Multiphysics-based reactor optimization methods, devices, equipment, and storage media

Abstract

This invention relates to the field of reactor technology, and discloses a reactor optimization method, apparatus, device, and storage medium based on multiphysics. The method includes: performing multi-level adaptive decomposition and reconstruction on the raw sensor data for acquiring multiphysics parameters; constructing implicit parameter relationships and dynamic evolution characteristics by performing spatiotemporal topological representation of multi-scale sensing features, and extracting multiphysics coupling features and performing feature causal analysis on the implicit parameter relationships and dynamic evolution characteristics between the generated physical field parameters; constructing an adaptive multiphysics coupling model and performing multi-objective collaborative optimization on the obtained multiphysics coupling features and feature causal relationship network; and performing full lifecycle data analysis and prediction on the generated final reactor optimization model for a target high-voltage toroidal hollow reactor, generating an optimization scheme for the full lifecycle management and maintenance of the reactor. This application combines multiphysics coupling and reactor optimization characteristics to improve the optimization effect of the reactor.

Description

Technical Field

[0001] This invention relates to the field of reactor technology, and in particular to a reactor optimization method, apparatus, device, and storage medium based on multiphysics. Background Technology

[0002] As power grids expand and operational requirements increase, reactors face more severe electromagnetic, thermal, and mechanical challenges. Accurately optimizing the design and operating parameters of high-voltage toroidal air-core reactors is crucial for power equipment manufacturers, grid operators, and research institutions, directly impacting equipment performance, reliability, and lifespan. Therefore, developing effective multiphysics-based optimization methods for designing and managing these reactors is essential for ensuring the stability and economy of power systems.

[0003] Currently, finite element analysis and numerical simulation techniques are used to analyze single physical fields, or artificial intelligence algorithms are applied to reactor parameter optimization to improve design efficiency. However, these methods still face challenges in integrating multi-physics coupling effects, handling nonlinear characteristics, and adapting to complex operating environments. Furthermore, these optimization methods often neglect unique characteristics of high-voltage toroidal air-core reactors, such as the influence of geometry, material nonlinearity, and external environmental factors, which can significantly impact reactor performance and lifespan. In other words, existing optimization methods for high-voltage toroidal air-core reactors ignore multi-physics coupling and the unique characteristics of reactors, resulting in low accuracy of the final optimization results and difficulty in meeting the stringent requirements of modern power systems. Summary of the Invention

[0004] The main objective of this invention is to address the problem that existing optimization methods for high-voltage toroidal air-core reactors neglect multi-physics coupling and some unique characteristics of reactors, resulting in low accuracy of the final optimization results and difficulty in meeting the stringent requirements of modern power systems.

[0005] The first aspect of this invention provides a reactor optimization method based on multiphysics. The method includes: acquiring original sensing data of multiphysics parameters corresponding to a target high-voltage toroidal hollow reactor; performing multi-level adaptive decomposition and reconstruction on the original sensing data to obtain multi-scale sensing features; constructing a spatiotemporal topological representation of the multi-scale sensing features to generate implicit parameter relationships and dynamic evolution features among the various physical field parameters; extracting multiphysics coupling features and performing feature causal analysis on the implicit parameter relationships and dynamic evolution features to obtain a multiphysics coupling feature and feature causal relationship network; constructing an adaptive multiphysics coupling model on the multiphysics coupling features and feature causal relationship network to obtain an initial reactor optimization model with dynamically optimized structure and parameters; performing multi-objective collaborative optimization on the initial reactor optimization model to obtain a final reactor optimization model; and performing full lifecycle data analysis and prediction on the final reactor optimization model to generate an optimization scheme for the full lifecycle management and maintenance of the target high-voltage toroidal hollow reactor.

[0006] Optionally, in a first implementation of the first aspect of the present invention, the step of performing multi-level adaptive decomposition and reconstruction on the original sensing data to obtain multi-scale sensing features includes: performing adaptive multi-scale decomposition on the original sensing data to obtain multiple sensing mode functions, and performing continuous wavelet transform on the sensing mode functions to obtain time-frequency sensing features; performing adaptive soft threshold denoising on the time-frequency sensing features to obtain denoised time-frequency sensing features, and performing singular value decomposition and reconstruction on the denoised time-frequency sensing features to obtain multiple sensing time-series components; performing tensor decomposition and feature compression on the sensing time-series components to obtain a compressed feature space representation, and performing constraint optimization of the compressed feature space representation based on each of the physical field parameters to obtain multi-scale sensing features.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the construction of spatiotemporal topological representation of the multi-scale sensing features, generating implicit parameter relationships and dynamic evolution features among various physical field parameters, includes: filtering the distances between feature data points in the multi-scale sensing features based on a preset point distance threshold, constructing a sensing feature border relationship graph, and performing simplex identification and boundary operator generation on the sensing feature border relationship graph to obtain a feature simplex chain complex; performing homology group calculation and persistence analysis on the feature simplex chain complex to generate a feature persistence map, and extracting features from the feature persistence map to obtain... Multi-resolution topological feature description; the multi-resolution topological feature description is subjected to sliding window continuous cohomology calculation and multi-scale diffusion map construction to obtain nonlinear dimensionality-reduced feature diffusion coordinates, and based on a preset mapping discretization strategy, the feature diffusion coordinates are subjected to coordinate projection and node discretization optimization to obtain a feature discrete topological skeleton; the feature discrete topological skeleton is subjected to time-varying continuous cohomology analysis and probability transmission calculation to obtain a feature similarity measure of the topological structure, and a preset multi-level graph neural network is used to analyze the feature similarity measure by multi-level graph convolution and temporal attention mechanism to generate the parameter implicit relationship and dynamic evolution characteristics between various physical field parameters.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the step of extracting multi-physics coupling features and performing feature causal analysis on the implicit parameter relationships and the dynamic evolution features to obtain a multi-physics coupling feature and feature causal relationship network includes: performing continuous cohomology analysis and multi-scale analysis on the implicit parameter relationships to obtain static topological feature descriptors; performing time series analysis on the dynamic evolution features to obtain dynamic topological feature descriptors; and performing descriptor fusion on the static topological feature descriptors and the dynamic topological feature descriptors to obtain a spatiotemporal topological feature representation; and constructing the discrete gradient corresponding to the spatiotemporal topological feature representation. The discrete gradient vector field is used to identify critical points and perform persistence pairing to obtain the reactance key points and reactance separation lines of the multiphysics field. Multidimensional probability density estimation and marginal distribution transformation are performed on the reactance key points and reactance separation lines to obtain the joint probability distribution of the multiphysics coupling. Multivariate empirical mode decomposition is then performed on the joint probability distribution to obtain the time-varying mode function of the multiphysics field. Nonlinear Granger causality tests are performed on the time-varying mode function to obtain a preliminary causal relationship network. Conditional mutual information analysis and graph structure learning are then performed on the preliminary causal relationship network to obtain the multiphysics coupling features and the causal relationship network between features.

