A Thermodynamic State Inversion Early Warning Method for Molten Salt Energy Storage Systems Based on Temperature Detection
By combining wireless temperature measurement on the outer wall with a multi-layer circular tube heat conduction model and data-driven residual correction, the risks of invasive monitoring and the problem of freezing and blockage lag in molten salt energy storage systems have been solved. This has enabled non-invasive panoramic monitoring and adaptive early warning of molten salt energy storage systems, improving the intelligence and safety of the system.
Patent Information
- Application Number
- CN202611600905.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-16
- Publication Date
- 2026-11-17
AI Technical Summary
Existing monitoring methods for molten salt energy storage systems rely on invasive sensors, which have problems such as leakage risk, limited monitoring dimensions, delayed early warning of freezing and blockage, lack of credibility of inversion results, and difficulty in models following system performance drift.
A non-invasive monitoring method based on temperature detection is adopted. By wirelessly measuring the temperature of the outer wall surface, combined with a multi-layer circular tube heat conduction model, data-driven residual correction and uncertainty quantification, the thermodynamic state of the molten salt system is inverted to achieve intelligent early warning of freezing blockage risk.
It achieves non-intrusive panoramic monitoring, adaptive multi-threshold early warning, and early identification of freezing blockage and leakage, thereby improving the system's intelligence level and operational safety.
Smart Images

Figure CN123111497A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of molten salt energy storage technology, and in particular to a method for thermodynamic state inversion and early warning of molten salt energy storage systems based on temperature detection. Background Technology
[0002] Currently, monitoring of molten salt tanks and pipelines in molten salt energy storage systems mainly relies on invasive temperature sensors and point measurements. Existing methods have the following drawbacks: Limited monitoring dimensions: It can only acquire the temperature of discrete measurement points and cannot know the key thermodynamic states such as the three-dimensional temperature field distribution inside the system and the location of the molten salt phase change interface, resulting in a one-sided and lagging understanding of the system's operating status.
[0003] Safety hazard: Invasive sensors require drilling holes in tanks or pipes for installation, increasing the risk of molten salt leakage.
[0004] Insufficient fault early warning capability: For critical faults such as freezing blockage and tank structure deformation, existing technologies mostly rely on simple temperature threshold alarms or require active intervention methods such as electric heating. The early warning time is late, and it is impossible to perform online and non-destructive diagnosis of early damage to the tank, pipeline and other structures themselves.
[0005] Therefore, there is an urgent need for a method that can non-invasively, online, and comprehensively sense the thermodynamic state of molten salt systems and achieve early and accurate fault warnings. Summary of the Invention
[0006] Therefore, the purpose of this invention is to provide a thermodynamic state inversion and early warning method for molten salt energy storage systems based on temperature detection, which breaks through the limitations of the existing "point temperature measurement + simple physical model" and solves the problems in the monitoring of existing molten salt energy storage systems, such as leakage risk due to invasive temperature measurement, limited monitoring dimensions, delayed freezing and blockage early warning, lack of credibility assessment of inversion results, difficulty in the model to follow system performance drift, and waste of heat tracing energy due to indiscriminate handling.
[0007] To achieve the above objectives, this invention provides a thermodynamic state inversion and early warning method for molten salt energy storage systems based on temperature detection, comprising the following steps: S1. Collect temperature data of the outer wall surface of multiple monitoring points on the outer wall surface of molten salt storage tanks and / or molten salt pipelines, and calculate their time derivatives; S2. Input the outer wall temperature data and time derivative into the thermodynamic state inversion model. The thermodynamic state inversion model includes a physical base value layer, a data-driven residual correction layer, and an uncertainty quantification layer. The physical base value layer explicitly couples the temperature-dependent molten salt physical property parameters to the multilayer circular tube heat conduction model in the form of a polynomial fitting function. The steady-state inversion temperature is obtained through fixed-point iteration and transient thermal hysteresis compensation, and the physical base value inversion result is finally output. The data-driven residual correction layer takes the physical base value inversion result and the outer wall temperature characteristics as input, uses a residual learning network, outputs the residual correction amount and superimposes it with the physical base value inversion result to obtain the corrected expected inversion temperature. The uncertainty quantification layer outputs the expected inversion temperature, the standard deviation of the expected residual, and the confidence interval of the expected inversion temperature. S3. Based on the desired inversion temperature, construct a continuous temperature curve along the molten salt flow direction, solve for the liquid-solid interface position, and propagate the error of the standard deviation of the desired inversion temperature using the same interpolation and root-finding process to obtain the confidence interval of the liquid-solid interface position; calculate the phase interface migration rate based on the phase interface position at adjacent sampling times. S4. Determine the rate of decrease in heat flux density through the wall based on the inversion temperature; normalize the phase interface movement rate and the rate of decrease in heat flux density respectively, and then sum them by weight to obtain the freezing blockage risk index; S5. Compare the freezing and blockage risk index with multiple thresholds to determine the freezing and blockage risk level, and execute differentiated response actions according to the determined freezing and blockage risk level. S6. Compare the standard deviation with the confidence threshold, and decide whether the thermodynamic state inversion model needs to be updated online based on the comparison result.
