and control setpoint, enabling the molten salt energy storage turbine power generation system to achieve a pre-set optimization objective. The simulation model is used to simulate the operating state of the molten salt energy storage turbine power generation system according to the target power output control strategy, obtaining simulation data. This simulation data is then subjected to safety verification based on pre-set unit operating constraints. If the safety verification passes, the molten salt energy storage turbine power generation system is configured according to the target power output control strategy and control setpoint. If the safety verification fails, the target power output control strategy is adjusted back until the safety verification passes.
[0014] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:
[0015] The process involves acquiring a simulation model, current operating data, and current load data for the molten salt energy storage turbine power generation system. Using a pre-set load forecasting model, the system predicts the load demand for the molten salt energy storage turbine power generation system over a pre-set future time period based on the current operating and load data. The corresponding control setpoint is then queried. This pre-set load forecasting model is obtained by training a spatiotemporal attention network based on historical operating and load data of the molten salt energy storage turbine power generation system. A near-end strategy optimization controller generates a target power output control strategy based on the load demand forecast and control setpoint, enabling the molten salt energy storage turbine power generation system to achieve a pre-set optimization objective. The simulation model is used to simulate the operating state of the molten salt energy storage turbine power generation system according to the target power output control strategy, obtaining simulation data. This simulation data is then subjected to safety verification based on pre-set unit operating constraints. If the safety verification passes, the molten salt energy storage turbine power generation system is configured according to the target power output control strategy and control setpoint. If the safety verification fails, the target power output control strategy is adjusted back until the safety verification passes.
[0016] The aforementioned monitoring method, device, computer equipment, computer-readable storage medium, and computer program product for molten salt energy storage turbine power generation system first utilizes a load forecasting model trained on a spatiotemporal attention network. By learning from historical operating data and historical load data, it predicts the load demand for a preset future time period and matches the corresponding control setpoint, overcoming the limitation of traditional PID control that can only respond to the current state and achieving proactive perception of load changes. Next, the near-end strategy optimization controller autonomously seeks the optimal target power output control strategy based on the load demand forecast results, rather than relying on fixed empirical parameters. This overcomes the insufficient adaptability of traditional PID control in nonlinear and hysteresis systems, significantly improving the dynamic matching accuracy between the molten salt energy storage unit and the turbine unit. Furthermore, a simulation model is used to simulate the operating state of the molten salt energy storage turbine power generation system under the target power output control strategy, identifying potential risk points that may cause abnormal operating conditions such as excessive turbine vibration in advance. This achieves pre-emptive safety verification of the target power output control strategy, ensuring the operational safety of the molten salt energy storage turbine power generation system while ensuring that the target power output control strategy meets load demand. This pre-verification mechanism effectively avoids efficiency losses and equipment damage caused by improper control. Thus, through the organic combination of load forecasting, strategy optimization, and safety verification, a technological leap from passive response to proactive regulation is achieved, enabling the system to accurately track load demand while maintaining optimal operating conditions, thereby significantly improving the overall operating efficiency of the molten salt energy storage turbine power generation system. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is an application environment diagram of the monitoring method for a molten salt energy storage turbine power generation system in one embodiment of this application;
[0019] Figure 2 This is a flowchart illustrating the fault identification process in one embodiment of this application;
[0020] Figure 3 This is a flowchart illustrating the fault sample data generation process in one embodiment of this application;
[0021] Figure 4 This is a structural block diagram of a monitoring device for a molten salt energy storage turbine power generation system in one embodiment;
[0022] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0024] Molten salt energy storage turbine power generation systems are receiving increasing attention from the global energy industry as a key technology for the efficient utilization of renewable energy and peak shaving of the power grid.
[0025] Molten salt energy storage turbine power generation system includes molten salt energy storage unit and turbine generator unit.
[0026] Molten salt energy storage units include electric heaters (or solar thermal collectors), cold molten salt storage tanks, hot molten salt storage tanks, molten salt pumps, and molten salt-working fluid heat exchangers. Turbine generator units include turbines, generators, and related auxiliary systems. In pure thermal energy storage power generation mode, during the energy storage phase, the molten salt energy storage unit can use electrical or solar energy to heat cold molten salt into high-temperature molten salt, which is then stored in a hot salt tank. During the discharge phase, the hot molten salt is pumped to the heat exchanger, transferring its heat energy to the working fluid (such as water / steam or air). The high-temperature, high-pressure working fluid obtained from the molten salt heat exchanger drives the turbine to rotate, thereby driving the generator to generate electricity. This effectively solves the problems of high volatility and difficulty in direct grid connection of renewable energy sources.
[0027] However, the system faces multiple core risks and technical challenges during operation: molten salt energy storage systems are susceptible to risks such as molten salt solidification and equipment corrosion, while turbine power generation systems are vulnerable to faults such as surge, overspeed, and blade erosion. These factors are crucial to the system's lifespan and operational safety.
[0028] At the technical implementation level, existing methods face four major constraints: First, molten salt energy storage systems are massive in scale (e.g., molten salt tanks in GW-level power plants can reach tens of thousands of tons), requiring the construction of full-size or half-size platforms for physical experiments. The material consumption, energy consumption, and time costs of a single experiment (such as testing the impact of molten salt flow / temperature regulation on turbine efficiency) are extremely high, often taking months to years. Second, physical experiments cannot effectively simulate extreme operating conditions (such as extreme low temperatures causing molten salt solidification and turbine overspeed) or complex dynamic processes, making it difficult to predict the system's performance under actual abnormal operating conditions. Third, traditional simulation models often rely on simplified assumptions (such as treating turbine efficiency as a constant value), which deviate significantly from the physical characteristics of the actual system and cannot accurately reflect the true dynamic laws of the "molten salt-turbine" coupled system. Fourth, due to the inherent strong coupling characteristics of the system, changes in the temperature and flow rate of the molten salt will directly affect the steam parameters (including pressure, temperature, and humidity) at the heat exchanger outlet, while changes in the turbine speed and valve opening will in turn affect the energy demand of the molten salt system (for example, when the turbine is deloaded, the molten salt flow rate needs to be reduced simultaneously, otherwise the molten salt will overheat in the storage tank).
[0029] Traditional control strategies primarily rely on empirical curves and controllers with fixed parameters, using real-time monitoring of system parameter changes for feedback adjustment. However, this approach struggles to handle the nonlinear characteristics and significant thermal inertia delays of molten salt energy storage turbine power generation systems. Furthermore, existing monitoring systems largely depend on sensor feedback data, making it impossible to predict parameter change trends in advance. For instance, a sudden drop in turbine inlet air temperature can lead to abnormal operating conditions such as excessive turbine vibration due to valve lag.
[0030] Therefore, in the actual operation of traditional technologies, molten salt energy storage turbine power generation systems often experience instability. This makes it difficult for the system to accurately respond to actual load changes, resulting in low operating efficiency and increased safety risks. These technical bottlenecks severely restrict the large-scale application and benefit realization of molten salt energy storage turbine power generation systems in the energy sector.
[0031] In one exemplary embodiment, such as Figure 1As shown, a monitoring method for a molten salt energy storage turbine power generation system is provided. This embodiment illustrates the application of this method to a terminal, where the terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can be smart speakers, smart TVs, smart air conditioners, smart vehicle devices, projection devices, etc. Portable wearable devices can be smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. It is understood that this method can also be applied to a server, and can also be applied to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps 102-112. Wherein:
[0032] Step 102: Obtain the simulation model, current operating data, and current load data of the molten salt energy storage turbine power generation system.
[0033] The current operating data refers to the operating data of the molten salt energy storage turbine power generation system within the current time period. The division of the time period can be determined according to actual needs and test results, etc., and this embodiment does not impose any restrictions on it.
[0034] The operating data of the molten salt energy storage turbine power generation system may include at least one of the following: operating condition data of the molten salt energy storage system, operating condition data of the turbine power generation system, and static attributes and dynamic operating parameters of the links between the molten salt energy storage system and the turbine power generation system.
[0035] The operating data of a molten salt energy storage system may include at least one of the following: mechanical parameters, flow and pressure parameters, and properties of the molten salt medium.