[0009] Optionally, in the fourth implementation of the first aspect of the present invention, the step of constructing an adaptive multiphysics coupling model for the multiphysics coupling features and feature causal relationship network to obtain an initial reactor optimization model with dynamically optimized structure and parameters includes: performing modular decomposition on the multiphysics coupling features and feature causal relationship network to obtain a structured representation of the multiphysics parameters, and encoding the structured representation of the multiphysics parameters using neural ordinary differential equations to obtain a feature description representation with consistent physical state; performing residual calculation and adaptive adjustment on the feature description representation to obtain dynamically optimized compensation data, and performing Bayesian inference and variational analysis on the dynamically optimized compensation data to obtain the compensation parameter distribution and uncertainty estimation parameters; performing model structure search and wavelet decomposition on the compensation parameter distribution and the uncertainty estimation parameters to obtain a multi-scale dynamic characteristic representation, and integrating and dynamically weighting the multi-scale dynamic characteristic representation to generate an initial reactor optimization model with dynamically optimized structure and parameters.

[0010] Optionally, in a fifth implementation of the first aspect of the present invention, the step of performing multi-objective collaborative optimization on the initial reactor optimization model to obtain a final reactor optimization model includes: performing multi-objective function transformation on the initial reactor optimization model and the feature causal relationship network to obtain a multi-dimensional reactor performance index function, and performing regression analysis on the multi-dimensional reactor performance index function for various performance parameters of the reactor to obtain a reactor performance prediction probability distribution; uniformly sampling the reactor performance prediction probability distribution to obtain a reactor parameter combination, and performing graph structure mapping and electromagnetic field iterative optimization on the reactor parameter combination to obtain a performance index response surface; and performing multi-objective reactor performance trade-off analysis and multi-criteria decision analysis on the performance index response surface to obtain a final reactor optimization model with corresponding high performance index.

[0011] Optionally, in the sixth implementation of the first aspect of the present invention, the step of performing full life-cycle data analysis and prediction of the target high-voltage toroidal air-core reactor on the final reactor optimization model to generate an optimization scheme for the full life-cycle management and maintenance of the target high-voltage toroidal air-core reactor includes: performing multi-scale data integration on the final reactor optimization model to obtain a global performance characterization of the reactor, and synchronizing the global performance characterization with time-series data to obtain consistent historical operating data of the target high-voltage toroidal air-core reactor; performing multi-level time-series analysis on the consistent historical operating data to obtain a future performance prediction of the target high-voltage toroidal air-core reactor, and performing multi-dimensional correlation analysis on the future performance prediction to obtain potential fault mode identification results; performing multi-scenario decision tree analysis and adaptive optimization processing on the potential fault mode identification results to obtain a dynamically adjusted maintenance strategy, and performing full-cycle data integration and correlation analysis on the dynamic maintenance strategy to obtain an optimization scheme for the full life-cycle management and maintenance of the target high-voltage toroidal air-core reactor.

[0012] A second aspect of the present invention provides a reactor optimization device based on multiphysics, comprising: a feature reconstruction module for acquiring original sensing data of multiphysics parameters corresponding to a target high-voltage toroidal hollow reactor, and performing multi-level adaptive decomposition and reconstruction on the original sensing data to obtain multi-scale sensing features; and a causal analysis module for constructing spatiotemporal topological representations of the multi-scale sensing features, generating implicit parameter relationships and dynamic evolution features between various physical field parameters, and extracting multiphysics coupling features and special features from the implicit parameter relationships and dynamic evolution features. The system employs a causal analysis approach to obtain multi-physics coupling characteristics and a characteristic causal relationship network. A collaborative optimization module is used to construct an adaptive multi-physics coupling model based on these characteristics and the characteristic causal relationship network, resulting in an initial reactor optimization model with dynamically optimized structure and parameters. This initial reactor optimization model is then subjected to multi-objective collaborative optimization to obtain a final reactor optimization model. An optimization module is used to perform full lifecycle data analysis and prediction on the final reactor optimization model, generating an optimization scheme for the full lifecycle management and maintenance of the target high-voltage toroidal air-core reactor.

[0013] A third aspect of the present invention provides a reactor optimization device based on multiphysics, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the reactor optimization device based on multiphysics to perform the various steps of the reactor optimization method based on multiphysics described above.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the steps of the above-described multiphysics-based reactor optimization method.

[0015] The above-described reactor optimization method, apparatus, equipment, and storage medium based on multiphysics are described above. In this embodiment, the original sensing data of the multiphysics parameters corresponding to the target high-voltage toroidal air-core reactor are acquired, and the original sensing data undergoes multi-level adaptive decomposition and reconstruction to obtain multi-scale sensing features. Spatiotemporal topological representations are constructed from the multi-scale sensing features to generate implicit parameter relationships and dynamic evolution features among the various physical field parameters. Multiphysics coupling feature extraction and feature causal analysis are performed on the implicit parameter relationships and dynamic evolution features to obtain a multiphysics coupling feature and feature causal relationship network. An adaptive multiphysics coupling model is constructed from the multiphysics coupling features and feature causal relationship network to obtain an initial reactor optimization model with dynamically optimized structure and parameters. Multi-objective collaborative optimization is performed on the initial reactor optimization model to obtain a final reactor optimization model. The final reactor optimization model is then used for full lifecycle data analysis and prediction of the target high-voltage toroidal air-core reactor to generate an optimization scheme for the full lifecycle management and maintenance of the target high-voltage toroidal air-core reactor. Compared to existing technologies, this application obtains multi-scale sensing features by performing multi-level adaptive decomposition and reconstruction on the original sensing data corresponding to the multi-physics parameters of the target high-voltage toroidal air-core reactor. Then, it constructs a spatiotemporal topological representation of these multi-scale sensing features, establishing implicit parameter relationships and dynamic evolution characteristics among the various physics parameters. Furthermore, it extracts multi-physics coupling features and performs causal analysis on these implicit relationships and dynamic evolution characteristics, obtaining the reactor's multi-physics coupling features and causal relationship network. Finally, it constructs an adaptive multi-physics coupling model and performs multi-objective collaborative optimization on this network, generating a final reactor optimization model. This final reactor optimization model is then used to analyze and predict the full lifecycle data of the target high-voltage toroidal air-core reactor, generating an optimized scheme for the reactor's full lifecycle management and maintenance. Therefore, in the optimization process of the high-voltage toroidal air-core reactor, by combining multi-physics coupling and the reactor's reactance characteristics, the final optimization effect of the reactor is improved.

[0016] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the first embodiment of the reactor optimization method based on multiphysics field in this invention;

[0019] Figure 2 This is a schematic diagram of one embodiment of the reactor optimization device based on multiphysics field in this invention;

[0020] Figure 3 This is a schematic diagram of one embodiment of the reactor optimization device based on multiphysics field in this invention. Detailed Implementation

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

[0022] The terms "comprising" and "having," and any variations thereof, used in the embodiments of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0023] To facilitate understanding of this embodiment, the specific process of this embodiment is described below. Please refer to [link / reference]. Figure 1 The first embodiment of the reactor optimization method based on multiphysics in this invention includes:

[0024] 101. Obtain the original sensing data of the multi-physics field parameters corresponding to the target high-voltage toroidal hollow reactor, and perform multi-level adaptive decomposition and reconstruction on the original sensing data to obtain multi-scale sensing features.

