Artificial intelligence-based cultural and tourism uniform light guide lighting fault diagnosis method and system

Abstract

The present invention provides a method and system for diagnosing faults of uniform light guide lighting for cultural tourism based on artificial intelligence, which relates to the field of smart cultural tourism technology, including collecting multi-source heterogeneous data of a uniform light guide lighting system for cultural tourism, processing multimodal data, and outputting abnormal feature vectors and quantum state entanglement matrices; inputting into a biologically inspired self-organizing neural network, generating a hierarchical representation of fault modes through dynamic weight adjustment and synaptic growth algorithms, constructing a fault topological landscape, and generating multi-scale continuous homology feature descriptors; inputting into a pre-constructed fuzzy cognitive graph reasoning system, modeling through the causal relationship between concept nodes, generating a fault scenario model, diagnosing the fault, updating a fault diagnosis knowledge base, obtaining a diagnosis result, and generating a fault diagnosis report.

Description

Technical Field

[0001] The present invention relates to the field of smart cultural tourism technology, and in particular to an artificial intelligence-based cultural tourism uniform light guide lighting fault diagnosis method and system. Background Art

[0002] With the development of the cultural tourism industry, the lighting systems in cultural tourism scenes are becoming increasingly complex and intelligent. Uniform light guide lighting technology has excellent display effects and is widely used in cultural exhibitions, art installations, tourist attractions and other fields. However, due to the complexity and variability of the use environment and the need for long-term continuous work, the uniform light guide lighting system is prone to failure, which seriously affects the lighting effect and tourist experience.

[0003] Traditional uniform light guide lighting fault diagnosis mainly relies on manual inspection and experience judgment, which has problems such as low efficiency, insufficient accuracy, and high maintenance cost. Attempts have been made to introduce technologies such as rule-based expert systems and signal processing-based fault detection algorithms, but they still face challenges such as difficulty in knowledge acquisition, poor adaptability, and limited diagnostic performance. With the development of artificial intelligence technology, by integrating expert knowledge and data-driven methods, using machine learning, deep learning and other technologies, the operating status of the lighting system can be automatically analyzed, potential faults can be discovered and located in a timely manner, and decision support can be provided. However, the existing uniform light guide lighting fault diagnosis methods based on artificial intelligence still have shortcomings: lack of global fault analysis, limited generalization ability of diagnostic models, lack of methods combining domain knowledge with data-driven, and imperfect application and feedback mechanisms of diagnostic results.

[0004] To sum up, there is an urgent need for an intelligent uniform light-guiding lighting fault diagnosis method that fully utilizes the advantages of artificial intelligence technology, integrates multi-source heterogeneous data and domain knowledge, builds an efficient, accurate, and adaptable diagnostic model, explores the inherent mechanism and evolution law of fault occurrence, and forms a closed-loop management of fault warning, diagnosis, decision-making, and feedback. The present invention can solve the problems in the prior art. Summary of the invention

[0005] The embodiments of the present invention provide a method and system for diagnosing faults of uniform light guide lighting in cultural tourism based on artificial intelligence, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] Provided is a method for diagnosing uniform light guide lighting faults in cultural tourism based on artificial intelligence, including:

[0008] Deploy a multimodal perception network to collect multi-source heterogeneous data of the cultural and tourism uniform light-guiding lighting system, process the multimodal data through a quantum annealing multimodal anomaly detection algorithm, detect data points that exceed a preset threshold in a high-dimensional feature space, and output anomaly feature vectors and a quantum state entanglement degree matrix;

[0009] The abnormal feature vector and quantum state entanglement matrix are input into a biologically inspired self-organizing neural network, and a hierarchical representation of the fault mode is generated through dynamic weight adjustment and synaptic growth algorithm. Based on the hierarchical representation, a fault topology landscape is constructed through topological data analysis to determine the internal relationship and evolution path between the fault modes, and generate a multi-scale continuous coherent feature descriptor;

[0010] The multi-scale continuous homophonic feature descriptors are input into the pre-built fuzzy cognitive graph reasoning system, and the causal relationship between concept nodes is modeled to generate a fault scenario model including concept node activation values ​​and relationship weight matrix. Based on the fault scenario model, the fault is diagnosed using a nonlinear fuzzy reasoning algorithm, and online learning is realized in combination with the adaptive resonance theory. The fault diagnosis knowledge base is updated to obtain the diagnosis results and generate a fault diagnosis report including the fault type, fault location and fault severity.

