Method, system, device and medium for performance diagnosis of a photovoltaic thermal power station based on mechanism-data fusion

Abstract

The application relates to the technical field of solar thermal power generation, in particular to a photothermal power station performance diagnosis method based on mechanism-data fusion, which comprises the following steps: acquiring multi-source heterogeneous data of a target photothermal power station in an operation process; driving a driving efficiency model corresponding to each target system based on the multi-source heterogeneous data, so as to obtain real-time dynamic efficiency corresponding to each target system; inputting the real-time dynamic efficiency corresponding to each target system into a pre-constructed fusion diagnosis model; performing deviation detection, target system contribution decomposition and / or efficiency bottleneck identification on the overall efficiency of the target photothermal power station based on an efficiency deviation detection module, a correlation analysis module and / or an efficiency bottleneck identification module of the fusion diagnosis model, so as to obtain a fusion diagnosis result; obtaining an overall performance diagnosis result of the target photothermal power station based on the fusion diagnosis result; and performing systematic diagnosis on the photothermal power station as a whole based on the overall efficiency of the power station, which represents the overall energy of the power station.

Description

Technical Field

[0001] This application relates to the field of solar thermal power generation technology, specifically to a method, system, equipment, and medium for performance diagnosis of solar thermal power plants based on mechanism-data fusion. Background Technology

[0002] Concentrating Solar Power (CSP) is a clean and sustainable form of energy that has been widely used globally in recent years. CSP power plants use solar mirrors to collect solar energy and heat the working fluid to drive a steam turbine generator set to generate electricity. Its core subsystems include a concentrating solar collector system, a thermal storage system, and a power generation system.

[0003] However, the performance evaluation, diagnosis and optimization of concentrated solar power (CSP) plants in related technologies usually only perform fragmented calculations and analyses on each subsystem contained in the CSP plant individually and in isolation. They do not consider the impact of the coupling of multiple subsystems as a whole on the operating efficiency of the CSP plant. At the plant level, they cannot identify the global efficiency bottleneck and the true source of performance loss when the efficiency of the CSP plant declines. Summary of the Invention

[0004] This application provides a method, system, equipment, and medium for performance diagnosis of solar thermal power plants based on mechanism-data fusion, in order to solve the technical problems existing in the performance evaluation, diagnosis, and optimization of solar thermal power plants.

[0005] Firstly, this application provides a performance diagnosis method for solar thermal power plants based on mechanism-data fusion. The performance diagnosis method for solar thermal power plants includes: Step S1: Acquire multi-source heterogeneous data of the target solar thermal power plant during operation; wherein, the multi-source heterogeneous data includes meteorological data and system operation data; the system operation data includes operating sub-data of the concentrating solar thermal collection system, operating sub-data of the thermal storage system, and operating sub-data of the power generation system; Step S2: Based on the multi-source heterogeneous data, drive the corresponding driving efficiency model of each target system to obtain the real-time dynamic efficiency of each target system; wherein, at least one driving efficiency model is obtained by fusing the mechanism submodule and the data submodule; Step S3: Input the real-time dynamic efficiency corresponding to each target system into the pre-built fusion diagnostic model. Based on the efficiency deviation detection module, correlation analysis module and / or efficiency bottleneck identification module of the fusion diagnostic model, perform deviation detection, target system contribution decomposition and / or efficiency bottleneck identification on the overall efficiency of the target solar thermal power plant to obtain the fusion diagnostic results. Step S4: Based on the fusion diagnostic results, obtain the overall performance diagnostic results of the target solar thermal power plant; the overall performance diagnostic results include efficiency bottleneck location, root cause analysis, and optimization and maintenance recommendations.

[0006] Further explanation is needed regarding step S3, which further includes: Step S31, using the efficiency deviation detection module to calculate the overall efficiency of the power station based on the pre-constructed overall energy balance formula of the power station, and performing deviation detection on the overall efficiency of the power station; Step S32, when the overall efficiency of the power station is lower than a preset threshold, based on the correlation analysis module, quantitatively calculating the percentage contribution of each efficiency deviation to the overall efficiency decline through efficiency deviation contribution decomposition, and automatically marking the target system with the largest contribution as the efficiency bottleneck; wherein, the percentage contribution of each system's efficiency deviation to the overall efficiency decline is calculated by the real-time dynamic efficiency of each target system and the corresponding historical dynamic efficiency; Step S33, the efficiency bottleneck identification module calculates the loss of each system and sorts them in real time, marking the target system whose loss ratio exceeds the preset threshold as the efficiency bottleneck system; Step S34, determining the fusion diagnosis result based on the efficiency bottleneck and / or efficiency bottleneck system.

[0007] It should be further noted that the correlation analysis module uses the following formula to decompose the contribution of efficiency deviation: In the formula, This indicates the overall efficiency deviation of the power plant. This indicates the overall efficiency of the power plant. This indicates the efficiency deviation of the receiver system. This indicates the real-time dynamic efficiency of the heat absorber system. This indicates the efficiency deviation of the thermal storage system. This indicates the real-time dynamic efficiency of the thermal storage system. This indicates the efficiency deviation of the thermal storage system. This indicates the real-time dynamic efficiency of the steam turbine system; To assist in the overall efficiency deviation of the auxiliary system, To improve the overall efficiency of the auxiliary system, the partial derivative coefficients were determined through sensitivity analysis based on historical data, and ε represents the unmodeled residual term.

[0008] It should be further noted that the pre-constructed overall energy balance formula for the power plant is derived from the following formula:

[0009] In the formula, The actual power output of the target solar thermal power plant. The overall efficiency of the target solar thermal power plant. ID card Indicates the direct normal radiation of the sun. For the purpose of a solar thermal power plant, The overall efficiency of the target solar thermal power plant is calculated using the following formula:

[0010] in, The overall efficiency of the target solar thermal power plant. For the heat absorber fusion efficiency, The overall efficiency of heat charging and discharging in the thermal storage system, For the efficiency of steam turbine integration, To improve the overall efficiency of the auxiliary system.

[0011] It should be further explained that the model-driven efficiency model includes a receiver efficiency sub-model; the receiver efficiency sub-model includes a first mechanism sub-module, a first data sub-module, and a first fusion efficiency calculation module; the first mechanism sub-module calculates the theoretical thermal efficiency of the receiver based on the lumped parameter method; the first data sub-module uses a long short-term memory network to dynamically predict the calculation residual of the first mechanism sub-module under transient conditions, and the input feature vector of the long short-term memory network model includes the mechanism calculation efficiency, the temperature field distribution of multiple regions on the receiver wall, the DNI time series, and the ambient wind speed; the first fusion efficiency calculation module sums the theoretical thermal efficiency of the receiver and the residual to obtain the real-time dynamic efficiency; and the first fusion efficiency calculation module is used to verify the rationality of the fusion efficiency of the fusion efficiency calculation unit. If the fusion efficiency exceeds the physically feasible range or the absolute value of the residual prediction exceeds a preset threshold, it will fall back to the mechanism calculation result and trigger model retraining.

[0012] It should be further noted that the driving efficiency model also includes a sub-model for calculating the dynamic heat loss of the thermal storage system. This sub-model includes a multi-physics sensing module, a dynamic heat loss calculation module, and a thermal storage and heat release calculation module. The multi-physics sensing module is used to collect the temperature gradient distribution inside the tank and the temperature distribution of the tank wall. The dynamic heat loss calculation module is used to input the temperature distribution at each height level of the tank, the molten salt level H, the ambient temperature Tamb, the molten salt inlet / outlet temperature, and the charging / releasing mode identifier. Under the constraints of the laws of heat transfer physics, it outputs the heat loss coefficient representing the heat loss per unit time of the thermal storage system at the current moment, which is the result of data fitting. The thermal storage and heat release calculation module is used to calculate the charging efficiency and the utilization efficiency on the power generation side.

