Equipment group collaborative fault prediction analysis method and system for complex industrial process

Abstract

The invention provides an equipment group collaborative fault prediction analysis method and system for a complex industrial process, and relates to the technical field of fault prediction, and the method comprises the steps: collecting equipment group data, and reconstructing a multi-scale time sequence feature; constructing a quantum walking enhanced dynamic association map; extracting spatial-temporal features and mapping the spatial-temporal features to a Riemannian manifold space; identifying an abnormal mode and generating hidden variable probability distribution; constructing a fault evolution model driven by a non-equilibrium state statistical fluctuation theory; and solving the cooperative field equation set to obtain a control parameter sequence. The method can effectively improve the accuracy and reliability of equipment group collaborative fault prediction and reduce the maintenance cost.

Description

Technical Field

[0001] The present invention relates to fault prediction technology, and in particular to a method and system for collaborative fault prediction analysis of equipment groups in complex industrial processes. Background Art

[0002] With the rapid development of industrial automation and intelligent manufacturing, the coordinated operation of equipment groups in complex industrial processes has become a key factor in ensuring production efficiency and product quality. There are complex physical connections and information interactions between equipment groups, and their operating status directly affects the stability and reliability of the entire production system. At present, data-driven fault diagnosis and prediction methods are widely used in the industrial field. By collecting equipment operation data and combining machine learning algorithms to identify fault patterns and provide early warnings, these methods mainly include traditional statistical analysis methods, deep learning methods, and intelligent diagnosis methods based on knowledge graphs. In practical applications, these methods have achieved certain results in single-device fault prediction and simple system fault diagnosis.

[0003] The main defects of the existing technology are: Existing fault prediction methods mainly focus on the fault feature extraction and pattern recognition of a single device, which makes it difficult to effectively characterize the complex correlation and co-evolution characteristics between device groups. When dealing with high-dimensional, nonlinear device group coordinated failures, traditional methods often ignore the integrity and correlation of the system, resulting in insufficient prediction accuracy.

[0004] When dealing with the dynamic evolution of device groups, existing technologies mostly use simple timing models or static network structures, which cannot accurately describe the dynamic change characteristics of device states at different time and space scales, and it is also difficult to capture the topological invariance characteristics of device groups during the development of faults.

[0005] Current fault prediction methods lack in-depth consideration of the system's non-equilibrium state. When faced with sudden failures and nonlinear evolution processes of equipment groups, it is difficult to accurately characterize the system's fluctuation characteristics and critical behaviors, which affects the timeliness and accuracy of fault warnings. Summary of the invention

[0006] The embodiments of the present invention provide a method and system for collaborative fault prediction and analysis of equipment groups in complex industrial processes, which can solve the problems in the prior art.

[0007] According to a first aspect of the embodiments of the present invention, Provides collaborative fault prediction and analysis methods for equipment groups in complex industrial processes, including: Collect the operating parameter data, status data, alarm data and maintenance record data of the equipment group in the complex industrial process, and perform noise reduction and time series reconstruction through adaptive wavelet transform to obtain multi-scale time series characteristics; based on the multi-scale time series characteristics, construct a quantum walk-enhanced dynamic correlation map; in the dynamic correlation map, use the quantum random walk algorithm to calculate the quantum correlation strength between the equipment nodes, and simulate the dynamic evolution process of the equipment group through the Schrödinger evolution equation to obtain the equipment group quantum correlation tensor with topological invariant characteristics; For the quantum correlation tensor of the device group, a bidirectional quantum gated recurrent network is used to extract the spatiotemporal features; a graph neural network based on a Dirac propagator is constructed, and the spatiotemporal features are mapped to a Riemann manifold space according to the graph neural network based on a Dirac propagator to obtain a manifold feature representation of the device group; the purity of the device group cooperative state in the manifold feature representation is evaluated by using quantum entanglement entropy, and a contrast loss function is constructed in combination with the Wasserstein metric to obtain an abnormal pattern recognition result; the abnormal pattern recognition result is input into a deep belief network based on a Boltzmann machine to generate a hidden variable probability distribution of the device group; Based on the hidden variable probability distribution, a fault evolution model driven by non-equilibrium statistical fluctuation theory is constructed; the state evolution equation of the equipment group is generated using the fault evolution model driven by non-equilibrium statistical fluctuation theory; the state evolution equation is mapped to the Lie group space, and the fault trajectory tensor is constructed through Lie algebra calculation; a random differential operator is applied to the fault trajectory tensor to establish a group of coordinated field equations; a Hamiltonian is constructed according to the group of coordinated field equations, and the corresponding group of canonical equations is solved to obtain a control parameter sequence of the equipment group.

[0008] Based on the multi-scale temporal characteristics, a dynamic correlation map of quantum walking enhancement is constructed, including: Constructing a high-order tensor based on the multi-scale time series features, wherein the dimensions of the high-order tensor include device dimension, parameter dimension and time dimension; performing tensor decomposition on the high-order tensor to obtain a tensor representing the characteristics of the device group and a factor matrix of each dimension; projecting the factor matrix into a complex Hilbert space to construct an initial quantum state of the device node, wherein the initial quantum state includes an amplitude component and a phase component; Constructing a non-Hermitian interaction operator based on the tensor characterizing the characteristics of the device group, wherein the non-Hermitian interaction operator characterizes the non-equilibrium quantum correlation between device nodes; combining the non-Hermitian interaction operator and the initial quantum state to construct a quantum system Hamiltonian; obtaining a dynamic coupling matrix based on the eigenvalue decomposition of the quantum system Hamiltonian, wherein the dynamic coupling matrix characterizes the correlation strength between device nodes; A quantum master equation containing noise terms is constructed based on the dynamic coupling matrix, wherein the quantum master equation includes a drift term determined by the non-Hermitian interaction operator and a quantum fluctuation diffusion term modulated by the dynamic coupling matrix; the drift term and the quantum fluctuation diffusion term are numerically solved based on the quantum Monte Carlo method to obtain a density operator that characterizes the evolution of the system; and a quantum walk-enhanced dynamic correlation map is constructed based on the density operator.

[0009] The method of constructing a quantum walk enhanced dynamic correlation map based on the density operator includes: The density operator is separated from system and environment related items by an adaptive quantum environment decoupling method to obtain a purified density operator; a conditional partial trace operation is performed on the purified density operator to obtain a reduced density matrix, the von Neumann entropy of the reduced density matrix is ​​calculated, and the quantum mutual information between node pairs is calculated based on the von Neumann entropy; the quantum mutual information is normalized to obtain a dynamic correlation weight matrix between device nodes; Accumulating the dynamic association weight matrix in different time windows in segments to obtain cumulative weight matrices at multiple time scales; constructing a corresponding time scale association map based on the cumulative weight matrix at each time scale, and combining the multiple time scale association maps to form a hierarchical multi-scale dynamic association map structure; calculating the degree-normalized Laplace matrix for each layer in the multi-scale dynamic association map, and performing eigenvalue decomposition to obtain a characteristic spectrum vector; The distance metric between the characteristic spectra of the normalized Laplace matrices of each layer is calculated, the distance metric is mapped to the kernel Hilbert space, and an inter-layer spectral distance matrix is ​​constructed; a dynamic singular spectrum analysis model is constructed by analyzing the inter-layer spectral distance matrix, a sliding window decomposition is performed on the matrix, and Hankel matrix embedding is applied to extract timing features; based on the timing features, the singular value sequence is mapped to the time-frequency domain to construct an energy spectrum, and the key evolution moments in the energy spectrum are identified by combining self-service sampling and cumulative sum detection algorithms; and a dynamic association map of the device group is reconstructed at the key evolution moments.

[0010] In the dynamic correlation graph, the quantum random walk algorithm is used to calculate the quantum correlation strength between device nodes, and the dynamic evolution process of the device group is simulated by the Schrödinger evolution equation to obtain the device group quantum correlation tensor with topological invariant characteristics, including: Mapping device nodes in a dynamic association graph to quantum ground states in a quantum Hilbert space, mapping the association strengths between device nodes in the dynamic association graph to quantum state superposition coefficients, and constructing a quantum state representation of the dynamic association graph; constructing a quantum walk operator based on the quantum state representation, wherein the quantum walk operator is represented by an exponential form of a system Hamiltonian, wherein the matrix elements of the system Hamiltonian are proportional to the association strengths between device nodes; Applying the quantum walk operator to the initial quantum state, solving the continuous-time quantum random walk evolution equation to obtain the quantum transfer probability between two device nodes at any time, and performing time-averaged calculation on the quantum transfer probability within the characteristic evolution period to obtain the quantum correlation strength between the device nodes; constructing a quantum dynamics model based on the Schrödinger equation, obtaining the evolution trajectory of the system quantum state by numerically solving the Schrödinger equation, and extracting the geometric phase of the system quantum state from the evolution trajectory as a topological invariant; Select three device nodes, calculate the quantum correlation strength between any two of the three device nodes, and multiply the quantum correlation strengths to obtain a correlation strength product; calculate the geometric phase of the loop formed by the three device nodes, and substitute the geometric phase into an exponential function to obtain a phase index; multiply the correlation strength product by the phase index to construct a tensor element; characterize the correlation strength between the device nodes based on the modulus of the tensor element, characterize the topological characteristics between the device nodes based on the phase of the tensor element, and obtain a third-order quantum correlation tensor that characterizes the three-body correlation characteristics between the three device nodes; determine the device group quantum correlation tensor based on the third-order quantum correlation tensors of multiple device nodes.

