Smoke analysis AI system based on quantum field theory tensor network

Abstract

The invention provides a smoke analysis AI system based on a quantum field theory tensor network, the algorithm is provided with a multi-camera fusion module, a quantum state coding module, a dynamic Hamiltonian construction module, a tensor network evolution module and a chaos detection module, and by monitoring a smoke diffusion Lyapunov index in real time, the sudden change from smoke to fire behavior is warned in advance; according to the algorithm, a quantum multi-body system theory and computer vision are combined, smoke diffusion is modeled into a phase change process of a quantum spin chain, high-dimensional quantum state information is compressed by using a tensor network, and a mutation behavior of a chaotic edge is detected in combination with a Lyapunov exponent. The method solves the problem of early fire hazard false alarm under the complex operation condition of the transformer substation in a traditional method, improves the initiative and accuracy of monitoring and prevention of the extremely early smoke fire hazard of the transformer substation, provides more reliable guarantee for the safe operation of the transformer substation, and has wide application prospects and important practical significance.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of the cross of artificial intelligence and quantum computing, in particular to a smoke analysis AI system based on a quantum field theory tensor network. BACKGROUND

[0002] With the continuous improvement and development of monitoring technology, the image analysis technology is used in security and protection, management, fire alarm and the like, and good effects are achieved, but in the image smoke analysis sensing technology, the traditional smoke analysis technology has many defects, for example, an optical method, such as infrared absorption spectroscopy, depends on specific wavelengths and is easily disturbed by complex environment such as water vapor; a traditional convolutional neural network algorithm needs a large amount of labeled data and is difficult to explain and track the diffusion and evolution process of smoke in real time; a light flow method has poor stability in a complex dynamic background environment, has a large amount of calculation and high calculation complexity, especially in spaces such as substations and basements, the amount and complexity of image analysis and calculation are large; with the continuous development of science and technology and the continuous breakthrough in the computing field, quantum computing as a new computing method has entered the fields of life, study and research of people, has strong parallel computing capability and modeling capability for complex systems, a quantum field theory tensor network combines the advantages of quantum field theory and tensor network, and provides a new idea and method for fire alarm smoke analysis; therefore, the applicant proposes the smoke analysis AI system based on the quantum field theory tensor network according to the image analysis requirements of fire alarm, combines quantum many-body system theory and computer vision, models smoke diffusion as a phase transition process of a quantum spin chain, compresses high-dimensional quantum state information by using a tensor network, detects mutation behavior of a chaotic edge by combining a Lyapunov index, and solves the false alarm problem of early fire hazards under the operation condition of a complex environment of the traditional method. SUMMARY

[0003] To solve the above technical problems, the smoke analysis AI system based on the quantum field theory tensor network is provided, a plurality of camera fusion, quantum state coding modules, dynamic Hamiltonian construction modules, tensor network evolution modules and chaotic detection modules are arranged, the Lyapunov index of smoke diffusion is monitored in real time, and the mutation of smoke to fire is warned in advance; the false alarm problem of early fire hazards under the operation condition of a complex environment of the traditional method is solved, the initiative and accuracy of early smoke fire hazard monitoring and prevention in a substation are improved, more reliable protection is provided for the safe operation of the substation, and the smoke analysis AI system based on the quantum field theory tensor network has wide application prospect and important practical significance.

[0004] To achieve the above purpose, the technical scheme adopted by the application is as follows:

[0005] The AI system for smoke analysis based on quantum field theory tensor network is characterized in that: the AI algorithm for smoke analysis based on quantum field theory tensor network sets a multi-camera fusion module, a quantum state coding module, a dynamic Hamiltonian construction module, a tensor network evolution module and a chaos detection module; specifically:

[0006] Multi-camera fusion: multiple monitoring videos are spliced into a joint quantum state;

[0007] Quantum state coding module: the quantum state module encodes the feature information such as the diffusion direction and concentration gradient of smoke into quantum states by mapping classical information to quantum state space, and uses the continuity and superposition of quantum states to preserve the continuous dynamic characteristics of smoke; the image pixel array is mapped to a quantum spin state, and the state vector is:

[0008] ∣ψ >=i=1⨂N ( ∣0 >+ >)

[0009] Where: is the normalized brightness;

[0010] =arctan(Gi / Bi) is the hue phase angle;

[0011] The specific process of mapping the image pixel array to the quantum spin state is as follows:

[0012] 1) Image preprocessing and parameter extraction;

[0013] Pixel vectorization: the pixel matrix of the input image is expanded into a one-dimensional vector by column, and the RGB value of each pixel is mapped to a three-dimensional vector, i.e. Ri, Gi, Bi;

[0014] Normalization brightness calculation: the pixel brightness is mapped to the quantum probability amplitude range through linear transformation, and is defined as , so that ai ∈ [0, 1];

[0015] Satisfy the normalization condition of quantum state: ;

[0016] Hue phase angle generation: based on color space mapping, the color information is encoded into the phase angle through =arctan(Gi / Bi), which reflects the position of the pixel on the color wheel;

[0017] 2) Quantum spin state construction;

[0018] Single-pixel quantum state definition: each pixel corresponds to a quantum bit, and the state vector is:

[0019] ∣ψ >=i=1⨂N ( ∣0 >+ >)

[0020] Among them: |0> and |1> are ground states;

[0021] To control brightness;

[0022] is the encoding hue‌;

[0023] Quantum superposition and entanglement: through parameters Realize the probabilistic superposition of brightness, Introducing quantum interference effects so that the phase correlation between adjacent pixels reflects color continuity;

[0024] Dynamic Hamiltonian building block: defines the quantum interaction of the smoke diffusion process, which is modeled as:

[0025] H = −

[0026] Where: coupling coefficient ;

[0027] τ is a thermodynamic time parameter that controls the rate at which the coupling strength decays with pixel differences. The larger τ is, the lower the difference sensitivity is.

[0028] μ is the smoke diffusion coefficient, which is linearly related to the diffusion speed of smoke. It simulates the lateral diffusion behavior of smoke particles at adjacent spatial locations.