[0013] It should be further explained that the driving efficiency model also includes a turbine variable operating condition dual-drive efficiency sub-model. The turbine variable operating condition dual-drive efficiency sub-model includes a second mechanism sub-module, a second data sub-module, and a second fusion efficiency calculation module. The second mechanism sub-module is established based on the reconstructed Flügel formula with the introduction of a temperature correction term. The second data sub-module is constructed based on a multi-layer deep neural network. The input feature vector of the second data sub-module includes meteorological parameters, real-time operating parameters of the turbine, and the calculation results of the second mechanism sub-module. The output of the second data sub-module is the turbine dynamic efficiency after physical constraint correction. The second fusion efficiency calculation module adopts an adaptive weighted fusion strategy to adaptively weight and fuse the second mechanism sub-module and the second data sub-module. According to the operating condition identification classifier, the current operating stage is determined. Under quasi-steady state, the weight of the second mechanism sub-module and the fusion weight are increased. Under transient variable operating conditions, the weight of the second data sub-module is increased.

[0014] Secondly, this application provides a performance diagnostic system for a concentrated solar power (CSP) plant based on mechanism-data fusion. The system includes: an acquisition module for acquiring multi-source heterogeneous data of the target CSP plant during operation; wherein the multi-source heterogeneous data includes meteorological data and system operation data; the system operation data includes sub-data of the concentrating solar collector system operation, sub-data of the thermal storage system operation, and sub-data of the power generation system operation; and a drive efficiency calculation module for driving the drive efficiency model corresponding to each target system based on the multi-source heterogeneous data, thereby obtaining the real-time dynamic efficiency corresponding to each target system; wherein at least one drive efficiency model is fused from a mechanism sub-module and a data sub-module. The target system includes a concentrating solar thermal power system, a thermal storage system, and a power generation system. A fusion diagnostic module is used to input the real-time dynamic efficiency of each target system into a pre-built fusion diagnostic model. Based on the efficiency deviation detection module, correlation analysis module, and efficiency bottleneck identification module of the fusion diagnostic model, deviation detection and target system contribution decomposition are performed on the overall efficiency of the target solar thermal power plant to obtain fusion diagnostic results. A performance diagnostic module is used to obtain the overall performance diagnostic results of the target solar thermal power plant based on the fusion diagnostic results. The overall performance diagnostic results include efficiency bottleneck location, root cause analysis, and optimization and maintenance suggestions.

[0015] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the steps of the mechanism-data fusion-based solar thermal power plant performance diagnosis method as described in any of the above embodiments.

[0016] Fourthly, this application provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the mechanism-data fusion-based solar thermal power plant performance diagnosis method as described in any of the above embodiments.

[0017] Compared with existing technologies, this application has the following advantages: This invention collects meteorological data, as well as full operational data from the three major subsystems: solar thermal collection, thermal storage, and power generation. Then, using a fusion model combining the mechanism and data, it calculates the real-time efficiency of each subsystem. The efficiency of all subsystems is input into a fusion diagnostic model, first comparing it with theoretical values ​​to identify efficiency deviations, then quantifying the impact weight of each subsystem on the overall power plant efficiency, and finally providing operation and maintenance recommendations. This overcomes the drawbacks of isolated analysis of subsystems in related technologies, avoiding the misjudgment of efficiency losses caused by coupling as faults in a single subsystem, and instead allowing the investigation to trace the cause of insufficient output from the upstream solar thermal collection system.

[0018] This invention achieves power plant-level positioning. Through contribution decomposition and coupling relationship analysis, it can distinguish the actual impact of efficiency losses in different subsystems on overall efficiency, directly targeting the most critical areas for optimization and improving operational efficiency. Based on real-time operational data-driven analysis, it can capture the causes of efficiency changes under transient conditions, such as the impact of short-term anomalies like cloud cover and load fluctuations, adapting to the operational needs of solar thermal power plants at different time scales. Attached Figure Description

[0019] To more clearly illustrate the technical solution of this patent, the drawings used in the description will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this patent. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating a mechanism-data fusion-based performance diagnosis method for a solar thermal power plant according to an embodiment of this application; Figure 2 This is a schematic diagram of the model structure of the PINN-based dynamic heat loss calculation module according to an embodiment of this application; Figure 3 This is a schematic block diagram of a solar thermal power plant performance diagnosis system based on mechanism-data fusion according to an embodiment of this application; Figure 4 A schematic diagram of the hardware structure of an electronic device for implementing the various embodiments of this application. Detailed Implementation

[0021] To make the objectives, features, and advantages of this patent more apparent and understandable, the technical solutions of this patent will be clearly and completely described below with reference to the accompanying drawings of the specific embodiments. Obviously, the embodiments described below are only some embodiments of this patent, and not all embodiments. Based on the embodiments of this patent, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this patent.

[0022] In view of this, this application proposes a mechanism-data fusion-based method for the performance diagnosis of solar thermal power plants, in order to solve at least one of the above-mentioned problems in the performance evaluation, diagnosis and optimization of solar thermal power plants in related technologies.

[0023] In a first aspect, according to embodiments of this application, a method for performance diagnosis of a solar thermal power plant based on mechanism-data fusion is provided; exemplarily, Figure 1 This is a flowchart illustrating a mechanism-data fusion-based performance diagnosis method for concentrated solar power plants according to an embodiment of this application, such as... Figure 1 As shown, the performance diagnosis methods for concentrated solar power (CSP) plants include: Step S1: Obtain multi-source heterogeneous data of the target solar thermal power plant during operation; wherein, the multi-source heterogeneous data includes meteorological data and system operation data; the system operation data includes operating sub-data of the concentrating solar thermal collection system, operating sub-data of the thermal storage system, and operating sub-data of the power generation system.

[0024] In this embodiment, the multi-source heterogeneous data includes meteorological data corresponding to the target solar thermal power plant and system operation data of multiple target systems contained in the target solar thermal power plant.

[0025] For example, a solar thermal power plant may include multiple systems; specifically, a solar thermal power plant may include a concentrating solar collector system, a thermal storage system, and a power generation system.

[0026] In one embodiment, the concentrating solar thermal system may include a heat absorber.

[0027] In one embodiment, the power generation system may include a steam turbine generator set.

[0028] One possible approach is to pre-deploy a multi-dimensional sensor network on the target solar thermal power plant system and key equipment to collect multi-source heterogeneous data.

[0029] For example, multi-source heterogeneous data may include meteorological data corresponding to the target solar thermal power plant and system operation data of multiple target systems contained in the target solar thermal power plant; wherein, the system operation data includes operation sub-data of the concentrating solar collector system, operation sub-data of the thermal storage system, and operation sub-data of the power generation system.

[0030] For example, meteorological parameters may include direct normal irradiance (DNI), ambient temperature, ambient humidity, wind speed, and wind direction.

[0031] For example, the parameters of the receiver system may include the inlet / outlet working fluid temperature, flow rate, pressure, temperature field distribution of multiple regions on the receiver wall, heat flux density of the receiver tube wall, and steam parameters at the receiver outlet; wherein, the temperature field distribution of multiple regions on the receiver wall can be obtained by an infrared thermal imager or a distributed thermocouple.

[0032] For example, the parameters of the thermal storage system may include the temperature gradient distribution of each height layer of the molten salt tank, the molten salt level, the molten salt flow rate, the temperature field distribution of the tank wall, the tank pressure, and the power of the circulating pump; wherein, the temperature gradient distribution of each height layer of the molten salt tank can be obtained by a pre-arranged array of multi-point temperature sensors.

[0033] For example, turbine system parameters may include main steam temperature, pressure, flow rate, extraction steam pressure, exhaust steam temperature, condenser vacuum, turbine speed, and generator power.

[0034] For example, auxiliary system energy consumption parameters may include the power consumption of the heliostat field tracking motor, the power consumption of the molten salt circulating pump, and the power consumption of the cooling tower.

[0035] In one embodiment, the raw data acquired from sensor data is preprocessed. During preprocessing, data points that are significantly outside the physical range are removed, and high-frequency noise is removed using median filtering or moving average. Z-score normalization is performed on sensor data of different dimensions. For some temperature field data with lower sampling frequencies, cubic spline interpolation is used to interpolate them to the same time baseline as the high-frequency data (such as DNI, pressure, etc.).