[0008] More preferably, in S1, the monitoring points are evenly distributed on the outer wall of the molten salt storage tank and / or the outer wall of the pipeline, and the temperature sensor is attached to the monitoring point for continuous data collection. The temperature sensor is an attached wireless platinum resistance temperature sensor.
[0009] Further preferably, in S2, the multi-layer circular tube heat conduction model described in the physical base layer, when the inner diameter of the j-th layer tube wall is r... j,i The outer diameter is r j,o The thermal conductivity is λ j and containing internal heat sources At steady state, the heat conduction equation is expressed by the following formula (1): Formula (1) By integrating the above equation twice and utilizing the interface temperature and heat flow continuity condition of each layer, the temperature distribution within the j-th layer can be obtained. After connecting multiple layers in series, by extrapolating from the outer wall temperature data Tw,o inward layer by layer, the general formula of the final multi-layer circular tube heat conduction model is obtained as follows: Formula (2) Where B is the total number of pipe wall layers; q′′ N This represents the outermost heat flux density. For internal heat sources, the value is zero when there is no electric heating; r j,i r j,oLet λ represent the inner and outer radii of the j-th layer, respectively; j h is the thermal conductivity of the j-th layer material; i The convective heat transfer coefficient between the molten salt inside the pipe and the inner wall surface; unit: W / (m²). 2 ·K); This represents the outer radius of the (j-1)th layer; Let be the heat flux density of the (j-1)th layer.
[0010] Further preferably, in S2, in the physical base layer, the temperature-dependent molten salt physical property parameters are explicitly coupled to the multilayer circular tube heat conduction model in the form of a polynomial fitting function. The molten salt physical property parameters include the molten salt thermal conductivity λ, specific heat capacity cp, density ρ, and dynamic viscosity η. Specifically, the process includes the following: using the measured physical property parameter table of the molten salt medium, a temperature-dependent polynomial fitting function is established. A quadratic polynomial can be used for fitting: Thermal conductivity: Formula (3) Specific heat capacity: Formula (4) density: Formula (5) Dynamic viscosity: Formula (6) Set the initial iteration temperature T f (0) =Tw,o+ΔT0; , , , , , , , , , , All coefficients are determined by fitting measured physical property parameter tables; The independent variable is temperature; In the k-th iteration, the current temperature Tf (k) Substitute each physical property parameter into the fitting formula to update the physical property parameters. Substitute the updated physical property parameters into the multilayer circular tube heat conduction model for inversion again. Repeat this process until convergence is obtained to obtain the steady-state inversion temperature. .
[0011] More preferably, the fixed-point iteration is based on = + The initial iteration temperature is [temperature value], at the [number]th iteration... In the iteration, the th Temperature obtained from -1st iteration The fitted functions of each physical property parameter are substituted to update the physical property parameters. The updated physical property parameters are then substituted back into the multilayer circular tube heat conduction model for further derivation until the absolute value of the difference between two adjacent iterations is less than the convergence criterion. Or reach the number of iterations Formula (7) Output physical basis value inversion results ; Formula (8) Formula (9) Where, τ sys The system's thermal time constant is determined by the thermal diffusivity and geometric dimensions of each layer of material. For steady-state inversion temperature, For the first Layer thickness, For the first Thermal diffusivity of the layer material.
[0012] More preferably, in the data-driven residual correction layer, the residual learning network adopts a PGRN (Physics-Guided Residual Network) network with a 4-layer fully connected residual structure, consisting of an input layer, a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, and a single-neuron output layer. Each hidden layer contains 64 nodes and uses the ReLU activation function. Furthermore, a skip connection residual branch is set between the second hidden layer and the fourth hidden layer to alleviate gradient vanishing. The network's input feature vector dimensions include: M-dimensional spatial temperature distribution, M-1-dimensional temperature difference between adjacent measuring points, 2-dimensional environmental parameters, 3-dimensional time-series statistics for the first K periods, and 1-dimensional physical basis value inversion result T. f,phys And a 2D running status identifier; the training label is set as the difference between the actual internal temperature and the temperature inverted by the physical model in the CFD simulation sample, and the training loss function is constructed as follows: Formula (10) Among them, L MSE L is the mean square error between the predicted residual value and the true residual label. physics As a physical consistency constraint, the mean of the residuals of all samples is forced to be zero to ensure that the correction process does not introduce systematic bias, and the tradeoff coefficient γ is set to 0.1.