[0029] The term describes the spin exchange interaction, which is analogous to momentum transfer between smoke particles or diffusion driven by concentration gradients;

[0030] Dynamic calculation based on external data such as temperature field and wind speed field;

[0031] Tensor Network Evolution Module: Real-time evolution of quantum states through matrix product states (MPS). The contraction algorithm is:

[0032] ⟨ψ(t)∣H∣ψ(t)>=Tr

[0033] in: is the local tensor of the kth grid point;

[0034] Optimize the quantum gate parameter θ by gradient descent and embed the quantum gate parameter θ into the evolution operator In the process, we construct a differentiable computational graph of the objective function with respect to θ; we use automatic differentiation techniques to calculate the gradient ∇θ ⟨H > and update the parameters through the optimizer; the physical indicators The ground state of the corresponding qubits, the virtual indices connect adjacent tensors to describe the entanglement structure, the global Hamiltonian H is decomposed into a sum of short-range interaction terms, and the time evolution operator The Hamiltonian H = ∑kHk is decomposed as:

[0035] ‌;

[0036] Chaos detection module: calculate the maximum Lyapunov exponent, the formula is:

[0037] λ max =

[0038] Where: ψ(t) is the system state vector;

[0039] ∂ψ(t) / ∂ψ(0) is the evolution matrix of the initial state perturbation over time;

[0040] ∥⋅∥ is the norm operator;

[0041] When λ max > 0, the system exhibits chaotic characteristics, and small initial differences lead to unpredictable long-term behavior, when λmax > 0.34, the smoke diffusion exhibits enhanced disordered motion and dramatic fluctuations in concentration distribution, and is determined to be a critical state of smoke diffusion; The Lyapunov exponent quantifies the average rate of exponential divergence of adjacent trajectories in phase space; accordingly, we can monitor the λ max of smoke diffusion in real time and provide early warning of smoke to fire mutation.

[0042] Further, the multi-camera fusion of the smoke analysis AI algorithm based on quantum field theory tensor network is specifically:

[0043] Single camera quantum state: map each frame of image pixel matrix to quantum state amplitude; let the single frame image matrix be M ∈ , after normalization, it is expanded to one-dimensional quantum , encoded as quantum state:

[0044]

[0045] Where: |k> is the computational ground state, vk satisfies the normalization condition ∑ =1;

[0046] Multi-camera fusion: four-way monitoring video is spliced into a joint quantum state, which is mathematically expressed as:

[0047] |ψtotal> = |ψ1> ⊗ |ψ2> ⊗ |ψ3> ⊗ |ψ4>

[0048] Wherein: | ψ1 >, | ψ2 >, | ψ3 >, | ψ4 > represent the quantum state of the video information captured by the four different cameras respectively;

[0049] | ψtotal > represents the joint or tensor product of the four quantum states;

[0050] According to the actual specific environment of the substation site, the four-way monitoring video is customized spliced, the size and position of the video picture are adjusted, and more accurate splicing effect is realized. The video information from multiple cameras is integrated into a single and unified view by using multi-camera fusion technology.

[0051] Further, the quantum state module of the smoke analysis AI algorithm based on quantum field theory tensor network encodes the diffusion direction, concentration gradient and other characteristic information of the smoke into quantum states by mapping classical information to quantum state space, and uses the continuity and superposition of quantum states to retain the continuous dynamic characteristics of the smoke: the specific process is:

[0052] Quantum state representation of smoke features: suppose the concentration feature of the smoke can be represented by a continuous function c(r, t), where r is the spatial position vector and t is the time; map this function to a quantum state; consider a simple quantum bit system, whose state can be represented by a two-dimensional complex vector:

[0053] | ψ > = α | 0 > + β | 1 >

[0054] Where: α and β are complex numbers, satisfying ;

[0055] | 0 > and | 1 > are two ground states;

[0056] To represent the continuous characteristics of the smoke, a quantum register is constructed using multiple quantum bits. Suppose N quantum bits are used, and the state of the quantum register is represented as:

[0057]

[0058] Where: is a complex coefficient;

[0059] | i > is the computational basis;

[0060] Discretize the concentration function c(r, t) of the smoke to obtain a sequence of discrete concentration values { }, and then encode these discrete concentration values into the coefficients of the quantum state;

[0061] Quantum operation preserving dynamic characteristics of smoke: in order to preserve dynamic characteristics of smoke such as diffusion direction and concentration gradient, quantum gate operations are used to simulate dynamic change process of smoke; considering diffusion process of smoke, a quantum evolution operator U(t) is used to describe change of quantum state of smoke with time, i.e.

[0062] |Ψ(t)> = U(t) |Ψ(0)>

[0063] Wherein |Ψ(0)> is initial quantum state of smoke;

[0064] |Ψ(t)> is quantum state at t time;

[0065] Quantum evolution operator U(t) is constructed according to physical model of smoke; diffusion of smoke is described by a diffusion equation; diffusion equation is converted into a sequence of quantum gate operations; diffusion process of smoke is simulated by continuously applying quantum gate operations to quantum state; for concentration gradient of smoke, it is obtained by measuring certain observable of quantum state; a quantum operator O related to concentration gradient is defined; a quantum system whose state space is related to space of smoke distribution is provided; smoke concentration distribution is discretized into a grid, each grid point corresponds to a quantum state basis vector; in three-dimensional space, space is divided into Nx×Ny×Nz small cubes, each small cube corresponds to a basis vector |ri>, wherein ri=( xi, yi, zi ) is center position of small cube; smoke concentration can be regarded as a classical field C(r,t); in quantum mechanics, concentration is represented by a diagonal operator C^, i.e. C^ |ri> = C(ri,t) |ri>; concentration gradient is described by using finite difference approximation; concentration gradient in x direction is represented as:

[0066] (∂x∂C)i≈ΔxC(ri+Δx, t )−C(ri, t )

[0067] Wherein Δx is interval of adjacent grid points in x direction;

[0068] Based on this approximation, a quantum operator O^x is defined to represent smoke concentration gradient in x direction, i.e.

[0069] O^x=i∑Δx∣ri+Δx>⟨ri+Δx∣−∣ri>⟨ri∣⊗∣x>⟨x∣

[0070] Here |x> is a quantum state of a mark direction, which is used to distinguish gradients in different directions; similarly, O^y and O^z are defined; more generally, a vector operator O^=(O^x,O^y,O^z) is defined, and its expectation value <O^>=(<O^x>,<O^y>,<O^z>) describes average value of smoke concentration gradient in each direction;

[0071] The expected value of the quantum state |Ψ(t)> under the operator O is measured, that is:

[0072] ⟨O >=⟨Ψ(t)∣O∣Ψ(t)>

[0073] Let O be a vector operator O^, then:

[0074] ⟨O^>=(⟨Ψ(t)∣O^x∣Ψ(t)>,⟨Ψ(t)∣O^y∣Ψ(t)>,⟨Ψ(t)∣O^z∣Ψ(t)>)

[0075] This expected value can reflect the information of the smoke concentration gradient, and each component of ⟨O^> represents the average rate of change of smoke concentration in the corresponding direction.