[0011] In an optional embodiment,

[0012] Deploy a multimodal perception network to collect multi-source heterogeneous data of the cultural tourism uniform light-guiding lighting system, process the multimodal data through the quantum annealing multimodal anomaly detection algorithm, detect data points exceeding the preset threshold in the high-dimensional feature space, and output the abnormal feature vector and quantum state entanglement degree matrix including:

[0013] The multimodal sensing network includes a spectrum analyzer, an acoustic sensor and an electromagnetic field detector, wherein the spectrum analyzer is used to collect visible spectrum and near infrared spectrum data of the light guide illumination, the acoustic sensor is used to collect ultrasonic signals and infrasonic signals, and the electromagnetic field detector is used to measure the electromagnetic field intensity and distribution;

[0014] Generate a multimodal data stream through the multimodal perception network, perform wavelet transform denoising, Fourier transform spectrum analysis and adaptive filtering on the multimodal data stream to obtain pre-processed multimodal data;

[0015] Construct a quantum annealing Hamiltonian, encode multimodal data into the initial superposition state of quantum bits as the starting point for anomaly detection, use adiabatic quantum computing method, gradually adjust annealing parameters to make the initial superposition state evolve toward the target state, and obtain the optimal quantum state. Through quantum measurement operations, project the optimal quantum state to the classical state to obtain preliminary anomaly detection results.

[0016] Based on the preliminary anomaly detection results, multimodal data fusion is performed, and the features of different modes are nonlinearly combined and dimensionally reduced using a deep belief network to obtain fused features; according to historical data and the current light guide lighting state, the anomaly detection threshold is adaptively adjusted, a dynamic threshold model is constructed, the fused features are compared with the dynamic threshold, potential anomalies are marked, and a density clustering algorithm is applied to analyze the distribution of the potential anomalies in the high-dimensional feature space and identify anomaly clusters; an anomaly association network is constructed based on graph theory to mine the spatiotemporal relationship between anomalies in the anomaly clusters;

[0017] Based on the abnormal cluster and the abnormal association network, a recurrent neural network model is trained to capture the temporal characteristics of the abnormal pattern, and key abnormal features are extracted from the recurrent neural network model in combination with an attention mechanism to generate an abnormal feature vector;

[0018] The quantum mutual information between different modal anomalies is calculated, and the quantum state entanglement matrix is ​​constructed by quantifying the correlation strength of multi-modal anomalies, and the anomaly feature vector and the quantum state entanglement matrix are output.

[0019] In an optional embodiment,

[0020] Construct a quantum annealing Hamiltonian, encode the multimodal data into the initial superposition state of quantum bits as the starting point for anomaly detection, use the adiabatic quantum computing method, gradually adjust the annealing parameters to make the initial superposition state evolve toward the target state, and obtain the optimal quantum state. Through quantum measurement operations, project the optimal quantum state to the classical state, and obtain preliminary anomaly detection results including:

[0021] Extract features from multimodal data to obtain feature vectors, map the feature vectors to the spin configuration of quantum bits, and construct a problem Hamiltonian, where each feature vector corresponds to the spin state of a quantum bit; construct a quadratic energy function to represent the problem Hamiltonian as a combination of interaction terms and monomer terms between quantum bits; select the transverse magnetic field as the driving Hamiltonian, and use the Pauli X operator to represent each quantum bit;

[0022] Combining the problem Hamiltonian and the driving Hamiltonian, controlling the relative strength of the problem Hamiltonian and the driving Hamiltonian in combination with the time-varying coefficient, determining the total Hamiltonian, and obtaining a quantum annealing Hamiltonian;

[0023] Based on the quantum annealing Hamiltonian, the multimodal data is normalized and converted into quantum state amplitude, the quantum state amplitude is converted into a quantum superposition state using a quantum gate operation, an initial quantum state is constructed, an annealing time and an annealing scheduling function are set, the time-varying coefficient in the total Hamiltonian is adjusted, and a time-varying total Hamiltonian is obtained, the initial quantum state and the time-varying total Hamiltonian are substituted into the Schrödinger equation, the Schrödinger equation is solved, and the evolution trajectory of the quantum state over time is calculated;

