Thermal energy storage system overall performance attenuation evaluation method

Abstract

The invention provides a thermal energy storage system overall performance attenuation evaluation method, and belongs to the technical field of thermal energy storage systems. Distributed optical fiber temperature sensors are arranged in a thermal storage medium, and an infrared thermal imaging device is combined to establish a multi-dimensional temperature measurement system; a dynamic sliding time window is used for continuously collecting temperature data to form a space-time temperature data matrix, and a thermal field reconstruction optimization model of a multi-layer coding and decoding architecture is combined with an attention mechanism and an information bottleneck principle to realize high-precision reconstruction of an internal complete three-dimensional temperature field. And dynamic performance parameters such as instantaneous heat storage efficiency, temperature uniformity coefficient and heat loss rate are calculated based on a reconstruction result, and finally a comprehensive performance attenuation evaluation index is formed to realize accurate judgment of the system state, so that the technical problem of insufficient performance attenuation evaluation precision caused by difficulty in accurate reconstruction of a complete three-dimensional temperature field in the heat energy storage system is solved.

Description

Technical Field

[0001] This invention belongs to the technical field of thermal energy storage systems, and more specifically, relates to a method for evaluating the overall performance degradation of a thermal energy storage system. Background Technology

[0002] Thermal energy storage systems, as a key technology in the new energy field, primarily rely on traditional temperature measurement methods for performance evaluation. This involves acquiring temperature data by installing temperature sensors on the surface of the storage container or at limited locations, and then combining this data with numerical simulations to analyze the system's thermal storage efficiency and heat loss characteristics. Traditional methods are widely used in solar thermal power plants, industrial waste heat recovery systems, and district heating systems, assessing system operation by monitoring temperature changes at key locations. However, traditional temperature measurement methods are limited by the number and spatial distribution of measurement points, only obtaining temperature information for localized areas of the thermal storage system. This makes it difficult to comprehensively reflect the complex internal heat transfer processes and temperature field distribution characteristics, leading to significant biases in performance evaluations based on limited measurement point data. In existing technologies, the lack of effective methods for reconstructing the complete internal temperature field makes it impossible to accurately obtain the true three-dimensional temperature distribution within the thermal storage system, resulting in insufficient accuracy and reliability in performance degradation assessments. In other words, existing technologies suffer from the technical problem of inaccurately reconstructing the complete three-dimensional temperature field within the thermal energy storage system, leading to insufficient accuracy in performance degradation assessments. Summary of the Invention

[0003] In view of this, the present invention provides a method for evaluating the overall performance degradation of a thermal energy storage system, which can solve the technical problem in the prior art that the complete three-dimensional temperature field inside the thermal energy storage system is difficult to reconstruct accurately, resulting in insufficient accuracy in performance degradation evaluation.

[0004] This invention is implemented as follows: It provides a method for evaluating the overall performance degradation of a thermal energy storage system. Distributed fiber optic temperature sensors are arranged radially and axially inside the thermal storage medium of the system, while an infrared thermal imaging device is installed on the outer surface of the storage container for surface temperature field measurement. A dynamic sliding time window is established, and temperature data from the distributed fiber optic temperature sensors and the infrared thermal imaging device are continuously collected through this window to form a spatiotemporal temperature data matrix. The spatiotemporal temperature data matrix is ​​processed using a thermal field reconstruction optimization model, and the complete three-dimensional temperature field distribution inside the thermal storage system is reconstructed using a numerical inversion algorithm to obtain the temperature field reconstruction result. Based on the temperature field reconstruction result, the dynamic performance degradation of the thermal storage system is calculated. The system identifies dynamic performance parameters, including instantaneous thermal storage efficiency, temperature uniformity coefficient, and heat loss rate. It constructs performance degradation curves by relating these dynamic performance parameters to the operating time of the thermal storage system, and updates the performance degradation model parameters in real time using online parameter identification methods. Cluster analysis is employed to identify the data aggregation and distribution patterns of dynamic performance parameters. When cluster analysis identifies these patterns, cluster center estimation and boundary determination algorithms are used to automatically classify and organize the dynamic performance parameters, establishing multimodal performance evaluation criteria. Finally, an overall performance degradation evaluation index for the thermal energy storage system is established, which integrates the instantaneous thermal storage efficiency degradation rate, temperature uniformity coefficient degradation rate, and heat loss rate growth rate.

[0005] The distributed optical fiber temperature sensors are arranged at intervals of 0.1 to 0.3 times the characteristic dimensions of the thermal storage container. The distributed optical fiber temperature sensors are temperature measurement devices based on the Raman scattering principle, which measure the temperature distribution along the optical fiber path by measuring the change in the backscattering intensity of the optical signal in the optical fiber.

[0006] The characteristic dimensions of the thermal storage container refer to the main geometric dimensions of the thermal storage container, including the maximum value of diameter, height or length, and the duration of the dynamic sliding time window is 1.5 to 3 times the thermal response time constant of the thermal storage system.

[0007] The dynamic sliding time window is a time series data processing technique that continuously monitors and analyzes the dynamic characteristics of the system by moving a fixed-length time window continuously on the time axis. The thermal response time constant of the thermal storage system is the characteristic time required for the thermal storage system to reach a steady state from receiving heat input.

[0008] The spatiotemporal temperature data matrix is ​​a three-dimensional data structure formed by arranging temperature measurement data according to spatial location and time order. The thermal field reconstruction optimization model is a temperature field reconstruction algorithm based on sparse representation theory and variational optimization. It reconstructs the complete temperature field from finite measurement point data by minimizing the weighted sum of reconstruction error and regularization term.

[0009] The numerical inversion algorithm is a calculation method that infers the internal state of the system from the observation data. In the temperature field reconstruction, the internal temperature distribution is obtained by solving the inverse heat transfer problem. The temperature field reconstruction result is the complete three-dimensional temperature field distribution data inside the thermal storage system obtained by the numerical inversion algorithm.

[0010] Wherein, the instantaneous heat storage efficiency is the ratio of stored heat to input heat, the temperature uniformity coefficient is the ratio of the standard deviation of the temperature field to the average temperature, and the heat loss rate is the ratio of dissipated heat to total stored heat.

[0011] The performance degradation curve is a functional relationship describing the change of dynamic performance parameters over time. The online parameter identification method is an adaptive algorithm for real-time estimation of system model parameters, which dynamically updates the model parameters through recursive least squares or Kalman filtering techniques.

[0012] Specifically, when the dispersion of instantaneous thermal storage efficiency data is less than the concentration requirement threshold of 0.05, a centralized processing method is adopted to improve efficiency; when the dispersion of temperature uniformity coefficient data is greater than the complexity processing threshold (i.e., 0.15), a distributed processing strategy is adopted.

[0013] The centralized processing method refers to a unified processing strategy used when the data is relatively discrete, which improves processing accuracy and efficiency by concentrating computing resources. The distributed processing strategy refers to a decentralized processing method used when the data is highly complex, which distributes computing tasks to multiple processing units for parallel execution.

[0014] The data clustering distribution pattern refers to the clustering distribution characteristics of dynamic performance parameters in the parameter space. The cluster center estimation and boundary determination algorithm is a data analysis method based on density clustering, which determines the cluster centers and boundaries by calculating the data point density and connectivity.

[0015] The multimodal performance evaluation criteria are evaluation standards established when performance parameters exhibit multiple concentrated distributions. They take into account the performance characteristics under different operating modes. The instantaneous thermal storage efficiency decay rate is the rate at which instantaneous thermal storage efficiency decreases over time, the temperature uniformity coefficient decay rate is the rate at which the temperature uniformity coefficient deteriorates over time, and the heat loss rate growth rate is the rate at which the heat loss rate increases over time.

[0016] Specifically, when the overall performance degradation assessment index of the thermal energy storage system is ∈ [0, 0.25), the system performance is normal; when the overall performance degradation assessment index of the thermal energy storage system is ∈ [0.25, 0.65), the system performance is slightly degraded and the operating parameters need to be optimized; when the overall performance degradation assessment index of the thermal energy storage system is ∈ [0.65, 1], the system performance is severely degraded and the thermal storage medium needs to be maintained or replaced.