[0076] Further, the specific operation process of the dynamic Hamiltonian construction module of the smoke analysis AI algorithm based on the quantum field theory tensor network is:

[0077] 1) Specification field construction: define SU(2) specification field to describe the interaction of smoke particles; that is:

[0078]

[0079] Among them: is the specification potential;

[0080] is the Pauli matrix;

[0081] 2) Covariant derivative: construct the coupling term of the material field, that is, the smoke particle and the specification field; and

[0082]

[0083] Among them: g=0.4 is the coupling constant;

[0084] Ψ is a two-component spinor field;

[0085] 3) Dynamic Hamiltonian construction, that is:

[0086]

[0087] It contains kinetic energy term, specification field intensity term and interaction potential V(ρ)=λ(ρ−ρ0)^4.

[0088] Further, the specific process of updating the dynamic Hamiltonian of the dynamic Hamiltonian construction module of the smoke analysis AI algorithm based on the quantum field theory tensor network is:

[0089] Dynamic model: model the smoke diffusion process as an evolution process in quantum field theory, and its dynamics is determined by the Hamiltonian Control, i.e.:

[0090] =−μ(t) + other interaction terms

[0091] where: and are field operators;

[0092] μ(t) is the diffusion coefficient, coupled with the real-time parameters of the ventilation system, wind speed, temperature;

[0093] Adjust the μ parameter in real time according to the ventilation system data, i.e.:

[0094]

[0095] The "wind speed" in the formula is converted into a dimensionless parameter by a normalization factor, i.e. 10 m / s, which is used to represent the coupling strength of kinetic energy and potential energy in the ventilation system; when the actual wind speed approaches 10 m / s, the adjustment range of μ reaches the maximum design value, μ = 0.7, at this time the potential energy term of the system is significantly enhanced, and the energy dissipation caused by turbulence is suppressed; the formula is a linear function, μ changes positively with wind speed, the coefficient is 0.07, which is suitable for the rapid response requirement under small range wind speed fluctuation; the constant term (-0.5) is used to modify the baseline coupling strength, to ensure that μ is in a reasonable interval under normal working conditions, such as wind speed = 5 m / s, μ = 0.2, to avoid over-damping of the system; Specifically: read the real-time wind speed data from the ventilation system, the sampling frequency is 10 Hz; calculate the current μ(t) and update the Hamiltonian, when the wind speed increases, μ(t) increases, indicating that the smoke diffusion rate accelerates; approximate solution of the evolution equation by quantum tensor network:

[0096]

[0097] Dynamic adjustment of the simulation trajectory of smoke diffusion improves the adaptability of the model to the real-time environment of the substation site.

[0098] Further, the tensor network evolution module of the smoke analysis AI algorithm based on quantum field theory tensor network contains a quantum-classical hybrid optimizer, which uses the alternating direction multiplier method to jointly optimize the tensor rank r and the Hamiltonian parameter , which satisfies the constraint condition:

[0099]

[0100] where: is the Hamiltonian matrix constructed by the coupling parameter ;

[0101] is the experimental observation value of the target Hamiltonian for local tensors in the tensor network;

[0102] denotes its rank constraint;

[0103] is a rank penalty coefficient for balancing accuracy and complexity.

[0104] Further, the quantum state encoding module of the smoke analysis AI algorithm based on the quantum field theory tensor network adopts a non-uniform quantization grid to map the RGB color domain to a Bloch sphere, and the formula for calculating the polar angle and the formula for calculating the azimuth angle are:

[0105] The polar angle calculation formula is:

[0106]

[0107] Quantized hue projection intensity on the blue-yellow axis, where the numerator highlights the contribution of the blue channel, and the denominator normalizes the brightness, through The dynamic range constraint ensures that the input value is within the range of [-1, 1], avoiding errors outside the domain of the inverse cosine function;

[0108] The azimuth angle calculation formula is:

[0109]

[0110] By capturing the red-green contrast through Gi−Ri, Balancing the modulation effect of the blue channel on the hue ring, using four-quadrant arctangent 2Eliminate angle jumps to ensure smooth transition of adjacent color blocks; accordingly, through the coordinated design of the polar angle and the azimuth angle, the RGB three-dimensional color domain information is completely preserved to the two-dimensional Bloch sphere, and the encoding parameters are adjusted in real time in combination with the quantum state fidelity index to adapt to complex and variable lighting environments.

[0111] Further, the operation process of the chaos detection module of the smoke analysis AI algorithm based on the quantum field theory tensor network is as follows:

[0112] 1) Phase space reconstruction: extract the quantum state probability distribution P(σ) from the MPS to construct a delay coordinate matrix:

[0113]

[0114] Where: the embedding dimension m=5;

[0115] The delay time τ=3 frames;

[0116] 2) Nearest neighbor search: for each phase point Xi, find the nearest neighbor Xj, and calculate the divergence speed:

[0117] dj(t) = ||Xj(t) - Xj(t)||2

[0118] 3) Exponential fitting: fitting by least squares:

[0119] λ max = .

[0120] The benefits brought by the present application are:

[0121] 1. The smoke analysis AI algorithm based on quantum field theory tensor network has high-dimensional data processing capability. Tensor network is good at processing multi-dimensional data, such as multi-spectral characteristics of smoke, spatio-temporal distribution information, and can realize efficient feature extraction through tensor decomposition technology to reduce computational complexity. In addition, quantum computing acceleration is adopted, combined with the mathematical framework of quantum field theory, which can take advantage of the parallel computing characteristics of quantum superposition state to significantly improve the speed of large-scale smoke simulation and real-time analysis and recognition.

[0122] 2. The smoke analysis AI algorithm based on quantum field theory tensor network has anti-interference ability and strong applicability in complex scenes. Quantum sensing technology can improve the detection sensitivity of weak smoke signals and reduce environmental noise interference through entangled state measurement principle. The topological structure of tensor network can model the dynamic field theory model of smoke diffusion flexibly, and is suitable for different environments such as transformer substations, forests, and complex air flow conditions in cities.