[0036] Mechanical parameters, primarily used to reflect the operating status and mechanical integrity of critical rotating and stationary equipment, may include: pump unit operating status parameters, such as valve opening, operating frequency, motor operating current, and bearing temperature of molten salt circulation pumps and feedwater pumps; critical interface temperatures of storage tanks, such as the outlet temperature of high-temperature molten salt tanks and the inlet temperature of low-temperature molten salt tanks; molten salt side performance monitoring parameters of heat exchangers, namely their molten salt inlet and outlet temperatures, used to directly calculate heat exchange and evaluate heat transfer efficiency; and pipeline system safety monitoring data, referring to multiple temperature points along molten salt pipelines, especially at critical locations such as bends and valves, used to monitor local overheating or solidification risks.
[0037] Flow and pressure parameters are mainly used to characterize the system's working fluid transport capacity, energy input rate, and energy storage status. They can include: core parameters that determine the system's energy transport, namely the outlet flow rate and pipeline pressure of the molten salt circulation pump, which together determine the heat input rate entering the heat exchanger; and parameters that reflect the system's real-time energy storage capacity, namely the liquid level of the high-temperature and low-temperature molten salt tanks, whose changes are directly related to the system's available energy and affect the turbine unit's sustainable operating time.
[0038] Molten salt medium properties refer to the physicochemical characteristics of the molten salt working medium itself. These properties change during operation and have a reciprocal effect on system performance. They mainly include the molten salt viscosity, thermal conductivity and impurity content monitored in real time.
[0039] Operating data of a turbine power generation system may include at least one of the following: steam parameters, turbine mechanical parameters, and power generation performance parameters.
[0040] Steam parameters are mainly used to characterize the working fluid state and energy grade, and are the basis for calculating the turbine's work capacity. They can include: steam temperature and steam pressure, which determine the turbine's enthalpy drop and output power; and exhaust pressure, which reflects the performance of the condenser's cold end and is directly related to the condenser's vacuum level.
[0041] Turbine mechanical parameters comprehensively reflect the unit's mechanical status, operating efficiency, and safety information, and may include: core operating status parameters, such as turbine speed and intake valve opening; mechanical safety monitoring data, such as shaft vibration signals, rotor eccentricity, and bearing lubricating oil temperature; thermal data, such as combustion chamber temperature, fuel characteristic data (including fuel calorific value and gas composition), and other thermal data (e.g., temperature sequences of combustion chamber, bearing, and exhaust gas measured at a frequency of 1 Hz, as well as pressure and flow signals such as steam pressure and lubricating oil flow that change relatively slowly); and load-related data, such as generator active power and grid frequency that reflect current load demand and grid status.
[0042] The shaft vibration signals can be divided into two categories: one is the vibration data transmitted from the outer ring of the bearing to the housing, collected by an accelerometer installed in the bearing housing bolt holes; the other is the blade and gear vibration signals, measured by an eddy current probe, reflecting the radial displacement of the rotor relative to the housing. The rotor eccentricity is also measured by a non-contact eddy current probe, which is typically one to two millimeters from the shaft surface, to monitor the radial displacement of the rotor relative to the housing.
[0043] Power generation performance parameters directly reflect the final output quality and energy conversion economy of the system, and may include: output power, voltage and current that characterize the degree of matching with grid dispatch; and heat rate that measures the efficiency of the turbine in converting heat energy into electrical energy.
[0044] The operation monitoring of turbine power generation systems can also be expanded to include the collection of auxiliary system parameters and external interference data as needed.
[0045] Auxiliary system parameters are used to ensure the stable operation of the main system and environmental safety. These parameters may include: turbine condenser parameters, such as circulating water flow rate, circulating water inlet and outlet temperatures, and system vacuum. These parameters collectively reflect the heat exchange efficiency and operating status of the condenser; feedwater system parameters, such as feedwater pump outlet flow rate, feedwater pressure, and feedwater temperature, which are crucial for maintaining the water and heat balance of the boiler or heat exchanger; and nitrogen sealing system parameters. The nitrogen sealing system specifically refers to the nitrogen system used for the protection of molten salt storage tanks, and its core parameters are the sealing nitrogen pressure and nitrogen humidity.
[0046] External interference data refers to environmental and power grid conditions from outside the system that directly affect its operation. These can include: environmental parameters, such as ambient temperature, ambient humidity, and ambient wind speed, which affect equipment heat dissipation and operating efficiency; and power grid parameters, mainly referring to the power grid frequency, whose fluctuations directly affect the speed control of the turbine generator.
[0047] Current load data refers to the actual electrical power output of a molten salt energy storage turbine power generation system to the grid within the current time period. As an example, current load data can be a power time series composed of multiple power values within the current time period, arranged in chronological order. As another example, current load data can also be a power curve composed of multiple power values within the current time period, arranged in chronological order.
[0048] For example, a pre-built simulation model capable of accurately reflecting the coupling dynamics of molten salt energy storage and turbine power generation is called from a model library. Simultaneously, current load data is collected in real time by a power measurement device deployed at the turbine power generation system outlet, and current operating data collected by at least one of the distributed control system and sensors is obtained through a data acquisition and monitoring control system.
[0049] In some feasible implementations, a simulation model of the molten salt energy storage turbine power generation system can be built in advance before monitoring the molten salt energy storage turbine power generation system.
[0050] In some feasible implementations, the simulation model of the molten salt energy storage turbine power generation system can be a digital twin model.
[0051] In some feasible implementations, the simulation model of the molten salt energy storage turbine power generation system may include a geometric model, a molten salt-steam heat transfer model, a turbine coupling model, and a control model, ultimately forming a multi-domain coupling model.
[0052] Among them, the geometric model can be a three-dimensional geometric model. The three-dimensional assemblic model can be constructed by computer-aided design software. The three-dimensional geometric model can include key components such as molten salt storage tank and its internal flow guiding structure, heat exchanger tube bundle arrangement, turbine stage interstage blades and shaft system, etc., to provide accurate geometric boundaries for subsequent flow field and physical process simulation.
[0053] For the molten salt-steam heat transfer model, computational fluid dynamics can be used to establish a heat transfer model between molten salt and steam, simulate the flow and heat exchange process of molten salt in the heat exchanger, quantitatively describe the influence of molten salt flow rate changes on steam parameters such as steam inlet temperature, steam inlet pressure and steam flow rate, and output the correspondence between molten salt temperature drop and steam parameters.
[0054] For turbine generator sets, a multiphysics coupled model integrating electrical, mechanical, and thermodynamic aspects is established. The electrical part can be described using Park's equations to describe the generator's electromagnetic transient process, where changes in the d-axis and q-axis flux linkages are jointly influenced by terminal voltage, resistance voltage drop, and cross-coupled potentials caused by rotor rotation. The mechanical part can be characterized by rotor dynamics equations to represent the dynamic balance between rotor speed changes and turbine mechanical torque, generator electromagnetic torque, and mechanical damping. The thermodynamic part can be based on stage variable operating condition theory to establish the functional relationship between turbine output power and steam flow rate, interstage enthalpy drop, and stage efficiency. By calculating state variables such as steam pressure and inlet / outlet enthalpy, the model reflects the entire process by which molten salt flow fluctuations are transmitted to steam parameters through the heat exchanger, thus affecting turbine output.
[0055] Single-physics models often have limitations. For example, using only the Park equations ignores the influence of thermal inertia on power response, and using only thermodynamic models cannot reflect electromagnetic transients during grid faults. Thermoelectric coupling manifests as changes in turbine enthalpy drop caused by variations in fuel characteristics or operating conditions, thus affecting mechanical torque and electrical power output. Electromechanical coupling manifests as changes in electromagnetic torque caused by sudden changes in grid load, which are fed back to the speed control system via rotor speed, driving fuel or steam regulation to achieve torque rebalancing. Coupled modeling can effectively overcome the limitations of single-physics models, capturing both the delayed effect of thermal inertia on power response and accurately simulating electromagnetic transients under grid faults, thereby providing a high-confidence simulation environment for the verification and optimization of system control strategies.