[0017] The specific structure of the thermal field reconstruction optimization model is a multi-layer encoding and decoding architecture. The encoder part uses a three-dimensional convolutional layer to extract the spatial features of the temperature field, the decoder part reconstructs the complete temperature field through a deconvolutional layer, and the intermediate layer uses an attention mechanism to enhance the feature representation of key areas. The attention weight parameters are determined by an adaptive adjustment factor based on three parameters: the thermal conductivity of the thermal storage medium, the feature size of the thermal storage container, and the density of temperature measurement points.

[0018] The establishment of the training dataset for the thermal field reconstruction optimization model includes collecting measured temperature field data of the thermal storage system under different operating conditions, generating a corresponding complete temperature field as the standard answer through numerical simulation, using the measured sparse temperature data as input and the complete temperature field as output to construct training sample pairs, and the dataset covers three typical operating stages: the heat charging stage, the heat preservation stage, and the heat release stage.

[0019] The training of the thermal field reconstruction optimization model includes adopting a representation learning framework based on the information bottleneck principle. It constrains the redundant information of the intermediate layer representation by minimizing the mutual information objective function, estimates the mutual information terms that are difficult to calculate directly using variational approximation techniques, and dynamically balances the trade-off between compression ratio and fidelity using the Lagrange multiplier method. During the training process, gradient pruning techniques and learning rate decay strategies are used to prevent overfitting.

[0020] This invention establishes a multi-dimensional performance evaluation system based on distributed fiber optic temperature sensing and a thermal field reconstruction optimization model. By combining dynamic sliding time window technology and adaptive processing strategies, it achieves high-precision reconstruction of the complete three-dimensional temperature field within a thermal energy storage system and accurate calculation of dynamic performance parameters. The method employs a multi-layered encoding-decoding architecture for the thermal field reconstruction optimization model. It extracts spatial features of the temperature field through three-dimensional convolutional layers and enhances the representation of key region features using an attention mechanism. This effectively solves the problems of incomplete temperature field information and low reconstruction accuracy in traditional methods, significantly improving the accuracy and reliability of performance evaluation. In summary, this invention solves the technical problem mentioned in the background art, where the inaccurate reconstruction of the complete three-dimensional temperature field within a thermal energy storage system leads to insufficient accuracy in performance degradation evaluation. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 The graph shows the dynamic performance parameters of the thermal storage system over time in the example.

[0023] Figure 3 The diagram shows the trend of heat loss rate and temperature field distribution characteristics of the thermal storage system in the example. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0025] like Figure 1 The diagram shown is a flowchart of a method for evaluating the overall performance degradation of a thermal energy storage system provided by this invention. This method includes the following steps: S01. Distributed fiber optic temperature sensors are arranged radially and axially inside the thermal storage medium of the thermal energy storage system. The spacing between the distributed fiber optic temperature sensors is 0.1 to 0.3 times the characteristic size of the thermal storage container. Meanwhile, an infrared thermal imaging device is installed on the outer surface of the thermal storage container for surface temperature field measurement. S02. Establish a dynamic sliding time window. The duration of the dynamic sliding time window is 1.5 to 3 times the thermal response time constant of the thermal storage system. Temperature data from distributed fiber optic temperature sensors and infrared thermal imaging devices are continuously collected through the dynamic sliding time window to form a spatiotemporal temperature data matrix. S03. The spatiotemporal temperature data matrix is ​​processed using the thermal field reconstruction optimization model, and the complete three-dimensional temperature field distribution inside the thermal storage system is reconstructed through numerical inversion algorithm to obtain the temperature field reconstruction result. S04. Calculate the dynamic performance parameters of the thermal storage system based on the temperature field reconstruction results, including instantaneous thermal storage efficiency, temperature uniformity coefficient and heat loss rate. The instantaneous thermal storage efficiency is the ratio of stored heat to input heat, the temperature uniformity coefficient is the ratio of temperature field standard deviation to average temperature, and the heat loss rate is the ratio of heat loss to total stored heat. S05. Construct a performance degradation curve by combining dynamic performance parameters with the operating time of the thermal storage system. Update the performance degradation model parameters in real time using an online parameter identification method. When the dispersion of instantaneous thermal storage efficiency data is less than the concentration requirement threshold of 0.05, adopt a centralized processing method to improve efficiency. When the dispersion of temperature uniformity coefficient data is greater than the complexity processing threshold of 0.15, adopt a distributed processing strategy. S06. Cluster analysis is used to identify the data clustering distribution pattern of dynamic performance parameters. When cluster analysis identifies the data clustering distribution pattern of dynamic performance parameters, cluster center estimation and boundary determination algorithms are used to realize the automatic classification and organization of dynamic performance parameters and establish multimodal performance evaluation criteria. S07. Establish an overall performance degradation assessment index for the thermal energy storage system. The overall performance degradation assessment index for the thermal energy storage system integrates the instantaneous thermal storage efficiency degradation rate, the temperature uniformity coefficient degradation rate, and the heat loss rate growth rate. When the overall performance degradation assessment index of the thermal energy storage system ∈ [0, 0.25), the system performance is normal; when the overall performance degradation assessment index of the thermal energy storage system ∈ [0.25, 0.65), the system performance is slightly degraded and the operating parameters need to be optimized; when the overall performance degradation assessment index of the thermal energy storage system ∈ [0.65, 1], the system performance is severely degraded and the thermal storage medium needs to be maintained or replaced.

[0026] The distributed fiber optic temperature sensor is a temperature measurement device based on the Raman scattering principle. It measures the temperature distribution along the fiber optic path by measuring the change in the backscattering intensity of the light signal in the fiber. The characteristic dimensions of the thermal storage container refer to its main geometric dimensions, including the maximum value of its diameter, height, or length. The dynamic sliding time window is a time series data processing technique that continuously monitors and analyzes the dynamic characteristics of the system by moving a fixed-length time window continuously along the time axis. The thermal response time constant of the thermal storage system is the characteristic time required for the system to reach steady state after receiving heat input.

[0027] The spatiotemporal temperature data matrix is ​​a three-dimensional data structure formed by arranging temperature measurement data according to spatial location and temporal order. The thermal field reconstruction optimization model is a temperature field reconstruction algorithm based on sparse representation theory and variational optimization. It reconstructs the complete temperature field from finite measurement point data by minimizing the weighted sum of reconstruction error and regularization terms. The numerical inversion algorithm is a computational method that infers the internal state of the system from observation data. In temperature field reconstruction, it obtains the internal temperature distribution by solving the inverse heat transfer problem. The temperature field reconstruction result is the complete three-dimensional temperature field distribution data inside the thermal storage system obtained through the numerical inversion algorithm.

[0028] The instantaneous thermal storage efficiency is a parameter characterizing the thermal storage system's ability to store heat at the current moment. The stored heat capacity is the total amount of thermal energy stored by the thermal storage medium at the current temperature. The input heat is the total thermal energy input to the thermal storage system. The temperature uniformity coefficient is a dimensionless parameter characterizing the uniformity of the temperature field distribution. The temperature field standard deviation is a measure of the dispersion of the temperature at each measuring point relative to the average temperature in the temperature field. The average temperature is the arithmetic mean of the temperatures at all measuring points in the temperature field. The heat loss rate is a parameter characterizing the degree of heat loss from the thermal storage system. The dissipated heat is the heat energy lost by the thermal storage system to the environment. The total stored heat capacity is the total heat capacity of the thermal storage system.

[0029] The performance degradation curve is a functional relationship describing the change of dynamic performance parameters over time. The online parameter identification method is an adaptive algorithm for real-time estimation of system model parameters, dynamically updating the model parameters through recursive least squares or Kalman filtering techniques. The performance degradation model parameters are mathematical parameters describing the shape and trend of the performance degradation curve. The concentration requirement threshold is a critical value for determining whether the data is suitable for centralized processing. The centralized processing method refers to a unified processing strategy used when the data dispersion is low, improving processing accuracy and efficiency by concentrating computing resources. The complexity processing threshold is a critical value for determining whether the data complexity requires distributed processing. The distributed processing strategy refers to a decentralized processing method used when the data complexity is high, distributing computational tasks to multiple processing units for parallel execution.

[0030] The data aggregation distribution pattern refers to the aggregation distribution characteristics of dynamic performance parameters in the parameter space. The cluster center estimation and boundary determination algorithm is a density-based clustering data analysis method that determines cluster centers and boundaries by calculating data point density and connectivity. The multimodal performance evaluation criterion is an evaluation standard established when performance parameters exhibit multiple concentrated distributions, considering performance characteristics under different operating modes. The instantaneous thermal storage efficiency decay rate is the rate at which instantaneous thermal storage efficiency decreases over time. The temperature uniformity coefficient decay rate is the rate at which the temperature uniformity coefficient deteriorates over time. The heat loss rate growth rate is the rate at which the heat loss rate increases over time.