[0025] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0026] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0027] In this embodiment, the aforementioned high-voltage toroidal air-core reactor is a reactor used to limit the magnitude of alternating current and ensure the stability of current in the power system. Its structural characteristics include being oil-free and coreless, with magnetic flux forming a loop through the air, hence the name air-core reactor. The toroidal design further enhances its structural strength and heat dissipation performance. The original sensing data is subjected to adaptive multi-scale decomposition to obtain multiple sensing mode functions. Continuous wavelet transform is then performed on the sensing mode functions to obtain time-frequency sensing features. Adaptive soft-threshold denoising is applied to the time-frequency sensing features to obtain denoised time-frequency sensing features. Singular value decomposition and reconstruction are then performed on the denoised time-frequency sensing features to obtain multiple sensing time-series components. Tensor decomposition and feature compression are performed on the sensing time-series components to obtain a compressed feature space representation. Based on the physical field parameters, the compressed feature space representation is optimized by physical field parameter constraints to obtain multi-scale sensing features.

[0028] In practical applications, the raw sensing data of multiple physical field parameters, such as electromagnetic field, thermal field, and mechanical stress field, of the target high-voltage toroidal hollow reactor are first acquired through a multi-scale sensor network. These sensors may include Hall effect sensors, fiber Bragg grating sensors, and piezoelectric sensors. Since the acquired raw data usually contains various noises and interferences, adaptive multi-scale decomposition is required. This is achieved by using variational mode decomposition (VMD) to decompose the complex nonlinear, non-stationary signal into a series of intrinsic eigenmode functions (IMFs), i.e., multiple sensing mode functions (where these IMFs represent the periodic patterns of the signal at different time scales, thus enabling the separation of information from different frequency components). Then, by selecting an appropriate mother wavelet function (such as the Morlet wavelet or Mexican wavelet), further optimization is performed. The HAT wavelet transform is applied to each sensing mode function to obtain the time-frequency representation of the signal, i.e., the time-frequency sensing features (these features reveal the energy distribution of the signal at different times and frequencies, helping to identify key time-frequency patterns). Adaptive soft-threshold denoising is then applied to these features, dynamically adjusting the threshold based on the local characteristics of the time-frequency sensing features. This effectively removes noise while preserving important information from the time-frequency sensing features, resulting in denoised time-frequency sensing features. Singular value decomposition (SVD) and reconstruction are then performed on the denoised time-frequency sensing features. This involves decomposing the signal into orthogonal basis vectors, ranking these basis vectors by importance, and selecting the most important singular values ​​and corresponding singular vectors for reconstruction to obtain the signal. The main components of the signal are identified, namely various sensing time-series components (which may include trend, periodic, and random components, representing different aspects of the signal). To further compress the data and extract more abstract features, Tucker decomposition is used to perform tensor decomposition on the sensing time-series components to capture the interactions between different physical quantities. Principal component analysis (PCA) is then used to compress the tensor decomposition results, mapping the high-dimensional data to a low-dimensional feature space. Furthermore, to ensure that the extracted features are consistent with the physical field parameters, a multi-objective optimization problem (i.e., a regularization term guided by physical knowledge) is constructed based on each physical field parameter to constrain and optimize the compressed feature space representation. The objective function is:

[0029] ;

[0030] in, It is the first Prediction function for each physical field parameter, These are the corresponding observed values. These are weights, used to adjust the impact of each error term. It is a regularization parameter used to balance the error term and the regularization term. The regularization function optimizes and obtains multi-scale sensing features. These features not only contain key information from the original data but also maintain physical consistency. For example, for a high-voltage toroidal hollow reactor, multi-physical field data such as current density, magnetic field strength, temperature distribution, and mechanical stress are collected. Through VMD, the complex temperature signal can be decomposed into multiple intrinsic mode functions (IMFs), including long-term trends, diurnal cycles, and short-term fluctuations. Continuous wavelet transform is performed on these IMFs to obtain the time-frequency characteristics of temperature changes, such as identifying rapid temperature rises caused by load changes. Multi-channel singular spectrum analysis extracts common time-series patterns from the multi-channel data, such as the correlation between current density and temperature. Finally, physical constraint optimization ensures that the extracted features conform to the physical characteristics of the reactor, such as satisfying Ampere's law and the heat conduction equation. Thus, the obtained multi-scale sensing features accurately reflect the current state of the reactor.

[0031] 102. Construct spatiotemporal topological representations of multi-scale sensing features, generate implicit parameter relationships and dynamic evolution features among various physical field parameters, and extract multi-physics coupling features and perform feature causal analysis on implicit parameter relationships and dynamic evolution features to obtain a multi-physics coupling feature and feature causal relationship network.

[0032] In this embodiment, based on a preset point distance threshold, the distances between various feature data points in the multi-scale sensing features are filtered to construct a sensing feature border relationship graph. Simplex identification and boundary operator generation are then performed on the sensing feature border relationship graph to obtain a feature simplex chain complex. Homology group calculation and persistence analysis are performed on the feature simplex chain complex to generate a feature persistence map. Feature extraction is then performed on the feature persistence map to obtain a multi-resolution topological feature description. Sliding window persistent homology calculation and multi-scale diffusion map construction are performed on the multi-resolution topological feature description to obtain nonlinearly reduced feature diffusion coordinates. Based on a preset mapping discretization strategy, coordinate projection and node discretization optimization are performed on the feature diffusion coordinates to obtain a feature discrete topological skeleton. Time-varying persistent homology analysis and probability transfer calculation are performed on the feature discrete topological skeleton to obtain a feature similarity measure of the topological structure. A preset multi-level graph neural network is used to analyze the feature similarity measure using multi-level graph convolution and temporal attention mechanisms to generate parameters between various physical field parameters. The method involves: implicit relationships and dynamic evolution characteristics; performing continuous coherence analysis and multi-scale analysis on the implicit relationships of the parameters to obtain static topological feature descriptors; performing time series analysis on the dynamic evolution characteristics to obtain dynamic topological feature descriptors; fusing the static and dynamic topological feature descriptors to obtain a spatiotemporal topological feature representation; constructing a discrete gradient vector field corresponding to the spatiotemporal topological feature representation; identifying critical points and performing persistent pairing on the discrete gradient vector field to obtain reactive key points and reactive separation lines of the multiphysics field; performing multidimensional probability density estimation and marginal distribution transformation on the reactive key points and reactive separation lines to obtain a joint probability distribution of multiphysics field coupling; performing multivariate empirical mode decomposition on the joint probability distribution to obtain time-varying mode functions of the multiphysics field; performing nonlinear Granger causality tests on the time-varying mode functions to obtain a preliminary causal relationship network; and performing conditional mutual information analysis and graph structure learning on the preliminary causal relationship network to obtain multiphysics field coupling characteristics and causal relationship networks between characteristics.