[0013] More preferably, in S2, the uncertainty quantification layer outputs the expected inversion temperature, the standard deviation of the expected residual, and the confidence interval of the expected inversion temperature, including: Expected inversion temperature Formula (11) Standard deviation of expected residuals Formula (12) The expected confidence interval for the inversion temperature is: Formula (13) In the formula, For the first The residual correction obtained from each forward propagation, where N is the total number of propagations and S is the independent variable representing the number of propagations.
[0014] Further preferred, the corrected expected inversion temperature output by S2 at each discrete monitoring section along the molten salt flow direction is used. ( , Using nodes as nodes, construct a cubic spline interpolation function. ( This yields a temperature curve that is continuously distributed along the molten salt flow direction, and the initial crystallization point temperature of the molten salt is determined. Substituting the temperature curve, we obtain the phase interface determination equation: Formula (14) The real roots of this equation in the axial coordinate interval are solved using numerical root-finding methods. ( ), which represents the current position of the molten salt-solid interface along the molten salt flow direction; the standard deviation obtained from S2 is... By propagating the error using the same interpolation and root-finding process, the confidence interval for the position of the liquid-solid interface is obtained, and the error propagation formula is as follows: Formula (15) The confidence interval for the location of the liquid-solid interface is: Formula (16) The phase interface movement rate is obtained by dividing the difference between the phase interface positions at two adjacent sampling times by the sampling interval. Formula (17) More preferably, in S4, the heat flux density through the wall is determined based on the distribution of the corrected inversion temperature along the molten salt flow direction output from S2. The rate of decrease in heat flux density is obtained by differentiating with respect to time. The phase interface movement rate ( ) and the rate of decrease of the heat flux density Normalized to their historical maximum values respectively: Formula (18) Normalized phase interface migration rate With normalized heat flux density decrease rate By weighting factor , Weighted summation yields the freezing and blockage risk index. : Formula (19) in, + =1, and , The value of is determined by experimental calibration.
[0015] Further preferably, in S5, the freezing and blockage risk index is... respectively with the first threshold Second threshold and the third threshold The comparison is used to determine the level of freezing and blockage risk, among which < < ,and , , The freezing risk index corresponding to the time of occurrence of historical freezing events is used as the calibration sample, and the distribution quantile of the calibration sample is used to determine the risk index. The differentiated response actions include: when ≤ < When the first sampling period is adjusted to a second sampling period that is shorter than the first sampling period, the preheating of the heat tracing device installed on the molten salt storage tank and / or molten salt pipeline is started. when ≤ < At that time, based on the previous level of response action, the molten salt heat extraction power is reduced and / or the molten salt flow rate is increased; when ≥ At that time, based on the previous level of response action, the molten salt pipe section where the liquid-solid interface is located is isolated and a freezing and blockage warning signal is output; In S6, the standard deviation is... Each with the first confidence threshold Second confidence threshold In comparison, among them < :when < At that time, the inversion temperature, liquid-solid interface position, and freezing risk index obtained from S2 to S5 are adopted; when ≤ < At that time, the expected inversion temperature is cross-validated using the outer wall temperature data from other monitoring points before acceptance; when ≥ When the inversion quality deteriorates, an alert is output, and the residual learning network is updated online using the outer wall temperature data collected in the current sampling period.
[0016] The thermodynamic state inversion and early warning method for molten salt energy storage systems based on temperature detection disclosed in this application has at least the following advantages compared with the prior art: 1. This application achieves non-invasive panoramic monitoring: by wirelessly measuring the temperature of the outer wall surface, the complete internal temperature field, heat flux density field, and phase interface are inferred, realizing a "penetrating view" of the system's thermodynamic state and avoiding the leakage risk caused by openings. The wirelessly attached sensor is easy to install, not limited by existing structures, and can be flexibly deployed in critical parts such as tanks and pipelines, making it suitable for new construction and renovation projects.