[0036] Step S2: Based on the multi-source heterogeneous data, drive the corresponding driving efficiency model of each target system to obtain the real-time dynamic efficiency of each target system; wherein, the at least one driving efficiency model is obtained by fusing the mechanism submodule and the data submodule; wherein, the target system includes a concentrating solar thermal system, a thermal storage system, and a power generation system.

[0037] As mentioned above, in the actual operation of concentrated solar power (CSP) plants, transient conditions such as rapid fluctuations in direct normal radiation (DNI), cloud cover, and variable loads frequently occur. Physical models struggle to accurately capture dynamic characteristics such as unsteady heat transfer and thermal inertia hysteresis, leading to increased performance calculation errors. To address this issue, this application uses multi-source heterogeneous data to drive the corresponding driving efficiency models for each target system. At least one driving efficiency model is obtained by fusing a mechanism submodule and a data submodule. In the aforementioned method of constructing driving efficiency models, the mechanism submodule considers the physical characteristics of each system, while the data submodule considers the numerical characteristics of each system as shown in the data. The mechanism submodule exhibits good accuracy under stable operating conditions of the CSP plant, while the data submodule considers the specific performance of the CSP plant under transient conditions such as rapid fluctuations in meteorological data. The two models complement each other during fusion to determine the real-time dynamic efficiency that characterizes the equipment-level operating status of the devices contained within each system, taking into account the physical characteristics and dynamic features of the CSP plant.

[0038] For example, after driving the driving efficiency model corresponding to each target system based on multi-source heterogeneous data, the real-time dynamic efficiency output by the driving efficiency model is physically verified; if the fused real-time dynamic efficiency exceeds the physically feasible range, the system falls back to the mechanism calculation result of the mechanism submodule and triggers model retraining for the data submodule.

[0039] For example, after obtaining the real-time dynamic efficiency, the equipment contained in the target system can be diagnosed based on the real-time dynamic efficiency, thereby realizing equipment-level performance diagnosis of the solar thermal power plant.

[0040] Step S3: Input the real-time dynamic efficiency corresponding to each target system into the pre-built fusion diagnostic model. Based on the efficiency deviation detection module, correlation analysis module and / or efficiency bottleneck identification module of the fusion diagnostic model, perform deviation detection, target system contribution decomposition and / or efficiency bottleneck identification on the overall efficiency of the target solar thermal power plant to obtain the fusion diagnostic results.

[0041] As mentioned above, the performance evaluation and diagnostic optimization methods for concentrated solar power (CSP) plants in related technologies typically only perform fragmented calculations and analyses on each subsystem within the CSP plant individually and in isolation. They do not consider the impact of multi-subsystem coupling on the operating efficiency of the CSP plant, and cannot identify the global efficiency bottleneck or the true source of performance loss at the plant level. Specifically, existing technologies cannot quantify the contribution of each system—the receiver, thermal storage, and turbine—to the overall efficiency decline of the plant. When the total power generation efficiency of the plant declines, maintenance personnel cannot quickly pinpoint whether the problem stems from insufficient heat collection, excessive thermal storage loss, or a decrease in power generation efficiency.

[0042] To address this issue, in this embodiment, the real-time dynamic efficiency corresponding to each target system is input into a pre-constructed fusion diagnostic model. Based on the efficiency deviation detection module, correlation analysis module, and efficiency bottleneck identification module of the fusion diagnostic model, deviation detection of the target system contribution and / or efficiency bottleneck identification are performed on the overall efficiency of the target solar thermal power plant to obtain the fusion diagnostic results. In one embodiment, the efficiency deviation detection module can be used to output the overall efficiency of the target solar thermal power plant; it is achieved by constructing the overall energy balance relationship of the power plant; the overall efficiency of the power plant can correspond to the working state of the solar thermal power plant, providing a basis for subsequent diagnosis of the solar thermal power plant.

[0043] Furthermore, after obtaining the overall efficiency of the power plant, the fusion diagnostic model uses its included efficiency deviation detection module to detect the deviation between the overall efficiency of the power plant and the design efficiency target value and / or the historical efficiency benchmark value. If the deviation between the overall efficiency of the power plant and the design efficiency target value and / or the historical efficiency benchmark value is greater than a preset value, it is confirmed that the overall efficiency of the power plant has decreased. At this time, the fusion diagnostic model uses the correlation analysis module to decompose the contribution of the system to quantitatively calculate the percentage contribution of each system efficiency deviation to the decrease in the overall efficiency of the power plant, and uses the efficiency bottleneck identification module to calculate the loss and efficiency of each system.

[0044] In one embodiment, after obtaining the overall power plant efficiency, the fusion diagnostic model, based on its included efficiency deviation detection module, detects the deviation between the overall power plant efficiency and the design efficiency target value and / or the historical efficiency benchmark value. If the deviation between the overall power plant efficiency and the design efficiency target value and / or the historical efficiency benchmark value is greater than a preset value, it is confirmed that the overall power plant efficiency has decreased. At this time, the fusion diagnostic model uses a correlation analysis module to decompose the system's contribution to quantitatively calculate the percentage contribution of each system's efficiency deviation to the decrease in the overall power plant efficiency. By comparing the efficiency deviation direction output by the driving efficiency model corresponding to each target system, The system's efficiency bottleneck is initially identified by its size and time correlation. Furthermore, the integrated diagnostic model employs an efficiency bottleneck identification module to quantitatively calculate the energy loss of each target system under current operating conditions from the perspective of energy transfer and energy analysis, automatically marking the current efficiency shortcomings. The above implementation method uses the overall efficiency of the power plant, which characterizes the overall energy of the power plant, as a benchmark. Through overall performance calculations, energy loss and energy efficiency assessments of the three core systems—the receiver, the thermal storage system, and the steam turbine—a systematic diagnosis of the solar thermal power plant is performed at the system level. The system with the bottleneck and shortcomings are automatically marked and identified, thus obtaining the integrated diagnostic results.

[0045] In one embodiment, after obtaining the overall power plant efficiency, the fusion diagnostic model uses its included efficiency deviation detection module to detect the deviation between the overall power plant efficiency and the design efficiency target value and / or the historical efficiency benchmark value. If the deviation between the overall power plant efficiency and the design efficiency target value and / or the historical efficiency benchmark value is greater than a preset value, it is confirmed that the overall power plant efficiency has decreased. At this time, the fusion diagnostic model uses the correlation analysis module to decompose the contribution of the system to quantitatively calculate the percentage contribution of each system efficiency deviation to the decrease in the overall power plant efficiency. By comparing the efficiency deviation direction, magnitude, and time correlation output by the three models, the system where the efficiency bottleneck is located is initially determined, thereby obtaining the fusion diagnostic result. The above implementation method uses the overall power plant efficiency, which characterizes the overall energy of the power plant, as the benchmark, and performs a systematic diagnosis of the solar thermal power plant by calculating the overall performance of the three core systems: receiver, thermal storage system, and steam turbine.

[0046] In one embodiment, if the deviation between the overall efficiency of the power plant and the design efficiency target value and / or the historical efficiency benchmark value is greater than a preset value, it is confirmed that the overall efficiency of the power plant has declined. At this time, the fusion diagnostic model adopts the efficiency bottleneck identification module: from the perspective of energy transfer and efficiency analysis, it quantitatively calculates the efficiency loss of each target system under the current operating conditions, automatically marks the efficiency bottleneck of the current operation, and thus obtains the fusion diagnostic results. The quantitative assessment of energy utilization quality can be achieved through efficiency loss and efficiency.

[0047] In one embodiment, the efficiency deviation detection module can monitor the numerical characteristics and / or trend characteristics of the real-time dynamic efficiency of each system relative to its health benchmark value in real time. When the real-time dynamic efficiency is lower than the health benchmark value, and / or when the trend of the real-time dynamic efficiency deviates from the trend of the health benchmark value, the module marks and judges the corresponding system health status.