[0033] In practical applications, firstly, based on a preset point distance threshold ε, the distances between various feature data points in the multi-scale sensing features are filtered to construct a sensing feature adjacency graph. This is achieved by calculating the Euclidean distance between pairs of points in the feature space and connecting pairs of points with a distance less than ε. Then, simplex identification and boundary operator generation are performed on this adjacency graph using methods such as Vietoris-Rips complex or Čech complex. Specifically, if the pairwise distance between n+1 points is less than a given threshold, these n+1 points form an n-dimensional simplex, thus generating a feature simplex chain complex. Finally, the Smith standard of the boundary matrix of the chain complex is calculated. This method employs homology group calculation on characteristic simplex chain complexes and persistence analysis by tracking the appearance and disappearance of topological features (such as connected components, loops, and holes) at different scales, generating a feature persistence map (where the persistence map can be represented by points on a two-dimensional plane, with the horizontal axis representing the scale of feature appearance and the vertical axis representing the scale of feature disappearance). Then, a persistence landscape function is used to extract features from this persistence map, obtaining a multi-resolution topological feature description (including persistence landscape function, persistence image, or persistence entropy, etc.). Finally, a sliding window persistence homology calculation is performed on the multi-resolution topological feature description to capture the topological features of time-varying data, and to calculate the settlement... The results are used to construct multi-scale diffusion maps to build diffusion kernels at different scales to describe the geometric structure of the data, thereby obtaining nonlinearly reduced feature diffusion coordinates. Then, based on a pre-defined mapping discretization strategy, an improved Mapper algorithm is used to perform coordinate projection and node discretization optimization on the feature diffusion coordinates, mapping high-dimensional feature diffusion coordinate data to a low-dimensional space. Discrete topological representations are then constructed through clustering and connectivity analysis to obtain a feature discrete topological skeleton. Finally, time-varying persistent cohomology analysis is performed on the feature discrete topological skeleton to capture the evolution of the topological structure over time, and the Wasserstein distance or bottleneck distance between persistent graphs is calculated. The results of continuous cohomology analysis are used for probability transfer calculations to quantify the similarity between different topologies, obtaining a feature similarity measure of the topologies. Then, a pre-defined multi-layer graph neural network is used to analyze the feature similarity measure through multi-layer graph convolution and temporal attention mechanisms. Specifically, graph convolution operations capture the local correlations between feature similarity measure nodes, while the temporal attention mechanism models long-term dependencies. By stacking multiple layers of graph convolution and attention layers, the network progressively extracts higher-level feature representations, ultimately revealing the complex implicit relationships and dynamic evolution characteristics between various physical field parameters, thus generating implicit parameter relationships and dynamic evolution characteristics between various physical field parameters. For example, firstly, topological representations of parameters such as current density, magnetic field strength, temperature, and mechanical stress are constructed.Through continuous cohomology analysis, a significant one-dimensional cyclic structure in the temperature field at a certain scale was discovered, which may reflect the thermal cycling pattern inside the reactor. Then, multi-scale diffusion maps were used to reduce the dimensionality of the high-dimensional physical field data to a lower-dimensional space while preserving key topological features. Furthermore, when constructing the feature discrete topological skeleton, tightly connected clusters of current density and temperature distribution were found in certain regions, which may indicate potential hotspot areas. Time-varying continuous cohomology analysis was then used to track the evolution of these hotspot areas over time. Finally, graph neural networks were used to learn the complex nonlinear relationships between current, magnetic field, temperature, and stress. For example, it may be discovered that sudden changes in current density affect the temperature distribution and mechanical stress in specific regions through a certain time delay. These in-depth insights can provide important basis for reactor performance optimization and fault prediction.

[0034] Secondly, continuous coherence analysis and multi-scale analysis are performed on the implicit relationships of parameters. By calculating the Betti number and persistence spectrum at different scales, the topological features of the data are captured, thereby obtaining static topological feature descriptors. At the same time, time series analysis is performed on the dynamic evolution features to obtain dynamic topological feature descriptors. Then, methods such as feature concatenation or attention mechanisms are used to fuse these two descriptors to obtain a comprehensive spatiotemporal topological feature representation. Furthermore, based on the spatiotemporal topological feature representation, a vector field is formed by calculating the local gradient of each point in the feature space, constructing a corresponding discrete gradient vector field. This discrete gradient vector field then undergoes critical point identification and persistent pairing. Singularities in the vector field are identified using Morse theory, and these singularities are paired using a persistent algorithm, ultimately obtaining the reactance key points and reactance dividing lines of the multiphysics field (where these key points and dividing lines represent the interaction and transformation regions between different physical fields in the reactor). Subsequently, multidimensional probability density estimation and marginal distribution transformation are performed on these reactance key points and dividing lines. Specifically, multidimensional probability density is estimated using kernel density estimation or Gaussian mixture models, and marginal distribution transformation is performed using methods such as the Copula function, thereby obtaining the multiphysics field. The coupled joint probability distribution (which describes the complex correlation between different physical quantities) is then subjected to multivariate empirical mode decomposition. Through an iterative screening process, the complex multivariate signal is decomposed into a series of intrinsic mode functions, resulting in time-varying mode functions of the multiphysics field. Nonlinear Granger causality tests are then performed on the time-varying mode functions to capture the nonlinear and time-varying causal relationships between variables, thus obtaining a preliminary causal relationship network. Conditional mutual information analysis is then performed on this preliminary causal relationship network to evaluate the strength of the direct causal relationship between each pair of variables by calculating the conditional mutual information. Graph structure learning algorithms (such as PC or FCI algorithms) are then used to optimize the network structure, ultimately yielding the multiphysics field coupling features and the causal relationship network between these features. For example, continuous cohomology analysis can reveal a significant one-dimensional cyclic structure in the electromagnetic field at a certain scale, which may reflect a periodic pattern in the current distribution. Simultaneously, dynamic analysis of the thermal field may reveal the evolution of temperature distribution over time, such as the formation and dissipation of hotspots. By fusing these static and dynamic features, a comprehensive spatiotemporal topological representation can be obtained, potentially showing the interaction patterns between the electromagnetic field, thermal field, and mechanical stress field. Further analysis of this spatiotemporal topological representation can identify several key points, such as the region with the highest current density, the point with the highest temperature, and the location of the greatest mechanical stress. The relationships between these key points can then be modeled using Copula functions, resulting in a joint probability distribution of multi-physics coupling. This distribution may reveal, for example, the probability distribution of temperature and mechanical stress increases when the current density exceeds a certain threshold.Finally, through nonlinear Granger causality tests and graph structure learning, it was discovered that changes in current density affect the temperature distribution in a specific region with a certain time delay, and changes in temperature distribution, in turn, affect the distribution of mechanical stress. These causal relationships can form a network to understand the complex physical processes inside the reactor, providing an important basis for optimized design and predictive maintenance.