[0017] 2. The early warning mechanism of this invention no longer relies on a fixed single temperature threshold, but is based on deep thermodynamic parameters obtained through inversion, thus achieving adaptive, multi-threshold intelligent early warning. For example, by monitoring regional sudden drops in heat flux density and phase interface movement, early warning of freezing blockage can be provided; by tracking abnormally rapid rises in local temperature fields and abnormal heat flow directions, early warning of local overheating can be provided; by analyzing unique "cold spot" diffusion patterns related to leakage appearing in the outer wall temperature field, potential precursors to molten salt leakage can be identified. This provides a longer response time window for the system to deal with various types of faults.
[0018] 3. This application enables early warning of structural damage such as tank deformation and crack initiation, truly preventing problems before they occur. It provides a higher-level and more forward-looking decision-making basis for the control systems of actuators such as molten salt pumps, electric heat tracing, and electric heaters, improving the intelligence level and operational safety of the entire energy storage system. Attached Figure Description
[0019] Figure 1 This is a schematic flowchart of the thermodynamic state inversion and early warning method for molten salt energy storage systems based on temperature detection, as described in this invention.
[0020] Figure 2 This is a schematic diagram of the inversion results of the thermodynamic state inversion model of the present invention. Detailed Implementation
[0021] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0022] like Figure 1 As shown, one embodiment of the present invention provides a thermodynamic state inversion and early warning method for molten salt energy storage systems based on temperature detection, comprising the following steps: S1, collect temperature data of the outer wall surface of multiple monitoring points on the outer wall surface of molten salt storage tank and / or molten salt pipeline, and calculate its time derivative.
[0023] More preferably, in S1, the monitoring points are evenly distributed on the outer wall of the molten salt storage tank and / or the outer wall of the pipeline, and the temperature sensor is attached to the monitoring point for continuous data collection. The temperature sensor is an attached wireless platinum resistance temperature sensor.
[0024] S2. Input the outer wall temperature data and time derivative into the thermodynamic state inversion model. The thermodynamic state inversion model includes a physical base value layer, a data-driven residual correction layer, and an uncertainty quantification layer. The physical base value layer explicitly couples the temperature-dependent molten salt physical property parameters to the multilayer circular tube heat conduction model in the form of a polynomial fitting function. The steady-state inversion temperature is obtained through fixed-point iteration and transient thermal hysteresis compensation, and the physical base value inversion result is finally output. Further, in the multi-layer circular tube heat conduction model, when the j-th layer of the tube wall has an inner diameter of r... j,i The outer diameter is r j,o The thermal conductivity is λ j and containing internal heat sources At steady state, the heat conduction equation is expressed by the following formula (1): Formula (1) As the independent variable representing temperature, The independent variable representing the radius of the multilayer circular tube heat conduction model; By integrating the above equation twice and utilizing the interface temperature and heat flow continuity condition of each layer, the temperature distribution within the j-th layer can be obtained. After connecting multiple layers in series, by extrapolating from the outer wall temperature data Tw,o inward layer by layer, the general formula of the final multi-layer circular tube heat conduction model is obtained as follows: Formula (2) Where B is the total number of pipe wall layers; q′′ N This represents the outermost heat flux density. For internal heat sources, the value is zero when there is no electric heating; r j,i r j,o Let λ represent the inner and outer radii of the j-th layer, respectively; j h is the thermal conductivity of the j-th layer material; i The convective heat transfer coefficient between the molten salt inside the pipe and the inner wall surface; unit: W / (m²). 2 ·K); This represents the outer radius of the (j-1)th layer; Let be the heat flux density of the (j-1)th layer.
[0025] In S2, in the physical base layer, the temperature-dependent molten salt physical property parameters are explicitly coupled to the multilayer circular tube heat conduction model in the form of a polynomial fitting function. The molten salt physical property parameters include the molten salt thermal conductivity λ, specific heat capacity cp, density ρ, and dynamic viscosity η. Specifically, the process includes the following: using the measured physical property parameter table of the molten salt medium, a temperature-dependent polynomial fitting function is established. A quadratic polynomial can be used for fitting: Thermal conductivity: Formula (3) Specific heat capacity: Formula (4) density: Formula (5) Dynamic viscosity: Formula (6) Set the initial iteration temperature T f (0) =Tw,o+ΔT0; , , , , , , , , , , All coefficients are determined by fitting measured physical property parameter tables; The independent variable is temperature; In the k-th iteration, the current temperature Tf (k) Substitute each physical property parameter into the fitting formula to update the physical property parameters. Substitute the updated physical property parameters into the multilayer circular tube heat conduction model for inversion again. Repeat this process until convergence is obtained to obtain the steady-state inversion temperature. .