[0024] During the quantum state evolution process, the system energy change rate is monitored. When the energy change rate is lower than a preset threshold, a quantum tunneling operator is introduced into the evolution trajectory to change the current quantum state; a quantum fluctuation operator is periodically applied to the evolution trajectory to introduce random phase disturbances;

[0025] The quantum bits of different modes are entangled together to obtain an entangled state, and controlled quantum gate operations are applied to the entangled state to perform parallel computing, calculate the entanglement entropy, quantify the degree of correlation between the modes, determine the quantum measurement basis based on the entangled state, perform projection measurement, project the final quantum state onto the quantum measurement basis, repeatedly perform the projection measurement, obtain the probability distribution of the measurement result, calculate the discreteness of the probability distribution, determine whether there is an abnormality according to a preset discreteness threshold, and output a preliminary abnormality detection result.

[0026] In an optional embodiment,

[0027] During the quantum state evolution process, the system energy change rate is monitored, and when the energy change rate is lower than a preset threshold, a quantum tunneling operator is introduced into the evolution trajectory to change the current quantum state; and periodically applying a quantum fluctuation operator to the evolution trajectory to introduce a random phase disturbance also includes:

[0028] During the quantum state evolution process, the changes in system energy and quantum state are continuously monitored, and the system energy change rate and quantum state overlap are calculated;

[0029] Based on the energy change rate and quantum state overlap of the system, the application intensity and frequency of the quantum tunneling operator and the quantum fluctuation operator are dynamically adjusted to form an adaptive quantum operator application strategy, wherein the adaptive quantum operator application strategy includes:

[0030] When the system energy change rate is lower than a first preset threshold, the application intensity of the quantum tunneling operator is increased to jump out of the local minimum; when the quantum state overlap is higher than a second preset threshold, the application frequency of the quantum fluctuation operator is increased;

[0031] According to the adaptive quantum operator application strategy, in the evolution trajectory of the quantum state, a quantum tunneling operator is applied to change the current quantum state, and a quantum fluctuation operator is applied to introduce a random phase disturbance;

[0032] According to the preset period, periodically monitor the changing trend of the system energy change rate and quantum state overlap;

[0033] When the rate of decrease of the energy change rate of the system is lower than a third preset threshold, or the amplitude of change of the quantum state overlap is lower than a fourth preset threshold, adjusting the parameters of the adaptive quantum operator application strategy, including adjusting the first preset threshold, the second preset threshold, the application intensity adjustment amplitude of the quantum tunneling operator, and the application frequency adjustment amplitude of the quantum fluctuation operator;

[0034] Repeat the steps until the preset number of iterations is reached.

[0035] In an optional embodiment,

[0036] The abnormal feature vector and quantum state entanglement matrix are input into the biologically inspired self-organizing neural network. Through dynamic weight adjustment and synaptic growth algorithm, a hierarchical representation of the fault mode is generated. Based on the hierarchical representation, through topological data analysis, a fault topological landscape is constructed to determine the internal relationship and evolution path between the fault modes. The multi-scale continuous coherent feature descriptors are generated, including:

[0037] Constructing the biologically inspired self-organizing neural network, including an input layer, a hidden layer, and an output layer, each layer consisting of a plurality of neurons, and establishing random initial connection weights between neurons;

[0038] The abnormal feature vector and the quantum state entanglement degree matrix are transmitted into the biologically inspired self-organizing neural network to trigger neuron activation, and the connection weights are dynamically adjusted according to the activity levels of adjacent neurons;

[0039] During the network training process, the activity of neurons is monitored, and synaptic growth and pruning are completed based on the synaptic growth algorithm. Through repeated iterative training, a multi-level structure is self-organized to generate a hierarchical representation of the fault mode.