[0035] 103. An adaptive multiphysics coupling model is constructed based on the multiphysics coupling characteristics and feature causal relationship network to obtain an initial reactor optimization model with dynamic optimization structure and parameters. The initial reactor optimization model is then subjected to multi-objective collaborative optimization to obtain the final reactor optimization model.

[0036] In this embodiment, the multi-physics coupling features and feature causal relationship network are modularly decomposed to obtain a structured representation of the multi-physics parameters. This structured representation is then encoded using neural ordinary differential equations to obtain a feature description representation with consistent physical states. Residual calculation and adaptive adjustment are performed on this feature description representation to obtain dynamically optimized compensation data. Bayesian inference and variational analysis are then performed on this dynamically optimized compensation data to obtain the compensation parameter distribution and uncertainty estimation parameters. Model structure search and wavelet decomposition are then performed on the compensation parameter distribution and uncertainty estimation parameters to obtain a multi-scale dynamic characteristic representation. Finally, this multi-scale dynamic characteristic representation is integrated and dynamically weighted. The process involves allocating and generating an initial reactor optimization model with dynamically optimized structure and parameters; performing multi-objective function transformation on the initial reactor optimization model and the characteristic causal relationship network to obtain a multi-dimensional reactor performance index function; performing regression analysis on the multi-dimensional reactor performance index function for various reactor performance parameters to obtain a reactor performance prediction probability distribution; uniformly sampling the reactor performance prediction probability distribution to obtain a reactor parameter combination; and performing graph structure mapping and electromagnetic field iterative optimization on the reactor parameter combination to obtain a performance index response surface; and performing multi-objective reactor performance trade-off analysis and multi-criteria decision analysis on the performance index response surface to obtain the final reactor optimization model with corresponding high performance index.

[0037] In practical applications, graph segmentation algorithms such as spectral clustering or the Louvain method are first used to modularly decompose the multiphysics coupling features and feature causal relationship network. This decomposes the complex network structure into several relatively independent but interconnected sub-modules, thereby obtaining a structured representation of the multiphysics parameters (this structure not only preserves the original network topology but also highlights the key coupling relationships between different physics fields). Then, the Neural ODE framework is used to encode the structured representation of the multiphysics parameters using Neural ODEs, transforming the discrete structured representation into a continuous-time dynamic system expression to represent the evolution of the multiphysics parameters over time. Simultaneously, physical consistency is maintained because it follows the continuous properties of the physical system, thus obtaining a characteristic description representation with consistent physical state. Then, residual calculation and adaptive adjustment are performed on this characteristic description representation to capture the difference between model predictions and actual observations. That is, by constructing an adaptive residual network to dynamically learn and compensate for model prediction errors, dynamic optimization compensation data (which reflects the direction and magnitude of model adjustment) is obtained. Then, using this compensation data and quantifying its uncertainty, variational Bayesian methods are employed to perform Bayesian inference and variational analysis on the dynamic optimization compensation data. This involves treating the compensation parameters as random variables and estimating their posterior distribution, where the objective function of variational inference is:

[0038] ;

[0039] Here, q(θ) is the approximate posterior distribution of parameter θ, p(x|θ) is the likelihood function, p(θ) is the prior distribution, and KL represents the KL divergence. By maximizing this objective function, the distribution estimate of the compensation parameter and the corresponding uncertainty quantification are obtained. Then, Neural Architecture Search (NAS) is used to search for the model structure based on the obtained compensation parameter distribution and uncertainty estimate parameters to automatically explore the optimal model structure. The results of the model structure search are then subjected to wavelet decomposition to decompose the signal into components of different scales, capturing multi-scale dynamic characteristics and thus obtaining the signal representation at different time scales. Dynamic ensemble learning methods, such as variants of Adaptive Boosting (AdaBoost), are then used to integrate and dynamically weight the multi-scale dynamic characteristic representations to dynamically adjust the weights of characteristics at different scales. Weight allocation can be based on the predictive performance and uncertainty estimate of each scale characteristic, ensuring that the model adaptively focuses on the most relevant and reliable characteristics, ultimately generating an initial reactor optimization model with dynamically optimized structure and parameters. For example, by encoding these representations using regular differential equations, a continuous-time model is obtained, describing how current density, temperature distribution, and mechanical stress dynamically change over time. Then, through a residual network, patterns of deviation between model predictions and actual conditions are learned. For instance, under certain load conditions, the model may systematically underestimate the temperature in certain areas. Bayesian inference and variational analysis are used to quantify the uncertainty of this compensation, revealing that the accuracy of compensation is higher under low load conditions, while uncertainty increases under high load conditions. Furthermore, through model structure search and wavelet decomposition, it is discovered that the temperature dynamics of the reactor exhibit different characteristics at different time scales. For example, on short time scales, temperature may be mainly affected by changes in current density, while on long time scales, changes in ambient temperature may play a more significant role. Finally, through dynamic weight allocation, the model can adaptively adjust its focus on characteristics at different time scales based on the current operating conditions, thus obtaining an optimized model that accurately describes short-term dynamics and captures long-term trends.

[0040] Secondly, by representing various performance indicators (such as inductance, loss, temperature rise, mechanical strength, etc.) as functions of model parameters and causal relationships, a multi-objective function transformation is performed on the initial reactor optimization model and the characteristic causal relationship network. For example, a multi-objective function of the following form can be used: F(x) = [f1(x), f2(x), ..., fk(x)], where x represents the design parameters of the reactor, and fi(x) represents the i-th performance indicator. This transformation yields a multi-dimensional reactor performance indicator function, which comprehensively describes the various performance aspects of the reactor. Regression analysis of various performance parameters of the reactor is then performed on this multi-dimensional reactor performance indicator function to capture complex nonlinear relationships and provide an estimate of the uncertainty of predictions, thereby obtaining the reactor performance prediction probability distribution. Latin hypercube sampling is then used to uniformly sample the obtained reactor performance prediction probability distribution to ensure uniform coverage in the high-dimensional parameter space, resulting in a reactor parameter combination. Finally, the sampled reactor parameter combination is subjected to graph structure mapping and electromagnetic field iterative optimization, using a graph neural network. A graph structure mapping is performed using Networks to encode the reactor's geometry and physical characteristics into a graph structure. Then, the finite element method (FEM) combined with gradient descent is used for iterative electromagnetic field optimization to obtain an accurate performance index response surface. Finally, the Pareto optimization method is employed to perform multi-objective reactor performance trade-off analysis and multi-criteria decision analysis on the obtained performance index response surface, i.e., by calculating the priority function between each pair of schemes (a, b):

[0041] ;