[0026] Fixed-point iteration = + The initial iteration temperature is [temperature value], at the [number]th iteration... In the iteration, the th Temperature obtained from -1st iteration The fitted functions of each physical property parameter are substituted to update the physical property parameters. The updated physical property parameters are then substituted back into the multilayer circular tube heat conduction model for further derivation until the absolute value of the difference between two adjacent iterations is less than the convergence criterion. Or reach the number of iterations Formula (7) Output physical basis value inversion results ; Formula (8) Formula (9) in, For the first The temperature obtained from the -1st iteration; τ sys The system's thermal time constant is determined by the thermal diffusivity and geometric dimensions of each layer of material. For steady-state inversion temperature, For the first Layer thickness, For the first The thermal diffusivity of the layer material, where R is the total number of layers.
[0027] The data-driven residual correction layer takes the physical base value inversion result and the outer wall temperature characteristics as input, uses a residual learning network, outputs the residual correction amount and superimposes it with the physical base value inversion result to obtain the corrected expected inversion temperature. The residual learning network adopts the PGRN (Physics-Guided Residual Network) network, which uses a 4-layer fully connected residual structure, consisting of an input layer, a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, and a single neuron output layer. Each hidden layer contains 64 nodes and uses the ReLU activation function. A skip connection residual branch is set between the second hidden layer and the fourth hidden layer to alleviate gradient vanishing. The network's input feature vector dimensions include: M-dimensional spatial temperature distribution, M-1-dimensional temperature difference between adjacent measuring points, 2-dimensional environmental parameters, 3-dimensional time-series statistics for the first K periods, and 1-dimensional physical basis value inversion result T. f,phys And a 2D running status identifier; the training label is set as the difference between the actual internal temperature and the temperature inverted by the physical model in the CFD simulation sample, and the training loss function is constructed as follows: Formula (10) Among them, L MSE L is the mean square error between the predicted residual value and the true residual label. physics As a physical consistency constraint, the mean of the residuals of all samples is forced to be zero to ensure that the correction process does not introduce systematic bias, and the tradeoff coefficient γ is set to 0.1.
[0028] In the formula, The weighting factor is 0.1; M is the number of training samples. For the first The residual correction for each sample; For the first The true residual label of each sample is determined by the difference between the true temperature inside the molten salt in the sample pair and the inversion result of the physical baseline value.
[0029] The uncertainty quantification layer outputs the expected inversion temperature, the standard deviation of the expected residual, and the confidence interval of the expected inversion temperature; specifically including: Expected inversion temperature Formula (11) Standard deviation of expected residuals Formula (12) The expected confidence interval for the inversion temperature is: Formula (13) In the formula, For the first The residual correction obtained from each forward propagation, where N is the total number of propagations.
[0030] S3. Based on the desired inversion temperature, construct a continuous temperature curve along the molten salt flow direction, solve for the liquid-solid interface position, and propagate the error of the standard deviation of the desired inversion temperature using the same interpolation and root-finding process to obtain the confidence interval of the liquid-solid interface position; calculate the phase interface migration rate based on the phase interface position at adjacent sampling times. The corrected expected inversion temperature output by S2 at each discrete monitoring section along the molten salt flow direction. ( , Using nodes as nodes, construct a cubic spline interpolation function. ( This yields a temperature curve that is continuously distributed along the molten salt flow direction, and the initial crystallization point temperature of the molten salt is determined. Substituting the temperature curve, we obtain the phase interface determination equation: Formula (14) The real roots of this equation in the axial coordinate interval are solved using numerical root-finding methods. ( ), which represents the current position of the molten salt-solid interface along the molten salt flow direction; the standard deviation obtained from S2 is... By propagating the error using the same interpolation and root-finding process, the confidence interval for the position of the liquid-solid interface is obtained, and the error propagation formula is as follows: Formula (15) The confidence interval for the location of the liquid-solid interface is: Formula (16) The phase interface movement rate is obtained by dividing the difference between the phase interface positions at two adjacent sampling times by the sampling interval. Formula (17).
[0031] S4. Determine the rate of decrease in heat flux density through the wall based on the inversion temperature; normalize the phase interface movement rate and the rate of decrease in heat flux density respectively, and then sum them by weight to obtain the freezing blockage risk index. The heat flux density through the wall is determined based on the distribution of the corrected inversion temperature along the molten salt flow direction from the S2 output. The rate of decrease in heat flux density is obtained by differentiating with respect to time. The phase interface movement rate ( ) and the rate of decrease of the heat flux density Normalized to their historical maximum values respectively: Formula (18) The normalized phase interface migration rate and the normalized heat flux density decrease rate were weighted by coefficients. , Weighted summation yields the freezing and blockage risk index. : Formula (19) in, + =1, and , The value of is determined by experimental calibration.