[0056] The control model aims to accurately simulate the response logic and behavior of the actual control unit in the system. This model can deeply integrate the basic control logic of a programmable logic controller (PLC) with advanced optimization algorithms, such as model predictive control (MMCC), to simulate the impact of actuator actions, such as valve opening adjustments or molten salt pump frequency changes, on key system process variables. These process variables include molten salt circulation flow rate, steam pressure and temperature, and ultimately, turbine output power. Molten salt energy storage turbine power generation systems contain multiple interconnected control loops, typically including molten salt flow control to maintain a set flow rate, turbine speed control to ensure stable unit operation, and generator power output control to respond to grid dispatch commands. Significant coupling exists between these loops; for example, adjustments to power output directly affect turbine speed, potentially triggering a reset of the molten salt flow rate. The core function of this control model is to support the dynamic verification of the complete closed loop of "load change - parameter response - control action." For instance, it can quantitatively assess the overshoot of turbine speed and the recovery time required for the system to return to a stable state when the grid load command suddenly drops by 10%.
[0057] Ultimately, the aforementioned multiple models can be coupled to form a multi-domain coupled model, realizing full-chain cross-domain dynamic coupling from molten salt fluid flow, heat transfer through heat exchangers, steam thermodynamic cycle, driving turbine mechanical motion, and finally power output. Specifically, fluctuations in the flow rate in the molten salt loop alter the heat transfer efficiency and energy balance within the heat exchanger, leading to changes in parameters such as steam pressure and temperature. These steam parameter changes directly affect the turbine's enthalpy drop and output mechanical power. Fluctuations in turbine power trigger changes in the generator's electromagnetic power and rotor speed; these changes in electromechanical state are reflected in the form of grid frequency. Frequency changes serve as feedback signals, captured by the control system, triggering recalculation and adjustment of the control strategy. Finally, the new control strategy generates correction commands, such as adjusting the opening of the molten salt flow regulating valve, thus forming a complete closed-loop dynamic process spanning fluid, thermodynamic, mechanical, and electrical domains. Through multi-domain coupling, the simulation can realistically reflect how a local disturbance or operation propagates across various physical domains and ultimately affects the behavior of the entire system.
[0058] Step 104: Using a preset load prediction model, based on current operating data and current load data, predict the load demand of the molten salt energy storage turbine power generation system in a preset future time period, obtain the load demand prediction result, and query the control setpoint corresponding to the load demand prediction result. The preset load prediction model is obtained by training the spatiotemporal attention network in advance based on the historical operating data and historical load data of the molten salt energy storage turbine power generation system.
[0059] Among them, the load forecasting model can refer to a pre-trained machine learning model used to predict the electricity load demand within a specific future time period.
[0060] The load forecasting model employs a spatiotemporal attention network, which is pre-trained using historical operating and load data from the molten salt energy storage turbine power generation system. The specific training method is similar to related technologies and will not be elaborated upon here. The spatiotemporal attention network is a special type of deep learning model that can simultaneously capture the dependencies between data in both temporal and spatial dimensions. For example, when forecasting load, the spatiotemporal attention network can simultaneously focus on "the load at the same time yesterday" (temporal attention) and "the current state of another related power station" (spatial attention).
[0061] The control setpoint refers to the target or reference value that the simulation model of a molten salt energy storage turbine power generation system needs to achieve and maintain under a specific load. It can include at least one of the following: valve setpoint, flow setpoint, power setpoint, temperature setpoint, speed setpoint, and pressure setpoint. The control setpoint serves as the input reference for the near-end strategy optimization controller, guiding the actions of actuators (such as valves and fuel injectors) in the molten salt energy storage turbine power generation system to stabilize actual operating parameters (such as power, temperature, and pressure) within the desired range.
[0062] The correspondence between load and control setpoint can be established and stored in advance based on actual needs and test results. In practical applications, the control setpoint corresponding to the load demand forecast result can be queried by directly calling this correspondence.
[0063] For example, the acquired current operating data and current load data are fed as input sequences into a pre-trained load forecasting model to calculate the load demand forecast for a preset future time period. Subsequently, based on the load demand forecast, the correspondence between preset loads and control setpoints is queried to obtain a series of control setpoints required to achieve the load demand forecast result.
[0064] Step 106: The near-end strategy optimization controller generates a target power output control strategy that enables the molten salt energy storage turbine power generation system to achieve the preset optimization target based on the load demand forecast results and control setpoint.
[0065] Among them, the proximal strategy optimization controller is a reinforcement learning algorithm used to learn and find the action strategy that yields the greatest long-term reward in a given environment (such as a power generation system). For example, the proximal strategy optimization controller can learn the optimal regulation strategy by continuously trying different valve opening commands and based on the feedback from the resulting power generation efficiency and safety.
[0066] The preset optimization objective can refer to the performance indicators that the molten salt energy storage turbine power generation system is expected to achieve, which are pre-set during the training phase of the near-end strategy optimization controller. The preset optimization objective may include at least one of maximizing thermoelectric conversion efficiency, minimizing equipment thermal stress, or extending equipment lifespan. The reward function and optimization objective in the near-end strategy optimization controller can be determined based on actual needs and test results, and this embodiment does not impose any restrictions on them.
[0067] In some feasible implementations, the controller's inputs include target guidance and system status.
[0068] The target guidance includes: load demand forecast results, that is, the target power that the system needs to achieve in a preset future time period, which is the ultimate goal that the PPO controller needs to track; and control setpoints, which are the safe operating ranges of key parameters obtained by the system that match the above-mentioned predicted load. These setpoints define a safe and efficient path for the control strategy.
[0069] System status includes: grid status parameters, such as real-time grid frequency, used to evaluate the effectiveness of the strategy in supporting the grid; steam and molten salt side status parameters, such as real-time main steam pressure, real-time molten salt flow rate and temperature, reflecting the current thermodynamic state of the system; and turbine side status parameters, such as real-time turbine output power, used to calculate the tracking error with the target power.
[0070] The core design idea of the reward function is to optimally track the load demand forecast result while satisfying the control setpoint constraints, which can be expressed as:
[0071]
[0072] in, Let be the total reward value at time t. The optimization objective of the PPO algorithm can be to maximize the cumulative total reward. ; , , , As the weight of the core reward items, satisfying the following conditions. > It can be determined through offline simulation; To reward grid frequency stability, while tracking the target power, the reward controller generates strategies that help stabilize the grid frequency. As a power point tracking (PPO) reward, it directly rewards the accurate tracking of actual power output to load demand forecast results. This requires the PPO controller to converge its strategy in a direction that satisfies the load demand forecast results. As a reward for system energy efficiency, it encourages the generation of power control strategies that enable the system to operate in the high-efficiency region; To control the reward smoothness of the strategy, it encourages gentle and stable power regulation; Penalty for exceeding the molten salt temperature limit; This is a penalty item for exceeding steam pressure limits; This is a penalty item for exceeding the molten salt flow rate limit.
[0073] Penalties are triggered when safety constraints are violated. Safety constraints are determined based on control setpoints.
[0074] As an example, the control setpoint may include the molten salt safety temperature boundary. The safety constraints of the control strategy are to prohibit the control strategy from causing the molten salt to overheat (e.g., the molten salt flow rate is too small, causing the molten salt to stay in the heat exchanger for too long, resulting in an increase in the molten salt temperature) or to be too cold (e.g., the molten salt flow rate is too large, causing insufficient heat exchange, resulting in an excessively low molten salt temperature). If the limit is exceeded, the total reward decreases, and the strategy is penalized.
[0075] As an example, the control setpoint may include the steam pressure boundary. The safety constraint of the control strategy is to avoid excessive flow fluctuations that could lead to a sudden increase in steam pressure (e.g., excessive molten salt flow leading to a surge in heat transfer, resulting in pressure exceeding limits) or a sudden decrease (e.g., insufficient molten salt flow leading to insufficient heat transfer, resulting in excessively low pressure). If the limit is exceeded, the total reward decreases, and the strategy is penalized.
[0076] As an example, the control setpoint may include the molten salt flow boundary. The safety constraint of the control strategy is to prohibit the flow rate from exceeding the equipment design range (such as the maximum output flow rate of the molten salt pump). If the limit is exceeded, the total reward decreases, and the strategy is penalized.
[0077] The target power output control strategy can be a set of detailed control command sequences, including at least one of the following: molten salt pump speed adjustment plan, timing of opening changes of various regulating valves, turbine guide vane angle adjustment scheme, etc., forming a complete system control trajectory.