[0031] The specific structure of the thermal field reconstruction optimization model is a multi-layer encoding and decoding architecture. The encoder part uses a three-dimensional convolutional layer to extract the spatial features of the temperature field, and the decoder part reconstructs the complete temperature field through a deconvolutional layer. The intermediate layer uses an attention mechanism to enhance the feature representation of key regions. The attention weight parameters are determined by an adaptive adjustment factor based on three parameters: the thermal conductivity of the thermal storage medium, the characteristic size of the thermal storage container, and the density of temperature measuring points. The thermal conductivity of the thermal storage medium is a parameter indicating its ability to conduct heat. The density of temperature measuring points is the number of distributed fiber optic temperature sensor measuring points per unit volume. The adaptive adjustment factor is a parameter that dynamically adjusts the attention weights according to the system operating status.

[0032] The steps for establishing the training dataset for the thermal field reconstruction optimization model specifically include collecting measured temperature field data of the thermal storage system under different operating conditions, generating a corresponding complete temperature field as the standard answer through numerical simulation, using the measured sparse temperature data as input and the complete temperature field as output to construct training sample pairs. The dataset covers three typical operating stages: the heat charging stage, the heat preservation stage, and the heat release stage. The heat charging stage is the operating stage where the thermal storage system receives and stores heat. The heat preservation stage is the operating stage where the thermal storage system maintains the stored heat. The heat release stage is the operating stage where the thermal storage system releases the stored heat.

[0033] The specific steps of training the thermal field reconstruction optimization model include employing a representation learning framework based on the information bottleneck principle. This involves minimizing the mutual information objective function to constrain redundant information in intermediate layer representations, using variational approximation techniques to estimate mutual information terms that are difficult to calculate directly, and dynamically balancing the trade-off between compression ratio and fidelity using the Lagrange multiplier method. During training, gradient pruning and learning rate decay strategies are employed to prevent overfitting. The information bottleneck principle is a theoretical framework for controlling the amount of information transmitted to achieve optimal representation learning. The mutual information objective function is a function that quantifies the statistical dependency between two random variables. The variational approximation technique is a mathematical method that approximates complex probability distributions using trainable parameters. The Lagrange multiplier method is a mathematical method for solving optimization problems under constraints. The compression ratio is a quantitative indicator of the degree of information compression. The fidelity is a quantitative indicator of the similarity between the reconstructed information and the original information. The gradient pruning technique is a training optimization method to prevent gradient explosion. The learning rate decay strategy is an optimization method for dynamically adjusting the model training step size.

[0034] The representation learning framework based on the information bottleneck principle controls the amount of information transmitted in the network, ensuring the complete preservation of important information while eliminating redundant and noisy information, thus achieving high-quality feature representations. The constraint mechanism of the mutual information objective function gives the intermediate representations learned by the model stronger generalization ability and robustness, avoiding overfitting to the training data. Variational approximation techniques solve the problem of accurately calculating mutual information in high-dimensional data, approximating complex probability distributions through trainable neural networks. The dynamic balancing mechanism of the Lagrange multiplier method automatically adjusts the importance weights of compression ratio and fidelity according to the current training state, achieving optimal model performance.

[0035] The adaptive selection mechanism between centralized and distributed processing strategies is based on the inherent patterns of data characteristics. When performance data exhibits high consistency, centralized processing can fully utilize the common characteristics of the data, achieving high-efficiency analysis and computation through a unified processing flow. When performance data exhibits significant diversity and complexity, the distributed processing strategy can fully explore the local features of the data, addressing data heterogeneity through parallel processing mechanisms. The combined application of clustering analysis and boundary determination algorithms enables intelligent organization and classification of complex performance data. By discovering the inherent structural patterns of the data, it provides a scientific data foundation for subsequent performance evaluation.

[0036] This technology enhances the adaptability and accuracy of thermal energy storage system performance evaluation under complex and variable operating conditions, significantly improving the accuracy and stability of temperature field reconstruction under dynamic conditions. This provides reliable technical support for intelligent operation and maintenance and performance optimization of thermal energy storage systems. By reducing interference from redundant information and enhancing the expressive power of key features, the method reduces computational complexity while maintaining evaluation accuracy, improving the feasibility of real-time performance evaluation. The introduction of an adaptive processing strategy enables the system to automatically select the optimal processing method based on data characteristics, achieving a dynamic balance between evaluation accuracy and computational efficiency.

[0037] This invention is also implemented by a computer to form an overall performance degradation assessment system for thermal energy storage systems. The computer is equipped with a readable storage medium, which stores program instructions. When the program instructions are run in the computer, they execute the above-described method.

[0038] The specific implementation methods of the above steps are described in detail below.

[0039] The specific implementation of step S01 involves first measuring the characteristic dimensions of the thermal storage container, including diameter, height, and length, and determining the maximum value as the characteristic dimension benchmark. Based on the characteristic dimensions, the spacing of the distributed fiber optic temperature sensors is calculated, and the spacing is determined to be 0.1 to 0.3 times the characteristic dimensions, preferably 0.2 times, to achieve the best balance between spatial resolution and economy. Inside the thermal storage medium, a grid-like arrangement is created in both radial and axial directions. In the radial direction, 3 to 5 layers of fiber optic sensors are evenly spaced from the center to the outer wall, and in the axial direction, 5 to 8 layers of fiber optic sensors are evenly spaced from the bottom to the top, forming a three-dimensional measurement point network. A distributed fiber optic temperature sensor based on the Raman scattering principle is used to measure temperature by utilizing the characteristic that the intensity ratio of Stokes and anti-Stokes scattered light in the fiber changes with temperature, achieving a temperature measurement accuracy of ±1℃ and a spatial resolution of 1 meter. An infrared thermal imaging device is installed on the outer surface of the thermal storage container. The device is equipped with an infrared detector array with a working wavelength range of 8 to 14 micrometers, a temperature measurement range of -20℃ to 500℃, a pixel resolution of no less than 320×240, and a frame rate of no less than 9Hz. The installation position is ensured to cover the entire outer surface of the thermal storage container, and the installation distance is 1.5 to 2 times the diameter of the container.

[0040] The specific implementation of step S02 involves first calculating the thermal response time constant using the physical parameters of the thermal storage system. This time constant is equal to the ratio of the thermal capacity of the storage medium to the overall heat transfer coefficient of the system. A dynamic sliding time window is established, with the window length set to 1.5 to 3 times the thermal response time constant, preferably 2 times, to ensure complete capture of the dynamic response process. A fixed window length is used to continuously slide along the time axis, with the sliding step size set to 10% to 20% of the window length, achieving continuous coverage of the time series. Temperature data from distributed fiber optic temperature sensors and infrared thermal imaging equipment are simultaneously acquired through a data acquisition system, with the sampling frequency set to 0.1Hz to 1Hz to ensure compliance with the Nyquist sampling theorem. The acquired temperature data is arranged according to spatial coordinates and time series to form a four-dimensional spatiotemporal temperature data matrix. The matrix dimension is three spatial coordinates plus one time dimension, and the data storage format adopts a tensor structure for easy subsequent processing. The acquired raw data is preprocessed, including outlier detection using the 3-standard-deviation criterion, missing value interpolation using a cubic spline interpolation algorithm, and noise filtering using a Kalman filter algorithm.

[0041] The specific implementation of step S03 involves inputting the preprocessed spatiotemporal temperature data matrix into the thermal field reconstruction optimization model for processing. This model employs a multi-layer encoder-decoder architecture. The encoder uses a three-dimensional convolutional neural network to extract the spatial features of the temperature field, with a kernel size of 3×3×3, a stride of 1, and zero-padding to maintain the feature map size. The encoder contains 4 to 6 convolutional layers, each followed by a batch normalization layer and an activation function layer. The activation function is a modified linear unit function. The decoder uses a three-dimensional deconvolutional layer to reconstruct the complete temperature field, with the deconvolution parameters maintaining a symmetrical relationship with the corresponding encoder layer. An attention mechanism layer is set between the encoder and decoder, with attention weights determined by an adaptive adjustment factor based on the thermal conductivity of the storage medium, the characteristic dimensions of the storage container, and the density of temperature measurement points. The numerical inversion algorithm uses a variational optimization method to solve the inverse heat transfer problem, estimating parameters by minimizing the mean square error between the observed temperature and the model-predicted temperature. The optimization algorithm uses an adaptive moment estimation algorithm, with a learning rate set to 0.001 to 0.01 and a batch size set to 16 to 64. Regularization constraints are used during the reconstruction process to prevent overfitting, and the regularization coefficient is set to [value missing]. to .