[0032] S5, compare the freezing and blockage risk index with multiple thresholds to determine the freezing and blockage risk level, and execute differentiated response actions according to the determined freezing and blockage risk level; The freezing and blockage risk index respectively with the first threshold Second threshold and the third threshold The comparison is used to determine the level of freezing and blockage risk, among which < < ,and , , The freezing risk index corresponding to the time of occurrence of historical freezing events is used as the calibration sample, and the distribution quantile of the calibration sample is used to determine the risk index. The differentiated response actions include: when ≤ < When the first sampling period is adjusted to a second sampling period that is shorter than the first sampling period, the preheating of the heat tracing device installed on the molten salt storage tank and / or molten salt pipeline is started. when ≤ < At that time, based on the previous level of response action, the molten salt heat extraction power is reduced and / or the molten salt flow rate is increased; when ≥ At that time, based on the previous level of response action, the molten salt pipe section where the liquid-solid interface is located is isolated and a freezing and blockage warning signal is output; S6. Compare the standard deviation with the confidence threshold, and decide whether the thermodynamic state inversion model needs to be updated online based on the comparison result.
[0033] In S6, the standard deviation is... Each with the first confidence threshold Second confidence threshold In comparison, among them < :when < At that time, the inversion temperature, liquid-solid interface position, and freezing risk index obtained from S2 to S5 are adopted; when ≤ < At that time, the expected inversion temperature is cross-validated using the outer wall temperature data from other monitoring points before acceptance; when ≥ When the inversion quality deteriorates, an alert is output, and the residual learning network is updated online using the outer wall temperature data collected in the current sampling period.
[0034] like Figure 2 As shown, this invention takes a typical DN200 molten salt pipe as an example (the thermal conductivity of the steel pipe is λ≈45W / (m·K), the outer diameter is 219mm, and the wall thickness is 8mm; the insulation layer is λQ≈0.05W / (m·K), and the thickness is 100mm). The external sensor measured Tw,o=220℃, and the ambient temperature was T=25℃. Using a pre-established thermodynamic state inversion model, T is obtained after inversion. f,phys =382.7℃ (converged after 3 property iterations) using PGRN residual correction Ultimately, the inversion occurs: =387.0±3.2℃ (95% confidence interval) Compared to a purely physical model (with a deviation of approximately -4.3℃), the hybrid inversion improves the accuracy to ±3.2℃ and provides uncertainty quantification. Therefore, this application demonstrates significant advantages over single physical models and pure data models in the following six dimensions, as shown in Table 1 below.
[0035] Table 1: Comparison between this application and a purely physical model Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.
Claims
1. A method for thermodynamic state inversion and early warning of molten salt energy storage systems based on temperature detection, characterized in that, Includes the following steps: S1, Collect temperature data of the outer wall surface of multiple monitoring points on the outer wall surface of molten salt storage tank and / or molten salt pipeline, and calculate its time derivative; S2, input the outer wall temperature data and time derivative into the thermodynamic state inversion model, the thermodynamic state inversion model includes a physical base value layer, a data-driven residual correction layer and an uncertainty quantification layer; The physical base value layer explicitly couples the temperature-dependent molten salt physical property parameters to the multilayer circular tube heat conduction model in the form of a polynomial fitting function. The steady-state inversion temperature is obtained through fixed-point iteration and transient thermal hysteresis compensation, and the physical base value inversion result is finally output. The data-driven residual correction layer takes the physical base value inversion result and the outer wall surface temperature characteristics as input, uses the residual learning network, outputs the residual correction amount and superimposes it with the physical base value inversion result to obtain the corrected expected inversion temperature. The uncertainty quantification layer outputs the expected inversion temperature, the standard deviation of the expected residual, and the confidence interval of the expected inversion temperature. S3. Based on the desired inversion temperature, construct a continuous temperature curve along the molten salt flow direction, solve for the liquid-solid interface position, and propagate the error of the standard deviation of the desired inversion temperature using the same interpolation and root-finding process to obtain the confidence interval of the liquid-solid interface position; calculate the phase interface migration rate based on the phase interface position at adjacent sampling times. S4. Determine the rate of decrease in heat flux density through the wall based on the inversion temperature; normalize the phase interface movement rate and the rate of decrease in heat flux density respectively, and then sum them by weight to obtain the freezing blockage risk index. S5, compare the freezing and blockage risk index with multiple thresholds to determine the freezing and blockage risk level, and execute differentiated response actions according to the determined freezing and blockage risk level; S6. Compare the standard deviation with the confidence threshold, and decide whether the thermodynamic state inversion model needs to be updated online based on the comparison result.