[0048] Step S4: Based on the fusion diagnostic results, obtain the overall performance diagnostic results of the target solar thermal power plant; the overall performance diagnostic results include efficiency bottleneck location, root cause analysis, and optimization and maintenance recommendations.

[0049] In this embodiment, after obtaining the fusion diagnostic results, the overall performance diagnostic results of the target solar thermal power plant are obtained based on the fusion diagnostics. Specifically, based on the analysis results of the fusion diagnostic model, a diagnostic report and operation and maintenance suggestions are automatically generated, including: the location of the efficiency bottleneck target system (absorber / thermal storage / turbine), root cause analysis suggestions (such as hot spots on the absorber surface, molten salt storage tank insulation layer failure, turbine scaling, etc.), and operable optimization control strategies and predictive maintenance suggestions.

[0050] This application presents a method for obtaining the overall performance diagnosis results of a target solar thermal power plant based on fusion diagnostic results. The fusion diagnostic results are generated by a fusion diagnostic model, which, through the linkage between an efficiency deviation detection module, a correlation analysis module, and / or an efficiency bottleneck identification module, places the concentrating solar collector system, thermal storage system, and power generation system into a unified diagnostic framework. It automatically locates efficiency bottlenecks through contribution decomposition and establishes quantitative correlations across systems through efficiency deviation contribution decomposition and Δ analysis. Using the overall power plant efficiency, which characterizes the overall energy of the power plant, as a benchmark, the method performs a systematic diagnosis of the solar thermal power plant by calculating the overall performance of the receiver, thermal storage system, and turbine.

[0051] At least one driving efficiency model in this application is obtained by fusing a mechanism submodule and a data submodule. The mechanism submodule is constructed considering the physical characteristics of each system, while the data submodule is constructed considering the numerical characteristics of each system in the data. The mechanism submodule can show good accuracy under stable operating conditions of the solar thermal power plant, while the data submodule can consider the specific performance of the solar thermal power plant under transient operating conditions such as rapid fluctuations in meteorological data. When the two models are fused, they complement each other to consider the physical characteristics and dynamic features of the solar thermal power plant and determine the real-time dynamic efficiency that characterizes the equipment-level operating status of the equipment contained in each system.

[0052] As an exemplary embodiment, step S3 further includes: step S31, using the efficiency deviation detection module to calculate the overall efficiency of the power station based on the pre-constructed overall energy balance formula of the power station, and performing deviation detection on the overall efficiency of the power station; step S32, when the overall efficiency of the power station is lower than a preset threshold, based on the correlation analysis module, quantitatively calculate the contribution percentage of each efficiency deviation to the overall efficiency decline through efficiency deviation contribution decomposition, and automatically mark the target system with the largest contribution as the efficiency bottleneck; wherein, the contribution percentage of each system efficiency deviation to the overall efficiency decline is calculated by the real-time dynamic efficiency of each target system and the corresponding historical dynamic efficiency; step S33, the efficiency bottleneck identification module calculates the loss of each system and sorts them in real time, and marks the target system whose loss ratio exceeds the preset threshold as the efficiency bottleneck system; step S34, determining the fusion diagnosis result based on the efficiency bottleneck and / or efficiency bottleneck system.

[0053] In this embodiment, the efficiency deviation detection module first calculates the overall efficiency of the power station based on a pre-built overall energy balance formula for the power station; wherein, the overall efficiency of the power station is calculated based on a predefined calculation formula, and the overall efficiency is expressed as the product of the efficiencies of each system.

[0054] For example, the overall energy balance relationship of the power plant is constructed using equation (1): (1) In equation (1), The actual power of the target solar thermal power plant. The overall efficiency of the target solar thermal power plant. ID card Indicates the direct normal radiation of the sun. A total The area of ​​the light-receiving opening of the heat absorber; For example, considering energy transfer losses, the overall efficiency of the target solar thermal power plant can be calculated by multiplying the efficiencies of each target system; specifically, the overall efficiency of the target solar thermal power plant can be calculated using equation (2): (2) In equation (2), The overall efficiency of the target solar thermal power plant. For the heat absorber fusion efficiency, The overall efficiency of heat charging and discharging in the thermal storage system, For the efficiency of steam turbine integration, To improve the overall efficiency of the auxiliary system.

[0055] Furthermore, when the overall efficiency of the power plant is lower than a preset threshold, the percentage contribution of each system's efficiency deviation to the overall efficiency decline is quantitatively calculated by decomposing the efficiency deviation contribution, and the target system with the largest contribution is automatically marked as the efficiency bottleneck.

[0056] In one embodiment, for detecting the overall efficiency of the power plant, a sliding time window with a preset duration can be predefined. The obtained overall efficiency of the power plant is sampled according to the sliding window. At each time step, the deviation between the overall efficiency of the power plant and the design efficiency target value or the historical efficiency benchmark value is calculated. Then, the deviation is used to determine whether the overall efficiency of the power plant is lower than a preset threshold.

[0057] For example, the preset duration can be 24 hours.

[0058] For example, the preset threshold can be the design efficiency target value of the target solar thermal power plant.

[0059] For example, the preset threshold can be a historical efficiency benchmark value for the same period.

[0060] Furthermore, when the overall efficiency of the power plant is lower than a preset threshold, the percentage contribution of each system's efficiency deviation to the overall efficiency decline is quantitatively calculated through efficiency deviation contribution decomposition, and the target system with the largest contribution is automatically marked as the efficiency bottleneck; specifically, the overall efficiency deviation can be decomposed using equation (3) to determine the percentage contribution of each system's efficiency deviation to the overall efficiency decline: (3) In equation (3), This indicates the overall efficiency deviation of the power plant. This indicates the overall efficiency of the power plant. This indicates the efficiency deviation of the receiver system. This indicates the real-time dynamic efficiency of the heat absorber system. This indicates the efficiency deviation of the thermal storage system. This indicates the real-time dynamic efficiency of the thermal storage system. This indicates the efficiency deviation of the thermal storage system. This indicates the real-time dynamic efficiency of the steam turbine system; To assist in the overall efficiency deviation of the auxiliary system, To assist in improving the overall system efficiency, the partial derivative coefficients are determined through sensitivity analysis based on historical data, and ε represents the unmodeled residual term. Through the above decomposition, the percentage contribution of each target system efficiency deviation to the overall power plant efficiency decline can be quantitatively calculated, thereby automatically identifying the subsystem with the largest current contribution as the efficiency bottleneck. Correlation analysis is then used to preliminarily determine the efficiency bottleneck.

[0061] Specifically, for each target system, the difference between the real-time dynamic efficiency and the historical efficiency of each target system is calculated as... , , and Furthermore, the difference is multiplied by the partial derivative coefficient of each target system to obtain the absolute contribution of each target system to the overall efficiency deviation of the power plant. Finally, the absolute contribution of each target system is divided by the overall deviation efficiency of the power plant to obtain the percentage contribution of the efficiency deviation of each target system to the overall efficiency decline.

[0062] In one embodiment, the efficiency deviation detection model monitors the trend of change in the system efficiency of each target system relative to its health baseline.

[0063] Simultaneously, or further, the efficiency loss of each system is calculated and sorted in real time, and the target system whose efficiency loss ratio exceeds a preset threshold is marked as an efficiency bottleneck system.

[0064] In this embodiment, the concept of exergy analysis is introduced to calculate the exergy loss and exergy efficiency of each target system, thereby achieving a quantitative assessment of the quality of energy utilization.

[0065] Specifically, the heat loss of the receiver system can be calculated using equation (4): Eloss,rec=Esolar,in(EHoutEHin)(4) In equation (4), Eloss,rec is the loss of the receiver system, Esolar,in is the loss of the incident solar energy, EHout is the loss of the receiver working fluid outlet, and EHin is the loss of the receiver working fluid inlet.