[0011] A quantum dynamics model based on the Schrödinger equation is constructed, the evolution trajectory of the system quantum state is obtained by numerically solving the Schrödinger equation, and the geometric phase of the system quantum state is extracted from the evolution trajectory as a topological invariant, including: Establish a wave function expression of the system in the quantum Hilbert space, and expand the wave function expression according to orthogonal basis vectors to obtain a state vector; construct a system Hamiltonian based on the state vector, wherein the system Hamiltonian consists of a kinetic energy term, a potential energy term, and an interaction term, wherein the kinetic energy term is represented by a quadratic form of a momentum operator, the potential energy term is represented by a function of a position operator, and the interaction term is characterized by a spin-orbit coupling term and an external field modulation term; substitute the system Hamiltonian and the wave function into the time-dependent Schrödinger equation to construct a system dynamics evolution equation, and introduce a system-environment interaction term and a quantum noise term into the system dynamics evolution equation to form a quantum dynamics model corrected by the decoherence effect; The quantum dynamics model corrected by the decoherence effect is subjected to space-time discretization, and the time domain and the space domain are discretized into finite-dimensional grid points to construct a time evolution operator; the time evolution operator is substituted into a fourth-order Runge-Kutta algorithm, and the discretized Schrödinger equation is iteratively solved through estimation steps and correction steps to calculate the dynamic evolution trajectory of the quantum state of the system, wherein the probability amplitude of the quantum state of the system is updated in each iterative step of the fourth-order Runge-Kutta algorithm, and the dynamic evolution trajectory is characterized by the time evolution sequence of the system energy expectation value, quantum coherence, entanglement and fidelity; Construct a system parameter space spanned by spin, orbital angular momentum and external field parameters, and track the dynamic evolution trajectory in the system parameter space; calculate the Berry phase generated by the evolution of the system quantum state around a closed path based on the dynamic evolution trajectory; use the Berry phase to calculate the Berry curvature in the system parameter space; integrate the Berry curvature on a closed surface to obtain the Chern number, and use the Chern number as a topological invariant that characterizes the overall topological characteristics of the system quantum state; and extract the geometric phase of the system quantum state by combining the Berry phase and the Chern number.

[0012] Constructing a graph neural network based on a Dirac propagator, mapping the spatiotemporal features to a Riemann manifold space according to the graph neural network based on the Dirac propagator, and obtaining a manifold feature representation of the device group, including: Construct a topological structure diagram of the device group, and define a graph Laplace operator based on the topological structure diagram; decompose the graph Laplace operator into a symmetric part and an antisymmetric part, construct a mass term based on the symmetric part, and construct a chiral term based on the antisymmetric part; combine the mass term and the chiral term to construct a Dirac operator, perform spectral decomposition on the Dirac operator to obtain eigenvalues ​​and eigenvectors; construct a Dirac propagator using the eigenvalues ​​and the eigenvectors, and the Dirac propagator describes the quantum characteristics of information propagation in the device group; A Chebyshev polynomial kernel function is constructed based on the real part and the imaginary part of the Dirac propagator, and a graph convolution layer is constructed based on the Chebyshev polynomial kernel function, wherein the graph convolution layer performs a tensor product operation on the input feature and the Dirac propagator; a multi-layer perceptron module is constructed, wherein the multi-layer perceptron module performs a nonlinear transformation on the output feature of the graph convolution layer; an attention mechanism layer is constructed, wherein the attention mechanism layer calculates the association weights between device nodes based on the output of the multi-layer perceptron module; and the graph convolution layer, the multi-layer perceptron module and the attention mechanism layer are cascaded to form a graph neural network model to extract the spatiotemporal features of the device group based on the input features; Construct a Riemann manifold space, wherein the Riemann manifold space has a local Euclidean metric structure; construct a tangent mapping using the spatiotemporal features extracted by the graph neural network model, wherein the tangent mapping maps the feature space of the device group to the tangent space of the Riemann manifold space; define a Riemann metric tensor in the tangent space, wherein the Riemann metric tensor describes the local geometric structure of the device group features in the Riemann manifold space; calculate a geodesic equation based on the Riemann metric tensor, map the spatiotemporal features of the device group to the Riemann manifold space by solving the geodesic equation, and obtain a manifold feature representation of the device group in the Riemann manifold space.

[0013] The purity of the device group cooperative state in the manifold feature representation is evaluated by using quantum entanglement entropy, and the contrast loss function is constructed in combination with the Wasserstein metric to obtain the abnormal pattern recognition results, including: The manifold feature representation is converted into a density matrix; the density matrix is ​​spectrally decomposed to obtain an eigenvalue sequence and an eigenvector sequence; a first reduced density matrix and a second reduced density matrix are constructed based on the eigenvalue sequence and the eigenvector sequence, wherein the first reduced density matrix and the second reduced density matrix respectively characterize the local subsystem states in the device group; the quantum entanglement entropy is calculated using the first reduced density matrix and the second reduced density matrix, wherein the quantum entanglement entropy is used to quantify the purity of the cooperative state of the device group; A positive sample pair and a negative sample pair are selected in the feature space corresponding to the manifold feature representation, wherein the positive sample pair is a feature representation of the same collaborative mode, and the negative sample pair is a feature representation of different collaborative modes; a first Wasserstein distance between two density matrices in the positive sample pair is calculated, and a second Wasserstein distance between two density matrices in the negative sample pair is calculated; a contrast loss function is constructed based on the first Wasserstein distance and the second Wasserstein distance, and the contrast loss function is used to optimize the distinguishability of feature representation; The quantum entanglement entropy and the contrast loss function are weightedly combined to obtain a discrimination score; a probability density function of the discrimination score is constructed based on historical data; the probability density function is solved according to a preset significance level to obtain an adaptive threshold; the discrimination score of a new sample is compared with the adaptive threshold, and when the discrimination score is greater than the adaptive threshold, it is determined that the device group is in an abnormal mode as an abnormal mode recognition result.

[0014] According to a second aspect of the embodiments of the present invention, Provides a collaborative fault prediction and analysis system for equipment groups in complex industrial processes, including: The first unit is used to collect the operating parameter data, status data, alarm data and maintenance record data of the equipment group in the complex industrial process, and perform noise reduction and time series reconstruction through adaptive wavelet transform to obtain multi-scale time series characteristics; based on the multi-scale time series characteristics, a quantum walk-enhanced dynamic correlation map is constructed; in the dynamic correlation map, the quantum random walk algorithm is used to calculate the quantum correlation strength between the equipment nodes, and the dynamic evolution process of the equipment group is simulated through the Schrödinger evolution equation to obtain the equipment group quantum correlation tensor with topological invariant characteristics; The second unit is used to extract the spatiotemporal features of the device group quantum correlation tensor using a bidirectional quantum gated recurrent network; construct a graph neural network based on a Dirac propagator, and map the spatiotemporal features to a Riemann manifold space according to the graph neural network based on a Dirac propagator to obtain a manifold feature representation of the device group; use quantum entanglement entropy to evaluate the purity of the device group cooperative state in the manifold feature representation, and construct a contrast loss function in combination with the Wasserstein metric to obtain an abnormal pattern recognition result; input the abnormal pattern recognition result into a deep belief network based on a Boltzmann machine to generate a hidden variable probability distribution of the device group; The third unit is used to construct a fault evolution model driven by non-equilibrium statistical fluctuation theory based on the hidden variable probability distribution; generate a state evolution equation of the equipment group using the fault evolution model driven by non-equilibrium statistical fluctuation theory; map the state evolution equation to the Lie group space, and construct a fault trajectory tensor through Lie algebra calculation; apply a random differential operator to the fault trajectory tensor to establish a group of coordinated field equations; construct a Hamiltonian according to the group of coordinated field equations, and solve the corresponding group of canonical equations to obtain a control parameter sequence of the equipment group.

[0015] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0016] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0017] The beneficial effects of this application are as follows: Through dynamic correlation graph analysis enhanced by adaptive wavelet transform and quantum walk, the complex correlation relationships and dynamic evolution characteristics between equipment groups can be accurately captured, which improves the recognition accuracy of collaborative failure modes of equipment groups and makes fault prediction more accurate and reliable.

[0018] By using a bidirectional quantum gated recurrent network and a graph neural network based on Dirac propagators, the spatiotemporal characteristics of the device group are mapped to the Riemann manifold space. Combined with quantum entanglement entropy and Wasserstein metric, abnormal patterns can be effectively identified, improving the early warning capability of collaborative failures of device groups.