2. The thermodynamic state inversion and early warning method for molten salt energy storage systems based on temperature detection according to claim 1, characterized in that, In S1, the monitoring points are evenly distributed on the outer wall of the molten salt storage tank and / or the outer wall of the pipeline. Temperature sensors are attached to the monitoring points for continuous data collection. The temperature sensors are attached wireless platinum resistance temperature sensors.
3. The thermodynamic state inversion and early warning method for molten salt energy storage systems based on temperature detection according to claim 1, characterized in that, In S2, the inner diameter of the j-th layer of the multi-layer circular tube heat conduction model described in the physical base layer is r. j,i The outer diameter is r j,o The thermal conductivity is λ j and containing internal heat sources At steady state, the heat conduction equation is expressed by the following formula (1): Official (1) As the independent variable representing temperature, The independent variable representing the radius of the multilayer circular tube heat conduction model; By integrating the above equation twice and utilizing the interface temperature and heat flow continuity condition of each layer, the temperature distribution within the j-th layer can be obtained. After connecting multiple layers in series, by extrapolating from the outer wall temperature data Tw,o inward layer by layer, the general formula of the final multi-layer circular tube heat conduction model is obtained as follows: Official (2) Where B is the total number of pipe wall layers; q′′ N This represents the outermost heat flux density. For internal heat sources, the value is zero when there is no electric heating; r j,i r j,o Let λ represent the inner and outer radii of the j-th layer, respectively; j h is the thermal conductivity of the j-th layer material; i The convective heat transfer coefficient between the molten salt inside the pipe and the inner wall surface; unit: W / (m²). 2 ·K); This represents the outer radius of the (j-1)th layer; Let be the heat flux density of the (j-1)th layer.
4. The thermodynamic state inversion and early warning method for molten salt energy storage systems based on temperature detection according to claim 3, characterized in that, In S2, within the physical base layer, the temperature-dependent molten salt physical property parameters are explicitly coupled to the multilayer circular tube heat conduction model in the form of a polynomial fitting function. These molten salt physical property parameters include the molten salt thermal conductivity λ, specific heat capacity cp, density ρ, and dynamic viscosity η. Specifically, this involves the following process: Using the measured physical property parameter table of molten salt medium, a temperature-dependent polynomial fitting function is established; A quadratic polynomial can be used for fitting: Thermal conductivity: Formula (3) Specific heat capacity: Formula (4) density: Formula (5) Dynamic viscosity: Formula (6) Set the initial iteration temperature T f (0) =Tw,o+ΔT0; , , , , , , , , , , All coefficients are determined by fitting measured physical property parameter tables; The independent variable is temperature; In the k-th iteration, the current temperature Tf (k) Substitute each physical property parameter into the fitting formula to update the physical property parameters. Substitute the updated physical property parameters into the multilayer circular tube heat conduction model for inversion again. Repeat this process until convergence is obtained to obtain the steady-state inversion temperature. .
5. The thermodynamic state inversion and early warning method for molten salt energy storage systems based on temperature detection according to claim 4, characterized in that, The fixed-point iteration is = + The initial iteration temperature is [temperature value], at the [number]th iteration... In the next iteration, the first Temperature obtained from -1st iteration The fitted functions of each physical property parameter are substituted to update the physical property parameters. The updated physical property parameters are then substituted back into the multilayer circular tube heat conduction model for further derivation until the absolute value of the difference between two adjacent iterations is less than the convergence criterion. Or reach the number of iterations Official (7) Output physical basis value inversion results : Official (8) Equation (9) in, For the first The temperature obtained from the -1st iteration; τ sys The system's thermal time constant is determined by the thermal diffusivity and geometric dimensions of each layer of material. For steady-state inversion temperature, For the first Layer thickness, For the first The thermal diffusivity of the layer material, where R is the total number of layers.