[0042] in It is the weight of the j-th criterion. It is based on the priority of a relative to b according to the j-th criterion. yes Compared to The overall priority is determined by k, which is the total number of criteria. The positive and negative flows of each scheme are calculated to obtain the final ranking, thereby generating the final reactor optimization model with corresponding high performance indicators, which includes the optimization order and optimization parameters. For example, considering the optimization problem of a high-voltage toroidal air-core reactor, the performance indicators that may be of interest include inductance, loss, temperature rise, mechanical strength, and electromagnetic field distribution. A multi-objective function transformation is used to obtain a multi-dimensional function describing these indicators. Then, deep Gaussian process regression is used to establish a complex nonlinear relationship model between these indicators and the reactor's geometric parameters (such as coil size and winding method) and material parameters (such as conductor material and insulation material). In the uniform sampling stage, thousands of different reactor design schemes are generated, and the geometry of each scheme is encoded as a graph using a graph neural network. This representation is then used for rapid performance prediction. In the multi-objective trade-off analysis stage, the optimal balance point needs to be determined based on the specific application scenario. A trade-off relationship between inductance and loss is discovered, and by comprehensively considering the importance of various performance indicators, the design scheme with the best overall performance is selected. The final optimization model not only provides the optimal geometric structure and material parameters of the reactor but also provides prediction ranges for performance indicators, providing a reliable basis for relevant decisions.

[0043] 104. Perform full life cycle data analysis and prediction on the final reactor optimization model for the target high-voltage toroidal air-core reactor, and generate an optimization scheme for the full life cycle management and maintenance of the target high-voltage toroidal air-core reactor.

[0044] In this embodiment, multi-scale data integration is performed on the final reactor optimization model to obtain a global performance characterization of the reactor. Time-series data synchronization is then performed on the global performance characterization to obtain consistent historical operating data of the target high-voltage toroidal air-core reactor. Multi-level time-series analysis is performed on the consistent historical operating data to obtain future performance predictions of the target high-voltage toroidal air-core reactor. Multi-dimensional correlation analysis is then performed on the future performance predictions to obtain potential fault mode identification results. Multi-scenario decision tree analysis and adaptive optimization processing are then performed on the potential fault mode identification results to obtain dynamically adjusted maintenance strategies. Finally, full-cycle data integration and correlation analysis are performed on the dynamic maintenance strategies to obtain an optimized scheme for the full lifecycle management and maintenance of the target high-voltage toroidal air-core reactor.

[0045] In practical applications, multi-resolution analysis (MRA) technology, combined with wavelet transform and empirical mode decomposition (EMD) methods, is employed to organically integrate microscale material performance data, mesoscale structural response data, and macroscale system performance data, thereby obtaining a global performance characterization of the reactor. This allows for multi-scale data integration of the final reactor optimization model, achieving the fusion of data from different time and spatial scales to obtain a global performance characterization of the reactor. Furthermore, the Dynamic Time Warping (DTW) algorithm, combined with Kalman filtering, is used to synchronize the time-series data of this global performance characterization, achieving alignment and fusion of time-series data from different sources and with different sampling rates. This results in consistent historical operating data for the target high-voltage toroidal hollow reactor (data that comprehensively reflects the reactor's performance changes under different operating conditions and environments). Finally, a method combining Long Short-Term Memory (LSTM) networks and attention mechanisms from deep learning is employed. A multi-layered LSTM network is constructed, with each layer capturing features at different time scales, while the attention mechanism adaptively adjusts the importance of features at different time scales to achieve consistency in these features. Multi-level time series analysis is performed on historical operating data to obtain future performance predictions for the target high-voltage toroidal air-core reactor (these predictions include not only short-term performance fluctuations but also long-term performance degradation trends). Then, by constructing a correlation network between performance indicators and analyzing the network's topology and dynamic evolution characteristics, potential fault modes are identified. Multi-dimensional correlation analysis is then performed on these future performance predictions (e.g., abnormally strong correlations between certain performance indicators may indicate the occurrence of specific types of faults), thus obtaining potential fault mode identification results. Furthermore, multi-scenario decision tree analysis and adaptive optimization are applied to the identified potential fault modes. The Monte Carlo Tree Search (MCTS) algorithm, combined with reinforcement learning techniques, simulates the long-term impact of different maintenance decisions under various possible scenarios to achieve multi-scenario decision tree analysis. Meta-learning methods are used to enable maintenance strategies to quickly adapt to new operating conditions and fault modes for adaptive optimization, resulting in dynamically adjusted maintenance strategies (where these strategies can adaptively adjust based on the reactor's real-time status and predicted future performance).Furthermore, knowledge graph and graph embedding technologies are employed to construct a comprehensive knowledge graph containing reactor design parameters, operating data, maintenance records, and performance indicators. TransE and other graph embedding methods are used to map entities and relationships in the graph to a low-dimensional vector space. This allows for in-depth analysis of the complex relationship between maintenance strategies and the reactor's lifecycle performance, enabling full-cycle data integration and correlation analysis of this dynamic maintenance strategy. This reveals potential optimization opportunities, such as the long-term impact of certain maintenance operations on extending equipment lifespan. Ultimately, an optimized scheme for the lifecycle management and maintenance of the target high-voltage toroidal air-core reactor is obtained. (This scheme considers not only the reactor's current state and historical performance but also includes predictions of future performance and assessments of potential risks. It can dynamically adjust dimensions based on real-time data.) (Protection strategies aim to maximize equipment lifespan and economic benefits while ensuring equipment reliability.) For example, for a high-voltage toroidal air-core reactor that has been operating for many years, multi-scale data integration may reveal a complex nonlinear relationship between the microstructural changes of its insulation material and its macroscopic electrical performance. Furthermore, during time-series data synchronization, it may be necessary to process data from different sensors and at different sampling rates, such as temperature, vibration, and partial discharge. Multi-level time-series analysis may reveal multiple periodic variations in reactor performance, including daily, seasonal, and annual variations, and may also predict long-term performance degradation trends. Multi-dimensional correlation analysis may reveal a significant time-lag correlation between temperature increases and increased partial discharge activity, which may indicate potential deterioration of the insulation material. Based on these analyses, multi-scenario decision trees may generate multiple possible maintenance plans, such as increasing detection frequency, partial repair, or complete replacement. Through adaptive optimization, the system may recommend preventative maintenance during the next planned power outage to avoid potential major failures. The ultimate lifecycle management solution may include a series of dynamically adjusted maintenance plans, such as automatically adjusting detection intervals based on load changes and environmental conditions, or triggering in-depth diagnostic processes when certain warning signals are detected. This comprehensive, dynamic, and predictive approach not only improves the reliability and lifespan of reactors but also optimizes maintenance costs and enhances the overall operational efficiency of the power system.