[0066] The thermal storage system loss is calculated using equation (5): Eloss,storage=Echarge,inEdischarge,out(5) In equation (5), Eloss,storage is the thermal storage system loss, Echarge,in is the thermal storage system working fluid inlet loss, and Edischarge,out is the thermal storage system working fluid outlet loss.

[0067] The turbine system loss is calculated using equation (6). Eloss,turb=Esteam,inEsteam,outWelec(6) In equation (6), Eloss,turb is the turbine system loss, Esteam,in is the turbine inlet loss, Esteam,out is the turbine outlet loss, and Welec is the generator power.

[0068] After calculating the loss of each system, the loss of each target system is sorted in real time, and the link with the largest current loss is automatically marked as the efficiency bottleneck.

[0069] Furthermore, the fusion diagnostic results are determined based on efficiency bottlenecks and / or efficiency shortcomings of the system.

[0070] In one embodiment, the fusion diagnosis result can be determined solely based on the efficiency bottleneck; specifically, through decomposition as shown in Equation (3), the percentage contribution of each target system efficiency deviation to the overall power plant efficiency decline can be quantitatively calculated, and the target system with the largest current contribution can be automatically marked as the efficiency bottleneck.

[0071] In one embodiment, the fusion diagnostic result can be determined solely based on the efficiency bottleneck system; specifically, when the efficiency loss ratio of a certain target system continuously exceeds a preset threshold (e.g., >40%), the system automatically outputs optimization suggestions, such as suggesting that the current receiver efficiency loss is too large and that the mirror field tracking accuracy or the surface dust accumulation of the receiver should be checked.

[0072] Using the above diagnostic methods, the overall performance diagnosis is performed in a three-tiered progression from equipment level to system level to power plant level. Specifically, for equipment-level diagnosis: the drive efficiency models corresponding to each of the aforementioned target systems are run respectively, and the real-time dynamic efficiency of each target system is output. or rec,fused , or storage,fused , or turb,fused And the equipment-level diagnostic conclusions within each system; furthermore, for system-level diagnostics, when the total power generation efficiency of the power plant... or plant,overallWhen a decline occurs, the integrated diagnostic model compares the deviation trends of the efficiency of the three target systems. If the efficiency deviation of target system A is significant while that of target systems B and C is normal, the problem is initially located in target system A. If the efficiency of multiple target systems deviates simultaneously, the primary and secondary causes are determined by the above-mentioned efficiency deviation contribution decomposition. Finally, power plant-level efficiency bottlenecks are identified: based on the efficiency analysis results, the efficiency losses of each target system are ranked and trend analyzed.

[0073] In one embodiment, after calculating the loss of each target system, the temporal changes of the loss of each target system are tracked. When the loss of a certain target system shows an accelerating growth trend, an early warning of performance degradation is issued to support predictive maintenance decisions.

[0074] As an exemplary embodiment, the model-driven efficiency model includes a receiver efficiency sub-model; the receiver efficiency sub-model includes a first mechanism sub-module, a first data sub-module, and a first fusion efficiency calculation module; the first mechanism sub-module calculates the theoretical thermal efficiency of the receiver based on the lumped parameter method; the first data sub-module uses a long short-term memory network to dynamically predict the calculation residual of the first mechanism sub-module under transient conditions, and the input feature vector of the long short-term memory network model includes the mechanism calculation efficiency, the temperature field distribution of multiple regions on the receiver wall, the DNI time series, and the ambient wind speed; the first fusion efficiency calculation module sums the theoretical thermal efficiency of the receiver and the residual to obtain the real-time dynamic efficiency.

[0075] As mentioned above, traditional physical models are insufficient in their accuracy under transient conditions when used for performance evaluation and diagnostic optimization of concentrated solar power (CSP) plants. Specifically, physical models based on thermodynamic laws have good accuracy at the design point and steady-state conditions, but in the actual operation of CSP plants, transient conditions such as rapid fluctuations in direct normal radiation (DNI), cloud obstruction, and variable loads frequently occur. Physical models struggle to accurately capture dynamic characteristics such as unsteady heat transfer and thermal inertia hysteresis, leading to increased performance calculation errors. To address this issue, in this embodiment, the receiver efficiency sub-model adopts a dual-drive architecture of mechanism calculation and artificial intelligence residual prediction. Specifically, the receiver efficiency sub-model includes a first mechanism sub-module, a first data sub-module, and a first fused efficiency calculation module.

[0076] For example, the first mechanism submodule calculates the theoretical thermal efficiency of the receiver based on the lumped parameter method; specifically, the mechanism calculation unit can be constructed based on equation (7): (7) In equation (7): or rec,physics Thermal efficiency calculated based on the mechanism. Q absorbed For the actual heat absorbed by the working fluid,Q solar,in For incident solar power, HTF For the working fluid flow rate, c p The specific heat capacity of the working fluid, T out This refers to the absorber outlet temperature. T in This refers to the inlet temperature of the absorber. ID card Normal direct irradiance, A aperture The area of ​​the light-receiving opening of the heat absorber. or opt,design To design optical efficiency.

[0077] Furthermore, considering the actual optical losses caused by cloud obstruction, mirror dust, tracking errors, etc., an optical efficiency correction factor as shown in equation (8) is introduced: (8) In equation (8), or opt,actua For actual optical efficiency, or opt,design To design optical efficiency, f cloud For cloud cover correction factor, f soiling For mirror dust correction factor f tracking This is the tracking error correction factor.

[0078] For example, the first data submodule uses a long short-term memory network to dynamically predict the computational residual of the first mechanism submodule under transient conditions. The input feature vector of the long short-term memory network model includes mechanism computation efficiency, temperature field distribution in multiple regions of the receiver wall, DNI time series and ambient wind speed.

[0079] Specifically, for the first data submodule, a Long Short-Term Memory (LSTM) network can be constructed as a dynamic residual prediction model. The input feature vector of the LSTM includes: the theoretical thermal efficiency ηrec,physics calculated and output by the first mechanism submodule, and the temperature field distribution of multiple regions on the absorber wall. DNI time series (first 10 time points), ambient wind speed V wind working fluid inlet temperature T in The LSTM model outputs the predicted residual value Δ. or AI, representing the dynamic deviation that the mechanism model cannot capture under transient conditions; for example, the residual prediction value can be expressed by equation (9): (9) In equation (9), or rec,physics The theoretical thermal efficiency calculated and output by the first mechanism submodule; T wall，1 ,…, T walln [This refers to the temperature field distribution across multiple regions of the absorber wall.] V wind For ambient wind speed, T in The inlet temperature of the working fluid. ID card (t-10: t ) represents the time series of DNI at the first 10 moments, Δ or AI Output the predicted residual values ​​for the LSTM model.

[0080] Furthermore, the fusion efficiency calculation sub-model sums the theoretical thermal efficiency of the receiver and the residual to obtain the real-time dynamic efficiency.

[0081] Meanwhile, to ensure the physical rationality of the prediction results, a physical constraint layer is introduced: the first fusion efficiency calculation module detects the real-time dynamic efficiency and residual prediction values ​​individually or jointly. If the calculated real-time dynamic efficiency exceeds the physically feasible range of [0,1], or the absolute value of the residual prediction value exceeds a preset threshold (such as |Δ... or AI If the value is greater than 0.15, the system will automatically mark it as an AI prediction anomaly, revert to the mechanism calculation result, and trigger the retraining of the residual prediction model.

[0082] For example, the first data submodule can be obtained by offline model training based on historical operating data of a solar thermal power plant; specifically, the training of the first data submodule can be achieved by the following steps: 1. Extract data segments from historical data during the normal operation of the receiver, and calculate ηrec,physics for each time step using the mechanism calculation unit.

[0083] 2. The residual between the measured value of the receiver outlet temperature and the predicted value of the mechanism model is used as the training target of the LSTM model.

[0084] 3. The LSTM network adopts a 2-layer structure with 128 hidden units in each layer. The dropout rate is set to 0.2 to prevent overfitting. The Adam optimizer is used for training, with the initial learning rate set to 0.001, the number of training epochs=100, and the batch size=64.