[0123] 3. The smoke analysis AI algorithm based on quantum field theory tensor network has the characteristics of model lightweight and low energy consumption. The algorithm uses chain tensor network technology to compress model parameters, reducing the demand for storage and computing resources, and is suitable for deployment on edge devices or unmanned aerial vehicle platforms. Quantum field theory simulation optimizes the algorithm design driven by physical laws, which can reduce energy consumption at the same accuracy compared with traditional deep learning models.

[0124] 4. The smoke analysis AI algorithm based on quantum field theory tensor network has good potential application extension, such as disaster early warning linkage and environmental monitoring, etc. Combined with the high-precision time synchronization capability of quantum time device, it can be linked with meteorological monitoring and fire fighting system to improve the response speed of early fire warning. Through multi-dimensional data fusion analysis, it supports pollutant tracing, air quality prediction and other derivative functions. BRIEF DESCRIPTION OF DRAWINGS

[0125] Figure 1 The figure is a schematic diagram of the algorithm analysis process of the present application;

[0126] Figure 2 The figure is a schematic diagram of the early warning application deployment architecture of the present application;

[0127] Figure 3 The figure is a schematic diagram of quantum computing simulation based on tensor network structure representation of the present application;

[0128] Figure 4 Mutation curve of Lyapunov exponent spectrum of the application in the smoke formation period;

[0129] Figure 5 Superimposed contrast diagram of thermodynamic entropy change diagram and visible light image of the application;

[0130] Figure 6 Quantum model Python part code block screenshot diagram of the application. DETAILED DESCRIPTION

[0131] The application will be further described in detail below in combination with the drawings and specific embodiments:

[0132] As shown in the figure, the figure is a smoke analysis AI system based on quantum field theory tensor network, and the smoke analysis AI algorithm based on quantum field theory tensor network is provided with a multi-camera fusion module, a quantum state encoding module, a dynamic Hamiltonian construction module, a tensor network evolution module and a chaos detection module. Specifically: Figures 1-6 Multi-camera fusion: splice multiple monitoring videos into a joint quantum state; taking four monitoring videos as an example, the specific steps are as follows:

[0133] Single camera quantum state: map each frame of image pixel matrix to quantum state amplitude; let the single frame image matrix be M∈

[0134] , after normalization, it is expanded into a one-dimensional quantum , and encoded into a quantum state:

[0135]

[0136] Where: |k> is the computational basis, vk satisfies the normalization condition ∑ =1;

[0137] Multi-camera fusion: splice four monitoring videos into a joint quantum state, which can be expressed as:

[0138] |ψtotal> = |ψ1> ⊗ |ψ2> ⊗ |ψ3> ⊗ |ψ4>

[0139] Where: |ψ1>, |ψ2>, |ψ3>, |ψ4> represent the quantum states of the video information captured by the four different cameras respectively;

[0140] |ψtotal> represents the joint or tensor product of the four quantum states;

[0141] ​According to the actual specific environment of the substation site, four-way monitoring video is self-defined splicing, the size and position of the video picture are adjusted, and more accurate splicing effect is realized. The video information from multiple cameras is integrated into a single and unified view by using multi-camera fusion technology. In the context of quantum physics, this fusion can be analogized as the superposition of quantum states. The joint state can capture the spatial correlation between multiple cameras and enhance the global representation ability of the smoke pattern. The tensor product operation naturally supports multi-modal data fusion and avoids geometric distortion in traditional splicing methods. Quantum state encoding preserves the continuous dynamic characteristics of smoke, such as diffusion direction and concentration gradient.

[0142] Quantum state encoding module: map the image pixel array to a quantum spin state, and the state vector is:

[0143] ∣ψ >=i=1⨂N ( ∣0 >+ >)

[0144] Where: is the normalized brightness;

[0145] =arctan(Gi / Bi) is the hue phase angle;

[0146] The specific process of mapping the image pixel array to a quantum spin state is as follows:

[0147] 1) Image preprocessing and parameter extraction;

[0148] Pixel vectorization: the pixel matrix of the input image is unfolded into a one-dimensional vector by column, and the RGB value of each pixel is mapped to a three-dimensional vector (Ri, Gi, Bi);

[0149] Normalization brightness calculation: map the pixel brightness to the quantum probability amplitude range through linear transformation, define , so that αi∈[0,1];

[0150] Satisfy the normalization condition of quantum state: ;

[0151] Hue phase angle generation: based on color space mapping, encode color information into phase angle through =arctan(Gi / Bi), which reflects the position of the pixel on the color wheel;

[0152] 2) Quantum spin state construction;

[0153] Single-pixel quantum state definition: each pixel corresponds to a quantum bit, and the state vector is:

[0154] ∣ψ >=i=1⨂N ( ∣0 >+ )

[0155] where |0> and |1> are the basis states;

[0156] to control the brightness;

[0157] to encode the hue;

[0158] Quantum superposition and entanglement: through the parameters to realize the probability superposition of brightness, introducing quantum interference effect, so that the phase correlation between adjacent pixels reflects the continuity of color;

[0159] The quantum state module encodes the feature information such as the diffusion direction and concentration gradient of the smoke into quantum states by mapping the classical information to the quantum state space, and uses the continuity and superposition of quantum states to preserve the continuous dynamic characteristics of the smoke: the specific process is as follows:

[0160] Quantum state representation of smoke features: suppose the concentration feature of the smoke can be represented by a continuous function c(r, t), where r is the spatial position vector and t is the time; map this function to a quantum state; consider a simple quantum bit system, whose state can be represented by a two-dimensional complex vector:

[0161] |ψ> = α|0> + β|1>

[0162] where α and β are complex numbers satisfying ;

[0163] |0> and |1> are the two basis states;

[0164] To represent the continuous characteristics of the smoke, a quantum register is constructed using multiple quantum bits, and suppose N quantum bits are used, then the state of the quantum register is represented as:

[0165]

[0166] where: is a complex coefficient;

[0167] | i > is the computational basis state;

[0168] Discretize the concentration function c(r, t) of the smoke to obtain a sequence of discrete concentration values { }, and then encode these discrete concentration values into the coefficients of the quantum state;

[0169] Quantum operation preserving smoke dynamic characteristics: in order to preserve the dynamic characteristics such as the diffusion direction and concentration gradient of smoke, quantum gate operations are used to simulate the dynamic change process of smoke; considering the diffusion process of smoke, a quantum evolution operator U(t) is used to describe the change of the quantum state of smoke with time, that is:

[0170] |Ψ(t)> = U(t) |Ψ(0)>

[0171] Wherein: |Ψ(0)> is the initial quantum state of smoke;