[0046] In this embodiment of the invention, multi-level adaptive decomposition and reconstruction are performed on the original sensing data corresponding to the multi-physics parameters of the target high-voltage toroidal air-core reactor to obtain multi-scale sensing features. Then, a spatiotemporal topological representation is constructed on these multi-scale sensing features to establish the implicit relationships and dynamic evolution characteristics between the various physics parameters. Multi-physics coupling features and causal analysis are then performed on these implicit relationships and dynamic evolution characteristics to obtain the reactor's multi-physics coupling features and causal relationship network. Adaptive multi-physics coupling model construction and multi-objective collaborative optimization are then performed on the multi-physics coupling features and causal relationship network to generate a final reactor optimization model. This final reactor optimization model is then used to analyze and predict the full lifecycle data of the target high-voltage toroidal air-core reactor, thereby generating an optimization scheme for the reactor's full lifecycle management and maintenance. Thus, in the optimization process of the high-voltage toroidal air-core reactor, the final optimization effect of the reactor is improved by combining multi-physics coupling and the reactor's reactance characteristics.

[0047] The multiphysics-based reactor optimization method in the embodiments of the present invention has been described above. The multiphysics-based reactor optimization device in the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 2 One embodiment of the reactor optimization device based on multiphysics in this invention includes:

[0048] The feature reconstruction module 201 is used to acquire the original sensing data of the multi-physics field parameters corresponding to the target high-voltage toroidal hollow reactor, and to perform multi-level adaptive decomposition and reconstruction on the original sensing data to obtain multi-scale sensing features.

[0049] The causal analysis module 202 is used to construct a spatiotemporal topological representation of the multi-scale sensing features, generate implicit parameter relationships and dynamic evolution features between various physical field parameters, and perform multi-physics coupling feature extraction and feature causal analysis on the implicit parameter relationships and dynamic evolution features to obtain a multi-physics coupling feature and feature causal relationship network.

[0050] The collaborative optimization module 203 is used to construct an adaptive multi-physics coupling model for the multi-physics coupling features and feature causal relationship network, obtain an initial reactor optimization model with dynamically optimized structure and parameters, and perform multi-objective collaborative optimization on the initial reactor optimization model to obtain the final reactor optimization model.

[0051] The optimization module 204 is used to perform full life cycle data analysis and prediction on the final reactor optimization model of the target high-voltage toroidal air-core reactor, and generate an optimization scheme for the full life cycle management and maintenance of the target high-voltage toroidal air-core reactor.

[0052] In this embodiment of the invention, multi-level adaptive decomposition and reconstruction are performed on the original sensing data corresponding to the multi-physics parameters of the target high-voltage toroidal air-core reactor to obtain multi-scale sensing features. Then, a spatiotemporal topological representation is constructed on these multi-scale sensing features to establish the implicit relationships and dynamic evolution characteristics between the various physics parameters. Multi-physics coupling features and causal analysis are then performed on these implicit relationships and dynamic evolution characteristics to obtain the reactor's multi-physics coupling features and causal relationship network. Adaptive multi-physics coupling model construction and multi-objective collaborative optimization are then performed on the multi-physics coupling features and causal relationship network to generate a final reactor optimization model. This final reactor optimization model is then used to analyze and predict the full lifecycle data of the target high-voltage toroidal air-core reactor, thereby generating an optimization scheme for the reactor's full lifecycle management and maintenance. Thus, in the optimization process of the high-voltage toroidal air-core reactor, the final optimization effect of the reactor is improved by combining multi-physics coupling and the reactor's reactance characteristics.

[0053] above Figure 2 The reactor optimization device based on multiphysics field in the embodiments of the present invention will be described in detail from the perspective of modular functional entities. The reactor optimization device based on multiphysics field in the embodiments of the present invention will be described in detail from the perspective of hardware processing.

[0054] Figure 3 This is a schematic diagram of a multiphysics-based reactor optimization device 300 provided in an embodiment of the present invention. The multiphysics-based reactor optimization device 300 can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the multiphysics-based reactor optimization device 300. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media 330 on the multiphysics-based reactor optimization device 300.

[0055] The multiphysics-based reactor optimization device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The illustrated reactor optimization device structure based on multiphysics does not constitute a limitation on the reactor optimization device based on multiphysics. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0056] The present invention also provides a reactor optimization device based on multiphysics, wherein the computer device includes a memory and a processor, the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor performs each step of the reactor optimization method based on multiphysics in the above embodiments.

[0057] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the various steps of the multiphysics-based reactor optimization method.

[0058] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0059] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0060] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0061] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A reactor optimization method based on multiphysics, characterized in that, The reactor optimization method based on multiphysics includes: The process involves acquiring raw sensing data corresponding to multiple physical field parameters of a target high-voltage toroidal hollow reactor, performing adaptive multi-scale decomposition on the raw sensing data to obtain various sensing mode functions, and then performing continuous wavelet transform on the sensing mode functions to obtain time-frequency sensing features. Adaptive soft-threshold denoising is then applied to the time-frequency sensing features to obtain denoised time-frequency sensing features. Singular value decomposition and reconstruction are then performed on the denoised time-frequency sensing features to obtain various sensing time-series components. Tensor decomposition and feature compression are then performed on the sensing time-series components to obtain a compressed feature space representation. Finally, based on each physical field parameter, the compressed feature space representation is optimized by physical field parameter constraints to obtain multi-scale sensing features. The spatiotemporal topological representation of the multi-scale sensing features is constructed to generate implicit parameter relationships and dynamic evolution features among various physical field parameters. Multi-physics coupling feature extraction and feature causal analysis are performed on the implicit parameter relationships and dynamic evolution features to obtain a multi-physics coupling feature and feature causal relationship network. An adaptive multiphysics coupling model is constructed based on the multiphysics coupling characteristics and the characteristic causal relationship network to obtain an initial reactor optimization model with dynamically optimized structure and parameters. A multi-objective function transformation is then performed on the initial reactor optimization model and the characteristic causal relationship network to obtain a multidimensional reactor performance index function. Regression analysis of various performance parameters of the reactor is then performed on the multidimensional reactor performance index function to obtain a reactor performance prediction probability distribution. Uniform sampling is then performed on the reactor performance prediction probability distribution to obtain reactor parameter combinations. Graph structure mapping and electromagnetic field iterative optimization are then performed on the reactor parameter combinations to obtain the performance index response surface. Finally, a multi-objective reactor performance trade-off analysis and multi-criteria decision analysis are performed on the performance index response surface to obtain the final reactor optimization model with corresponding high-performance indices. The final reactor optimization model is used to perform full life cycle data analysis and prediction of the target high-voltage toroidal air-core reactor, and an optimization scheme for the full life cycle management and maintenance of the target high-voltage toroidal air-core reactor is generated.