[0078] For example, the near-end policy optimization controller is used as the decision-making brain, and its input state is the predicted load demand and the corresponding control setpoint. Based on its internally trained policy network, the near-end policy optimization controller outputs a specific sequence of control commands, namely the target power output control strategy. The generation process of this target power output control strategy is guided by a built-in reward function, aiming to enable the system to achieve a preset optimization target when responding to this load demand.
[0079] The proximal policy optimization controller was trained using reinforcement learning in a simulation environment before application, enabling it to generate control policies that balance performance optimization and constraint satisfaction. Those skilled in the art can use open-source reinforcement learning frameworks to train the proximal policy optimization controller and integrate it into the system control architecture.
[0080] Step 108: Using a simulation model, the operating status of the molten salt energy storage turbine power generation system is simulated according to the target power output control strategy to obtain operating status simulation data. The operating status simulation data is then verified for safety based on preset unit operating constraints.
[0081] Among them, operational simulation data refers to the data set generated by simulating the dynamic response of a molten salt energy storage turbine power generation system under a specific control strategy through a simulation model, including the predicted values of various key parameters of the molten salt energy storage turbine power generation system over time. For example, in a molten salt-turbine system, operational simulation data can include time series data of key operational parameters such as molten salt temperature field distribution, steam parameter changes, turbine speed fluctuations, and pressure changes at various levels.
[0082] Preset unit operating constraints can refer to the set of boundary conditions for the safe operation of a molten salt energy storage turbine power generation system, which may include at least one of the following: equipment physical limits, operating procedure requirements, and safety protection thresholds. For example, the minimum flow temperature of molten salt, the maximum allowable vibration value of the turbine, and valve regulation rate limits.
[0083] For example, in pure thermal energy storage power generation mode, after obtaining the target power output control strategy, this can be used as input to the simulation model to drive its operation, simulating the complete dynamic process of the molten salt energy storage turbine power generation system following the target power output control strategy within a preset future time period. After the simulation, the operating state simulation data is extracted, including the time series of parameters such as temperature, pressure, flow rate, vibration, and stress at key nodes. Subsequently, this operating state simulation data is compared with preset unit operating constraints for safety verification: checking whether all operating state simulation data meet safety boundary requirements, such as whether the molten salt temperature is higher than the freezing point, whether the turbine vibration is lower than the allowable threshold, and whether the pressure change rate is within the equipment's tolerance range. Safety verification can use a preset rule base and threshold table to automatically evaluate the simulation results.
[0084] Step 110: If the safety verification passes, configure the molten salt energy storage turbine power generation system according to the target power output control strategy and control setpoint.
[0085] For example, if the safety verification confirms that all simulation data of the operating status meet the preset unit operating constraints, the target power output control strategy and the corresponding control setpoint can be converted into specific equipment control commands. These commands are then sent to the actuators of the molten salt energy storage turbine power generation system, including the molten salt pump frequency converter, actuators of various regulating valves, and the turbine control system, through the industrial control system. These commands precisely guide the operation of the molten salt energy storage turbine power generation system according to preset timing and parameters, achieving accurate response to load demand forecast results. The transmission and execution of control commands follow industrial standard communication protocols to ensure real-time performance and reliability.
[0086] Step 112: If the safety check fails, the target power output control strategy is adjusted according to feedback until the safety check passes.
[0087] For example, if the safety verification finds that at least some of the simulated operating data violates the preset unit operating constraints, the system does not execute the current target power output control strategy. Instead, it activates a feedback adjustment mechanism to automatically adjust the key parameters in the target power output control strategy and generate a corrected target power output control strategy. Subsequently, the system repeats the above simulation and safety verification process to evaluate the safety of the corrected target power output control strategy. This process is executed iteratively until the generated target power output control strategy can simultaneously meet the load demand and all safety constraints. The feedback adjustment can employ rule-based adjustment algorithms or optimization search methods to ensure the safety and effectiveness of the final strategy. The specific choice can be made based on actual needs and test results, and this embodiment does not impose any limitations on this.
[0088] In some feasible implementations, in molten salt turbine power generation systems, the control strategies generated by the PPO (Proximal Policy Optimization) controller must be subject to strict safety constraints to ensure that parameters such as valve opening, steam temperature, exhaust temperature, speed, vibration, and blade damage do not exceed safety thresholds during dynamic adjustments. This ensures that control actions do not lead to mechanical resonance, thermal stress fatigue, sensor malfunctions, or partial equipment failures, and that the unit can still operate safely.
[0089] The safety constraints embedded in the PPO controller can be expressed as follows:
[0090] a safe =clip(a PPO ,a min (s),a max (s)
[0091] Among them, a safeFor control actions after safety constraints have been applied; a PPO The original action generated by PPO; clip(·) is the clipping function, which forces a PPO The final value falls within [a] min (s),a max Within the interval [s], if a PPO If the action exceeds this safe range, it will be "trimmed" to the boundary value; amin(s) and amax(s) are the safe action boundary limits under the current operating state s, where the boundary limits can be determined based on real-time operating conditions by looking up a table. This safety constraint restricts the obtained actions to the safe range under the current state to avoid dangerous operations. For example, in afterburning or auxiliary heating mode, if the control strategy generated by the PPO controller is the fuel valve opening a PPO =70%, but the current exhaust temperature is too high. The allowable fuel valve opening range is [50%, 65%]. Therefore, the final fuel valve opening a safe =clip(70%,50%,65%)=65%, which is fed back to the PPO controller.
[0092] The aforementioned monitoring method for molten salt energy storage turbine power generation systems first utilizes a load forecasting model trained on a spatiotemporal attention network. By learning from historical operating and load data, it predicts the load demand for a preset future time period and matches the corresponding control setpoint. This overcomes the limitation of traditional PID control, which can only respond to the current state, and achieves proactive perception of load changes. Next, a near-end strategy optimization controller autonomously seeks the optimal target power output control strategy based on the load demand forecast, rather than relying on fixed empirical parameters. This overcomes the insufficient adaptability of traditional PID control in nonlinear and hysteresis systems, significantly improving the dynamic matching accuracy between the molten salt energy storage unit and the turbine unit. Furthermore, a simulation model is used to simulate the operating state of the molten salt energy storage turbine power generation system under the target power output control strategy. This identifies potential risk points that may cause abnormal operating conditions such as excessive turbine vibration, achieving a pre-emptive safety verification of the target power output control strategy. This ensures that the target power output control strategy meets load demand while guaranteeing the operational safety of the molten salt energy storage turbine power generation system. This pre-emptive verification mechanism effectively avoids efficiency losses and equipment damage caused by improper control. In this way, by organically combining the three links of load forecasting, strategy optimization and safety verification, a technological leap from passive response to active control is achieved, enabling the system to maintain a better operating state while accurately tracking load demand, thereby significantly improving the overall operating efficiency of the molten salt energy storage turbine power generation system.
[0093] In one exemplary embodiment, such as Figure 2As shown, the current operating data includes vibration signals and thermal data; after configuring the molten salt energy storage turbine power generation system according to the target power output control strategy and control setpoint, the method further includes steps 202 to 206. Wherein:
[0094] Step 202: Obtain the physical structure adjacency matrix of the molten salt energy storage turbine power generation system.
[0095] It should be noted that there are multiple core risks and technical challenges in the operation of molten salt energy storage turbine power generation systems: molten salt energy storage systems face risks such as molten salt solidification and equipment corrosion, while turbine power generation systems are susceptible to faults such as surge, overspeed, and blade erosion. These factors are crucial to the system's lifespan and operational safety.
[0096] However, molten salt energy storage systems are large in scale (such as molten salt storage tanks in GW-level power plants that can reach tens of thousands of tons). Physical tests require the construction of full-size or half-size platforms. The material consumption, energy consumption, and time costs of a single test (such as testing the impact of molten salt flow / temperature regulation on turbine efficiency) are extremely high, often taking several months to several years.