[0040] Based on the hierarchical representation of the failure mode, each neuron in the biologically inspired self-organizing neural network is regarded as a data point in a high-dimensional space, and the position is determined according to the connection weight of the neuron;

[0041] A connection relationship is established between the data points using the k-nearest neighbor method to form a simple complex representing the topological structure of the data, topological data analysis is performed on the simple complex, a fault topological landscape is constructed, the connection relationship and transition paths of different regions in the fault topological landscape are analyzed, the internal relationship and evolution path between fault modes are determined, the persistence homology of the simple complex is calculated, the appearance and disappearance of topological features are recorded to form a persistence graph, and based on the persistence graph, the topological features of the fault mode are described to generate a multi-scale persistence homology feature descriptor.

[0042] In an optional embodiment,

[0043] The synapse growth algorithm includes:

[0044] Continuously monitoring the activity of each neuron in the biologically inspired self-organizing neural network, and quantifying the activity by recording the number of activations of each neuron within a predetermined time window;

[0045] Set high activity threshold and inactive threshold to determine the state of neurons;

[0046] Taking neurons whose activity exceeds the high-activity threshold as high-activity neurons, randomly selecting positions in the surrounding space of the high-activity neurons to generate new neurons, establishing initial connections between the new neurons and the high-activity neurons, and establishing probabilistic connections with neighboring neurons of the high-activity neurons;

[0047] Neurons with activity levels lower than the inactive threshold are regarded as inactive neurons, and the inactive neurons are marked as being in a state to be removed. When an inactive neuron is marked as being in a state to be removed multiple times within a plurality of consecutive predetermined time windows, the inactive neuron is removed from the biologically inspired self-organizing neural network, and the connection weights with neighboring neurons are proportionally distributed to surrounding neurons;

[0048] Update the connection matrix of the biologically inspired self-organizing neural network according to the newly added and removed neurons, and recalculate the positional relationship of all neurons in the biologically inspired self-organizing neural network;

[0049] Based on the overall activity status, the high activity threshold and the inactivity threshold are periodically adjusted.

[0050] In an optional embodiment,

[0051] The multi-scale continuous homology feature descriptors are input into the pre-built fuzzy cognitive graph reasoning system, and the causal relationship between concept nodes is modeled to generate a fault scenario model containing concept node activation values ​​and relationship weight matrices. Based on the fault scenario model, the nonlinear fuzzy reasoning algorithm is used to diagnose the fault, and the adaptive resonance theory is combined to realize online learning, update the fault diagnosis knowledge base, obtain the diagnosis results, and generate a fault diagnosis report including fault type, fault location and fault severity.

[0052] Construct a fuzzy cognitive graph reasoning system, which includes multiple concept nodes and inter-node links. The concept nodes represent fault characteristics and fault states, and the inter-node links represent the causal impact of faults.

[0053] Mapping the multi-scale continuous homology feature descriptors to the concept nodes of the fuzzy cognitive graph reasoning system and assigning initial activation values ​​to the corresponding concept nodes; analyzing the mutual influence between the concept nodes, determining the connection strength and influence direction between the nodes, and establishing a causal relationship model;

[0054] Based on the initial activation value of the concept node and the causal relationship model, a fault scenario model is generated, wherein the fault scenario model includes a concept node activation value vector and a relationship weight matrix;

[0055] The fault scenario model is analyzed using a nonlinear fuzzy inference algorithm, wherein the analysis includes initializing a system state, applying a relationship weight matrix to calculate mutual influences between nodes, and updating node activation values ​​using a nonlinear conversion function, and repeatedly iterating the analysis until the fault scenario model reaches a stable state, thereby obtaining a final concept node activation value;

[0056] Based on the final concept node activation value, diagnose the fault, determine the fault type, location and severity, and obtain a preliminary fault result;

[0057] Combining adaptive resonance theory to realize online learning, compare the similarity between the preliminary fault result and the known fault mode, update the fault diagnosis knowledge base according to the similarity, update the existing fault mode information when the similarity is higher than the preset threshold, and add the preliminary fault result as a new fault mode to the fault diagnosis knowledge base when the similarity is lower than the preset threshold;

[0058] Based on the preliminary fault diagnosis result and the updated fault diagnosis knowledge base, a final diagnosis result is obtained, and a fault diagnosis report is generated according to the final diagnosis result. The fault diagnosis report includes the fault type, fault location and fault severity.