[0172] |Ψ(t)> is the quantum state at time t;

[0173] The quantum evolution operator U(t) is constructed according to the physical model of smoke. The diffusion of smoke is described by a diffusion equation. The diffusion equation is converted into a sequence of quantum gate operations. By continuously applying these quantum gate operations to the quantum state, the diffusion process of smoke is simulated. For the concentration gradient of smoke, some observables of the quantum state are measured to obtain it. A quantum operator O related to the concentration gradient is defined. A quantum system whose state space is related to the space of smoke distribution is set. The smoke concentration distribution is discretized into a grid, and each grid point corresponds to a quantum state basis vector. In three-dimensional space, the space is divided into Nx×Ny×Nz small cubes, and each small cube corresponds to a basis vector |ri>, where ri=( xi, yi, zi ) is the center position of the small cube. The smoke concentration can be regarded as a classical field C(r,t). In quantum mechanics, the concentration is represented by a diagonal operator C^, that is, C^ |ri> = C(ri,t) |ri>. The concentration gradient is described by using finite difference approximation. The concentration gradient in the x direction is expressed as:

[0174] (∂x∂C)i≈ΔxC(ri+Δx, t )−C(ri, t )

[0175] Wherein: Δx is the interval of adjacent grid points in the x direction;

[0176] Based on this approximation, a quantum operator O^x is defined to represent the smoke concentration gradient in the x direction, that is:

[0177] O^x=i∑Δx∣ri+Δx>⟨ri+Δx∣−∣ri>⟨ri∣⊗∣x>⟨x∣

[0178] Here |x> is a quantum state that marks the direction, which is used to distinguish the gradient in different directions. Similarly, O^y and O^z are defined. More generally, we define a vector operator O^=(O^x,O^y,O^z), and its expectation value ⟨O^>=(⟨O^x>,⟨O^y>,⟨O^z>) describes the average value of the smoke concentration gradient in each direction;

[0179] Measure the expectation value of the quantum state |Ψ(t)> under the operator O, that is:

[0180] ⟨O >=⟨Ψ(t)|O|Ψ(t)>

[0181] Let O be a vector operator O^, then:

[0182] ⟨O^>=(⟨Ψ(t)∣O^x∣Ψ(t)>,⟨Ψ(t)∣O^y∣Ψ(t)>,⟨Ψ(t)∣O^z∣Ψ(t)>)

[0183] This expected value can reflect the information of the smoke concentration gradient. Each component of ⟨O^> represents the average rate of change of smoke concentration in the corresponding direction, thus helping us analyze the diffusion trend of smoke.

[0184] Dynamic Hamiltonian building block: defines the quantum interaction of the smoke diffusion process, which is modeled as:

[0185] H = −

[0186] Where: coupling coefficient ;

[0187] τ is a thermodynamic time parameter that controls how quickly the coupling strength decays with pixel differences; the larger τ is, the lower the difference sensitivity is);

[0188] μ is the smoke diffusion coefficient, which is linearly related to the diffusion speed of smoke. It simulates the lateral diffusion behavior of smoke particles at adjacent spatial locations.

[0189] The term describes the spin exchange interaction, which is analogous to momentum transfer between smoke particles or diffusion driven by concentration gradients;

[0190] It is a dynamic calculation based on external data such as temperature field and wind speed field;

[0191] The specific operation process of the dynamic Hamiltonian construction module is as follows:

[0192] 1) Gauge field construction: Define the SU(2) gauge field to describe the interaction between smoke particles; that is:

[0193]

[0194] in: For the normative potential;

[0195] is the Pauli matrix;

[0196] 1.2) Covariant derivatives: construct the coupling term between the matter field, i.e., smoke particles, and the gauge field; and

[0197]

[0198] where: g=0.4 is the coupling constant;

[0199] Ψ is a two-component spinor field;

[0200] 2.3) Dynamic Hamiltonian construction, i.e.:

[0201]

[0202] which contains kinetic term, gauge field strength term and interaction potential V(ρ)=λ(ρ−ρ0)^4;

[0203] 3. The dynamic Hamiltonian update process of the dynamic Hamiltonian construction module is as follows:

[0204] Dynamic model: model the smoke diffusion process as an evolution process in quantum field theory, whose dynamics is controlled by the Hamiltonian , i.e.:

[0205] =−μ(t) + other interaction terms

[0206] where: and are field operators;

[0207] μ(t) is the diffusion coefficient, which is coupled with the real-time parameters of the ventilation system, such as wind speed and temperature;

[0208] According to the real-time data of the ventilation system, the μ parameter is adjusted, i.e.:

[0209]

[0210] In the formula, the "wind speed" is converted into a dimensionless parameter through a normalization factor, i.e. 10 m / s, which is used to represent the coupling strength of kinetic energy and potential energy in the ventilation system; when the actual wind speed approaches 10 m / s, the adjustment range of μ reaches the maximum design value, μ=0.7, at this time the potential energy term of the system is significantly enhanced, which suppresses the energy dissipation caused by turbulence; the formula is a linear function, μ changes positively with wind speed, the coefficient is 0.07, which is suitable for the rapid response requirement under small range wind speed fluctuation; the constant term (-0.5) is used to modify the baseline coupling strength, to ensure that μ is in a reasonable range under normal working conditions, such as wind speed=5 m / s, μ=0.2, to avoid over-damping of the system; Specifically: read the real-time wind speed data from the ventilation system, the sampling frequency is 10 Hz; calculate the current μ(t) and update the Hamiltonian, when the wind speed increases, μ(t) increases, indicating that the smoke diffusion rate accelerates; the evolution equation is approximately solved by quantum tensor network:

[0211]

[0212] Dynamic adjustment of the simulated trajectory of smoke diffusion, improving the adaptability of the model to the real-time environment of the substation site;

[0213] Tensor network evolution module: real-time evolution of quantum state through matrix product state-MPS, whose contraction algorithm is:

[0214] ⟨ψ(t)∣H∙∣ψ(t)>=Tr

[0215] Where: Tk is the local tensor of the kth lattice point;

[0216] Optimize quantum gate parameters θ by gradient descent, embed quantum gate parameters θ into evolution operator , construct the differentiable computational graph of the objective function for θ; use automatic differentiation technology to calculate the gradient ∇θ ⟨H >, and update the parameters through the optimizer; physical indicators correspond to the ground state of the quantum bit, the virtual index connects adjacent tensors to describe the entanglement structure, and the global Hamiltonian H is decomposed into the sum of short-range interaction terms, and the time evolution operator is approximated as the product of local gate operations through Trotter-Suzuki decomposition; the decomposition form of the Hamiltonian H = ∑kHk is:

[0217] ‌;

[0218] Chaos detection module: calculate the maximum Lyapunov exponent, the formula is:

[0219] λ max =

[0220] Where: ψ(t) is the system state vector;

[0221] ∂ψ(t) / ∂ψ(0) is the evolution matrix of the initial state perturbation over time;

[0222] ∥⋅∥ is the norm operator;

[0223] The specific process of quantum Lyapunov exponent calculation is:

[0224] 1) Phase space reconstruction: extract the quantum state probability distribution P(σ) from MPS, and construct the delay coordinate matrix:

[0225]

[0226] Where: the embedding dimension m = 5;

[0227] Delay time τ = 3 frames;

[0228] 2) Nearest neighbor search: For each phase point Xi, find the nearest neighbor Xj, and calculate the divergence velocity:

[0229] dj(t) = ||Xj(t) - Xi(t)||2

[0230] 3) Exponential fitting: Fit by least squares:

[0231] λ max =

[0232] When λ max > 0, the system exhibits chaotic behavior, and small initial differences lead to unpredictable long-term behavior. When λ max > 0.34, the smoke diffusion exhibits enhanced disordered motion and dramatic fluctuations in concentration distribution, indicating a critical state of smoke diffusion. The Lyapunov exponent quantifies the average rate of exponential divergence of adjacent trajectories in phase space. Based on this, we can monitor λ max of smoke diffusion in real time and provide early warning of the mutation of smoke to fire;

[0233] The tensor network evolution module of the quantum field theory-based tensor network smoke analysis AI algorithm shown comprises a quantum-classical hybrid optimizer, which uses the alternating direction multiplier method to jointly optimize the tensor rank r and the Hamiltonian parameter , satisfying the constraint condition:

[0234]

[0235] where: is the Hamiltonian matrix constructed by the coupling parameter ;

[0236] is the experimental observation value of the target Hamiltonian is the local tensor in the tensor network;

[0237] , which represents its rank constraint;

[0238] is the rank penalty coefficient, used to balance accuracy and complexity.

[0239] The quantum state encoding module of the quantum field theory-based tensor network smoke analysis AI algorithm shown uses a non-uniform quantization grid to map the RGB color domain to the Bloch sphere, and the formula for calculating the polar angle and the formula for calculating the azimuth angle are:

[0240] Polar angle calculation formula:

[0241]

[0242] Quantize the projection intensity of hue on the blue-yellow axis, where the numerator highlights the contribution of the blue channel, and the denominator normalizes the luminance, through Ensure the dynamic range constraint that the input value is in [−1, 1], to avoid the error outside the domain of the inverse cosine function;

[0243] Azimuth angle calculation formula:

[0244]

[0245] Capture the red-green contrast through Gi−Ri, Balance the modulation effect of the blue channel on the hue ring, and use the four-quadrant inverse tangent 2Eliminate angle jumps and ensure smooth transitions between adjacent color blocks; accordingly, through the coordinated design of polar angle and azimuth angle, the RGB three-dimensional color gamut information is completely preserved to the two-dimensional Bloch sphere, and the quantum state fidelity index is used to real-time control the encoding parameters, and adapt to complex and variable lighting environment scenes.

[0246] The real-time monitoring of the Lyapunov exponent λmax of smoke diffusion is shown, which can early warn the mutation of smoke to fire, and a quantum-classical hybrid early warning decision engine is constructed to realize real-time monitoring and early warning system; Specifically:

[0247] 1) Four threshold early warning mechanisms: set the warning threshold λwarn = 0.35, the general threshold λgeneral = 0.37, the serious threshold λserious = 0.39, and the critical threshold λdanger = 0.41; When λmax>0.34, it is determined that the smoke diffusion is in a critical state point, triggering multi-level early warning-general / serious / danger;

[0248] 2) Bayesian update: combine historical data to correct the prediction probability, that is:

[0249]

[0250] Where: P0(fire)= 0.01 is the prior probability.

[0251] The AI algorithm edge computing analysis deployment shown: the server uses a domestic Huawei Taishan 2280 server, installs the EulerOS operating system independently developed by Huawei, and realizes tensor slicing calculation on the operating system, with a delay of less than 200ms;

[0252] Input: 4K video stream (3840×2160, 30fps);

[0253] Tensor slicing: Extract sub-regions from high-dimensional tensors, spatio-temporal segments in video streams, local regions in images, extract 100 candidate region slices per frame for target detection, slice generation by Huawei Taishan 2280 server multi-core parallel tensor slicing calculation, decompose high-dimensional joint quantum state | ψtotal> into local tensor slices, use multi-core CPU parallel computation to shrink the path;

[0254] Delay decomposition: data loading, 20ms (NVMe SSD→memory); slice calculation: 150ms (64-core parallel);

[0255] Result returned to the front-end monitoring client: 10ms (RDMA transmission), total frame processing time total delay is 180ms, lower than 200ms.

[0256] The above description is only a preferred embodiment of the present application, not any other form of limitation on the present application, and any modification or equivalent change made according to the technical essence of the present application still belongs to the scope of the present application.