[0097] The physical adjacency matrix refers to a mathematical matrix that represents the physical connections between components in a molten salt energy storage turbine power generation system. The matrix element values reflect the connection strength or type. For example, the connection between the turbine rotor and various bearings and blades can be represented by non-zero elements in the adjacency matrix. Rows and columns in the matrix correspond to different components, and element values reflect the tightness of the connection. The physical adjacency matrix mathematically and precisely describes the connection topology between key components such as the molten salt pump and drive motor, the turbine rotor and support bearings, and the gearbox and coupling, providing a structural basis for subsequent analysis of fault propagation paths.
[0098] For example, a three-dimensional structural model in a pre-built digital structural model or simulation model of a molten salt energy storage turbine power generation system can be invoked to extract the physical connection relationships between the main components in the molten salt energy storage turbine power generation system and generate a physical structure adjacency matrix.
[0099] Step 204: Obtain subband energy features by wavelet packet decomposition of vibration signal, extract envelope spectrum kurtosis features of vibration signal, and generate thermal field spatial gradient matrix based on thermal data and physical structure adjacency matrix.
[0100] Vibration signals can refer to data collected by vibration measurement devices such as accelerometers, reflecting the operating status of mechanical components in a molten salt energy storage turbine power generation system. For example, acceleration signals collected by vibration sensors installed on turbine bearing housings are used to analyze mechanical faults such as rotor imbalance and misalignment.
[0101] Thermal data refers to data collected by sensors such as temperature and pressure that reflects the thermal process state of a molten salt energy storage turbine power generation system. For example, parameters such as the inlet and outlet temperatures of the molten salt tank, steam pressure, and turbine exhaust temperature are used to analyze the system's thermal performance and fault conditions.
[0102] Wavelet packet decomposition is a signal processing method that simultaneously subdivides a signal into low-frequency and high-frequency bands to identify fault characteristics in different frequency ranges. For example, a vibration signal can be decomposed into multiple sub-signals with different frequency bands, enabling fine extraction of the signal's frequency domain features.
[0103] Subband energy characteristics refer to the energy distribution features of a signal within each frequency band. They can be used to distinguish early wear from normal noise, thereby accurately identifying impact faults. For example, the frequency band energy distribution reflecting the fault characteristics can be obtained by calculating the energy values of the signals in each frequency band after wavelet packet decomposition.
[0104] Envelope spectrum kurtosis is an indicator that characterizes the intensity of the impact component in the signal envelope spectrum. High kurtosis values typically indicate the presence of periodic impacts and are an important indicator of early failures. For example, weak impact signals generated by early wear of the bearing inner ring will exhibit high kurtosis values in the envelope spectrum, even if the original vibration amplitude is still within the normal range. Envelope spectrum kurtosis can enhance the significance of weak impact features, preserve local defect features in the original waveform, and can be used in models to identify blade crack features.
[0105] The spatial gradient matrix of a thermal field can be a mathematical representation of the rate of change of the temperature field in space, used to characterize the propagation path of local hot spots or overpressure points. For example, by using data from multiple temperature measuring points, the temperature gradient between adjacent measuring points can be calculated to form a matrix reflecting the anomalous distribution of the thermal field.
[0106] For example, vibration and thermal data are first extracted from the current operating data. Then, wavelet packet decomposition is performed on the vibration signal to divide it into different frequency bands, and the energy value of each band is calculated, thus obtaining sub-band energy characteristics reflecting the frequency domain distribution of vibration energy. Simultaneously, envelope analysis is performed on the vibration signal to calculate the kurtosis value of its envelope spectrum, obtaining the envelope spectrum kurtosis characteristic, which is sensitive to early impact failures. Furthermore, based on the thermal data and combined with the spatial relationships of components described by the physical structure adjacency matrix, the temperature change rate between adjacent components or measuring points is calculated, constructing a thermal field spatial gradient matrix reflecting the abnormal distribution of the system's thermal field.
[0107] Step 206: Using a preset fault prediction model, fault prediction is performed on the molten salt energy storage turbine power generation system based on current operating data, sub-band energy characteristics, envelope spectrum kurtosis characteristics, and thermal field spatial gradient matrix. The fault prediction results include the fault probability and fault type of the molten salt energy storage turbine power generation system in a preset future time period. The preset fault prediction model is trained based on the fault sample data of the molten salt energy storage turbine power generation system.
[0108] Among them, the fault prediction model can refer to the computational model that predicts the probability and type of future equipment failures by constructing a machine learning algorithm.
[0109] Fault sample data can include datasets of equipment operating under normal and various fault conditions, which can be used to train fault prediction models.
[0110] As an example, fault sample data can be obtained by collecting historical operating data of molten salt energy storage turbine power generation systems.
[0111] As another example, fault sample data can be obtained by simulating different faults.
[0112] For example, the extracted sub-band energy features, envelope spectrum kurtosis features, and thermal field spatial gradient matrix, together with the directly collected current operating data, are used as inputs, concatenated into a feature vector, and fed into a pre-trained fault prediction model to output fault prediction results, including the types of faults that the molten salt energy storage turbine power generation system may experience in a preset future time period, as well as the probability of occurrence of each type of fault.
[0113] In this embodiment, wavelet packet decomposition and envelope spectrum kurtosis analysis can extract early fault features from real-time vibration signals, enabling accurate identification of progressive faults such as turbine blade erosion and bearing wear, avoiding the delayed diagnostic problems of traditional methods that rely on physical experiments. Simultaneously, by fusing thermal data with the adjacency matrix of the physical structure, a digital mapping of the propagation path of system thermal anomalies can be established. Furthermore, by using a predictive model trained on historical fault samples and fusing multi-source features, early warning of system fault risks is achieved. This predictive maintenance strategy based on digital features can achieve accurate assessment of system status without large-scale physical experiments, significantly reducing the material, energy, and time costs required for physical experiments, and providing an economical and effective technical guarantee for the safe operation of the system.
[0114] In an exemplary embodiment, the current operating data includes time series of sensor data collected by multiple sensors; the preset fault prediction model includes a fault diagnosis model, an operating condition prediction model, a thermal anomaly detection model, and a multimodal fusion layer; using the preset fault prediction model, based on the current operating data, sub-band energy characteristics, envelope spectrum kurtosis characteristics, and thermal field spatial gradient matrix, fault prediction is performed on the molten salt energy storage turbine power generation system to obtain fault prediction results, including:
[0115] The fault diagnosis model identifies the local target fault type based on subband energy characteristics and envelope spectrum kurtosis characteristics.
[0116] Based on the time series of various sensor data, the operating condition prediction model is used to predict the target operating condition type of the molten salt energy storage turbine power generation system in a preset future time period.
[0117] Based on the thermal field spatial gradient matrix, the target system-level thermal anomaly region in the molten salt energy storage turbine power generation system is identified by the thermal anomaly detection model, and the thermal anomaly propagation path corresponding to the target system-level thermal anomaly region is predicted.
[0118] By using a multimodal fusion layer, a cross-force attention mechanism is employed to fuse the target's local fault type, target operating condition type, target system-level thermal anomaly region, and thermal anomaly propagation path to obtain fault prediction results.
[0119] It should be noted that analysis based on data from a single type of sensor is insufficient for effectively identifying early-stage complex faults. Due to the complex system structure and strong coupling between components, local faults often propagate in multiple physical fields and exhibit different characteristics. This results in shortcomings in the accuracy and timeliness of traditional methods for fault prediction, making it impossible to achieve early warning and precise fault location.
[0120] Among them, the fault diagnosis model can refer to the algorithm model used to identify the fault type of a specific component of the equipment.
[0121] Operating condition prediction model refers to an algorithmic model that predicts the future operating state of a molten salt energy storage turbine power generation system.
[0122] A thermal anomaly detection model can refer to an algorithmic model that identifies the distribution area of abnormal thermal parameters of equipment and the propagation path of thermal anomalies.
[0123] A multimodal fusion layer can refer to a computational layer that integrates features of different types of data to coordinate information from multiple sources.
[0124] Cross-force attention mechanism can refer to a multimodal feature fusion algorithm that integrates information by calculating the correlation weights between features of different modalities.
[0125] The target local target failure type can refer to a specific failure mode of a particular component or location. For example, specific failure types such as turbine first-stage blade cracks, molten salt pump bearing wear, or seal failure.