[0019] A fault evolution model is constructed based on the non-equilibrium statistical fluctuation theory. Through Lie group space mapping and solving the cooperative field equations, the fault development trend of equipment groups can be accurately predicted, providing a scientific basis for equipment maintenance decisions and significantly improving the efficiency and accuracy of predictive maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of a process flow of a method for collaborative fault prediction and analysis of equipment groups in a complex industrial process according to an embodiment of the present invention; Figure 2 It is a schematic diagram of the structure of a system for collaborative fault prediction and analysis of equipment groups in a complex industrial process according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0023] Figure 1 FIG. 1 is a flow chart of a method for collaborative fault prediction and analysis of equipment groups in a complex industrial process according to an embodiment of the present invention. Figure 1 As shown, the method includes: S11. Collect the operating parameter data, status data, alarm data and maintenance record data of the equipment group in the complex industrial process, and perform noise reduction and time series reconstruction through adaptive wavelet transform to obtain multi-scale time series characteristics; based on the multi-scale time series characteristics, construct a quantum walk-enhanced dynamic correlation map; in the dynamic correlation map, use the quantum random walk algorithm to calculate the quantum correlation strength between the equipment nodes, and simulate the dynamic evolution process of the equipment group through the Schrödinger evolution equation to obtain the equipment group quantum correlation tensor with topological invariant characteristics; S12. For the quantum correlation tensor of the device group, a bidirectional quantum gated recurrent network is used to extract the spatiotemporal features; a graph neural network based on a Dirac propagator is constructed, and the spatiotemporal features are mapped to a Riemann manifold space according to the graph neural network based on a Dirac propagator to obtain a manifold feature representation of the device group; the purity of the device group cooperative state in the manifold feature representation is evaluated by using quantum entanglement entropy, and a contrast loss function is constructed in combination with the Wasserstein metric to obtain an abnormal pattern recognition result; the abnormal pattern recognition result is input into a deep belief network based on a Boltzmann machine to generate a hidden variable probability distribution of the device group; S13. Based on the hidden variable probability distribution, construct a fault evolution model driven by non-equilibrium statistical fluctuation theory; use the fault evolution model driven by non-equilibrium statistical fluctuation theory to generate the state evolution equation of the equipment group; map the state evolution equation to the Lie group space, and construct the fault trajectory tensor through Lie algebra calculation; apply random differential operators to the fault trajectory tensor to establish a set of coordinated field equations; construct the Hamiltonian according to the set of coordinated field equations, and solve the corresponding set of canonical equations to obtain the control parameter sequence of the equipment group.

[0024] In an optional implementation, based on the multi-scale temporal features, a dynamic correlation map of quantum walk enhancement is constructed, including: Constructing a high-order tensor based on the multi-scale time series features, wherein the dimensions of the high-order tensor include device dimension, parameter dimension and time dimension; performing tensor decomposition on the high-order tensor to obtain a tensor representing the characteristics of the device group and a factor matrix of each dimension; projecting the factor matrix into a complex Hilbert space to construct an initial quantum state of the device node, wherein the initial quantum state includes an amplitude component and a phase component; Constructing a non-Hermitian interaction operator based on the tensor characterizing the characteristics of the device group, wherein the non-Hermitian interaction operator characterizes the non-equilibrium quantum correlation between device nodes; combining the non-Hermitian interaction operator and the initial quantum state to construct a quantum system Hamiltonian; obtaining a dynamic coupling matrix based on the eigenvalue decomposition of the quantum system Hamiltonian, wherein the dynamic coupling matrix characterizes the correlation strength between device nodes; A quantum master equation containing noise terms is constructed based on the dynamic coupling matrix, wherein the quantum master equation includes a drift term determined by the non-Hermitian interaction operator and a quantum fluctuation diffusion term modulated by the dynamic coupling matrix; the drift term and the quantum fluctuation diffusion term are numerically solved based on the quantum Monte Carlo method to obtain a density operator that characterizes the evolution of the system; and a quantum walk-enhanced dynamic correlation map is constructed based on the density operator.

[0025] The specific implementation process of constructing a quantum walk-enhanced dynamic correlation map based on multi-scale temporal features is as follows: First, a high-order tensor is constructed for the multi-scale time series features. The equipment operation data is organized according to the equipment dimension, parameter dimension and time dimension to form a three-order tensor structure. Specifically, key operating parameters such as temperature, pressure, and flow can be selected to collect 24 hours of continuous time series data for each device with a sampling interval of 5 minutes, thereby constructing data samples of 288 time points. The high-order tensor is decomposed by the Tucker decomposition method to obtain the core tensor and the factor matrix corresponding to the three dimensions. The core tensor reflects the essential characteristics of the equipment group, and the factor matrix contains the main pattern information of each dimension.

[0026] Next, the obtained factor matrix is ​​mapped to the complex Hilbert space. By mapping the numerical size of the factor matrix to the amplitude of the quantum state and the changing trend of the factor matrix to the phase of the quantum state, the initial quantum state corresponding to each device node is constructed. The amplitude component reflects the relative importance of the device parameters, and the phase component reflects the evolution characteristics of the parameters over time.

[0027] Then, a non-Hermitian interaction operator is constructed based on the core tensor that characterizes the characteristics of the device group. The off-diagonal elements of this operator represent the non-equilibrium quantum correlation strength between device nodes, and the diagonal elements represent the characteristics of the nodes themselves. This operator is combined with the initial quantum state to construct the quantum Hamiltonian of the system. By performing eigenvalue decomposition on the Hamiltonian, a dynamic coupling matrix is ​​obtained, and the element values ​​of this matrix characterize the dynamic correlation strength between device nodes.

[0028] Based on the above dynamic coupling matrix, a quantum master equation containing noise terms is constructed. This equation contains two key parts: one is the drift term determined by the non-Hermitian interaction operator, which describes the deterministic evolution of the system; the other is the quantum fluctuation diffusion term modulated by the dynamic coupling matrix, which describes the random fluctuation characteristics of the system. The quantum Monte Carlo method is used to numerically solve the equation by generating a large number of random trajectories and averaging them to obtain a density operator that characterizes the evolution of the system. Finally, a quantum walk-enhanced dynamic correlation map is constructed based on this density operator to intuitively display the correlation structure and strength between devices.

[0029] The solution of this application can: By constructing and decomposing high-order tensors, key features in multi-scale time series data can be effectively extracted, data redundancy can be reduced, and the accuracy and computational efficiency of feature extraction can be improved. Mapping the factor matrix to the quantum state can make full use of the superposition property of the quantum state to achieve efficient expression of the dynamic characteristics of the complex system. The non-Hermitian interaction operator is used to describe the non-equilibrium correlation between devices, which can more accurately characterize the asymmetric coupling characteristics of the system than the traditional method. By constructing and solving the quantum Hamiltonian, the dynamic evolution law of the system can be accurately obtained, and the accuracy of correlation analysis can be improved. The introduction of quantum fluctuation diffusion terms can effectively deal with random disturbances and uncertainties in the system. The quantum Monte Carlo method is used for numerical solution to obtain stable and reliable calculation results. The quantum walk enhanced dynamic correlation map finally constructed can intuitively display the dynamic correlation characteristics of the device group, providing strong support for device management and fault diagnosis.

[0030] In an optional implementation, the construction of a quantum walk-enhanced dynamic association map based on the density operator includes: The density operator is separated from system and environment related items by an adaptive quantum environment decoupling method to obtain a purified density operator; a conditional partial trace operation is performed on the purified density operator to obtain a reduced density matrix, the von Neumann entropy of the reduced density matrix is ​​calculated, and the quantum mutual information between node pairs is calculated based on the von Neumann entropy; the quantum mutual information is normalized to obtain a dynamic correlation weight matrix between device nodes; Accumulating the dynamic association weight matrix in different time windows in segments to obtain cumulative weight matrices at multiple time scales; constructing a corresponding time scale association map based on the cumulative weight matrix at each time scale, and combining the multiple time scale association maps to form a hierarchical multi-scale dynamic association map structure; calculating the degree-normalized Laplace matrix for each layer in the multi-scale dynamic association map, and performing eigenvalue decomposition to obtain a characteristic spectrum vector; The distance metric between the characteristic spectra of the normalized Laplace matrices of each layer is calculated, the distance metric is mapped to the kernel Hilbert space, and an inter-layer spectral distance matrix is ​​constructed; a dynamic singular spectrum analysis model is constructed by analyzing the inter-layer spectral distance matrix, a sliding window decomposition is performed on the matrix, and Hankel matrix embedding is applied to extract timing features; based on the timing features, the singular value sequence is mapped to the time-frequency domain to construct an energy spectrum, and the key evolution moments in the energy spectrum are identified by combining self-service sampling and cumulative sum detection algorithms; and a dynamic association map of the device group is reconstructed at the key evolution moments.

[0031] First, the density operator is separated from the system and environment correlation terms through the adaptive quantum environment decoupling method to obtain the purified density operator. This step uses dynamic decoupling technology to gradually eliminate the influence of environmental noise on the quantum system through an iterative optimization process. In specific implementation, the quantum principal component analysis method can be used to identify the main environmental correlation patterns, and then the corresponding decoupling pulse sequence can be designed to suppress these correlations. For example, for a two-bit quantum system, a π / 2 pulse sequence can be applied to offset the influence of static magnetic field noise.

[0032] Next, a conditional partial trace operation is performed on the purified density operator to obtain a reduced density matrix. This step reduces the system dimensionality by integrating over the subsystems of no interest, retaining the quantum correlation information of interest. For example, for a three-qubit system, if only the correlation between the first two qubits is of interest, a partial trace operation can be performed on the third qubit.