[0085] 4. During training, 80% of the data is used as the training set and 20% as the validation set. An early stopping mechanism is used to prevent overfitting.

[0086] In one embodiment, the receiver efficiency sub-model further includes a receiver anomaly diagnosis module. The receiver anomaly diagnosis module detects the residual value output by the first data sub-module. If the residual value output by the first data sub-module is continuously negative and its absolute value exceeds a first threshold, then a second threshold is compared based on the temperature sensor reading of a local area on the receiver wall with the historical average value of that area. If the temperature sensor reading of a local area on the receiver wall is continuously higher than the second threshold, the receiver anomaly diagnosis module outputs a diagnosis result of hot spots or uneven heat flux density on the receiver surface.

[0087] In one embodiment, if the residual value output by the first data submodule is consistently negative but the local temperature rise is not significant, it is diagnosed as overall contamination of the absorber surface or a decrease in the reflectivity of the mirror field.

[0088] In related technologies, methods for performance evaluation and diagnostic optimization of concentrated solar power (CSP) plants often use the average heat loss coefficient for calculating thermal storage efficiency. This fails to reflect the dynamic heat loss distribution caused by changes in the internal temperature gradient of the storage tank and local failures of the insulation layer. Furthermore, purely data-driven models may deviate from physical laws. To address this issue, as an exemplary embodiment, the driving efficiency model also includes a dynamic heat loss calculation sub-model for the thermal storage system. This sub-model includes a multiphysics sensing module, a dynamic heat loss calculation module, and a thermal storage and heat release calculation module. The multiphysics sensing module collects the temperature gradient distribution inside the storage tank and the temperature distribution of the tank wall. The dynamic heat loss calculation module takes into account the temperature distribution at each height level of the storage tank, the molten salt level H, the ambient temperature Tamb, the molten salt inlet / outlet temperature, and the charging / releasing mode identifier. Under the constraints of the laws of heat transfer physics, it outputs the heat loss coefficient representing the heat loss per unit time of the thermal storage system at the current moment, after data fitting. The thermal storage and heat release calculation module calculates the charging efficiency and the utilization efficiency on the power generation side.

[0089] In this embodiment, the multiphysics sensing module is used to collect the temperature field gradient distribution inside the tank and the temperature distribution of the tank wall. In one embodiment, a temperature sensor array can be arranged at different height layers of the molten salt tank to obtain the temperature field gradient distribution inside the tank [Tlayer1,Tlayer2,...,TlayerM], while a wall temperature sensor is arranged on the outer wall of the tank to obtain the temperature distribution of the tank wall.

[0090] In one embodiment, the dynamic heat loss calculation module can be constructed based on a Physics-Informed Neural Network (PINN). For example, in order for the dynamic heat loss calculation module to satisfy the physical laws of heat transfer while fitting data, the design of the loss function of PINN includes both a data fitting term and a physical constraint term. The data fitting term is used to represent the mean square error between the heat loss coefficient value output by the network and the measured heat loss value. The physical constraint term is constructed according to the one-dimensional unsteady heat conduction differential equation of the thermal storage system as shown in equation (10). (10) By incorporating the residuals (left side minus right side) of the above partial differential equations into the loss function, the prediction results of PINN satisfy the laws of heat transfer physics while fitting the data.

[0091] The input features of the PINN model include: temperature distribution at each height layer of the tank [Tlayer1,…,TlayerM], molten salt level H, ambient temperature Tamb, molten salt inlet / outlet temperature, and charging / discharging mode identifier.

[0092] The model output is a heat loss coefficient kloss(t) that changes dynamically over time, representing the heat loss per unit time of the thermal storage system at the current moment.

[0093] For example, Figure 2 This is a schematic diagram of the model structure of the PINN-based dynamic heat loss calculation module according to an embodiment of this application, as shown below. Figure 2 As shown, under this model structure, the dynamic heat loss calculation module can be obtained through offline model training based on historical operating data of the solar thermal power plant; specifically, the training of the first data sub-module can be achieved using the following steps: A physical information neural network model of a molten salt storage tank system is constructed. The network structure is as follows: input layer (feature dimension din=15) → hidden layer 1 (128 neurons, tanh activation) → hidden layer 2 (128 neurons, tanh activation) → hidden layer 3 (64 neurons, tanh activation) → output layer (1 neuron, linear activation).

[0094] The composite loss function is set as: Ltotal = Ldata + λphy·Lphysics, where Ldata is the mean square error between the predicted heat loss coefficient and the heat loss coefficient calculated by back-calculation through energy balance, Lphysics is the mean square error of the residual of the heat transfer differential equation, and λphy is the physical constraint weight coefficient (it is recommended to take λphy = 0.1~0.5).

[0095] Use automatic differentiation tools, such as TensorFlow's GradientTape or PyTorch's autograd, to compute the residuals of differential equations.

[0096] Training strategy: First, train for 100 rounds with a small λphy (e.g., 0.01) to allow the model to fit the data distribution; then gradually increase λphy to 0.3 to strengthen the physical constraints, and train for another 200 rounds.

[0097] In one embodiment, the thermal storage and heat release calculation module further includes a thermal storage system anomaly diagnosis module; the thermal storage system anomaly diagnosis module determines the state of the thermal storage and heat release system based on the difference between the heat loss coefficient output by the thermal storage and heat release calculation module and the benchmark value during the historical healthy operation period; specifically, when the heat loss coefficient kloss(t) output by the PINN model is continuously higher than the benchmark value during the historical healthy operation period, the system triggers an abnormal warning for thermal storage efficiency.

[0098] For example, the benchmark value can be obtained by statistically analyzing the average value of historical healthy operating periods; specifically, the benchmark value can be the statistical mean + 2σ.

[0099] In one embodiment, after the system triggers an abnormal warning of thermal storage efficiency, the thermal storage system abnormal diagnosis module performs a refined diagnosis by combining the temperature field gradient distribution; specifically, if the wall temperature of a certain height layer area of ​​the storage tank rises abnormally, and the temperature gradient of that area is significantly greater than that of other areas of the storage tank, and the heat loss coefficient remains high, then it is diagnosed as a local failure of the insulation layer in that area of ​​the molten salt storage tank.

[0100] If the wall temperature of all areas of the storage tank is higher than the benchmark value, and the heat loss coefficient increases overall, but the temperature gradient is uniformly distributed, then the overall insulation performance of the molten salt storage tank is diagnosed as declining.

[0101] If the temperature gradient at the top of the storage tank increases sharply and is accompanied by a rapid drop in liquid level, it is diagnosed as abnormal molten salt circulation or deterioration of molten salt pump performance.

[0102] The performance evaluation and diagnostic optimization methods for concentrated solar power plants in related technologies suffer from insufficient accuracy in turbine efficiency calculations under varying operating conditions. Furthermore, some schemes employing purely data-driven models may violate the second law of thermodynamics. Therefore, as an exemplary embodiment, the driving efficiency model further includes a dual-drive efficiency sub-model for turbines under varying operating conditions. This sub-model comprises a second mechanism sub-module, a second data sub-module, and a second fused efficiency calculation module. The second mechanism sub-module is established based on the reconstructed Flügel formula with the introduction of a temperature correction term. The second data submodule is constructed based on a multi-layer deep neural network. The input feature vector of the second data submodule includes meteorological parameters, real-time operating parameters of the steam turbine, and the calculation results of the second mechanism submodule. The output of the second data submodule is the steam turbine dynamic efficiency after physical constraint correction. The second fusion efficiency calculation module adopts an adaptive weighted fusion strategy to adaptively weight and fuse the second mechanism submodule and the second data submodule. Based on the operating condition identification classifier, the current operating stage is determined. Under quasi-steady state, the fusion weight of the second mechanism submodule is increased, and under transient changing operating conditions, the weight of the second data submodule is increased.