[0042] The specific implementation of step S04 involves calculating the dynamic performance parameters of the thermal storage system based on the temperature field reconstruction results. Instantaneous thermal storage efficiency is calculated as the ratio of stored heat to input heat. Stored heat is obtained by integrating the specific heat capacity, density, volume, and temperature difference of the thermal storage medium. Input heat is obtained by measuring the heat flow rate and time entering the thermal storage system. The temperature uniformity coefficient is calculated as the ratio of the temperature field standard deviation to the average temperature. The standard deviation is calculated using the Bessel formula, and the average temperature is the arithmetic mean of all reconstructed temperature points. The heat loss rate is calculated as the ratio of dissipated heat to total stored heat. Dissipated heat is calculated using the heat transfer coefficient, surface area, and temperature difference of the thermal storage container surface. Total stored heat is the maximum heat capacity of the thermal storage system. Numerical integration is used to process spatially distributed data during the calculation process. The integration algorithm uses Simpson's integral rule, and the integration step size is determined based on the accuracy of the temperature field reconstruction. The calculation results are filtered using a moving average filtering algorithm to eliminate high-frequency noise. The filtering window length is set to 5 to 10 times the data sampling period.

[0043] The specific implementation of step S05 involves constructing a performance degradation curve by combining the calculated dynamic performance parameters with the operating time of the thermal storage system. The least squares method is used to fit the changing trend of the performance parameters over time, with the fitting function employing an exponential decay model or a polynomial model. The performance degradation model parameters are updated in real-time using an online parameter identification method. The parameter identification algorithm uses recursive least squares, with a forgetting factor set between 0.95 and 0.99, and the diagonal elements of the initial covariance matrix set between 100 and 1000. A data dispersion assessment mechanism is established, using the coefficient of variation as a quantitative indicator of dispersion. The coefficient of variation is equal to the ratio of the standard deviation to the mean. When the coefficient of variation of the instantaneous thermal storage efficiency data is less than the concentration requirement threshold of 0.05, a centralized processing approach is adopted, improving processing accuracy and efficiency through unified computing resources and algorithms. When the coefficient of variation of the temperature uniformity coefficient data is greater than the complexity processing threshold of 0.15, a distributed processing strategy is adopted, dividing the data into multiple subsets and allocating them to different processing units for parallel computation. The processing strategy selection employs an adaptive mechanism, dynamically adjusting the processing method based on real-time data characteristics.

[0044] The specific implementation of step S06 involves using cluster analysis to identify the data aggregation distribution patterns of dynamic performance parameters. The clustering algorithm employs a density-based spatial clustering algorithm, which can discover clusters of arbitrary shapes and is robust to noise. In the algorithm parameter settings, the neighborhood radius is determined based on the data distribution density, typically set to 1.5 to 2 times the average distance between data points, and the minimum number of points is set to 2% to 5% of the total data. Cluster center estimation uses a centroid calculation method, calculating the geometric center of each data point within a cluster as the cluster center. The boundary determination algorithm calculates the distance distribution from data points to the cluster centers and uses statistical methods to determine the cluster boundaries, with the boundary radius set to the 95th percentile of the distance from points within the cluster to the center. When cluster analysis identifies a significant multimodal distribution, multimodal performance evaluation criteria are established, and evaluation standards are formulated for different operating modes. The multimodal discrimination criterion uses a combination of the silhouette coefficient and the Davis-Boudin index; a silhouette coefficient greater than 0.5 and a Davis-Boudin index less than 2 are considered valid multimodal distributions.

[0045] The specific implementation of step S07 involves establishing an overall performance degradation assessment index for the thermal energy storage system. This index comprehensively considers three key parameters: instantaneous thermal energy storage efficiency degradation rate, temperature uniformity coefficient degradation rate, and heat loss rate growth rate. The degradation rate is calculated using the absolute value of the ratio of the current value to the initial value minus 1, and the growth rate is calculated using the ratio of the current value to the initial value minus 1. The overall performance degradation assessment index is calculated using a weighted average method, with weights allocated based on the importance of each parameter to the system performance. The weight for the instantaneous thermal energy storage efficiency degradation rate is set to 0.5, the weight for the temperature uniformity coefficient degradation rate is set to 0.3, and the weight for the heat loss rate growth rate is set to 0.2. A tiered assessment standard is established. When the overall performance degradation assessment index is within the range of 0 to 0.25, the system performance is considered normal, and no action is required. When the index is within the range of 0.25 to 0.65, the system performance is considered slightly degraded, requiring optimization of operating parameters, including adjusting the thermal energy storage medium flow rate and optimizing temperature control strategies. When the index is within the range of 0.65 to 1, the system performance is considered severely degraded, requiring system maintenance or replacement of the thermal energy storage medium. The assessment results are displayed in real time through a visual interface, including trend charts, warning signals, and maintenance recommendations.

[0046] Further explanation is needed regarding the thermal field reconstruction optimization model, which employs a deep learning architecture. The overall structure includes a data preprocessing module, a feature extraction module, an attention mechanism module, and a reconstruction output module. The data preprocessing module standardizes the input sparse temperature data, using a max-min normalization method to map the temperature data to the range of 0 to 1. Simultaneously, data augmentation is performed, including adding Gaussian noise, random occlusion, and rotation transformation. The feature extraction module uses a three-dimensional convolutional neural network architecture, containing six convolutional layers. The first layer has 32 kernels, followed by layers with 64, 128, 256, 512, and 1024 kernels respectively. Each convolutional layer is followed by a batch normalization layer and an activation function layer. The pooling layer uses max pooling with a 2×2×2 kernel size and a stride of 2 to reduce the feature map dimensionality and computational complexity. The attention mechanism module combines spatial attention and channel attention. Spatial attention enhances key regions by calculating importance weights at different locations in the feature map, while channel attention selects features by calculating importance weights for different feature channels. Attention weights are calculated using a combination of fully connected layers and activation functions. The weight parameters are determined by a linear combination of three parameters: the thermal conductivity of the storage medium, the characteristic dimensions of the storage container, and the density of temperature measurement points. The adaptive adjustment factor is optimized using a gradient descent algorithm. The reconstructed output module employs a three-dimensional deconvolutional neural network with a structure symmetrical to the encoder. It restores the spatial resolution of the temperature field through upsampling and deconvolution operations. Finally, the output layer uses a linear activation function to ensure the continuity of temperature values.

[0047] The training dataset establishment comprises four stages: data collection, data generation, data annotation, and data validation. The data collection stage involves acquiring measured temperature field data through actual operation of the thermal storage system, covering three typical operating stages: the charging stage, the insulation stage, and the heat release stage. At least 1000 data samples are collected for each stage. The data collection time for the charging stage is the complete process of the thermal storage system heating up from room temperature to the set temperature; the data collection time for the insulation stage is the process of the thermal storage system maintaining the set temperature for at least 4 hours; and the data collection time for the heat release stage is the complete process of the thermal storage system cooling down from the set temperature to room temperature. The data generation stage uses computational fluid dynamics simulation software to establish a three-dimensional numerical model of the thermal storage system. Boundary conditions are set to be consistent with actual operating conditions, and a structured mesh is used, with a minimum number of meshes. The simulation uses a unit time step of 0.1 to 1 second. The heat transfer equation is solved using the finite volume method, with a second-order upwind scheme for convection, a central difference scheme for diffusion, and an implicit Euler scheme for time progression. In the data annotation phase, the simulated complete temperature field is used as the standard answer, and sparse measured temperature data at corresponding locations are used as input to construct training samples corresponding to the input and output. The sample data format uses a tensor structure, with the input tensor dimension being three spatial dimensions plus one temporal dimension, and the output tensor dimension being the complete three-dimensional spatial temperature field. In the data validation phase, cross-validation is used to evaluate the dataset quality. The dataset is divided into training, validation, and test sets in an 8:1:1 ratio, and mean squared error and structural similarity index are used as evaluation metrics.