2. The reactor optimization method based on multiphysics as described in claim 1, characterized in that, The construction of spatiotemporal topological representation of the multi-scale sensing features, generating implicit parameter relationships and dynamic evolution characteristics among various physical field parameters, includes: Based on a preset point distance threshold, the distances between each feature data point in the multi-scale sensing features are filtered to construct a sensing feature boundary relationship diagram. Then, simplex identification and boundary operator generation are performed on the sensing feature boundary relationship diagram to obtain the feature simplex chain complex. Homology group calculation and persistence analysis are performed on the feature simplex chain complex to generate a feature persistence map, and feature extraction is performed on the feature persistence map to obtain a multi-resolution topological feature description. The multi-resolution topological feature description is subjected to sliding window continuous cohomology calculation and multi-scale diffusion map construction to obtain nonlinear dimensionality reduction feature diffusion coordinates. Based on a preset mapping discretization strategy, the feature diffusion coordinates are subjected to coordinate projection and node discretization optimization to obtain the feature discrete topological skeleton. Time-varying continuous cohomology analysis and probability transfer calculation are performed on the feature discrete topological skeleton to obtain the feature similarity measure of the topological structure. Then, using a preset multi-level graph neural network, multi-level graph convolution and temporal attention mechanism analysis are performed on the feature similarity measure to generate the implicit parameter relationships and dynamic evolution features between various physical field parameters.

3. The reactor optimization method based on multiphysics as described in claim 1, characterized in that, The step of extracting multi-physics coupling features and performing feature causal analysis on the implicit relationships of the parameters and the dynamic evolution features to obtain a multi-physics coupling feature and feature causal relationship network includes: The implicit relationships of the parameters are subjected to continuous coherence analysis and multi-scale analysis to obtain static topological feature descriptors, and the dynamic evolution features are subjected to time series analysis to obtain dynamic topological feature descriptors. The static topological feature descriptors and the dynamic topological feature descriptors are then fused to obtain a spatiotemporal topological feature representation. A discrete gradient vector field corresponding to the spatiotemporal topological feature representation is constructed, and critical point identification and persistent pairing are performed on the discrete gradient vector field to obtain the reactance key points and reactance separation lines of the multiphysics field. Multidimensional probability density estimation and marginal distribution transformation are performed on the key points of the reactance and the reactance dividing line to obtain the joint probability distribution of the multi-physics coupling, and multivariate empirical mode decomposition is performed on the joint probability distribution to obtain the time-varying mode function of the multi-physics. The time-varying mode function is subjected to a nonlinear Granger causality test to obtain a preliminary causal relationship network. Conditional mutual information analysis and graph structure learning are then performed on the preliminary causal relationship network to obtain multi-physics coupling features and causal relationship networks between features.

4. The reactor optimization method based on multiphysics as described in claim 1, characterized in that, The adaptive multiphysics coupling model construction based on the multiphysics coupling features and feature causal relationship network, resulting in an initial reactor optimization model with dynamically optimized structure and parameters, includes: The multi-physics coupling features and feature causal relationship network are modularly decomposed to obtain a structured representation of the multi-physics parameters. Then, the structured representation of the multi-physics parameters is encoded by a neural ordinary differential equation to obtain a feature description representation with consistent physical state. The feature description representation is subjected to residual calculation and adaptive adjustment to obtain dynamic optimization compensation data. Bayesian inference and variational analysis are then performed on the dynamic optimization compensation data to obtain the compensation parameter distribution and uncertainty estimation parameters. The compensation parameter distribution and the uncertainty estimation parameters are subjected to model structure search and wavelet decomposition to obtain a multi-scale dynamic characteristic representation. The multi-scale dynamic characteristic representation is then integrated and dynamically weighted to generate an initial reactor optimization model with dynamically optimized structure and parameters.

5. The reactor optimization method based on multiphysics as described in claim 1, characterized in that, The step involves performing full lifecycle data analysis and prediction on the final reactor optimization model to generate an optimization scheme for the full lifecycle management and maintenance of the target high-voltage toroidal air-core reactor, including: Multi-scale data integration is performed on the final reactor optimization model to obtain the global performance characterization of the reactor, and time-series data synchronization is performed on the global performance characterization to obtain consistent historical operating data of the target high-voltage toroidal air-core reactor. Multi-level time series analysis is performed on the consistent historical operating data to obtain the future performance prediction of the target high-voltage toroidal air-core reactor, and multi-dimensional correlation analysis is performed on the future performance prediction to obtain the potential fault mode identification results. The potential fault mode identification results are subjected to multi-scenario decision tree analysis and adaptive optimization processing to obtain a dynamically adjusted maintenance strategy. The dynamically adjusted maintenance strategy is then subjected to full-cycle data integration and correlation analysis to obtain an optimized scheme for the full life cycle management and maintenance of the target high-voltage toroidal air-core reactor.

6. A reactor optimization device based on multiphysics, characterized in that, The reactor optimization device based on multiphysics includes: The feature reconstruction module is used to acquire the original sensing data of the target high-voltage toroidal hollow reactor corresponding to multiple physical field parameters, and to perform adaptive multi-scale decomposition on the original sensing data to obtain multiple sensing mode functions. Then, continuous wavelet transform is performed on the sensing mode functions to obtain time-frequency sensing features. Adaptive soft-threshold denoising is performed on the time-frequency sensing features to obtain denoised time-frequency sensing features. Singular value decomposition and reconstruction are then performed on the denoised time-frequency sensing features to obtain multiple sensing time-series components. Tensor decomposition and feature compression are performed on the sensing time-series components to obtain a compressed feature space representation. Finally, based on each physical field parameter, the compressed feature space representation is optimized by physical field parameter constraints to obtain multi-scale sensing features. The causal analysis module is used to construct the spatiotemporal topological representation of the multi-scale sensing features, generate the implicit parameter relationships and dynamic evolution features between various physical field parameters, and perform multi-physics coupling feature extraction and feature causal analysis on the implicit parameter relationships and dynamic evolution features to obtain a multi-physics coupling feature and feature causal relationship network. The collaborative optimization module is used to construct an adaptive multiphysics coupling model for the multiphysics coupling features and feature causal relationship network, obtaining an initial reactor optimization model with dynamically optimized structure and parameters. It then performs multi-objective function transformation on the initial reactor optimization model and the feature causal relationship network to obtain a multidimensional reactor performance index function. Furthermore, it performs regression analysis on the multidimensional reactor performance index function for various reactor performance parameters to obtain a reactor performance prediction probability distribution. The module then performs uniform sampling on the reactor performance prediction probability distribution to obtain a reactor parameter combination, and performs graph structure mapping and electromagnetic field iterative optimization on the reactor parameter combination to obtain a performance index response surface. Finally, it performs multi-objective reactor performance trade-off analysis and multi-criteria decision analysis on the performance index response surface to obtain the final reactor optimization model with corresponding high-performance indexes. The optimization module is used to perform full life cycle data analysis and prediction on the final reactor optimization model of the target high-voltage toroidal air-core reactor, and generate an optimization scheme for the full life cycle management and maintenance of the target high-voltage toroidal air-core reactor.

7. A reactor optimization device based on multiphysics, characterized in that, The reactor optimization device based on multiphysics includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the multiphysics-based reactor optimization device to perform the steps of the multiphysics-based reactor optimization method as described in any one of claims 1-5.

8. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the steps of the multiphysics-based reactor optimization method as described in any one of claims 1-5.

Images