6. The method for thermodynamic state inversion and early warning of molten salt energy storage system based on temperature detection according to claim 1, characterized in that, In the data-driven residual correction layer, the residual learning network adopts the PGRN network, which uses a 4-layer fully connected residual structure, namely the input layer, the first hidden layer, the second hidden layer, the third hidden layer, the fourth hidden layer and the single neuron output layer. Each hidden layer contains 64 nodes and uses the ReLU activation function. A skip connection residual branch is set between the second hidden layer and the fourth hidden layer to alleviate gradient vanishing. The input feature vector dimension of the network comprises: M-dimensional space temperature distribution, M-1-dimensional adjacent measuring point temperature difference, 2-dimensional environmental parameters, 3-dimensional time sequence statistics of the previous K periods, 1-dimensional physical base value inversion result T f,phys and 2-dimensional running state identification; the training label is set as the difference between the internal real temperature in the CFD simulation sample and the physical model inversion temperature, and the training loss function is constructed as: Official (10) where L MSE is the mean square error between the residual prediction and the true residual label, L physics is the physical consistency constraint term, which forces the mean of all sample residuals to be zero to ensure that the correction process does not introduce systematic bias, with a trade-off coefficient γ of 0.
1.
7. The thermodynamic state inversion and early warning method for molten salt energy storage systems based on temperature detection according to claim 5, characterized in that, In S2, the uncertainty quantification layer outputs the expected inversion temperature, the standard deviation of the expected residual, and the confidence interval of the expected inversion temperature, including: Expected inversion temperature Formula (11) Standard deviation of expected residuals Formula (12) The expected confidence interval for the inversion temperature is: Formula (13) In the formula, For the first The residual correction obtained from each forward propagation, where N is the total number of propagations.
8. The method for thermodynamic state inversion and early warning of molten salt energy storage system based on temperature detection according to claim 1, characterized in that, The corrected expected inversion temperature output by S2 at each discrete monitoring section along the molten salt flow direction. ( , Using nodes as nodes, construct a cubic spline interpolation function. ( This yields a temperature curve that is continuously distributed along the molten salt flow direction, and the initial crystallization point temperature of the molten salt is determined. Substituting the temperature curve, we obtain the phase interface determination equation: Official (14) The real roots of this equation in the axial coordinate interval are solved using numerical root-finding methods. ( ), which represents the current position of the molten salt-solid interface along the molten salt flow direction; the standard deviation obtained from S2 is... By propagating the error using the same interpolation and root-finding process, the confidence interval for the position of the liquid-solid interface is obtained, and the error propagation formula is as follows: Formula (15) The confidence interval for the location of the liquid-solid interface is: Official (16) The phase interface movement rate is obtained by dividing the difference between the phase interface positions at two adjacent sampling times by the sampling interval. Formula (17).
9. The method for thermodynamic state inversion and early warning of molten salt energy storage system based on temperature detection according to claim 8, characterized in that, In S4, the heat flux density through the wall is determined based on the corrected inversion temperature distribution along the molten salt flow direction output from S2. The rate of decrease in heat flux density is obtained by differentiating with respect to time. ; The phase interface movement rate ( ) and the rate of decrease of the heat flux density Normalized to their historical maximum values respectively: Official (18) Normalized phase interface migration rate With normalized heat flux density decrease rate By weighting factor , Weighted summation yields the freezing and blockage risk index. : Official (19) in, + =1, and , The value of is determined by experimental calibration.
10. The method for thermodynamic state inversion and early warning of molten salt energy storage system based on temperature detection according to claim 8, characterized in that, In S5, the freezing and blockage risk index is... respectively with the first threshold Second threshold and the third threshold The comparison is used to determine the level of freezing and blockage risk, among which < < ,and , , The freezing risk index corresponding to the time of occurrence of historical freezing events is used as the calibration sample, and the distribution quantile of the calibration sample is used to determine the risk index. The differentiated response actions include: when ≤ < When the first sampling period is adjusted to a second sampling period that is shorter than the first sampling period, the preheating of the heat tracing device installed in the molten salt storage tank and / or molten salt pipeline is started. when ≤ < At that time, based on the previous level of response action, the molten salt heat extraction power is reduced and / or the molten salt flow rate is increased; when ≥ At that time, based on the previous level of response action, the molten salt pipe section where the liquid-solid interface is located is isolated and a freezing and blockage warning signal is output; In S6, the standard deviation is... Each with the first confidence threshold Second confidence threshold In comparison, among them < :when < At that time, the inversion temperature, liquid-solid interface position, and freezing risk index obtained from S2 to S5 are adopted; when ≤ < At that time, the expected inversion temperature is cross-validated using the outer wall temperature data from other monitoring points before acceptance; when ≥ When the inversion quality deteriorates, an alert is output, and the residual learning network is updated online using the outer wall temperature data collected in the current sampling period.