[0048] The model training employs a representation learning framework based on the information bottleneck principle, achieving optimal feature representation by controlling the amount of information transmitted within the network. The objective function includes a reconstruction error term and an information constraint term. The reconstruction error is calculated using mean squared error, and the information constraint term uses mutual information estimation. Mutual information calculation employs variational approximation, approximating a complex probability distribution through a trainable discriminant network. The discriminant network uses a fully connected neural network structure with 3 to 5 hidden layers, each containing 128 to 512 neurons. The training process uses the Lagrange multiplier method to dynamically balance the trade-off between compression ratio and fidelity. The initial Lagrange multiplier is set to 1 and dynamically adjusted using a gradient optimization algorithm. The optimization algorithm uses adaptive moment estimation, with an initial learning rate of 0.001 and a cosine annealing learning rate decay strategy, with a decay period set to 10% of the total training epochs. Gradient pruning is used during training to prevent gradient explosion, with a pruning threshold set between 1.0 and 5.0. An early stopping mechanism is used to prevent overfitting, stopping training when the validation set loss shows no improvement for 10 consecutive epochs. Model evaluation uses multiple metrics for comprehensive judgment, including mean squared error, mean absolute error, peak signal-to-noise ratio, and structural similarity index. The mean squared error of the trained model on the test set should be less than 0.01, and the structural similarity index should be greater than 0.95.

[0049] It should be noted that the key technical ideas of this invention are mainly reflected in three aspects: deep learning temperature field reconstruction technology based on the information bottleneck principle, adaptive data processing strategy selection mechanism, and multimodal performance evaluation system.

[0050] Deep learning-based temperature field reconstruction technology, based on the information bottleneck principle, controls the amount of information transmitted in the intermediate layers of a neural network. This ensures the complete preservation of important temperature features while effectively eliminating redundant and noisy information, thus achieving high-quality temperature field reconstruction. Compared to traditional interpolation methods and empirical models, this technology can automatically learn the complex spatial distribution patterns of the temperature field without requiring manual parameter setting or assumptions, significantly improving the accuracy and robustness of temperature field reconstruction. Traditional methods are usually based on linear assumptions or simplified physical models, making it difficult to accurately describe the complex nonlinear heat transfer phenomena in thermal storage systems. In contrast, deep learning methods can adaptively fit arbitrarily complex temperature distribution patterns, showing significant advantages, especially when dealing with multiphase media, phase transition processes, and irregular geometric structures.

[0051] The adaptive data processing strategy selection mechanism automatically chooses the optimal processing method based on the inherent characteristics of the performance data. When the data exhibits high consistency, centralized processing is used to improve efficiency; when the data exhibits significant complexity, distributed processing is employed to address heterogeneity. Compared to a fixed data processing method, this mechanism can dynamically adjust the processing strategy according to actual operating conditions, achieving an optimal balance between computational efficiency and processing accuracy. Traditional performance evaluation methods typically employ a single data processing flow, which cannot adapt to the changing data characteristics of the thermal storage system under different operating modes, leading to low computational efficiency or insufficient accuracy under certain conditions. The adaptive mechanism, however, can specifically optimize the processing flow, significantly improving overall performance.

[0052] The multimodal performance evaluation system identifies multiple distribution patterns of performance parameters through cluster analysis, establishing corresponding evaluation criteria for different operating states. This approach is more scientific and reasonable than traditional single evaluation standards. Traditional methods typically rely on a single average value or threshold for performance judgment, ignoring the performance characteristics differences of thermal storage systems under different operating conditions, which may lead to misjudgments or omissions. The multimodal evaluation system can identify and distinguish different operating modes, and formulate specific evaluation criteria for each mode, improving the accuracy and reliability of performance evaluation. It is particularly suitable for complex thermal storage systems with multiple operating modes.

[0053] The synergistic effect of these three key technological approaches forms a complete intelligent performance evaluation system. Deep learning technology provides a high-precision foundation for temperature field reconstruction, the adaptive processing mechanism ensures optimal data processing, and the multimodal evaluation system achieves accurate performance judgment. These three technologies work together to overcome the limitations of traditional methods in terms of accuracy, efficiency, and adaptability, achieving a technological leap from static and fixed to dynamic and adaptive performance evaluation of thermal storage systems. This provides a new technical approach for the intelligent operation and maintenance and performance optimization of thermal storage systems, significantly improving the reliability and practicality of evaluation under complex operating conditions.

[0054] It should be noted that this invention also addresses the following technical problem: the lack of adaptability in selecting data processing strategies for thermal storage system performance parameters. In existing thermal storage system performance evaluations, a fixed data processing strategy is typically used regardless of changes in data characteristics. When data exhibits high consistency, complex distributed processing leads to resource waste; conversely, when data exhibits significant complexity, simple centralized processing reduces analytical accuracy. This invention establishes an adaptive selection mechanism based on data dispersion. When the dispersion of instantaneous thermal storage efficiency data is less than 0.05, centralized processing is automatically adopted to improve efficiency; when the dispersion of temperature uniformity coefficient data is greater than 0.15, a distributed processing strategy is automatically adopted to address complexity, thus achieving intelligent selection of processing strategies. Simultaneously, this invention also solves the technical problem of insufficient uniformity in performance evaluation standards under multiple operating modes of thermal storage systems. By identifying the data aggregation and distribution patterns of dynamic performance parameters through cluster analysis, and employing cluster center estimation and boundary determination algorithms, automatic classification and organization of dynamic performance parameters are achieved. Multimodal performance evaluation criteria for different operating modes are established, effectively solving the problem that traditional single evaluation standards are difficult to adapt to complex operating conditions.

[0055] Specifically, the principle of this invention is as follows: This invention can solve the technical problem of accurately reconstructing the complete three-dimensional temperature field inside a thermal energy storage system, mainly based on the following technical principles. First, by arranging distributed fiber optic temperature sensors radially and axially inside the thermal storage medium, combined with infrared thermal imaging equipment on the outer surface of the thermal storage container, a multi-dimensional, high-density temperature measurement system is established, providing sufficient spatial temperature information for temperature field reconstruction. Second, the established thermal field reconstruction optimization model adopts a multi-layer encoding and decoding architecture. The three-dimensional convolutional layer of the encoder can effectively extract the spatial features of the temperature field, and the deconvolutional layer of the decoder achieves accurate reconstruction of the complete temperature field. The attention mechanism of the intermediate layer adaptively adjusts the weights according to key parameters such as the thermal conductivity of the thermal storage medium, the feature size of the thermal storage container, and the density of temperature measurement points, ensuring that the features of key areas are fully expressed during the reconstruction process. Third, the representation learning framework based on the information bottleneck principle constrains the redundant information of the intermediate layer representation by minimizing the mutual information objective function. Combined with variational approximation techniques and the Lagrange multiplier method, a dynamic balance between compression ratio and fidelity is achieved, effectively avoiding overfitting and improving the model's generalization ability and reconstruction accuracy. Finally, the dynamic sliding time window technology, combined with an adaptive processing strategy, can automatically select centralized or distributed processing methods based on data characteristics, which not only ensures calculation accuracy but also improves processing efficiency, providing a reliable technical basis for accurately assessing the dynamic performance degradation of thermal storage systems.

[0056] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0057] In this embodiment, the specific implementation of step S01 is the same as described above, and will not be repeated in detail here.

[0058] The specific implementation of step S02 involves establishing a dynamic sliding time window and acquiring a spatiotemporal temperature data matrix. The formula for calculating the duration of the dynamic sliding time window is as follows: ; In the formula, The length of the dynamic sliding time window, in units of ; This is the time window coefficient, with an empirical value of 1.5 to 3.0; The thermal response time constant of the thermal storage system is given in units of 1. .in, The calculation formula is expressed as follows: ; In the formula, Density of the thermal storage medium, in units of This can be obtained through experimental determination or by consulting property tables; Specific heat capacity of the heat storage medium, in units of This can be obtained through experimental determination or by consulting property tables; The volume of the thermal storage medium is expressed in units of 1. Obtained through geometric measurements; The overall heat transfer coefficient of the system is expressed in units of 1000 ppm. It is obtained through heat transfer experiments or theoretical calculations; Total heat transfer area, in units of This data is obtained through geometric measurements. The formula for constructing the spatiotemporal temperature data matrix is ​​as follows: ; In the formula, This is a spatiotemporal temperature data matrix; For the first The spatial measuring point is at the _ ... Temperature values ​​at each time point, in units of The temperature was obtained through measurements using distributed fiber optic temperature sensors and infrared thermal imaging equipment. The total number of spatial measurement points is calculated using the spacing of the fiber optic sensors. The total number of time points is calculated using the sampling frequency and the time window length.