[0033] Then, the von Neumann entropy of the reduced density matrix is ​​calculated, and the quantum mutual information between node pairs is calculated based on the von Neumann entropy. The von Neumann entropy reflects the purity of the quantum state, while the quantum mutual information measures the total correlation strength between the two subsystems. The eigenvalue decomposition method of the density matrix can be used for calculation. For a two-qubit system, if the eigenvalues ​​of the reduced density matrix are 0.7 and 0.3, its von Neumann entropy is about 0.61.

[0034] The quantum mutual information is normalized to obtain the dynamic correlation weight matrix between device nodes. Normalization can be done by using methods such as maximum and minimum value scaling or softmax function, so that the weight value falls between 0 and 1. This facilitates subsequent graph construction and analysis.

[0035] The dynamic association weight matrix is ​​accumulated in different time windows in segments to obtain the cumulative weight matrix at multiple time scales. The exponentially growing time window size can be selected, such as 1 minute, 5 minutes, 30 minutes, etc., to capture dynamic features at different scales. The accumulation process can use methods such as sliding average or exponential weighted average.

[0036] Based on the cumulative weight matrix at each time scale, the corresponding time scale correlation map is constructed, and multiple time scale correlation maps are combined to form a hierarchical multi-scale dynamic correlation map structure. These maps can be drawn and analyzed using graphic visualization tools such as Gephi or NetworkX. Maps at different levels can be distinguished by color coding or spatial layout.

[0037] The degree-normalized Laplace matrix is ​​calculated for each layer in the multi-scale dynamic association graph, and the eigenvalue decomposition is performed to obtain the eigenspectral vector. Degree normalization can eliminate the influence of node degree differences and highlight the topological structure characteristics. Eigenvalue decomposition can be efficiently solved using numerical methods such as power iteration or Lanczos algorithm.

[0038] Calculate the distance metric between the characteristic spectra of the normalized Laplacian matrices of each layer, map the distance metric to the kernel Hilbert space, and construct the inter-layer spectral distance matrix. The distance metric can be selected from Euclidean distance, Manhattan distance, or KL divergence. Kernel mapping can use common kernel functions such as Gaussian kernel or polynomial kernel.

[0039] A dynamic singular spectrum analysis model is constructed by analyzing the inter-layer spectral distance matrix, performing sliding window decomposition on the matrix and applying Hankel matrix embedding to extract time series features. The sliding window size can be selected based on the data characteristics, such as selecting a window size that can cover a complete cycle. Hankel matrix embedding can capture the dynamic characteristics and periodic patterns of time series.

[0040] Based on the time series characteristics, the singular value sequence is mapped to the time-frequency domain to construct the energy spectrum, and the key evolution moments in the energy spectrum are identified by combining the self-service sampling and cumulative sum detection algorithm. The time-frequency domain mapping can adopt methods such as short-time Fourier transform or wavelet transform. Self-service sampling can estimate the confidence interval of the energy spectrum, while the cumulative sum detection algorithm can accurately locate the mutation point.

[0041] Finally, the dynamic association graph of the device group is reconstructed at the key evolution moment. The reconstruction process can use the multi-scale graph information obtained in the previous step, combined with graph fusion algorithms such as graph matching or spectral clustering, to generate a comprehensive association graph that can reflect the changes in the key states of the system.

[0042] The solution of this application can: This method achieves efficient analysis and visualization of complex quantum systems through quantum walk-enhanced dynamic correlation graph construction technology. It can effectively capture the multi-scale dynamic characteristics of the system and provide a novel method for characterizing quantum correlation structures. This method combines quantum information theory and graph analysis technology, which can effectively reduce data complexity while retaining quantum correlation information. By constructing a multi-scale dynamic correlation graph, the evolution characteristics of the system at different time and space scales can be intuitively displayed, providing a powerful tool for the analysis and control of complex quantum systems. This method introduces dynamic singular spectrum analysis and energy spectrum construction technology, which can accurately identify key moments and mutation points in the system evolution process. This provides new ideas for anomaly detection and predictive maintenance of quantum systems, and helps to improve the reliability and stability of quantum devices.

[0043] In an optional implementation, in the dynamic correlation graph, the quantum random walk algorithm is used to calculate the quantum correlation strength between device nodes, and the dynamic evolution process of the device group is simulated by the Schrödinger evolution equation to obtain a device group quantum correlation tensor with topological invariant characteristics, including: Mapping device nodes in a dynamic association graph to quantum ground states in a quantum Hilbert space, mapping the association strengths between device nodes in the dynamic association graph to quantum state superposition coefficients, and constructing a quantum state representation of the dynamic association graph; constructing a quantum walk operator based on the quantum state representation, wherein the quantum walk operator is represented by an exponential form of a system Hamiltonian, wherein the matrix elements of the system Hamiltonian are proportional to the association strengths between device nodes; Applying the quantum walk operator to the initial quantum state, solving the continuous-time quantum random walk evolution equation to obtain the quantum transfer probability between two device nodes at any time, and performing time-averaged calculation on the quantum transfer probability within the characteristic evolution period to obtain the quantum correlation strength between the device nodes; constructing a quantum dynamics model based on the Schrödinger equation, obtaining the evolution trajectory of the system quantum state by numerically solving the Schrödinger equation, and extracting the geometric phase of the system quantum state from the evolution trajectory as a topological invariant; Select three device nodes, calculate the quantum correlation strength between any two of the three device nodes, and multiply the quantum correlation strengths to obtain a correlation strength product; calculate the geometric phase of the loop formed by the three device nodes, and substitute the geometric phase into an exponential function to obtain a phase index; multiply the correlation strength product by the phase index to construct a tensor element; characterize the correlation strength between the device nodes based on the modulus of the tensor element, characterize the topological characteristics between the device nodes based on the phase of the tensor element, and obtain a third-order quantum correlation tensor that characterizes the three-body correlation characteristics between the three device nodes; determine the device group quantum correlation tensor based on the third-order quantum correlation tensors of multiple device nodes.

[0044] When calculating the quantum correlation strength between device nodes in a dynamic correlation graph, it is necessary to first construct a quantum state representation. Each device node is mapped to a basis state vector in the quantum Hilbert space, and the correlation strength between nodes corresponds to the superposition coefficient of the quantum state. For example, for a system containing ten device nodes, each node is represented as a ten-dimensional basis vector, and the initial correlation strength between nodes can be determined after normalization based on actual operating data such as the communication frequency and data interaction volume between devices.

[0045] When constructing a quantum walk operator, it is constructed based on the Hamiltonian of the system. In specific implementation, the correlation strength between device nodes can be used as the weight factor of the Hamiltonian matrix element. For example, when the correlation strength is 0.8, the corresponding matrix element value is 0.8. The quantum walk operator constructed in this way can accurately reflect the correlation relationship between device nodes in the system.

[0046] When calculating the quantum transfer probability, it is necessary to simulate the quantum random walk process. Set an appropriate evolution time step, such as 0.01, and the total evolution time is 10 units of time. At each time step, calculate the transfer probability between two device nodes, and perform cumulative averaging over the entire evolution cycle to finally obtain the quantum correlation strength in the steady state.

[0047] For the solution of the Schrödinger equation, a numerical method is used for time evolution. The time step can be selected as 0.001 and the total evolution time is 100 unit time. During the evolution process, the trajectory information of the system quantum state is recorded and the geometric phase is extracted as the topological invariant. In actual calculations, the geometric phase information can be obtained by recording the change in the inner product of the quantum state between adjacent time steps.

[0048] When constructing a third-order quantum correlation tensor, you first need to select three device nodes in the system. Calculate the quantum correlation strength between the three nodes. Assuming that the three correlation strengths are 0.7, 0.8, and 0.6, respectively, the product of the correlation strengths is 0.336. At the same time, calculate the geometric phase of the loop formed by these three nodes. For example, the phase value is 1.5, and the phase index is obtained after conversion through an exponential function. Multiply the correlation strength product by the phase index to obtain a tensor element, whose modulus represents the magnitude of the correlation strength, and the phase reflects the topological characteristics.

[0049] Finally, by calculating the third-order tensors corresponding to all possible three-node combinations in the system, a complete device group quantum correlation tensor is constructed. This tensor contains the multi-body correlation information between all device nodes in the system.

[0050] The solution of this application can: By mapping device nodes to quantum states and introducing a quantum random walk algorithm, the dynamic correlation between device nodes can be described more accurately, overcoming the limitations of traditional correlation analysis methods and improving the accuracy of correlation strength calculation. The Schrödinger equation is used to describe system evolution and the geometric phase is introduced as a topological invariant, so that the system can capture the overall correlation characteristics between device groups and enhance the stability and reliability of the analysis results. By constructing a quantum correlation tensor, the correlation strength and topological characteristics are unified into the same mathematical framework, which not only retains the local correlation information between nodes, but also includes the overall topological characteristics at the group level, providing a more comprehensive basis for the dynamic management and optimization of device groups.