Claims

1. A smoke analysis AI system based on quantum field theory tensor networks, characterized by: The collaborative system of the smoke analysis AI system based on quantum field theory tensor network is equipped with multi-camera fusion, quantum state encoding module, dynamic Hamiltonian construction module, tensor network evolution module and chaos detection module; specifically: Multi-camera fusion: stitching multiple surveillance videos into a joint quantum state; Quantum State Encoding Module: The quantum state module encodes characteristic information such as the diffusion direction and concentration gradient of smoke into quantum states by mapping classical information into quantum state space. It utilizes the continuity and superposition of quantum states to preserve the continuous dynamic characteristics of smoke. Map the image pixel array into a quantum spin state, whose state vector is: ∣ψ >=i=1⨂N ( ∣0 >+ >); in: is the normalized brightness; =arctan(Gi / Bi) is the hue phase angle; The specific process of mapping the image pixel array into quantum spin states is as follows: 1) Image preprocessing and parameter extraction; Pixel vectorization: The pixel matrix of the input image is expanded into a one-dimensional vector by column, and the RGB value of each pixel is mapped into a three-dimensional vector, namely: Ri, Gi, Bi; ‌Normalized brightness calculation‌: Map pixel brightness to the quantum probability amplitude range through linear transformation, defining , so that αi∈[0,1]; Satisfy the normalization conditions of the quantum state: ; ‌Hue phase angle generation‌: Based on chromaticity space mapping, through =arctan(Gi / Bi) encodes color information as a phase angle, reflecting the position of the pixel on the color wheel; 2) Quantum spin state construction; Single pixel quantum state definition: Each pixel corresponds to a quantum bit, and its state vector is: ∣ψ >=i=1⨂N ( ∣0 >+ >); Among them: |0> and |1> are ground states; To control brightness; is the encoding hue‌; Quantum superposition and entanglement: through parameters Realize the probabilistic superposition of brightness, Introducing quantum interference effects so that the phase correlation between adjacent pixels reflects color continuity; Dynamic Hamiltonian building block: defines the quantum interaction of the smoke diffusion process, which is modeled as: H = − ； Where: coupling coefficient ; τ is a thermodynamic time parameter that controls the rate at which the coupling strength decays with pixel differences. The larger τ is, the lower the difference sensitivity is. μ is the smoke diffusion coefficient, which is linearly related to the diffusion speed of smoke. It simulates the lateral diffusion behavior of smoke particles at adjacent spatial locations. The term describes the spin exchange interaction, which is analogous to momentum transfer between smoke particles or diffusion driven by concentration gradients; Dynamic calculation based on external data such as temperature field and wind speed field; Tensor Network Evolution Module: Real-time evolution of quantum states through matrix product states (MPS). The contraction algorithm is: ⟨ψ(t)∣H∣ψ(t)>=Tr ; in: is the local tensor of the kth grid point; Optimize the quantum gate parameter θ by gradient descent and embed the quantum gate parameter θ into the evolution operator In the process, we construct a differentiable computational graph of the objective function with respect to θ; we use automatic differentiation techniques to calculate the gradient ∇θ ⟨H > and update the parameters through the optimizer; the physical indicators Corresponding to the ground state of the quantum bit, the virtual index connects the adjacent tensors to describe the entanglement structure, and the global Hamiltonian H is decomposed into the sum of short-range interaction terms. The time evolution operator is decomposed into It is approximately the product of local department operations; the Hamiltonian H = ∑kHk decomposition form is: ‌; Chaos detection module: Calculates the maximum Lyapunov exponent, the formula is: λmax = ; Where: ψ(t) is the system state vector; ∂ψ(t) / ∂ψ(0) is the time evolution matrix of the initial state perturbation; ∥⋅∥ is the norm operator; When λ max > 0, the system exhibits chaotic characteristics, with small initial differences leading to unpredictable long-term behavior. When λ max > 0.34, smoke diffusion exhibits increased disordered motion and violent fluctuations in concentration distribution, indicating a critical state of smoke diffusion. The Lyapunov exponent quantifies the average rate of exponential divergence of adjacent trajectories in phase space. Based on this, we can monitor the λ max of smoke diffusion in real time and provide early warning of sudden changes from smoke to fire.

2. The smoke analysis AI system based on quantum field theory tensor network according to claim 1 is characterized by: The multi-camera fusion of the smoke analysis AI algorithm based on quantum field theory tensor network is specifically as follows: Single camera quantum state: Map the pixel matrix of each frame image to the quantum state amplitude; let the single frame image matrix be M ∈ , after normalization, it is expanded into a one-dimensional quantity , encoded as a quantum state: ； Where: |k> is the calculation basis state, vk satisfies the normalization condition ∑ =1; Multi-camera fusion: The four-channel surveillance video is spliced ​​into a joint quantum state, which is expressed mathematically as follows: ∣ψtotal>=∣ψ1>⊗∣ψ2>⊗∣ψ3>⊗∣ψ4>; Among them: ∣ψ1>, ∣ψ2>, ∣ψ3>, ∣ψ4> represent the quantum states of the video information captured by four different cameras; ∣ψtotal> represents the union or tensor product of these four quantum states; Customize the stitching of four surveillance videos based on the actual substation environment, adjust the size and position of the video images to achieve a more precise stitching effect, and use multi-camera fusion technology to integrate video information from multiple cameras into a single, unified view.

3. The smoke analysis AI system based on quantum field theory tensor network according to claim 1 is characterized by: The quantum state module of the smoke analysis AI algorithm based on quantum field theory tensor network has the following specific process: Quantum state representation of smoke characteristics: Assume that the concentration characteristics of smoke are represented by a continuous function c(r,t), where r is the spatial position vector and t is time. Map this function to a quantum state. Consider a simple quantum bit system whose state can be represented by a two-dimensional complex vector: ∣ψ>=α∣0>+β∣1> Among them: α and β are complex numbers, satisfying ; |0> and |1> are two ground states; To represent the continuous characteristics of smoke, multiple quantum bits are used to form a quantum register. Assuming that N quantum bits are used, the state of the quantum register is expressed as: ； in: is a complex coefficient; ∣i> is the calculation ground state; Discretize the smoke concentration function c(r, t) to obtain a discrete concentration value sequence { }, and then encode these discretized concentration values ​​into the coefficients of the quantum state middle; Quantum operations that preserve the dynamic characteristics of smoke: To preserve the dynamic characteristics of smoke, such as diffusion direction and concentration gradient, quantum gate operations are used to simulate the dynamic changes of smoke. Considering the diffusion process of smoke, a quantum evolution operator U(t) is used to describe the change of the smoke quantum state over time, namely: |Ψ(t)>=U(t)|Ψ(0)>; Where: |Ψ(0)> is the initial quantum state of the smoke; ∣Ψ(t)> is the quantum state at time t; The quantum evolution operator U(t) is constructed based on the physical model of smoke. The diffusion of smoke is described by a diffusion equation, which is converted into a sequence of quantum gate operations. These quantum gate operations are continuously applied to the quantum state to simulate the diffusion process of smoke. The concentration gradient of smoke is obtained by measuring certain observables of the quantum state. A quantum operator O related to the concentration gradient is defined. Assume a quantum system whose state space is related to the spatial distribution of smoke. The smoke concentration distribution is discretized into a grid, with each grid point corresponding to a quantum state basis vector. In three-dimensional space, the space is divided into Nx×Ny×Nz small cubes, each of which corresponds to a basis vector |ri>, where ri = (xi, yi, zi) is the center position of the cube. The smoke concentration can be regarded as a classical field C(r,t). In quantum mechanics, the concentration is represented by a diagonal operator C^, that is, C^|ri>=C(ri,t)|ri>. The concentration gradient is described using the finite difference approximation. The concentration gradient in the x direction is expressed as: (∂x∂C)i≈ΔxC(ri+Δx , t )−C(ri , t ); Where: Δx is the distance between adjacent grid points in the x direction; Based on this approximation, a quantum operator O^x is defined to represent the smoke concentration gradient in the x direction, namely: O^x=i∑Δx∣ri+Δx>⟨ri+Δx∣−∣ri>⟨ri∣⊗∣x>⟨x∣; Here, |x> is a quantum state that marks the direction and is used to distinguish gradients in different directions. Similarly, O^y and O^z are defined. More generally, a vector operator O^=(O^x,O^y,O^z) is defined, whose expected value ⟨O^>=(⟨O^x>,⟨O^y>,⟨O^z>) describes the average value of the smoke concentration gradient in each direction. Measure the expectation value of the quantum state |Ψ(t)> under the operator O, that is: ⟨O >=⟨Ψ(t)∣O∣Ψ(t)>; Let O be a vector operator O^, then: ⟨O^>=(⟨Ψ(t)∣O^x∣Ψ(t)>,⟨Ψ(t)∣O^y∣Ψ(t)>,⟨Ψ(t)∣O^z∣Ψ(t)>); This expected value reflects the information of the smoke concentration gradient, and each component of ⟨O^> represents the average rate of change of smoke concentration in the corresponding direction.