[0126] The target operating condition type can refer to the category of operating states that the system will enter in the future. For example, high load stable operating condition, load rapidly rising operating condition, or start-stop transition operating condition, etc.
[0127] A target system-level thermal anomaly region can refer to an area of abnormal temperature or heat flux distribution identified at the system scale. For example, a localized overheated area on the inner wall of a turbine cylinder or an area with insufficient cooling in molten salt pipes.
[0128] The propagation path of thermal anomalies can refer to the direction and path of thermal anomalies spreading or propagating in a molten salt energy storage turbine power generation system. For example, the heat conduction path of heat spreading from a locally overheated area of a turbine blade to an adjacent area.
[0129] For example, anomaly detection can be performed in parallel using a fault diagnosis model, a condition prediction model, and a thermal anomaly detection model. The fault diagnosis model receives sub-band energy features obtained through wavelet packet decomposition and kurtosis features obtained through envelope spectrum analysis as input. Through training, the model learns the mapping relationship between vibration features and fault types, and outputs the identification result of specific component fault types, such as bearing wear, gear pitting, or blade cracks. The condition prediction model receives time series data from sensors as input. By analyzing the trends and periodic patterns in the time series data, the model predicts the type of operating condition the system may be in within a preset time period, such as typical operating states like startup, steady-state operation, load changes, or shutdown. The thermal anomaly detection model can receive the spatial gradient matrix of the thermal field as input. By analyzing the spatial distribution characteristics of the temperature field, the thermal anomaly detection model can identify the thermal anomaly region in the molten salt energy storage turbine power generation system. At the same time, based on the physical structure relationship and heat conduction characteristics of the molten salt energy storage turbine power generation system represented by the spatial gradient matrix of the thermal field, the model can predict the propagation path and development trend of thermal anomalies in the molten salt energy storage turbine power generation system.
[0130] The outputs of the fault diagnosis model, the operating condition prediction model, and the thermal anomaly detection model are all input into the multimodal fusion layer. This multimodal fusion layer uses a cross-force attention mechanism to generate a comprehensive fault prediction result by weighted fusion of the interrelationships and importance weights between different modal features learned during training. The fault prediction result may include at least one of the following: fault type, probability of occurrence of each fault type, possible impact range, and severity assessment.
[0131] In some feasible implementations, the fault diagnosis model can use the 1D-ResNet (1D Residual Network) model; the operating condition prediction model can use the TST (Time Series Transformer) model; and the thermal anomaly detection model can use the GAT (Graph Attention Network) model.
[0132] In this context, by performing feature extraction on vibration signals through refined feature engineering, not only is the original waveform required by 1D-ResNet preserved, but statistical features are also extracted for use by the TST model. The system can transform the original sensor data into high-value features suitable for different model processing, ultimately achieving precise load control and early fault warning.
[0133] In an exemplary embodiment, in the case of gas turbine bearing overheating early warning, the 1D-ResNet model detected a sudden increase in energy in the bearing vibration signal at the 6kHz frequency band (an early wear characteristic). The TimeSeriesTransformer model analyzed historical temperature curves and predicted that the bearing temperature would exceed the threshold within the next 10 minutes. The GAT model found that the abnormal bearing temperature had propagated to adjacent gearboxes through the lubrication circuit. The multimodal fusion layer calculated a comprehensive confidence level of 92%, triggering a level-two early warning; the control system automatically reduced the load to 75% to avoid overheating damage.
[0134] This embodiment utilizes the collaborative work of three models to enable the system to capture both microscopic mechanical defects and macroscopic thermodynamic state evolution, ultimately achieving comprehensive health management of molten salt turbine generator sets. In actual deployment, the model structure and input characteristics need to be adjusted according to the unit type.
[0135] In this embodiment, in-depth analysis of vibration characteristics through a fault diagnosis model enables accurate identification of early local faults; analysis of time-series data through a condition prediction model reveals the development trend of system operating status; a thermal anomaly detection model enables spatial location and propagation prediction of system-level thermal anomalies; and finally, the cross-force attention mechanism of the multimodal fusion layer intelligently fuses fault information from different sources and of different natures to generate comprehensive and accurate fault prediction results. This multi-level, multimodal fault prediction method significantly improves the accuracy and reliability of fault early warning for molten salt energy storage turbine power generation systems.
[0136] In one exemplary embodiment, such as Figure 3 As shown, the generation of fault sample data includes steps 302 to 304. Wherein:
[0137] Step 302: Obtain the fault injection parameter values of the molten salt energy storage turbine power generation system under different fault types.
[0138] It should be noted that obtaining sufficient and realistic fault sample data is a core challenge in the development of fault prediction models for molten salt energy storage turbine power generation systems. Due to the large scale and high cost of the system, conducting fault injection experiments on actual systems is extremely risky, costly, and may cause irreversible equipment damage, resulting in a scarcity of fault sample data, which severely restricts the training and application of high-precision prediction models.
[0139] The fault injection parameter value refers to the model input parameter set in the simulation model to simulate a specific fault. For example, to simulate turbine blade fouling, the parameter "blade surface roughness" is adjusted in the simulation model to increase its value.
[0140] For example, firstly, the set of fault types to be studied can be determined according to actual needs, such as bearing wear, molten salt pump efficiency reduction, and heat exchanger fouling. For each fault type, based on its physical mechanism or engineering experience, one or more key parameters, i.e., fault injection parameters, are determined for simulating the fault in the simulation model, and one or more values that can characterize different fault severity are assigned to them. The mapping relationship between these parameter values and corresponding fault types is organized and stored to provide input for subsequent simulations.
[0141] Step 304: Based on the fault injection parameter values, the simulation model is used to simulate the operating status of the molten salt energy storage turbine power generation system under different fault types, and fault sample data is obtained.
[0142] Fault sample data can refer to system operation data containing fault characteristics used to train and test fault prediction models. For example, a complete data record containing changes in system parameters such as vibration, temperature, and pressure when a bearing wear fault occurs.
[0143] For example, after determining the fault injection parameter values, they can be input one by one into the simulation model of the molten salt energy storage turbine power generation system. The simulation model is then driven to run under set operating conditions, simulating the system's dynamic response process after a fault injection. The simulation model will output all operating data that can be measured by each virtual sensor in the system under this fault state. These data, together with the fault type label, constitute the fault sample data used to train and test the fault prediction model.
[0144] In this embodiment, a fault injection method based on simulation models can safely, economically, and efficiently simulate various faults and their different degrees in the digital space, thereby generating a large amount of sample data covering multiple fault scenarios. This provides a fundamental way to solve the bottleneck problem of model performance being restricted by the scarcity of fault data, and lays a solid data foundation for training high-precision and high-reliability fault prediction models, thus effectively improving the accuracy of fault early warning.
[0145] In one exemplary embodiment, the current operating data includes sensing data collected by the sensor; after acquiring the current operating data, the method further includes:
[0146] Preprocess the currently running data, including at least one of the following:
[0147] The current running data is denoised using a Kalman filter;
[0148] The measurement error of the sensor data is corrected through a prediction-correction mechanism;
[0149] The timestamps of each sensor are unified through a hardware time synchronization protocol;
[0150] Dynamic time warping is performed on lagging sensor data;
[0151] Given that the current operating data includes steady-state operating condition data, the standard deviation standardization method is used to process the steady-state operating condition data;
[0152] Given that the current operating data includes transient operating condition data, the range normalization method is used to process the steady-state operating condition data.
[0153] It should be noted that in the operation monitoring of molten salt energy storage turbine power generation systems, the raw data directly collected from sensors often contains problems such as noise, measurement errors, time asynchrony, and inconsistencies in units. These data quality issues directly affect the accuracy of subsequent load forecasting, state assessment, and control strategy generation, causing the system to fail to reach its optimal operating state, and may even lead to misjudgments and erroneous control.
[0154] The Kalman filter is a recursive estimation algorithm that estimates the system state and filters out noise by combining the system model and measurement data through two steps: prediction and update.
[0155] The prediction-correction mechanism is a method that corrects errors by combining predicted and measured values. It first predicts the current value based on historical data or models, and then corrects the measurement error based on the difference between the actual measured value and the predicted value.
[0156] Hardware time synchronization protocols are technical protocols that use hardware signals to synchronize the clocks of multiple devices.