[0051] In an optional implementation, a quantum dynamics model based on the Schrödinger equation is constructed, the evolution trajectory of the system quantum state is obtained by numerically solving the Schrödinger equation, and the geometric phase of the system quantum state is extracted from the evolution trajectory as a topological invariant, including: Establish a wave function expression of the system in the quantum Hilbert space, and expand the wave function expression according to orthogonal basis vectors to obtain a state vector; construct a system Hamiltonian based on the state vector, wherein the system Hamiltonian consists of a kinetic energy term, a potential energy term, and an interaction term, wherein the kinetic energy term is represented by a quadratic form of a momentum operator, the potential energy term is represented by a function of a position operator, and the interaction term is characterized by a spin-orbit coupling term and an external field modulation term; substitute the system Hamiltonian and the wave function into the time-dependent Schrödinger equation to construct a system dynamics evolution equation, and introduce a system-environment interaction term and a quantum noise term into the system dynamics evolution equation to form a quantum dynamics model corrected by the decoherence effect; The quantum dynamics model corrected by the decoherence effect is subjected to space-time discretization, and the time domain and the space domain are discretized into finite-dimensional grid points to construct a time evolution operator; the time evolution operator is substituted into a fourth-order Runge-Kutta algorithm, and the discretized Schrödinger equation is iteratively solved through estimation steps and correction steps to calculate the dynamic evolution trajectory of the quantum state of the system, wherein the probability amplitude of the quantum state of the system is updated in each iterative step of the fourth-order Runge-Kutta algorithm, and the dynamic evolution trajectory is characterized by the time evolution sequence of the system energy expectation value, quantum coherence, entanglement and fidelity; Construct a system parameter space spanned by spin, orbital angular momentum and external field parameters, and track the dynamic evolution trajectory in the system parameter space; calculate the Berry phase generated by the evolution of the system quantum state around a closed path based on the dynamic evolution trajectory; use the Berry phase to calculate the Berry curvature in the system parameter space; integrate the Berry curvature on a closed surface to obtain the Chern number, and use the Chern number as a topological invariant that characterizes the overall topological characteristics of the system quantum state; and extract the geometric phase of the system quantum state by combining the Berry phase and the Chern number.

[0052] In quantum Hilbert space, we first construct the wave function expression of the system. This wave function can be expanded using orthogonal complete basis vectors, and the expansion coefficient is the probability amplitude of the quantum state. Based on this state vector, the Hamiltonian of the system is constructed, which contains three main parts: the kinetic energy term is expressed in the quadratic form of the momentum operator; the potential energy term uses the functional form of the position operator; and the interaction term includes spin-orbit coupling and external field modulation.

[0053] When constructing the system dynamics evolution equation, the above Hamiltonian and wave function are substituted into the time-dependent Schrödinger equation. In order to describe the quantum decoherence effect in the actual system, the interaction term between the system and the environment and the quantum noise term describing random fluctuations are introduced into the evolution equation. In specific implementation, the Lindblad form of the master equation can be selected to describe the evolution of the open quantum system.

[0054] The modified quantum dynamics equations are discretized in time and space. In the time domain, the time domain is divided into equal intervals, and the typical time step can be selected as 1 femtosecond; the space domain is divided into three-dimensional grid points, and the grid spacing can be taken as 0.1 nanometers. The time evolution operator is constructed based on the discretized grid.

[0055] The fourth-order Runge-Kutta algorithm is used to solve the discretized Schrödinger equation. In each iterative step, the probability amplitude of the quantum state is updated through two stages: estimation and correction. The estimation step first calculates the quantum state derivative under the current state of the system, and the correction step corrects the quantum state according to the estimation result. By recording the time evolution sequence of physical quantities such as the system energy expectation value, quantum coherence, quantum entanglement, and quantum state fidelity, the dynamic evolution trajectory of the system is obtained.

[0056] The evolution trajectory is tracked in the parameter space composed of spin, orbital angular momentum and external field parameters. When the quantum state evolves along a closed path in the parameter space, its acquired Berry phase can be calculated. Based on the Berry phase distribution on the evolution trajectory, the Berry curvature field in the parameter space is calculated. By integrating the Berry curvature on the closed surface, the Chern number that characterizes the topological properties of the system can be obtained. Finally, the geometric phase information of the quantum state of the system is extracted by combining the Berry phase and the Chern number.

[0057] The solution of this application can: By introducing the interaction between the system and the environment and the quantum noise term, the quantum dynamics model is made more consistent with the characteristics of the actual physical system, and the accuracy and reliability of the model are improved. At the same time, the time-space discretization processing method is adopted to enable the complex quantum dynamics equations to be numerically solved on the computer. The fourth-order Runge-Kutta algorithm is used for numerical solution, and the calculation accuracy is improved by coordinating the estimation step and the correction step. By tracking the evolution of multiple physical quantities, the dynamic behavior of the quantum system is fully characterized, providing a reliable data basis for the subsequent analysis of the topological characteristics of the system. Extracting the geometric phase of the system quantum state in the parameter space, combined with the Berry phase and Chern number, can effectively characterize the overall topological characteristics of the quantum system. This method can not only reveal the intrinsic topological properties of the system, but also provide theoretical guidance for quantum state manipulation and quantum computing.

[0058] In an optional implementation, a graph neural network based on a Dirac propagator is constructed, and the spatiotemporal features are mapped to a Riemann manifold space according to the graph neural network based on the Dirac propagator to obtain a manifold feature representation of the device group, including: Construct a topological structure diagram of the device group, and define a graph Laplace operator based on the topological structure diagram; decompose the graph Laplace operator into a symmetric part and an antisymmetric part, construct a mass term based on the symmetric part, and construct a chiral term based on the antisymmetric part; combine the mass term and the chiral term to construct a Dirac operator, perform spectral decomposition on the Dirac operator to obtain eigenvalues ​​and eigenvectors; construct a Dirac propagator using the eigenvalues ​​and the eigenvectors, and the Dirac propagator describes the quantum characteristics of information propagation in the device group; A Chebyshev polynomial kernel function is constructed based on the real part and the imaginary part of the Dirac propagator, and a graph convolution layer is constructed based on the Chebyshev polynomial kernel function, wherein the graph convolution layer performs a tensor product operation on the input feature and the Dirac propagator; a multi-layer perceptron module is constructed, wherein the multi-layer perceptron module performs a nonlinear transformation on the output feature of the graph convolution layer; an attention mechanism layer is constructed, wherein the attention mechanism layer calculates the association weights between device nodes based on the output of the multi-layer perceptron module; and the graph convolution layer, the multi-layer perceptron module and the attention mechanism layer are cascaded to form a graph neural network model to extract the spatiotemporal features of the device group based on the input features; Construct a Riemann manifold space, wherein the Riemann manifold space has a local Euclidean metric structure; construct a tangent mapping using the spatiotemporal features extracted by the graph neural network model, wherein the tangent mapping maps the feature space of the device group to the tangent space of the Riemann manifold space; define a Riemann metric tensor in the tangent space, wherein the Riemann metric tensor describes the local geometric structure of the device group features in the Riemann manifold space; calculate a geodesic equation based on the Riemann metric tensor, map the spatiotemporal features of the device group to the Riemann manifold space by solving the geodesic equation, and obtain a manifold feature representation of the device group in the Riemann manifold space.

[0059] This embodiment provides a graph neural network method based on Dirac propagator, which is used to map spatiotemporal features to Riemann manifold space, thereby obtaining a manifold feature representation of a device group. The method includes the following main steps: First, build a topological structure graph of the device group. In this step, each device is regarded as a node in the graph, and the connection relationship between devices is represented as an edge. For example, for a group containing 10 devices, a 10-node graph structure can be built based on the physical connection or logical relationship between the devices.

[0060] Next, we define the graph Laplacian based on the constructed topology graph. The graph Laplacian is a matrix whose diagonal elements represent the degree of the nodes and the off-diagonal elements represent the connection relationship between the nodes. For example, for a simple three-node graph, its Laplacian may be a 3x3 matrix.

[0061] Then, the graph Laplacian operator is decomposed into a symmetric part and an antisymmetric part. The symmetric part reflects the undirected nature of the graph, while the antisymmetric part captures the directed information that may exist in the graph. Based on the symmetric part, a quality term is constructed, which describes the relative importance of the nodes in the graph. Based on the antisymmetric part, a chiral term is constructed, which reflects the directionality of information propagation in the graph.

[0062] The mass term and the chiral term are combined to construct the Dirac operator. The Dirac operator is a more complex operator that not only contains the topological information of the graph, but also introduces concepts from quantum mechanics. The Dirac operator is spectrally decomposed to obtain eigenvalues ​​and eigenvectors. These eigenvalues ​​and eigenvectors contain the essential characteristics of the graph structure.

[0063] The obtained eigenvalues ​​and eigenvectors are used to construct the Dirac propagator. The Dirac propagator is a complex function that describes the quantum characteristics of information propagation in a group of devices. This propagator can be regarded as a probability distribution of information propagation in the graph.

[0064] The Chebyshev polynomial kernel function is constructed based on the real and imaginary parts of the Dirac propagator. Chebyshev polynomials can effectively approximate complex functions, so it is a good choice to use it to process the Dirac propagator. This kernel function will be used in subsequent graph convolution operations.

[0065] The graph convolution layer is constructed using the constructed Chebyshev polynomial kernel function. The main function of the graph convolution layer is to perform a tensor product operation on the input features and the Dirac propagator. This process can be understood as a special convolution operation on the graph structure, which can effectively extract local and global features in the graph.