[0059] According to a second aspect of the embodiments of the present invention,

[0060] Provided is a cultural tourism uniform light guide lighting fault diagnosis system based on artificial intelligence, including:

[0061] The first unit is used to deploy a multimodal perception network, collect multimodal data, process the multimodal data through a quantum annealing multimodal anomaly detection algorithm, detect data points exceeding a preset threshold in a high-dimensional feature space, and output an abnormal feature vector and a quantum state entanglement degree matrix;

[0062] The second unit is used to input the abnormal feature vector and the quantum state entanglement matrix into the biologically inspired self-organizing neural network, generate a hierarchical representation of the fault mode through dynamic weight adjustment and synaptic growth algorithm, and build a fault topology landscape based on the hierarchical representation through topological data analysis, determine the internal relationship and evolution path between the fault modes, and generate a multi-scale continuous coherent feature descriptor;

[0063] The third unit is used to input the multi-scale continuous homology feature descriptors into the pre-built fuzzy cognitive graph reasoning system, model the causal relationship between concept nodes, generate a fault scenario model including concept node activation values ​​and relationship weight matrix, and diagnose the fault based on the fault scenario model using nonlinear fuzzy reasoning algorithm, combine adaptive resonance theory to realize online learning, update the fault diagnosis knowledge base, obtain the diagnosis results, and generate a fault diagnosis report including fault type, fault location and fault severity.

[0064] According to a third aspect of the embodiments of the present invention,

[0065] An electronic device is provided, comprising:

[0066] processor;

[0067] a memory for storing processor-executable instructions;

[0068] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0069] A fourth aspect of the embodiments of the present invention is:

[0070] 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.

[0071] In an embodiment of the present invention, various types of data are collected through a multimodal perception network to improve the comprehensive perception ability of the cultural and tourism uniform light-guiding lighting system; a quantum annealing multimodal anomaly detection algorithm is used to effectively process complex multimodal data and improve the accuracy of data anomaly detection; abnormal data points are detected in a high-dimensional feature space to ensure accurate capture and analysis of complex features; the output of abnormal feature vectors and quantum state entanglement matrix helps to more deeply understand the relationship between data anomalies and system states; by inputting abnormal feature vectors and quantum state entanglement matrix into a neural network, in-depth analysis of fault modes is achieved to improve the accuracy of anomaly identification; dynamic weight adjustment and synaptic growth algorithms are applied to optimize network learning capabilities and enhance the recognition and representation of complex fault modes; a hierarchical representation of fault modes is generated to facilitate understanding of the relationship and relative importance between different faults. , providing a clear view of the fault structure; by constructing a fault topology landscape, revealing the internal relationship and evolution path between fault modes, it helps to predict and prevent potential faults; generating multi-scale continuous coherent feature descriptors, so that fault analysis can be carried out at different levels, enhancing the comprehensive understanding of the behavior of complex systems; through the fuzzy cognitive graph reasoning system, the multi-scale continuous coherent feature descriptors are converted into fault scenario models to improve the understanding of complex fault relationships; using nonlinear fuzzy reasoning algorithms for fault diagnosis, it improves the accuracy and flexibility of diagnosis and adapts to different fault characteristics; combining adaptive resonance theory to achieve online learning, ensure that the fault diagnosis knowledge base is continuously updated to adapt to new data and fault modes; generating detailed fault diagnosis reports covering fault type, location and severity, providing clear decision support information, and facilitating subsequent processing and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a flow chart of a method for diagnosing faults of uniform light guide lighting for cultural tourism based on artificial intelligence according to an embodiment of the present invention;

[0073] Figure 2 It is a structural schematic diagram of a cultural and tourism uniform light guide lighting fault diagnosis system based on artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION

[0074] 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.

[0075] 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.

[0076] Figure 1 FIG. 1 is a flow chart of a method for diagnosing uniform light guide lighting faults in a tourism industry based on artificial intelligence according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0077] S101. Deploy a multimodal sensing network, collect multi-source heterogeneous data of the cultural tourism uniform light-guiding lighting system, process the multimodal data through a quantum annealing multimodal anomaly detection algorithm, detect data points exceeding a preset threshold in a high-dimensional feature space, and output an abnormal feature vector and a quantum state entanglement degree matrix;

[0078] In this embodiment, various types of data are collected through a multimodal perception network to improve the comprehensive perception ability of the cultural and tourism uniform light-guiding lighting system; a quantum annealing multimodal anomaly detection algorithm is used to effectively process complex multimodal data and improve the accuracy of data anomaly detection; abnormal data points are detected in a high-dimensional feature space to ensure accurate capture and analysis of complex features; and the output of abnormal feature vectors and quantum state entanglement matrices helps to gain a deeper understanding of the relationship between data anomalies and system states.