4. The smoke analysis AI system based on quantum field theory tensor network according to claim 1 is characterized by: The specific operation process of the dynamic Hamiltonian construction module of the smoke analysis AI algorithm based on quantum field theory tensor network is as follows: 1) Gauge field construction: Define the SU(2) gauge field to describe the interaction between smoke particles; that is: ； in: For the normative potential; is the Pauli matrix. 5.2) Covariant derivatives: construct the coupling term between the matter field, i.e., smoke particles, and the gauge field; and ； in: g = 0.4 is the coupling constant; Ψ is a two-component spinor field. 6.3) Dynamic Hamiltonian construction, namely: ； It includes kinetic energy term and gauge field intensity term And the interaction potential V(ρ)=λ(ρ−ρ0)^4.

7. The smoke analysis AI system based on quantum field theory tensor network according to claim 1 is characterized by: The specific process of updating the dynamic Hamiltonian of the dynamic Hamiltonian building module of the smoke analysis AI algorithm based on quantum field theory tensor network is as follows: Dynamic model: The smoke diffusion process is modeled as an evolutionary process in quantum field theory, whose dynamics is represented by the Hamiltonian Control, that is: =−μ(t) + Other interaction terms in: and is a field operator; μ(t) is the diffusion coefficient, which is coupled in real time with the ventilation system parameters wind speed and temperature; Adjust the μ parameter in real time according to the ventilation system data, namely: ； In the formula, "wind speed" is converted into a dimensionless parameter using a normalization factor of 10 m / s, representing the coupling strength between the kinetic energy and potential energy of the airflow within the ventilation system. When the actual wind speed approaches 10 m / s, the adjustment range of μ reaches the designed maximum, μ = 0.

7. At this point, the system potential energy term is significantly enhanced, suppressing energy dissipation caused by turbulence. The formula is a linear function, with μ positively correlated with wind speed with a coefficient of 0.07, making it suitable for rapid response requirements under small wind speed fluctuations. The constant term (-0.5) is used to correct the baseline coupling strength to ensure that μ is within a reasonable range under normal operating conditions. For example, when the wind speed is 5 m / s, μ = 0.2 to avoid overdamping of the system. Specifically, real-time wind speed data is read from the ventilation system with a sampling frequency of 10 Hz. The current μ(t) is calculated and the Hamiltonian is updated. As the wind speed increases, μ(t) increases, indicating an accelerated smoke diffusion rate. The evolution equation is approximated using a quantum tensor network: ； Dynamically adjust the simulated trajectory of smoke diffusion to improve the model's adaptability to the real-time substation environment.

8. The smoke analysis AI system based on quantum field theory tensor network according to claim 1 is characterized by: The tensor network evolution module of the smoke analysis AI algorithm based on quantum field theory tensor network includes a quantum-classical hybrid optimizer, which uses the alternating direction multiplier method to jointly optimize the tensor rank r and Hamiltonian parameters. , satisfying the constraints: ； in: The coupling parameters The constructed Hamiltonian matrix; is the experimental observation value of the target Hamiltonian is the local tensor in the tensor network; represents its rank constraint; is the rank penalty coefficient, which is used to balance accuracy and complexity.

9. The smoke analysis AI system based on quantum field theory tensor network according to claim 1 is characterized by: The quantum state encoding module of the smoke analysis AI algorithm based on quantum field theory tensor network uses a non-uniform quantized grid to map the RGB color space to the Bloch sphere. The angle calculation formula and azimuth angle calculation formula are as follows: Polar angle calculation formula: ； Quantify the projection intensity of hue on the blue-yellow axis, where the numerator highlights the contribution of the blue channel and the denominator realizes brightness normalization. Ensure that the input value is within the dynamic range constraint of [−1,1] to avoid errors outside the domain of the arccosine function; Azimuth calculation formula: ； Capturing red-green contrast through Gi−Ri, Balance the blue channel's modulation of the hue circle using the four-quadrant inverse tangent 2. Eliminate angle jumps to ensure smooth transitions between adjacent color block mappings. Based on this, through the coordinated design of polar angles and azimuth angles, the RGB three-dimensional color gamut information is fully preserved on the two-dimensional Bloch sphere. Combined with the quantum state fidelity index, the encoding parameters are controlled in real time to adapt to complex and changing lighting environment scenes.

10. The smoke analysis AI system based on quantum field theory tensor network according to claim 1 is characterized by: The operation process of the chaos detection module of the smoke analysis AI algorithm based on quantum field theory tensor network is as follows: 1) Phase space reconstruction: Extract the quantum state probability distribution P(σ) from the MPS and construct the delay coordinate matrix: ； Where: embedding dimension m=5; Delay time τ = 3 frames; 2) Nearest neighbor search: For each phase point Xi, find the nearest neighbor point Xj and calculate the divergence speed: dj(t)=∥Xj(t)−Xi(t)∥2; 3) Exponential fitting: fitting by least squares method: λmax = 。

Images