[0066] Construct a multi-layer perceptron module to perform nonlinear transformation on the output features of the graph convolution layer. The multi-layer perceptron consists of multiple fully connected layers, and nonlinear activation functions (such as ReLU) are used between each layer. This step increases the expressiveness of the model, enabling it to learn more complex feature representations.

[0067] Design an attention mechanism layer to calculate the association weights between device nodes based on the output of the multi-layer perceptron module. The attention mechanism can adaptively adjust the importance of different nodes to better capture the complex interactions between nodes.

[0068] The graph convolution layer, multi-layer perceptron module and attention mechanism layer are cascaded to form a complete graph neural network model. This model can extract the spatiotemporal features of a device group based on the input features. For example, for a group of 100 devices, the input may be multi-dimensional time series data of each device, and the output is a vector that can represent the spatiotemporal features of the entire group.

[0069] Construct a Riemannian manifold space that has a local Euclidean metric structure. A Riemannian manifold can be viewed as a complex surface that approximates Euclidean space near each point, but the overall structure may be very complex.

[0070] The tangent map is constructed using the spatiotemporal features extracted by the graph neural network model. The function of the tangent map is to map the feature space of the device group to the tangent space of the Riemann manifold space. The tangent space can be understood as a local linear approximation of a point on the Riemann manifold.

[0071] The Riemann metric tensor is defined in the tangent space. The Riemann metric tensor describes the local geometric structure of the device group characteristics in the Riemann manifold space. It determines how distances and angles are measured on the manifold.

[0072] The geodesic equation is calculated based on the Riemann metric tensor. A geodesic can be understood as the shortest path between two points on a manifold. By solving the geodesic equation, the spatiotemporal characteristics of a device group can be mapped to the Riemann manifold space.

[0073] Finally, we obtain the manifold feature representation of the device group in the Riemannian manifold space, which captures the essential characteristics of the device group, including its topological structure, temporal dynamics, and spatial distribution.

[0074] Through the above steps, the complex spatiotemporal features of a device group can be mapped into a Riemannian manifold space with good geometric properties, thereby obtaining a more compact and meaningful feature representation.

[0075] The solution of this application can: By introducing Dirac propagators and Riemann manifolds, this method can better capture the complex topological structures and nonlinear relationships in device groups. This representation method not only takes into account the direct connection between devices, but also reflects deeper interactions, thereby improving the accuracy and comprehensiveness of feature representation. Mapping spatiotemporal features to the Riemann manifold space gives the feature representation good geometric properties. This representation method retains the essential characteristics of the original data and is also convenient for subsequent analysis and processing. For example, clustering or classification operations on Riemann manifolds may be more effective and accurate than those in the original feature space. This method combines the advantages of graph neural networks and differential geometry and has strong versatility and extensibility. It is not only suitable for feature extraction of device groups, but can also be extended to the analysis of other complex network structures, such as social networks, transportation networks, etc. This method provides a new perspective and tool for the modeling and analysis of complex systems.

[0076] In an optional implementation, the purity of the device group cooperative state in the manifold feature representation is evaluated by using quantum entanglement entropy, and a contrast loss function is constructed in combination with the Wasserstein metric to obtain an abnormal pattern recognition result, including: The manifold feature representation is converted into a density matrix; the density matrix is ​​spectrally decomposed to obtain an eigenvalue sequence and an eigenvector sequence; a first reduced density matrix and a second reduced density matrix are constructed based on the eigenvalue sequence and the eigenvector sequence, wherein the first reduced density matrix and the second reduced density matrix respectively characterize the local subsystem states in the device group; the quantum entanglement entropy is calculated using the first reduced density matrix and the second reduced density matrix, wherein the quantum entanglement entropy is used to quantify the purity of the cooperative state of the device group; A positive sample pair and a negative sample pair are selected in the feature space corresponding to the manifold feature representation, wherein the positive sample pair is a feature representation of the same collaborative mode, and the negative sample pair is a feature representation of different collaborative modes; a first Wasserstein distance between two density matrices in the positive sample pair is calculated, and a second Wasserstein distance between two density matrices in the negative sample pair is calculated; a contrast loss function is constructed based on the first Wasserstein distance and the second Wasserstein distance, and the contrast loss function is used to optimize the distinguishability of feature representation; The quantum entanglement entropy and the contrast loss function are weightedly combined to obtain a discrimination score; a probability density function of the discrimination score is constructed based on historical data; the probability density function is solved according to a preset significance level to obtain an adaptive threshold; the discrimination score of a new sample is compared with the adaptive threshold, and when the discrimination score is greater than the adaptive threshold, it is determined that the device group is in an abnormal mode as an abnormal mode recognition result.

[0077] The present invention proposes an abnormal pattern recognition method based on quantum entanglement entropy and Wasserstein metric, which mainly includes the following steps: First, the manifold feature representation is converted into a density matrix. Specifically, the kernel principal component analysis (Kernel PCA) method can be used to map the original high-dimensional features to the low-dimensional manifold space, and then the manifold features are converted into a density matrix through exponential mapping. For example, for a given manifold feature vector x, the elements of the density matrix can be calculated by exp(-||x-xi||^2 / σ^2), where xi is the training sample and σ is the bandwidth parameter.

[0078] Next, perform spectral decomposition on the density matrix to obtain the eigenvalue sequence and eigenvector sequence. This can be achieved using numerical methods such as singular value decomposition (SVD) or eigenvalue decomposition. Assume that the obtained eigenvalue sequence is λ_1,λ_2,...,λ_n, and the corresponding eigenvector sequence is v_1,v_2,...,v_n.

[0079] Then, the first reduced density matrix and the second reduced density matrix are constructed based on the eigenvalue sequence and the eigenvector sequence. Here, a partial trace operation can be used to reduce the density matrix of the entire system to a local subsystem. For example, the eigenvectors corresponding to the first k largest eigenvalues ​​can be selected to construct the first reduced density matrix, and the eigenvectors corresponding to the last nk eigenvalues ​​can be selected to construct the second reduced density matrix.

[0080] The quantum entanglement entropy is calculated using the first reduced density matrix and the second reduced density matrix. The quantum entanglement entropy can be measured by von Neumann entropy, that is, S=-Tr(ρ log ρ), where ρ is the reduced density matrix and Tr represents the trace operation of the matrix. The larger the quantum entanglement entropy, the lower the purity of the device group cooperative state and the more likely it is that there is anomaly.

[0081] Positive sample pairs and negative sample pairs are selected in the feature space corresponding to the manifold feature representation. Positive sample pairs can be selected by selecting feature representations at different time points under the same collaborative mode, while negative sample pairs can be selected by selecting feature representations under different collaborative modes. For example, a sliding window method can be used to select sample pairs in time series data.

[0082] Calculate the first Wasserstein distance between the two density matrices in the positive sample pair, and the second Wasserstein distance between the two density matrices in the negative sample pair. The Wasserstein distance can be calculated by solving the optimal transmission problem, and a fast approximation method such as the Sinkhorn algorithm can be used. The Wasserstein distance can effectively measure the difference between probability distributions and is suitable for comparing density matrices.

[0083] Construct a contrast loss function based on the first Wasserstein distance and the second Wasserstein distance. The goal of the contrast loss function is to make the distance between positive sample pairs as small as possible and the distance between negative sample pairs as large as possible. It can be in a form similar to triplet loss, that is, L=max(0,m+d_pos-d_neg), where m is the boundary parameter, d_pos is the distance of the positive sample pair, and d_neg is the distance of the negative sample pair.

[0084] The discriminant score is obtained by weighted combination of quantum entanglement entropy and contrast loss function. A linear combination method can be used, that is, Score=α*S+(1-α)*L, where α is a weight parameter and the optimal value can be determined by cross-validation and other methods. The discriminant score comprehensively considers the purity of the cooperative state and the distinguishability of the feature representation.

[0085] Construct a probability density function of the discriminant score based on historical data. Non-parametric methods such as kernel density estimation (KDE) can be used to estimate the probability distribution of the discriminant scores based on a large number of historical samples. The KDE method can adaptively capture the distribution characteristics of the discriminant scores without assuming a specific distribution form.

[0086] The probability density function is solved according to the preset significance level to obtain the adaptive threshold. The significance level can be set according to the needs of the actual application scenario, such as 0.05 or 0.01. By solving the inverse function of the cumulative distribution function, the threshold corresponding to the significance level can be obtained. This method can adaptively adjust the threshold to adapt to different data distribution characteristics.

[0087] Finally, the discrimination score of the new sample is compared with the adaptive threshold. When the discrimination score is greater than the adaptive threshold, the device group is judged to be in abnormal mode, which is used as the abnormal mode recognition result. This judgment method based on statistical significance can effectively control the false alarm rate and improve the reliability of anomaly detection.

[0088] In actual applications, the status of the device group can be continuously monitored by sliding windows. For example, the discrimination score is calculated every certain time interval (such as 5 minutes) and compared with the current adaptive threshold. If the discrimination scores of multiple consecutive time windows exceed the threshold, an abnormal alarm can be triggered to prompt the operation and maintenance personnel to conduct further inspection and processing.