[0103] In this embodiment, the second mechanism submodule is established based on the reconstructed Freuger formula with the introduction of a temperature correction term; specifically, based on the reconstructed Freuger formula, a temperature correction term is introduced to establish the second mechanism submodule as shown in equation (13): (13) In equation (13), or 0 represents the turbine's design efficiency. m and m 0 represents the steam mass flow rate under actual operating conditions and the design operating conditions, respectively. P in and P out These are the actual inlet and outlet steam pressures under operating conditions. P in0 and P out0 The value corresponding to the design working condition. T in This refers to the actual steam inlet temperature under operating conditions. T in0 This refers to the inlet steam temperature under design operating conditions.

[0104] In this embodiment, the second data submodule is constructed based on a multi-layer deep neural network; specifically, a three-layer deep neural network can be constructed (with hidden layer dimensions of 128, 64, and 32 respectively), and the input feature vector is: ; in Metadata Includes meteorological data such as temperature, air pressure, humidity, and wind speed. fcond The output of the first data module is the condenser state parameters; the output of the second data module is the turbine dynamic efficiency after physical constraint correction. or turb,AI During the training process of the neural network, a thermodynamic constraint regularization term is introduced: if the model's prediction efficiency... or turb,AI Efficiency higher than the theoretical limit calculated based on the second law of thermodynamics or carnot If this is not the case, then an additional penalty loss is imposed to ensure that the prediction results of the second data submodule conform to the fundamental thermodynamic constraints.

[0105] In this embodiment, the second fusion efficiency calculation module adopts an adaptive weighted fusion strategy to adaptively weight the second mechanism submodule and the second data submodule; specifically, the second data submodule can adaptively weight the second mechanism submodule and the second data submodule using equation (14): (14) In equation (14), or turb,fused The real-time dynamic efficiency output by the second fusion efficiency calculation module; or turb,physics The efficiency output by the second mechanism submodule, w 1 represents the adaptive weighted fusion weights of the calculation model in the second mechanism submodule. w 2. Adaptive weighted fusion weights of the second data submodule or turb,AI This refers to the efficiency output by the second data submodule.

[0106] Among them, the weighting coefficient w 1 and w 2. Adaptive adjustment based on operating conditions: Under quasi-steady-state conditions close to the design conditions, increase the weight of the second mechanism submodule; under transient and variable operating conditions, increase the weight of the second data submodule.

[0107] For example, the weight adjustment is achieved through a condition identification classifier that determines the current operating stage based on the fluctuation range and rate of change of steam parameters.

[0108] For example, the second data submodule can be obtained by offline model training based on historical operating data of a solar thermal power plant; specifically, the training of the second data submodule can be achieved using the following steps: 1. A three-layer fully connected neural network is used, with hidden layer dimensions of 128, 64, and 32, respectively, and ReLU is used as the activation function.

[0109] 2. The training data covers the typical variable operating conditions of the steam turbine (50%~100% rated load) to ensure the model's generalization ability under variable operating conditions.

[0110] 3. The loss function is the mean square error between the predicted efficiency and the measured efficiency, with an additional physical constraint regularization term: if the predicted efficiency ηpred > ηcarnot, then an additional penalty loss Lphy = α·max(0, ηpredηcarnot)^2 is applied, where α is the penalty coefficient.

[0111] 4. L2 weight regularization is used during training, with a regularization coefficient λ=0.0001.

[0112] In one embodiment, the turbine variable operating condition dual-drive efficiency sub-model further includes a turbine anomaly diagnosis module; the turbine anomaly diagnosis module determines the state of the turbine system based on the difference between the real-time dynamic efficiency output by the turbine variable operating condition dual-drive efficiency sub-model and the benchmark value during the historical healthy operation period; specifically, when the real-time dynamic efficiency output by the turbine variable operating condition dual-drive efficiency sub-model is continuously higher than the benchmark value during the historical healthy operation period, the system triggers a turbine performance degradation warning.

[0113] In one embodiment, after the system triggers a turbine performance degradation warning, the turbine anomaly diagnosis module performs a refined diagnosis by combining multi-dimensional parameters: if the main steam parameters (pressure, temperature) are normal, but the exhaust steam temperature rises, the condenser vacuum decreases, and the efficiency decreases, then the diagnosis is a decrease in condenser heat exchange efficiency or performance degradation of the air-cooled island.

[0114] If the main steam temperature and pressure are normal, the exhaust parameters are normal, but the efficiency decreases, and this decrease shows an irreversible trend of accumulation over time, then the diagnosis is scaling in the turbine's flow passage or blade wear. If the main steam temperature and pressure deviate abnormally from the design values, the operating status of the upstream receiver and steam generation system should be checked first.

[0115] Secondly, this application provides a mechanism-data fusion-based solar thermal power plant performance diagnosis system. This mechanism-data fusion-based solar thermal power plant performance diagnosis system belongs to the same inventive concept as the mechanism-data fusion-based solar thermal power plant performance diagnosis method in the above embodiments. For details not described in detail in the embodiments of the mechanism-data fusion-based solar thermal power plant performance diagnosis system, please refer to the embodiments of the above mechanism-data fusion-based solar thermal power plant performance diagnosis method.

[0116] For example, Figure 3 This is a schematic block diagram of a solar thermal power plant performance diagnosis system based on mechanism-data fusion according to an embodiment of this application, such as... Figure 3 As shown, the system includes: The acquisition module 301 is used to acquire multi-source heterogeneous data of the target solar thermal power plant during its operation; wherein, the multi-source heterogeneous data includes meteorological data corresponding to the target solar thermal power plant and system operation data of multiple target systems contained in the target solar thermal power plant.

[0117] The driving efficiency calculation module 302 is used to drive the driving efficiency model corresponding to each target system based on multi-source heterogeneous data, and obtain the real-time dynamic efficiency of each target system; wherein, at least one driving efficiency model is obtained by fusing the mechanism submodule and the data submodule.

[0118] The fusion diagnosis module 303 is used to input the real-time dynamic efficiency corresponding to each of the target systems into the pre-built fusion diagnosis model. Based on the efficiency deviation detection module, correlation analysis module and efficiency bottleneck identification module of the fusion diagnosis model, the overall efficiency of the target solar thermal power plant is decomposed into deviation and the contribution of the target system, and the fusion diagnosis results are obtained.

[0119] The performance diagnostic module 304 is used to obtain the overall performance diagnostic results of the target solar thermal power plant based on the fusion diagnostic results; the overall performance diagnostic results include efficiency bottleneck location, root cause analysis and optimization maintenance suggestions.

[0120] The mechanism-data fusion-based solar thermal power plant performance diagnosis system in this embodiment is used to implement a mechanism-data fusion-based solar thermal power plant performance diagnosis system method.

[0121] This application also provides an electronic device for implementing the various embodiments of this application. Figure 4 To illustrate the hardware structure of an electronic device according to various embodiments of this application, as shown in the following diagram... Figure 4 As shown, the electronic device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor.

[0122] This application also provides a storage medium storing a program product capable of implementing embodiments of the method of the present invention. In some possible implementations, various aspects of this application may also be implemented as a program product including program code, which, when run on a terminal device, causes the terminal device to perform the steps of the various exemplary embodiments of the present application described in the foregoing exemplary method section of this specification.

[0123] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A performance diagnosis method for solar thermal power plants based on mechanism-data fusion, characterized in that, The performance diagnosis method for the solar thermal power plant includes: Step S1: Acquire multi-source heterogeneous data of the target solar thermal power plant during operation; wherein, the multi-source heterogeneous data includes meteorological data and system operation data; the system operation data includes operating sub-data of the concentrating solar thermal collection system, operating sub-data of the thermal storage system, and operating sub-data of the power generation system; Step S2: Based on the multi-source heterogeneous data, drive the corresponding driving efficiency model of each target system to obtain the real-time dynamic efficiency of each target system; wherein, at least one driving efficiency model is obtained by fusing the mechanism submodule and the data submodule; wherein, the target system includes a concentrating solar thermal system, a thermal storage system, and a power generation system; Step S3: Input the real-time dynamic efficiency corresponding to each target system into the pre-built fusion diagnostic model. Based on the efficiency deviation detection module, correlation analysis module and / or efficiency bottleneck identification module of the fusion diagnostic model, perform deviation detection, target system contribution decomposition and / or efficiency bottleneck identification on the overall efficiency of the target solar thermal power plant to obtain the fusion diagnostic result. Step S4: Based on the fusion diagnostic results, obtain the overall performance diagnostic results of the target solar thermal power plant; wherein, the overall performance diagnostic results include efficiency bottleneck location, root cause analysis, and optimization and maintenance recommendations.