[0059] The specific implementation of step S03 involves processing the spatiotemporal temperature data matrix using a thermal field reconstruction optimization model. The objective function formula of the thermal field reconstruction optimization model is expressed as follows: ; In the formula, This is the total loss function; The first reconstructed model The spatial location is at the first Temperature values ​​at each time point, in units of It is obtained through deep learning model calculation; For reference temperature, the default value is 298. ; Here is the regularization coefficient, with an empirical value of [value missing]. ~ ; For regularization terms, the following is usually used: Norm; These are the model parameters, obtained through optimization during the training process. The formula for calculating the attention weights is as follows: ; In the formula, Attention weights; These are weighting coefficients, typically with values ​​of 0.4, 0.3, or 0.3. Thermal conductivity of the heat storage medium, in units of This can be obtained through experimental determination or by consulting property tables; For reference thermal conductivity, the default value is 1. ; These are the characteristic dimensions of the thermal storage container, in units of... Obtained through geometric measurements; For reference size, the default value is 1. ; Density of temperature measuring points, in units of It is obtained by calculating the ratio of the number of measuring points to the volume. The default value for the density of the reference measuring points is 1. .

[0060] The specific implementation of step S04 involves calculating the dynamic performance parameters of the thermal storage system based on the temperature field reconstruction results. The formula for calculating the instantaneous thermal storage efficiency is as follows: ; In the formula, For instantaneous heat storage efficiency; For heat storage, the unit is... ; For input heat, the unit is This is obtained through measurement using a heat flow sensor; For reference calories, the default is... .in, The calculation formula is expressed as follows: ; In the formula, Spatial location The reconstruction temperature at the location, in units of The result was obtained through the thermal field reconstruction optimization model. Ambient temperature, unit: The temperature was obtained by measuring the ambient temperature sensor. The volume of the thermal storage medium is expressed in units of 1. The formula for calculating the temperature uniformity coefficient is as follows: ; In the formula, This is the temperature uniformity coefficient; The standard deviation of the temperature field is expressed in units of 1000 ppm. ; Average temperature, in units of .in, The calculation formula is expressed as follows: ; In the formula, For the first The temperature values ​​at each reconstructed temperature point, in units of... The temperature reconstruction results at all spatial locations were obtained through the thermal field reconstruction optimization model. The calculation formula is expressed as follows: ; The formula for calculating the heat loss rate is as follows: ; In the formula, This refers to the heat loss rate; Heat loss, in units of ; Total heat storage, in units of This represents the maximum heat capacity of the thermal storage system. The calculation formula is expressed as follows: ; In the formula, Surface heat transfer coefficient, unit: It is obtained through experimental determination or theoretical calculation; Surface temperature, unit: , obtained through infrared thermal imaging equipment; Surface area, unit: ; The time interval is expressed in units of 1 / 2. .

[0061] The specific implementation of step S05 involves constructing a performance degradation curve and performing online parameter identification. The exponential degradation model formula for the performance degradation curve is expressed as follows: ; In the formula, For time The performance parameter value at any given time can be any one of the instantaneous thermal storage efficiency, temperature uniformity coefficient, or heat loss rate; These are the initial performance parameter values; Runtime, in units of ; For reference time, the default is 86400. ; The decay time constant is expressed in units of 1. ; These are steady-state performance parameters. The formula for calculating the coefficient of variation is as follows: ; In the formula, The coefficient of variation; The standard deviation of the performance parameters refers specifically to the standard deviation of the instantaneous thermal storage efficiency or temperature uniformity coefficient within a time window. This refers to the average value of performance parameters, specifically the average value of instantaneous thermal storage efficiency or temperature uniformity coefficient within a time window. For reference performance, the default value is 1. The parameter update formula for the recursive least squares method is expressed as follows: ; In the formula, For the first The parameter estimation vector at time step; For the first The parameter estimation vector at time step; This is the gain matrix; For the first The observed value at a given time, i.e., the actual measured performance parameter value; The default value is 1, which is used as a reference observation. For the first The regression vector at time step (k) contains information about the input variables. The formula for calculating the gain matrix is ​​as follows: ; In the formula, For the first The covariance matrix at time t; The forgetting factor has an empirical value of 0.95 to 0.99.

[0062] The specific implementation of step S06 involves using cluster analysis to identify the data aggregation distribution patterns of dynamic performance parameters. The density calculation formula for the density-based clustering algorithm is expressed as follows: ; In the formula, For data points The density; For the first The feature vector of each data point contains a three-dimensional vector composed of three performance parameters: instantaneous thermal storage efficiency, temperature uniformity coefficient, and heat loss rate. For the first Feature vectors of data points; This represents the total number of data points. Euclidean distance; For reference distance, the default value is 1; The standard deviation of the density kernel function is given, with an empirical value of 0.5–2.0. The formula for calculating the cluster centers is as follows: ; In the formula, For the first The center vectors of each cluster; For the first The number of data points contained in each cluster; For the first A set of data points for each cluster.

[0063] The specific implementation of step S07 involves establishing an overall performance degradation assessment index for the thermal energy storage system. The calculation formula for the overall performance degradation assessment index is as follows: ; In the formula, As an indicator for evaluating overall performance degradation; These are the weighting coefficients, with default values ​​of 0.5, 0.3, and 0.2 respectively. The initial instantaneous thermal storage efficiency was obtained through measurements during the initial stage of system commissioning. The initial temperature uniformity coefficient is obtained through measurements during the initial stage of system commissioning. This is the initial heat loss rate, obtained through measurements during the initial stage of system operation.

[0064] It should be explained that the principle of the formula for calculating the length of the dynamic sliding time window is based on the thermal inertia characteristics of the thermal storage system, through the thermal response time constant. The formula characterizes the time from receiving heat input to reaching steady state. It comprehensively considers the thermophysical properties of the heat storage medium and the heat transfer characteristics of the system. Compared with the fixed time window method, it can adaptively adjust the sampling window according to the dynamic characteristics of different heat storage systems, which significantly improves the accuracy and adaptability of dynamic monitoring.

[0065] The objective function of the thermal field reconstruction optimization model is based on variational optimization theory, which minimizes a weighted combination of reconstruction error and regularization constraints. To achieve optimal temperature field estimation, the dimensionless processing in this formula eliminates the influence of parameters of different magnitudes, and the regularization term prevents overfitting and improves generalization ability. Compared with traditional interpolation methods, it can accurately reconstruct the complete three-dimensional temperature field from sparse measurement point data, which greatly improves the accuracy and robustness of temperature field reconstruction.

[0066] The principle behind the attention weight calculation formula is based on a multi-factor fusion mechanism, which uses a linear combination of three key parameters: thermal conductivity of the thermal storage medium, container characteristic dimensions, and measurement point density. By determining the spatial attention distribution, this formula enables adaptive adjustment to the characteristics of different thermal storage systems. Compared with the fixed weight method, it can dynamically allocate computing resources according to the physical characteristics of the system, which significantly improves the reconstruction accuracy and computational efficiency.

[0067] The principle behind the instantaneous heat storage efficiency calculation formula is based on the law of conservation of energy, using the ratio of stored heat to input heat. This formula, used for quantitative assessment of thermal storage performance, employs a volume integral form. Accurate calculation of the total heat storage in the three-dimensional temperature field can more accurately reflect the actual performance state of the thermal storage system compared to approximate methods based on average temperature, providing a reliable quantitative indicator for performance evaluation.

[0068] The principle behind the formula for calculating the temperature uniformity coefficient is based on statistical analysis of variance, using the ratio of the standard deviation to the mean. This formula quantifies the uniformity of temperature distribution and can sensitively reflect the non-uniformity of temperature distribution within the thermal storage system. Compared with traditional methods based on temperature difference, it has stronger statistical significance and comparability, providing an important evaluation basis for optimizing the operation of thermal storage systems.

[0069] The principle behind the heat loss rate calculation formula is based on heat transfer theory. It calculates the heat loss power of the heat storage container through surface integration, using the ratio of heat loss to total stored heat. Quantification is performed, where heat loss is achieved through surface integral. The calculations show that the formula takes into account the spatial distribution of surface heat transfer coefficient and temperature difference. Compared with the simplified method based on average parameters, it can more accurately assess the actual heat loss and provides a precise quantitative analysis tool for system energy efficiency optimization.