[0089] In addition, in order to improve the robustness and generalization ability of the method, an online learning strategy can be adopted to regularly use new normal samples to update the probability density function and adaptive threshold. This allows the anomaly detection model to adapt to the dynamic changes in the behavior patterns of the device group and improve the stability and accuracy of long-term operation.

[0090] The solution of this application can: This method evaluates the purity of the device group cooperative state by introducing quantum entanglement entropy, which can effectively capture the complex interactions and correlations between multiple devices. As an information theory metric, quantum entanglement entropy can reveal hidden structures and nonlinear dependencies in the system, thereby improving the sensitivity and accuracy of anomaly detection. This method based on quantum information theory provides a new perspective and tool for traditional anomaly detection technology, which helps to discover subtle abnormal patterns that are difficult to identify with traditional methods. The contrast loss function is constructed using the Wasserstein metric, which can better measure the similarities and differences between samples in the feature representation space. The Wasserstein distance takes into account the geometric structure of the probability distribution and is more suitable for processing high-dimensional sparse data and multi-peak distributions than traditional metric methods such as the Euclidean distance. By minimizing the distance between positive sample pairs and maximizing the distance between negative sample pairs, this method can learn more discriminative feature representations and improve the ability to identify abnormal patterns. This method combines adaptive threshold technology to dynamically adjust the discrimination criteria according to the statistical characteristics of historical data. This threshold setting method based on probability density function and significance level can effectively control the false alarm rate while maintaining a high detection rate. The adaptive threshold technology enables the method to adapt to different application scenarios and data distribution characteristics, and has good generalization ability and practicality. In general, the method proposed in this invention has significant advantages in improving the accuracy of anomaly detection, enhancing model interpretability and enhancing system adaptability, and provides strong technical support for intelligent monitoring and predictive maintenance of industrial equipment groups.

[0091] Figure 2 FIG. 1 is a schematic diagram of a structure of a collaborative fault prediction and analysis system for equipment groups in a complex industrial process according to an embodiment of the present invention. Figure 2 As shown, the system comprises: The first unit is used to collect the operating parameter data, status data, alarm data and maintenance record data of the equipment group in the complex industrial process, and perform noise reduction and time series reconstruction through adaptive wavelet transform to obtain multi-scale time series characteristics; based on the multi-scale time series characteristics, a quantum walk-enhanced dynamic correlation map is constructed; in the dynamic correlation map, the quantum random walk algorithm is used to calculate the quantum correlation strength between the equipment nodes, and the dynamic evolution process of the equipment group is simulated through the Schrödinger evolution equation to obtain the equipment group quantum correlation tensor with topological invariant characteristics; The second unit is used to extract the spatiotemporal features of the device group quantum correlation tensor using a bidirectional quantum gated recurrent network; construct a graph neural network based on a Dirac propagator, and map the spatiotemporal features to a Riemann manifold space according to the graph neural network based on a Dirac propagator to obtain a manifold feature representation of the device group; use quantum entanglement entropy to evaluate the purity of the device group cooperative state in the manifold feature representation, and construct a contrast loss function in combination with the Wasserstein metric to obtain an abnormal pattern recognition result; input the abnormal pattern recognition result into a deep belief network based on a Boltzmann machine to generate a hidden variable probability distribution of the device group; The third unit is used to construct a fault evolution model driven by non-equilibrium statistical fluctuation theory based on the hidden variable probability distribution; generate a state evolution equation of the equipment group using the fault evolution model driven by non-equilibrium statistical fluctuation theory; map the state evolution equation to the Lie group space, and construct a fault trajectory tensor through Lie algebra calculation; apply a random differential operator to the fault trajectory tensor to establish a group of coordinated field equations; construct a Hamiltonian according to the group of coordinated field equations, and solve the corresponding group of canonical equations to obtain a control parameter sequence of the equipment group.

[0092] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0093] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0094] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A collaborative fault prediction and analysis method for equipment groups in complex industrial processes, characterized in that: include: Collect operating parameter data, status data, alarm data and maintenance record data of equipment groups in complex industrial processes, and perform noise reduction and time series reconstruction through adaptive wavelet transform to obtain multi-scale time series features; Based on the multi-scale time series characteristics, a quantum walk-enhanced dynamic correlation map is constructed; in the dynamic correlation map, the quantum random walk algorithm is used to calculate the quantum correlation strength between device nodes, and the dynamic evolution process of the device group is simulated by the Schrödinger evolution equation to obtain a device group quantum correlation tensor with topological invariant characteristics; For the quantum correlation tensor of the device group, a bidirectional quantum gated recurrent network is used to extract the spatiotemporal characteristics; A graph neural network based on Dirac propagator is constructed, and the spatiotemporal features are mapped to Riemann manifold space according to the graph neural network based on Dirac propagator to obtain the manifold feature representation of the device group; the purity of the cooperative state of the device group in the manifold feature representation is evaluated by using quantum entanglement entropy, and a contrast loss function is constructed in combination with Wasserstein metric to obtain an abnormal pattern recognition result; the abnormal pattern recognition result is input into a deep belief network based on Boltzmann machine to generate a hidden variable probability distribution of the device group; Based on the hidden variable probability distribution, a fault evolution model driven by non-equilibrium statistical fluctuation theory is constructed; the state evolution equation of the device group is generated by using the fault evolution model driven by non-equilibrium statistical fluctuation theory; The state evolution equation is mapped to the Lie group space, and the fault trajectory tensor is constructed through Lie algebra calculation; a random differential operator is applied to the fault trajectory tensor to establish a set of coordinated field equations; a Hamiltonian is constructed according to the set of coordinated field equations, and the corresponding canonical equations are solved to obtain a control parameter sequence of the device group.

2. The method according to claim 1, characterized in that Based on the multi-scale temporal characteristics, a dynamic correlation map of quantum walking enhancement is constructed, including: Constructing a high-order tensor based on the multi-scale time series features, wherein the dimensions of the high-order tensor include device dimension, parameter dimension and time dimension; performing tensor decomposition on the high-order tensor to obtain a tensor representing the characteristics of the device group and a factor matrix of each dimension; projecting the factor matrix into a complex Hilbert space to construct an initial quantum state of the device node, wherein the initial quantum state includes an amplitude component and a phase component; Constructing a non-Hermitian interaction operator based on the tensor characterizing the characteristics of the device group, wherein the non-Hermitian interaction operator characterizes the non-equilibrium quantum correlation between device nodes; combining the non-Hermitian interaction operator and the initial quantum state to construct a quantum system Hamiltonian; obtaining a dynamic coupling matrix based on the eigenvalue decomposition of the quantum system Hamiltonian, wherein the dynamic coupling matrix characterizes the correlation strength between device nodes; A quantum master equation containing noise terms is constructed based on the dynamic coupling matrix, wherein the quantum master equation includes a drift term determined by the non-Hermitian interaction operator and a quantum fluctuation diffusion term modulated by the dynamic coupling matrix; the drift term and the quantum fluctuation diffusion term are numerically solved based on the quantum Monte Carlo method to obtain a density operator that characterizes the evolution of the system; and a quantum walk-enhanced dynamic correlation map is constructed based on the density operator.

3. The method according to claim 2, characterized in that The method of constructing a quantum walk enhanced dynamic correlation map based on the density operator includes: The density operator is separated from system and environment related items by an adaptive quantum environment decoupling method to obtain a purified density operator; a conditional partial trace operation is performed on the purified density operator to obtain a reduced density matrix, the von Neumann entropy of the reduced density matrix is ​​calculated, and the quantum mutual information between node pairs is calculated based on the von Neumann entropy; the quantum mutual information is normalized to obtain a dynamic correlation weight matrix between device nodes; Accumulating the dynamic association weight matrix in different time windows in segments to obtain cumulative weight matrices at multiple time scales; constructing a corresponding time scale association map based on the cumulative weight matrix at each time scale, and combining the multiple time scale association maps to form a hierarchical multi-scale dynamic association map structure; calculating the degree-normalized Laplace matrix for each layer in the multi-scale dynamic association map, and performing eigenvalue decomposition to obtain a characteristic spectrum vector; The distance metric between the characteristic spectra of the normalized Laplace matrices of each layer is calculated, the distance metric is mapped to the kernel Hilbert space, and an inter-layer spectral distance matrix is ​​constructed; a dynamic singular spectrum analysis model is constructed by analyzing the inter-layer spectral distance matrix, a sliding window decomposition is performed on the matrix, and Hankel matrix embedding is applied to extract timing features; based on the timing features, the singular value sequence is mapped to the time-frequency domain to construct an energy spectrum, and the key evolution moments in the energy spectrum are identified by combining self-service sampling and cumulative sum detection algorithms; and a dynamic association map of the device group is reconstructed at the key evolution moments.