[0157] Dynamic time warping is an algorithm that processes time axis deformation in time series data. It aligns sequences of different lengths or speeds by nonlinearly stretching or compressing the time axis.
[0158] The standard deviation standardization method is a data standardization method that involves subtracting the mean and dividing by the standard deviation.
[0159] Range normalization is a method of scaling data to a specific range through linear transformation.
[0160] Steady-state operating data refers to data collected by a molten salt energy storage turbine power generation system under relatively stable operating conditions, with minimal parameter variations. For example, temperature and pressure data of a molten salt energy storage turbine power generation system operating continuously for more than 4 hours under constant load.
[0161] Transient operating condition data refers to data collected during the transition of operating conditions in a molten salt energy storage turbine power generation system, where parameters change drastically. For example, data on changes in molten salt temperature and flow rate during the system's transition from cold start to rated load.
[0162] For example, in the denoising process, the state equation and observation equation are first established based on the dynamic characteristics of the molten salt energy storage turbine power generation system, and the covariance matrices of process noise and measurement noise are determined. Then, the Kalman filter algorithm is applied to the current operating data collected by the sensors: the state estimate and error covariance are initialized; at each time step, state prediction is performed first, predicting the current state based on the system model; then, measurement updates are performed, and the predicted value is corrected by combining the actual measured value to obtain the optimal estimate. The Kalman filter is particularly suitable for processing random noise that follows a Gaussian distribution, and it has low computational complexity, making it suitable for real-time processing. For example, for the molten salt temperature signal at the heat exchanger outlet, the process noise covariance matrix and measurement noise covariance matrix are set, and the state estimate is initialized; at each sampling period, the current state is first predicted based on the system model, and then the estimate is updated by combining the actual measured value to obtain the denoised temperature sequence. For slowly varying signals such as temperature and pressure, the Kalman filter can effectively eliminate random noise interference, improve data quality, and maintain computational efficiency, making it suitable for deployment in resource-constrained industrial control systems.
[0163] In the process of measurement error correction, a predictive model of the sensor's readings can first be established based on the sensor's physical characteristics or historical normal data. The predicted value is then compared with the actual measured value to calculate the error. Subsequently, a preset correction algorithm (such as linear compensation or adaptive adjustment) is used to compensate for this systematic error in real time, thereby improving the absolute accuracy of the measurement data. For example, for thermocouple temperature measurement, electromagnetic interference often leads to systematic deviations. The theoretical temperature value can be predicted using a thermodynamic model; then, the difference between the predicted value and the actual measured value is calculated, and the proportional-integral correction mechanism is applied to dynamically adjust the correction coefficient.
[0164] During time synchronization, a high-precision master clock can be pre-deployed and a precision clock synchronization protocol can be run. This protocol uses network message exchange to accurately calculate and compensate for transmission delays, allowing the clock devices to adjust their local clocks accordingly. This ensures that all data acquisition devices assign a unified and accurate timestamp to each data sample, providing a strict time alignment basis for multi-source data fusion.
[0165] When a sensor data signal with an inherent delay is identified, a dynamic time warping algorithm can be used to calculate the optimal nonlinear matching path between the two signals, based on the delayed sensor data signal and the reference signal. This path is then used to warp the time axis of the lagging sensor data signal, aligning its feature points with the reference signal, thereby eliminating non-causal time differences and ensuring data synchronization in physical events.
[0166] The acquired current operating data can be further differentiated into steady-state and transient operating conditions. Data is considered steady-state if its fluctuation range remains stable within a preset threshold range; for example, molten salt flow fluctuation ≤ ±2% and steam pressure fluctuation ≤ ±0.5%. The mean and standard deviation of each variable within this steady-state data segment are calculated. Then, for each data point, the mean is subtracted and the standard deviation is divided to standardize the standard deviation. This method eliminates the influence of different dimensions and absolute values of the parameters, transforming them into a distribution with a mean of zero and a standard deviation of one, facilitating subsequent analysis models that focus on relative fluctuation patterns.
[0167] When current operating data undergoes significant changes within a short period, it is classified as steady-state data. For example, the molten salt flow rate decreases from its rated value to 80% within 10 seconds, and the steam pressure decreases from 10 MPa to 9.2 MPa within 30 seconds. The minimum and maximum values of each variable within this data segment are identified, and then the range normalization formula is used to linearly transform each data point to the interval [0, 1]. This method eliminates the influence of dimensions while preserving the complete form and trend of the transient process to the greatest extent possible, avoiding characteristic distortion caused by excessive differences in the amplitudes of different variables.
[0168] In this embodiment, Kalman filtering and a prediction-correction mechanism are used to process random noise and systematic errors respectively, thereby improving the accuracy and reliability of the data. Hardware time synchronization and dynamic time warping improve the consistency of multi-source heterogeneous data on the time axis. By using standard deviation standardization and range normalization methods according to the data characteristics of steady-state and transient conditions respectively, the statistical fluctuation characteristics of steady-state data are preserved, while the complete trend of transient data is maintained. Through these data preprocessing, high-quality and highly consistent input data can be provided for subsequent load forecasting, fault diagnosis, and advanced control, thereby significantly improving the accuracy and stability of the entire power generation system monitoring and management.
[0169] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0170] Based on the same inventive concept, this application also provides a molten salt energy storage turbine power generation system monitoring device for implementing the above-mentioned monitoring method for molten salt energy storage turbine power generation systems. The solution provided by this device is similar to the solution described in the above-described method. Therefore, the specific limitations of one or more embodiments of the molten salt energy storage turbine power generation system monitoring device provided below can be found in the limitations of the molten salt energy storage turbine power generation system monitoring method described above, and will not be repeated here.
[0171] In one exemplary embodiment, such as Figure 4 As shown, a monitoring device for a molten salt energy storage turbine power generation system is provided, comprising: an acquisition module 402, a load prediction module 404, a strategy generation module 406, a simulation module 408, a configuration module 410, and a feedback adjustment module 412, wherein:
[0172] The acquisition module 402 is used to acquire the simulation model, current operating data and current load data of the molten salt energy storage turbine power generation system;
[0173] The load forecasting module 404 is used to predict the load demand of the molten salt energy storage turbine power generation system in a preset future time period based on the current operating data and the current load data through a preset load forecasting model, obtain the load demand forecasting result, and query the control setpoint corresponding to the load demand forecasting result. The preset load forecasting model is obtained by training the spatiotemporal attention network in advance based on the historical operating data and historical load data of the molten salt energy storage turbine power generation system.
[0174] The strategy generation module 406 is used to generate a target power output control strategy for the molten salt energy storage turbine power generation system to achieve the preset optimization target by using the near-end strategy optimization controller, based on the load demand forecast results and control setpoint.
[0175] The simulation module 408 is used to simulate the operating state of the molten salt energy storage turbine power generation system according to the target power output control strategy through the simulation model, obtain the operating state simulation data, and perform safety verification on the operating state simulation data according to the preset unit operating constraints.
[0176] Configuration module 410 is used to configure the molten salt energy storage turbine power generation system according to the target power output control strategy and control setpoint if the safety verification passes.
[0177] The feedback adjustment module 412 is used to adjust the target power output control strategy if the safety check fails, until the safety check passes.
[0178] In an exemplary embodiment, the current operating data includes vibration signals and thermal data. The molten salt energy storage turbine power generation system monitoring device also includes a fault identification module. After configuring the molten salt energy storage turbine power generation system according to the target power output control strategy and control setpoint, the fault identification module is used to: obtain the physical structure adjacency matrix of the molten salt energy storage turbine power generation system; obtain sub-band energy features by wavelet packet decomposition of vibration signals, extract the envelope spectrum kurtosis features of vibration signals, and generate a thermal field spatial gradient matrix based on thermal data and the physical structure adjacency matrix; and perform fault prediction on the molten salt energy storage turbine power generation system based on the current operating data, sub-band energy features, envelope spectrum kurtosis features, and thermal field spatial gradient matrix using a preset fault prediction model, and obtain fault prediction results. The fault prediction results include the fault probability and fault type of the molten salt energy storage turbine power generation system in a preset future time period. The preset fault prediction model is trained based on fault sample data of the molten salt energy storage turbine power generation system.