[0070] The principle of the exponential model of performance degradation curves is based on the physical laws of material aging and system performance degradation, and is expressed through an exponential function. This model describes the nonlinear decay process of performance parameters over time. It includes three key parameters: initial performance, decay time constant, and steady-state performance. Compared with the linear decay model, it can more realistically reflect the long-term performance evolution of thermal storage systems and provides a scientific mathematical basis for predictive maintenance.

[0071] The principle of the recursive least squares parameter update formula is based on adaptive filtering theory, through an online update mechanism. This algorithm tracks system parameter changes in real time and has the advantages of low computational complexity and fast convergence speed. Compared with batch processing methods, it can realize real-time parameter estimation and online model correction, providing an efficient computational method for parameter identification of dynamic systems.

[0072] The principle of density clustering algorithm is based on the local density distribution of points in the data space, using a Gaussian kernel function. The algorithm calculates the density value around each data point and is able to discover clusters of arbitrary shapes and is robust to noise. Compared with distance-based clustering methods, it can better handle non-spherical data patterns and provides a powerful analytical tool for pattern recognition of complex performance data.

[0073] The principle of the overall performance degradation assessment index formula is based on multi-dimensional comprehensive assessment theory, using a weighted average method. By integrating the attenuation information of three key performance parameters—instantaneous thermal storage efficiency, temperature uniformity, and heat loss rate—this formula uses relative change rates to eliminate the influence of different parameter magnitudes. The weight allocation reflects the importance of each parameter to the overall performance. Compared with single-index evaluation methods, it can comprehensively reflect the overall performance status of the thermal storage system and provide a scientific quantitative basis for system maintenance decisions.

[0074] It should be noted that the variables involved in this invention are explained in detail in Table 1.

[0075] Table 1. Variable Explanation Table

[0076] To better understand and implement this invention, the following is a specific application scenario of embodiment 2: A certain thermal storage device uses mixed molten salt as the thermal storage medium. The thermal storage container has a cylindrical structure with a diameter of 8.5m and a height of 12.3m, and the designed thermal storage capacity reaches... J. The heat storage medium is a mixed molten salt of 60% sodium nitrate and 40% potassium nitrate, with an operating temperature range of 280℃ to 560℃.

[0077] The technical team first deployed distributed fiber optic temperature sensors radially and axially inside the thermal storage container. Based on the container's characteristic dimension of 12.3m, the spacing between the distributed fiber optic temperature sensors was chosen to be 2.5m, meeting the requirement of 0.2 times the characteristic dimension. Four fiber optic lines were deployed radially, evenly distributed from the container's center outwards, while five fiber optic lines were deployed axially, evenly spaced from bottom to top. Simultaneously, an infrared thermal imaging device was installed on the outer surface of the thermal storage container, employing a 640×512 pixel resolution thermal imager with a temperature measurement accuracy of ±0.5℃, covering the entire outer surface of the container.

[0078] Based on the thermal response time constant of 1800s for the thermal storage system, the technical team established a dynamic sliding time window with a duration of 3600s, twice the thermal response time constant. Temperature data from distributed fiber optic temperature sensors and infrared thermal imaging equipment were continuously acquired through this dynamic sliding time window, with a sampling frequency set to 0.1Hz. Each time window contained temperature data at 360 time points. Spatially, the fiber optic sensors provided temperature data from 45 measuring points, and the infrared thermal imaging equipment provided surface temperature data from 327,680 pixels, forming a spatiotemporal temperature data matrix with dimensions of 45×360×3 (internal measuring points) and 327,680×360×3 (surface measuring points).

[0079] The technical team utilized a thermal field reconstruction optimization model to process the spatiotemporal temperature data matrix. This model employs a multi-layer encoder-decoder architecture. The encoder consists of three 3D convolutional layers with kernel sizes of 5×5×5, 3×3×3, and 3×3×3, all with a stride of 1, and using ReLU as the activation function. The decoder reconstructs the complete temperature field through three deconvolutional layers, with the deconvolution kernel size corresponding to that of the encoder. An attention mechanism is used in the intermediate layers, with attention weights based on the molten salt thermal conductivity of 0.52. The thermal storage container has a characteristic dimension of 12.3m and a temperature measuring point density of 0.024 points / day. The weight allocation of each region is determined by an adaptive adjustment factor of 0.76.

[0080] During the model training dataset creation process, the technical team collected measured temperature field data from a thermal storage system that had been running continuously for 180 days, covering three typical operating stages: the charging stage, the insulation stage, and the heat release stage. The charging stage lasted for 8 hours, with an input thermal power of... W; the heat preservation stage lasts for 12 hours; the heat release stage lasts for 4 hours, and the output heat power is [not specified]. W. A complete temperature field was generated through numerical simulation as the standard answer, and 43,200 training sample pairs were constructed.

[0081] The thermal field reconstruction optimization model was trained using a representation learning framework based on the information bottleneck principle. The constraint parameter β of the mutual information objective function was set to 0.85. The weights balancing compression ratio and fidelity were dynamically adjusted using the Lagrange multiplier method. The initial learning rate was 0.001, and an exponential decay strategy with a decay rate of 0.95 was employed. The gradient clipping threshold was set to 1.0, the training batch size was 64, and the total number of training epochs was 500. The trained model achieved a mean square error of 1.2℃ in temperature field reconstruction on the validation set, meeting the requirements for engineering applications.

[0082] The complete three-dimensional temperature field distribution inside the thermal storage system was reconstructed using a numerical inversion algorithm. The reconstruction results show that the temperature distribution inside the storage container exhibits a clear stratification during the charging phase, with a top temperature of 548℃, a middle temperature of 526℃, and a bottom temperature of 503℃. During the heat preservation phase, the temperature distribution tends to be uniform, with an average temperature of 535℃ and a temperature gradient of 0.8℃ / m. During the heat release phase, the temperature distribution again shows stratification, ranging from 485℃ at the top to 432℃ at the bottom.

[0083] Based on the temperature field reconstruction results, the technical team calculated the dynamic performance parameters of the thermal storage system. During the charging phase, the instantaneous thermal storage efficiency gradually increased from the initial 0.92 to 0.96, and the stored heat capacity reached [value missing]. J, Input heat J. The temperature uniformity coefficient was 0.082 in the initial stage of heating and decreased to 0.034 in the later stage, indicating that the temperature distribution tended to be uniform. The heat loss rate remained at around 0.028 in the heating stage, increased to 0.045 in the heat preservation stage, and decreased to 0.021 in the heat release stage.

[0084] The technical team constructed a performance degradation curve by linking dynamic performance parameters with the operating time of the thermal storage system. The performance degradation model parameters were updated in real time using an online parameter identification method, with the forgetting factor for the recursive least squares method set to 0.98. When the instantaneous thermal storage efficiency data dispersion was 0.033, which is less than the concentration requirement threshold of 0.05, a centralized processing approach was adopted to uniformly process data from all measuring points, improving computational efficiency. When the temperature uniformity coefficient data dispersion was 0.187, which is greater than the complexity processing threshold of 0.15, a distributed processing strategy was adopted, distributing the computational tasks to four parallel processing units.

[0085] The technical team used cluster analysis to identify the data aggregation and distribution patterns of dynamic performance parameters, as shown in Table 2.

[0086] Table 2. Cluster analysis results of dynamic performance parameters

[0087] Cluster analysis identified four main data aggregation patterns, corresponding to different system operating states. Cluster center estimation and boundary determination algorithms were used to automatically classify and organize dynamic performance parameters, establishing a multimodal performance evaluation criterion. Cluster center 1 corresponds to the optimal system operating state, while cluster center 4 corresponds to a state with significant system performance degradation.

[0088] The technical team established an overall performance degradation assessment index for the thermal energy storage system. This index combines the instantaneous thermal storage efficiency degradation rate (0.024 / day), the temperature uniformity coefficient degradation rate (0.018 / day), and the heat loss rate growth rate (0.012 / day), and calculates the comprehensive assessment index value through weighted calculation. During the initial 30 days of operation, the overall system performance degradation assessment index was 0.187, within the range of [0, 0.25), indicating normal system performance. After 60 days of operation, the assessment index rose to 0.334, entering the range of [0.25, 0.65), indicating a slight degradation in system performance requiring optimization of operating parameters. The technical team adjusted the molten salt circulation flow rate from... Increase to The heat charging temperature curve was optimized, reducing the evaluation index to 0.268.