4. The method according to claim 1, characterized in that: In the dynamic correlation graph, the quantum random walk algorithm is used to calculate the quantum correlation strength between device nodes, and the dynamic evolution process of the device group is simulated by the Schrödinger evolution equation to obtain the device group quantum correlation tensor with topological invariant characteristics, including: Mapping device nodes in a dynamic association graph to quantum ground states in a quantum Hilbert space, mapping the association strengths between device nodes in the dynamic association graph to quantum state superposition coefficients, and constructing a quantum state representation of the dynamic association graph; constructing a quantum walk operator based on the quantum state representation, wherein the quantum walk operator is represented by an exponential form of a system Hamiltonian, wherein the matrix elements of the system Hamiltonian are proportional to the association strengths between device nodes; Applying the quantum walk operator to the initial quantum state, solving the continuous-time quantum random walk evolution equation to obtain the quantum transfer probability between two device nodes at any time, and performing time-averaged calculation on the quantum transfer probability within the characteristic evolution period to obtain the quantum correlation strength between the device nodes; constructing a quantum dynamics model based on the Schrödinger equation, obtaining the evolution trajectory of the system quantum state by numerically solving the Schrödinger equation, and extracting the geometric phase of the system quantum state from the evolution trajectory as a topological invariant; Select three device nodes, calculate the quantum correlation strength between any two of the three device nodes, and multiply the quantum correlation strengths to obtain a correlation strength product; calculate the geometric phase of the loop formed by the three device nodes, and substitute the geometric phase into an exponential function to obtain a phase index; multiply the correlation strength product by the phase index to construct a tensor element; characterize the correlation strength between the device nodes based on the modulus of the tensor element, characterize the topological characteristics between the device nodes based on the phase of the tensor element, and obtain a third-order quantum correlation tensor that characterizes the three-body correlation characteristics between the three device nodes; determine the device group quantum correlation tensor based on the third-order quantum correlation tensors of multiple device nodes.

5. The method according to claim 4, characterized in that A quantum dynamics model based on the Schrödinger equation is constructed, the evolution trajectory of the system quantum state is obtained by numerically solving the Schrödinger equation, and the geometric phase of the system quantum state is extracted from the evolution trajectory as a topological invariant, including: Establish a wave function expression of the system in the quantum Hilbert space, and expand the wave function expression according to orthogonal basis vectors to obtain a state vector; construct a system Hamiltonian based on the state vector, wherein the system Hamiltonian consists of a kinetic energy term, a potential energy term, and an interaction term, wherein the kinetic energy term is represented by a quadratic form of a momentum operator, the potential energy term is represented by a function of a position operator, and the interaction term is characterized by a spin-orbit coupling term and an external field modulation term; substitute the system Hamiltonian and the wave function into the time-dependent Schrödinger equation to construct a system dynamics evolution equation, and introduce a system-environment interaction term and a quantum noise term into the system dynamics evolution equation to form a quantum dynamics model corrected by the decoherence effect; The quantum dynamics model corrected by the decoherence effect is subjected to space-time discretization, and the time domain and the space domain are discretized into finite-dimensional grid points to construct a time evolution operator; the time evolution operator is substituted into a fourth-order Runge-Kutta algorithm, and the discretized Schrödinger equation is iteratively solved through estimation steps and correction steps to calculate the dynamic evolution trajectory of the quantum state of the system, wherein the probability amplitude of the quantum state of the system is updated in each iterative step of the fourth-order Runge-Kutta algorithm, and the dynamic evolution trajectory is characterized by the time evolution sequence of the system energy expectation value, quantum coherence, entanglement and fidelity; Construct a system parameter space spanned by spin, orbital angular momentum and external field parameters, and track the dynamic evolution trajectory in the system parameter space; calculate the Berry phase generated by the evolution of the system quantum state around a closed path based on the dynamic evolution trajectory; use the Berry phase to calculate the Berry curvature in the system parameter space; integrate the Berry curvature on a closed surface to obtain the Chern number, and use the Chern number as a topological invariant that characterizes the overall topological characteristics of the system quantum state; and extract the geometric phase of the system quantum state by combining the Berry phase and the Chern number.

6. The method according to claim 1, characterized in that Constructing a graph neural network based on a Dirac propagator, mapping the spatiotemporal features to a Riemann manifold space according to the graph neural network based on the Dirac propagator, and obtaining a manifold feature representation of the device group, including: Construct a topological structure diagram of the device group, and define a graph Laplace operator based on the topological structure diagram; decompose the graph Laplace operator into a symmetric part and an antisymmetric part, construct a mass term based on the symmetric part, and construct a chiral term based on the antisymmetric part; combine the mass term and the chiral term to construct a Dirac operator, perform spectral decomposition on the Dirac operator to obtain eigenvalues ​​and eigenvectors; construct a Dirac propagator using the eigenvalues ​​and the eigenvectors, and the Dirac propagator describes the quantum characteristics of information propagation in the device group; A Chebyshev polynomial kernel function is constructed based on the real part and the imaginary part of the Dirac propagator, and a graph convolution layer is constructed based on the Chebyshev polynomial kernel function, wherein the graph convolution layer performs a tensor product operation on the input feature and the Dirac propagator; a multi-layer perceptron module is constructed, wherein the multi-layer perceptron module performs a nonlinear transformation on the output feature of the graph convolution layer; an attention mechanism layer is constructed, wherein the attention mechanism layer calculates the association weights between device nodes based on the output of the multi-layer perceptron module; and the graph convolution layer, the multi-layer perceptron module and the attention mechanism layer are cascaded to form a graph neural network model to extract the spatiotemporal features of the device group based on the input features; Construct a Riemann manifold space, wherein the Riemann manifold space has a local Euclidean metric structure; construct a tangent mapping using the spatiotemporal features extracted by the graph neural network model, wherein the tangent mapping maps the feature space of the device group to the tangent space of the Riemann manifold space; define a Riemann metric tensor in the tangent space, wherein the Riemann metric tensor describes the local geometric structure of the device group features in the Riemann manifold space; calculate a geodesic equation based on the Riemann metric tensor, map the spatiotemporal features of the device group to the Riemann manifold space by solving the geodesic equation, and obtain a manifold feature representation of the device group in the Riemann manifold space.

7. The method according to claim 1, characterized in that The purity of the device group cooperative state in the manifold feature representation is evaluated by using quantum entanglement entropy, and the contrast loss function is constructed in combination with the Wasserstein metric to obtain the abnormal pattern recognition results, including: The manifold feature representation is converted into a density matrix; the density matrix is ​​spectrally decomposed to obtain an eigenvalue sequence and an eigenvector sequence; a first reduced density matrix and a second reduced density matrix are constructed based on the eigenvalue sequence and the eigenvector sequence, wherein the first reduced density matrix and the second reduced density matrix respectively characterize the local subsystem states in the device group; the quantum entanglement entropy is calculated using the first reduced density matrix and the second reduced density matrix, wherein the quantum entanglement entropy is used to quantify the purity of the cooperative state of the device group; A positive sample pair and a negative sample pair are selected in the feature space corresponding to the manifold feature representation, wherein the positive sample pair is a feature representation of the same collaborative mode, and the negative sample pair is a feature representation of different collaborative modes; a first Wasserstein distance between two density matrices in the positive sample pair is calculated, and a second Wasserstein distance between two density matrices in the negative sample pair is calculated; a contrast loss function is constructed based on the first Wasserstein distance and the second Wasserstein distance, and the contrast loss function is used to optimize the distinguishability of feature representation; The quantum entanglement entropy and the contrast loss function are weightedly combined to obtain a discrimination score; a probability density function of the discrimination score is constructed based on historical data; the probability density function is solved according to a preset significance level to obtain an adaptive threshold; the discrimination score of a new sample is compared with the adaptive threshold, and when the discrimination score is greater than the adaptive threshold, it is determined that the device group is in an abnormal mode as an abnormal mode recognition result.

8. A collaborative fault prediction and analysis system for equipment groups in complex industrial processes, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to collect the operating parameter data, status data, alarm data and maintenance record data of equipment groups in complex industrial processes, and perform noise reduction and time series reconstruction through adaptive wavelet transform to obtain multi-scale time series features; Based on the multi-scale time series characteristics, a quantum walk-enhanced dynamic correlation map is constructed; in the dynamic correlation map, the quantum random walk algorithm is used to calculate the quantum correlation strength between device nodes, and the dynamic evolution process of the device group is simulated by the Schrödinger evolution equation to obtain a device group quantum correlation tensor with topological invariant characteristics; The second unit is used to extract the spatiotemporal characteristics of the device group quantum correlation tensor using a bidirectional quantum gated recurrent network; A graph neural network based on Dirac propagator is constructed, and the spatiotemporal features are mapped to Riemann manifold space according to the graph neural network based on Dirac propagator to obtain the manifold feature representation of the device group; the purity of the cooperative state of the device group in the manifold feature representation is evaluated by using quantum entanglement entropy, and a contrast loss function is constructed in combination with Wasserstein metric to obtain an abnormal pattern recognition result; the abnormal pattern recognition result is input into a deep belief network based on Boltzmann machine to generate a hidden variable probability distribution of the device group; The third unit is used to construct a fault evolution model driven by non-equilibrium statistical fluctuation theory based on the hidden variable probability distribution; and generate a state evolution equation of a device group using the fault evolution model driven by non-equilibrium statistical fluctuation theory; The state evolution equation is mapped to the Lie group space, and the fault trajectory tensor is constructed through Lie algebra calculation; a random differential operator is applied to the fault trajectory tensor to establish a set of coordinated field equations; a Hamiltonian is constructed according to the set of coordinated field equations, and the corresponding canonical equations are solved to obtain a control parameter sequence of the device group.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images