[0179] In an exemplary embodiment, the current operating data includes time series of sensor data collected by multiple sensors; the preset fault prediction model includes a fault diagnosis model, an operating condition prediction model, a thermal anomaly detection model, and a multimodal fusion layer; the fault identification module is further used to: identify the target local target fault type based on subband energy characteristics and envelope spectrum kurtosis characteristics using the fault diagnosis model; predict the target operating condition type of the molten salt energy storage turbine power generation system in a preset future time period based on the time series of each sensor data using the operating condition prediction model; identify the target system-level thermal anomaly region in the molten salt energy storage turbine power generation system based on the thermal field spatial gradient matrix using the thermal anomaly detection model, and predict the thermal anomaly propagation path corresponding to the target system-level thermal anomaly region; and fuse the target local target fault type, target operating condition type, target system-level thermal anomaly region, and thermal anomaly propagation path using the multimodal fusion layer with a cross-force attention mechanism to obtain the fault prediction result.
[0180] In an exemplary embodiment, the fault diagnosis model adopts a one-dimensional residual network model; the operating condition prediction model adopts a time-series transformer model; and the thermal anomaly detection model adopts a graph attention network model.
[0181] In an exemplary embodiment, the fault identification module is further configured to: acquire fault injection parameter values of the molten salt energy storage turbine power generation system under different fault types; and simulate the operating status of the molten salt energy storage turbine power generation system under different fault types using a simulation model based on the fault injection parameter values to obtain fault sample data.
[0182] In an exemplary embodiment, the current operating data includes sensor data collected by sensors; the molten salt energy storage turbine power generation system monitoring device further includes a preprocessing module. After acquiring the current operating data, the preprocessing module is used to: preprocess the current operating data, the preprocessing including at least one of the following: denoising the current operating data using a Kalman filter; correcting the measurement error of the sensor data using a prediction-correction mechanism; unifying the timestamps of each sensor using a hardware time synchronization protocol; performing dynamic time warping on the lagging sensor data; processing the steady-state operating data using a standard deviation normalization method when the current operating data includes steady-state operating data; and processing the steady-state operating data using a range normalization method when the current operating data includes transient operating data.
[0183] Each module in the aforementioned molten salt energy storage turbine power generation system monitoring device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.
[0184] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a monitoring method for a molten salt energy storage turbine power generation system. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.
[0185] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0186] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.
[0187] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.
[0188] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.
[0189] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.
[0190] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.
[0191] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.
[0192] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A monitoring method for a molten salt energy storage turbine power generation system, characterized in that, The method includes: Obtain the simulation model, current operating data, and current load data of the molten salt energy storage turbine power generation system; By using a preset load prediction model, based on the current operating data and current load data, the load demand of the molten salt energy storage turbine power generation system is predicted in a preset future time period, and the load demand prediction result is obtained. The control setpoint corresponding to the load demand prediction result is then queried. The preset load prediction model is obtained by training a spatiotemporal attention network in advance based on the historical operating data and historical load data of the molten salt energy storage turbine power generation system. The near-end strategy optimization controller generates a target power output control strategy that enables the molten salt energy storage turbine power generation system to achieve a preset optimization target based on the load demand prediction results and the control setpoint. The simulation model is used to simulate the operating state of the molten salt energy storage turbine power generation system according to the target power output control strategy, and the operating state simulation data is obtained. The operating state simulation data is then verified for safety according to the preset unit operating constraints. If the safety verification passes, the molten salt energy storage turbine power generation system is configured according to the target power output control strategy and the control setpoint. If the safety check fails, the target power output control strategy will be adjusted accordingly until the safety check passes.
2. The method according to claim 1, characterized in that, The current operating data includes vibration signals and thermal data; after configuring the molten salt energy storage turbine power generation system according to the target power output control strategy and the control setpoint, the method further includes: Obtain the physical structure adjacency matrix of the molten salt energy storage turbine power generation system; The vibration signal is decomposed by wavelet packet to obtain subband energy features, the envelope spectrum kurtosis features of the vibration signal are extracted, and a thermal field spatial gradient matrix is generated based on the thermal data and the physical structure adjacency matrix. By using a preset fault prediction model, fault prediction is performed on the molten salt energy storage turbine power generation system based on the current operating data, the sub-band energy characteristics, the envelope spectrum kurtosis characteristics, and the thermal field spatial gradient matrix, and the fault prediction results are obtained. The fault prediction results include the fault probability and fault type of the molten salt energy storage turbine power generation system in a preset future time period. The preset fault prediction model is trained based on the fault sample data of the molten salt energy storage turbine power generation system.
3. The method according to claim 2, characterized in that, The current operating data includes time series of sensor data collected by various sensors; the preset fault prediction model includes a fault diagnosis model, an operating condition prediction model, a thermal anomaly detection model, and a multimodal fusion layer; the fault prediction of the molten salt energy storage turbine power generation system is performed using the preset fault prediction model based on the current operating data, the sub-band energy characteristics, the envelope spectrum kurtosis characteristics, and the thermal field spatial gradient matrix, to obtain fault prediction results, including: The fault diagnosis model identifies the local target fault type based on the subband energy characteristics and the envelope spectrum kurtosis characteristics. Based on the time series of the sensor data, the operating condition prediction model is used to predict the target operating condition type of the molten salt energy storage turbine power generation system in a preset future time period. Based on the thermal field spatial gradient matrix, the thermal anomaly detection model identifies the target system-level thermal anomaly region in the molten salt energy storage turbine power generation system and predicts the thermal anomaly propagation path corresponding to the target system-level thermal anomaly region. Through the multimodal fusion layer, the target local target fault type, the target operating condition type, the target system-level thermal anomaly region, and the thermal anomaly propagation path are fused using a cross-force attention mechanism to obtain the fault prediction result.
4. The method according to claim 3, characterized in that, The fault diagnosis model adopts a one-dimensional residual network model; the operating condition prediction model adopts a time-series transformer model; and the thermal anomaly detection model adopts a graph attention network model.
5. The method according to claim 2, characterized in that, The generation of the fault sample data includes: Obtain the fault injection parameter values of the molten salt energy storage turbine power generation system under different fault types; The simulation model is used to simulate the operating status of the molten salt energy storage turbine power generation system under different fault types based on the fault injection parameter values, and fault sample data is obtained.
6. The method according to any one of claims 1 to 5, characterized in that, The current operating data includes sensor data collected by the sensors; After obtaining the current running data, the method further includes: The currently running data is preprocessed, and the preprocessing includes at least one of the following: The current running data is denoised using a Kalman filter; The measurement error of the sensor data is corrected through a prediction-correction mechanism; The timestamps of all the sensors are unified through a hardware time synchronization protocol; Dynamic time warping is performed on lagging sensor data; If the current operating data includes steady-state operating data, the steady-state operating data is processed using the standard deviation standardization method. When the current operating data includes transient operating condition data, the steady-state operating condition data is processed using the range normalization method.
7. A monitoring device for a molten salt energy storage turbine power generation system, characterized in that, The device includes: The acquisition module is used to acquire the simulation model, current operating data, and current load data of the molten salt energy storage turbine power generation system; The load forecasting module is used to predict the load demand of the molten salt energy storage turbine power generation system in a preset future time period based on the current operating data and the current load data through a preset load forecasting model, obtain the load demand forecasting result, and query the control setpoint corresponding to the load demand forecasting result. The preset load forecasting model is obtained by training a spatiotemporal attention network in advance based on the historical operating data and historical load data of the molten salt energy storage turbine power generation system. The strategy generation module is used to generate a target power output control strategy for the molten salt energy storage turbine power generation system to achieve a preset optimization target by using the near-end strategy optimization controller, based on the load demand prediction results and the control setpoint. The simulation module is used to simulate the operating state of the molten salt energy storage turbine power generation system according to the target power output control strategy through the simulation model, obtain operating state simulation data, and perform safety verification on the operating state simulation data according to the preset unit operating constraints. A configuration module is used to configure the molten salt energy storage turbine power generation system according to the target power output control strategy and the control setpoint if the safety verification passes. The feedback adjustment module is used to adjust the target power output control strategy if the safety verification fails, until the safety verification passes.
8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.