[0089] like Figure 2 As shown, the instantaneous thermal energy storage efficiency of the thermal energy storage system exhibits a gradual decreasing trend during the 180-day operation period. The efficiency remains above 0.95 for the first 30 days, then begins to decline slowly, dropping to 0.91 at 90 days and further to 0.87 at 150 days. The temperature uniformity coefficient shows the opposite trend to the thermal energy storage efficiency, initially remaining around 0.035, gradually increasing with operating time, reaching 0.128 at 180 days, indicating a deterioration in the uniformity of temperature distribution.

[0090] like Figure 3 As shown, the heat loss rate remained relatively stable at around 0.028 during the initial stage of operation, but began to rise significantly after 90 days, reaching 0.065 after 180 days, indicating a decline in the thermal insulation performance of the thermal storage system. This trend is closely related to changes in the thermophysical properties of the thermal storage medium, the aging of the container insulation material, and the reduction in the system's sealing performance.

[0091] The technical team evaluated the system status at different operating stages using multimodal performance evaluation criteria, as shown in Table 3.

[0092] Table 3 Multimodal performance evaluation results

[0093] Throughout the operational cycle, the technical team promptly identified trends in system performance through real-time monitoring and evaluation. When an abnormal increase in the temperature uniformity coefficient was detected, the molten salt circulation system was immediately inspected, revealing a 13% decrease in circulation pump efficiency, which was promptly addressed. When the heat loss rate exceeded the expected threshold, inspection revealed localized detachment of the vessel's external insulation layer, which was also promptly repaired. These preventative maintenance measures effectively slowed the rate of system performance degradation.

[0094] Compared to traditional periodic inspections and experience-based maintenance, the evaluation method provided by this invention has significant technical advantages. Traditional methods mainly rely on manual inspections and fixed-point temperature measurements, which cannot obtain a complete temperature field distribution inside the thermal storage system and make it difficult to detect local performance anomalies in a timely manner. This invention achieves continuous monitoring of the internal temperature of the thermal storage medium through distributed fiber optic sensors, obtains the external surface temperature distribution through infrared thermal imaging equipment, and obtains a complete three-dimensional temperature field through thermal field reconstruction technology, providing a comprehensive and accurate data foundation for performance evaluation.

[0095] Traditional performance evaluation methods rely on single indicators or simple parameter calculations, lacking in-depth analysis of the system's dynamic characteristics. The multi-parameter comprehensive evaluation system established in this invention can comprehensively evaluate system performance from multiple dimensions such as thermal storage efficiency, temperature uniformity, and heat loss, and accurately captures the system's transient characteristics through dynamic sliding time window technology. The thermal field reconstruction model based on the information bottleneck principle can effectively reduce computational complexity while ensuring reconstruction accuracy, making it suitable for online real-time applications.

[0096] Traditional methods cannot adaptively select processing strategies based on data characteristics, and are prone to misjudgment when dealing with complex operating conditions. The centralized and distributed adaptive processing mechanism introduced in this invention can automatically select the optimal processing method based on the dispersion and complexity of the data. Cluster analysis technology can identify performance characteristics under different operating modes and establish targeted evaluation criteria. This intelligent processing method significantly improves the accuracy and reliability of evaluation results, providing a scientific basis for the precise operation and maintenance of thermal storage systems.

[0097] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for evaluating the overall performance degradation of a thermal energy storage system, characterized in that, Distributed fiber optic temperature sensors are arranged radially and axially inside the thermal storage medium of the thermal energy storage system, while an infrared thermal imaging device is installed on the outer surface of the thermal storage container for surface temperature field measurement. A dynamic sliding time window is established, and temperature data from the distributed fiber optic temperature sensors and the infrared thermal imaging device are continuously collected through the dynamic sliding time window to form a spatiotemporal temperature data matrix. The spatiotemporal temperature data matrix is ​​processed using a thermal field reconstruction optimization model, and the complete three-dimensional temperature field distribution inside the thermal storage system is reconstructed through a numerical inversion algorithm to obtain the temperature field reconstruction result. Based on the temperature field reconstruction result, the dynamic performance parameters of the thermal storage system are calculated, including instantaneous thermal storage efficiency, temperature uniformity coefficient, and heat loss rate. The dynamic performance parameters are compared with the operating time of the thermal storage system to construct a performance degradation curve, and the performance degradation model parameters are updated in real time through an online parameter identification method. Cluster analysis is used to identify the data aggregation and distribution patterns of dynamic performance parameters. When cluster analysis identifies the data aggregation and distribution patterns of dynamic performance parameters, cluster center estimation and boundary determination algorithms are used to automatically classify and organize the dynamic performance parameters and establish multimodal performance evaluation criteria. An overall performance degradation evaluation index for thermal energy storage systems is established, which integrates the instantaneous thermal storage efficiency degradation rate, temperature uniformity coefficient degradation rate, and heat loss rate growth rate.

2. The method for evaluating the overall performance degradation of a thermal energy storage system according to claim 1, characterized in that, The spacing between the distributed optical fiber temperature sensors is 0.1 to 0.3 times the characteristic size of the thermal storage container. The distributed optical fiber temperature sensor is a temperature measurement device based on the Raman scattering principle, which measures the temperature distribution along the optical fiber path by measuring the change in the backscattering intensity of the light signal in the optical fiber.

3. The method for evaluating the overall performance degradation of a thermal energy storage system according to claim 2, characterized in that, The characteristic dimensions of the thermal storage container refer to the main geometric dimensions of the thermal storage container, including the maximum value of diameter, height, or length. The duration of the dynamic sliding time window is 1.5 to 3 times the thermal response time constant of the thermal storage system.

4. The method for evaluating the overall performance degradation of a thermal energy storage system according to claim 3, characterized in that, The dynamic sliding time window is a time series data processing technology that continuously monitors and analyzes the dynamic characteristics of the system by moving a fixed-length time window continuously on the time axis. The thermal response time constant of the thermal storage system is the characteristic time required for the thermal storage system to reach a steady state from receiving heat input.

5. The method for evaluating the overall performance degradation of a thermal energy storage system according to claim 4, characterized in that, The spatiotemporal temperature data matrix is ​​a three-dimensional data structure formed by arranging temperature measurement data according to spatial location and time order. The thermal field reconstruction optimization model is a temperature field reconstruction algorithm based on sparse representation theory and variational optimization. It reconstructs the complete temperature field from finite measurement point data by minimizing the weighted sum of reconstruction error and regularization term.

6. The method for evaluating the overall performance degradation of a thermal energy storage system according to claim 5, characterized in that, The numerical inversion algorithm is a calculation method that infers the internal state of the system from the observation data. In the temperature field reconstruction, the internal temperature distribution is obtained by solving the inverse heat transfer problem. The temperature field reconstruction result is the complete three-dimensional temperature field distribution data inside the thermal storage system obtained by the numerical inversion algorithm.

7. The method for evaluating the overall performance degradation of a thermal energy storage system according to claim 6, characterized in that, The instantaneous heat storage efficiency is the ratio of stored heat to input heat, the temperature uniformity coefficient is the ratio of the standard deviation of the temperature field to the average temperature, and the heat loss rate is the ratio of dissipated heat to total stored heat.

8. The method for evaluating the overall performance degradation of a thermal energy storage system according to claim 7, characterized in that, The performance degradation curve is a functional relationship describing the change of dynamic performance parameters over time. The online parameter identification method is an adaptive algorithm for real-time estimation of system model parameters, which dynamically updates the model parameters through recursive least squares or Kalman filtering techniques.

9. The method for evaluating the overall performance degradation of a thermal energy storage system according to claim 8, characterized in that, When the dispersion of instantaneous thermal storage efficiency data is less than the concentration requirement threshold of 0.05, a centralized processing method is adopted to improve efficiency. When the dispersion of temperature uniformity coefficient data is greater than 0.15, a distributed processing strategy is adopted to deal with complexity.

10. The method for evaluating the overall performance degradation of a thermal energy storage system according to claim 9, characterized in that, The centralized processing method refers to a unified processing strategy adopted when the data is relatively discrete, which improves processing accuracy and efficiency by concentrating computing resources. The distributed processing strategy refers to a decentralized processing method adopted when the data is highly complex, which distributes computing tasks to multiple processing units for parallel execution.

Images