2. The performance diagnosis method for solar thermal power plants based on mechanism-data fusion according to claim 1, characterized in that, Step S3 further includes: Step S31: The efficiency deviation detection module calculates the overall efficiency of the power station based on the pre-built overall energy balance formula of the power station, and performs deviation detection on the overall efficiency of the power station. Step S32: When the overall efficiency of the power plant is lower than the preset threshold, the contribution percentage of each efficiency deviation to the overall efficiency decline is quantitatively calculated by the correlation analysis module through efficiency deviation contribution decomposition, and the target system with the largest contribution is automatically marked as the efficiency bottleneck; wherein, the contribution percentage of each system efficiency deviation to the overall efficiency decline is calculated by the real-time dynamic efficiency of each target system and the corresponding historical dynamic efficiency. Step S33: The efficiency bottleneck identification module calculates the efficiency loss of each system and sorts them in real time, marking the target system whose efficiency loss ratio exceeds the preset threshold as the efficiency bottleneck system. Step S34: Determine the fusion diagnostic results based on efficiency bottlenecks and efficiency shortcomings.

3. The performance diagnosis method for solar thermal power plants based on mechanism-data fusion according to claim 2, characterized in that, The correlation analysis module uses the following formula to decompose the contribution of efficiency deviation: In the formula, This indicates the overall efficiency deviation of the power plant. This indicates the overall efficiency of the power plant. This indicates the efficiency deviation of the receiver system. This indicates the real-time dynamic efficiency of the heat absorber system. This indicates the efficiency deviation of the thermal storage system. This indicates the real-time dynamic efficiency of the thermal storage system. This indicates the efficiency deviation of the thermal storage system. This indicates the real-time dynamic efficiency of the steam turbine system; To assist in the overall efficiency deviation of the auxiliary system, To improve the overall efficiency of the auxiliary system; The partial derivative coefficients were determined through sensitivity analysis based on historical data, and ε represents the unmodeled residual term.

4. The performance diagnosis method for solar thermal power plants based on mechanism-data fusion according to claim 2, characterized in that, The pre-constructed overall energy balance formula for the power plant is derived from the following formula: In the formula, The actual power output of the target solar thermal power plant. The overall efficiency of the target solar thermal power plant. DNI Indicates the direct normal radiation of the sun. The area of ​​the light-receiving opening of the heat absorber. The overall efficiency of the target solar thermal power plant is calculated using the following formula: in, The overall efficiency of the target solar thermal power plant. For the heat absorber fusion efficiency, The overall efficiency of heat charging and discharging in the thermal storage system, For the efficiency of steam turbine integration, To improve the overall efficiency of the auxiliary system.

5. The performance diagnosis method for solar thermal power plants based on mechanism-data fusion according to claim 1, characterized in that, The model-driven efficiency model includes a receiver efficiency sub-model; the receiver efficiency sub-model includes a first mechanism sub-module, a first data sub-module, and a first fusion efficiency calculation module. The first mechanism submodule calculates the theoretical thermal efficiency of the receiver based on the lumped parameter method; The first data submodule uses a long short-term memory network to dynamically predict the computational residual of the first mechanism submodule under transient conditions. The input feature vector of the long short-term memory network model includes mechanism computation efficiency, temperature field distribution in multiple regions of the receiver wall, DNI time series and ambient wind speed. The first fusion efficiency calculation module sums the theoretical thermal efficiency of the absorber and the residual to obtain the real-time dynamic efficiency; and the first fusion efficiency calculation module is used to verify the rationality of the fusion efficiency of the fusion efficiency calculation unit. If the fusion efficiency exceeds the physically feasible range or the absolute value of the residual prediction exceeds the preset threshold, it will fall back to the mechanism calculation result and trigger model retraining.

6. The performance diagnosis method for solar thermal power plants based on mechanism-data fusion according to claim 1, characterized in that, The driving efficiency model also includes a sub-model for calculating dynamic heat loss of the thermal storage system; the sub-model for calculating dynamic heat loss of the thermal storage system includes a multi-physics sensing module, a dynamic heat loss calculation module, and a thermal storage and heat release calculation module. The multi-physics sensing module is used to collect the temperature gradient distribution inside the storage tank and the temperature distribution of the tank wall; The dynamic heat loss calculation module is used to input the temperature distribution of each height layer of the storage tank, the molten salt level, the ambient temperature, the molten salt inlet / outlet temperature, and the heat charging / discharging mode identifier. Under the constraints of the laws of heat transfer physics, it outputs the heat loss coefficient representing the heat loss per unit time of the thermal storage system at the current moment after data fitting. The thermal storage and heat release calculation module is used to calculate charging efficiency and power generation utilization efficiency.

7. The performance diagnosis method for solar thermal power plants based on mechanism-data fusion according to claim 1, characterized in that, The driving efficiency model also includes a turbine variable operating condition dual-drive efficiency sub-model, which includes a second mechanism sub-module, a second data sub-module, and a second fusion efficiency calculation module. The second mechanism submodule is established based on the reconstructed Freuger formula with the introduction of a temperature correction term; The second data submodule is constructed based on a multi-layer deep neural network. The input feature vector of the second data submodule includes meteorological parameters, real-time operating parameters of the steam turbine, and the calculation results of the second mechanism submodule. The output of the second data submodule is the steam turbine dynamic efficiency after physical constraint correction. The second fusion efficiency calculation module adopts an adaptive weighted fusion strategy to adaptively weight and fuse the second mechanism submodule and the second data submodule. Based on the working condition identification classifier to determine the current operating stage, the fusion weight of the second mechanism submodule is increased in the quasi-steady state, and the weight of the second data submodule is increased in the transient changing working condition.

8. A performance diagnostic system for solar thermal power plants based on mechanism-data fusion, characterized in that, The system is used to implement the mechanism-data fusion-based performance diagnosis method for solar thermal power plants as described in any one of claims 1 to 7; the system includes: The acquisition module is used to acquire multi-source heterogeneous data of the target solar thermal power plant during its operation; wherein, the multi-source heterogeneous data includes meteorological data and system operation data; the system operation data includes sub-data of the concentrating solar collector system operation, sub-data of the thermal storage system operation, and sub-data of the power generation system operation. The driving efficiency calculation module drives the driving efficiency model corresponding to each of the target systems based on the multi-source heterogeneous data, and obtains the real-time dynamic efficiency of each target system; wherein, the at least one driving efficiency model is obtained by fusing the mechanism submodule and the data submodule; wherein, the target system includes a concentrating solar thermal system, a thermal storage system, and a power generation system; The fusion diagnostic module is used to input the real-time dynamic efficiency of each target system into the pre-built fusion diagnostic model. Based on the efficiency deviation detection module, correlation analysis module and efficiency bottleneck identification module of the fusion diagnostic model, the overall efficiency of the target solar thermal power plant is decomposed into deviation and target system contribution, and the fusion diagnostic results are obtained. The performance diagnostic module is used to obtain the overall performance diagnostic results of the target solar thermal power plant based on the fused diagnostic results; wherein, the overall performance diagnostic results include efficiency bottleneck location, root cause analysis and optimization and maintenance suggestions.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor is used to execute a computer program, it implements the steps of the mechanism-data fusion-based solar thermal power plant performance diagnosis method as described in any one of claims 1 to 7.

10. A storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the mechanism-data fusion-based solar thermal power plant performance diagnosis method as described in any one of claims 1 to 7.

Images