[0079] In an optional embodiment, a multimodal sensing network is deployed to collect multi-source heterogeneous data of a cultural tourism uniform light-guiding lighting system, and the multimodal data is processed by a quantum annealing multimodal anomaly detection algorithm, and data points exceeding a preset threshold are detected in a high-dimensional feature space, and the output of an abnormal feature vector and a quantum state entanglement degree matrix includes:

[0080] The multimodal sensing network includes a spectrum analyzer, an acoustic sensor and an electromagnetic field detector, wherein the spectrum analyzer is used to collect visible spectrum and near infrared spectrum data of the light guide illumination, the acoustic sensor is used to collect ultrasonic signals and infrasonic signals, and the electromagnetic field detector is used to measure the electromagnetic field intensity and distribution;

[0081] Generate a multimodal data stream through the multimodal perception network, perform wavelet transform denoising, Fourier transform spectrum analysis and adaptive filtering on the multimodal data stream to obtain pre-processed multimodal data;

[0082] Construct a quantum annealing Hamiltonian, encode multimodal data into the initial superposition state of quantum bits as the starting point for anomaly detection, use adiabatic quantum computing method, gradually adjust annealing parameters to make the initial superposition state evolve toward the target state, and obtain the optimal quantum state. Through quantum measurement operations, project the optimal quantum state to the classical state to obtain preliminary anomaly detection results.

[0083] Based on the preliminary anomaly detection results, multimodal data fusion is performed, and the features of different modes are nonlinearly combined and dimensionally reduced using a deep belief network to obtain fused features; according to historical data and the current light guide lighting state, the anomaly detection threshold is adaptively adjusted, a dynamic threshold model is constructed, the fused features are compared with the dynamic threshold, potential anomalies are marked, and a density clustering algorithm is applied to analyze the distribution of the potential anomalies in the high-dimensional feature space and identify anomaly clusters; an anomaly association network is constructed based on graph theory to mine the spatiotemporal relationship between anomalies in the anomaly clusters;

[0084] Based on the abnormal cluster and the abnormal association network, a recurrent neural network model is trained to capture the temporal characteristics of the abnormal pattern, and key abnormal features are extracted from the recurrent neural network model in combination with an attention mechanism to generate an abnormal feature vector;

[0085] The quantum mutual information between different modal anomalies is calculated, and the quantum state entanglement matrix is ​​constructed by quantifying the correlation strength of multi-modal anomalies, and the anomaly feature vector and the quantum state entanglement matrix are output.

[0086] The sampling frequency and wavelength range of the spectrum analyzer are set to cover the visible light and near-infrared spectrum regions; the spectrum analyzer is aimed at the light guide illumination to continuously collect spectrum data within a certain period of time; the collected spectrum data is converted into digital signals and stored in a time series format.

[0087] Select appropriate acoustic sensors to ensure that their frequency response range covers the ultrasonic and infrasonic frequency bands; place acoustic sensors at key locations to collect acoustic signals in the light-guiding lighting environment; perform analog-to-digital conversion on the collected acoustic signals to obtain digital time series data.

[0088] Select highly sensitive electromagnetic field detectors to ensure that their measurement range meets the requirements; arrange multiple electromagnetic field detectors around the light guide to form a detection array; synchronously collect the electromagnetic field intensity data of each detector to form electromagnetic field data with temporal and spatial distribution.

[0089] The data of each mode is decomposed by wavelet respectively to obtain wavelet coefficients at different scales; according to the statistical characteristics of the wavelet coefficients, an appropriate threshold is set and the coefficients are subjected to soft threshold processing; the processed wavelet coefficients are reconstructed by wavelet to obtain denoised time series data.

[0090] Perform Fourier transform on the denoised time series data to obtain data in the frequency domain. Calculate the amplitude and phase information of each frequency component to obtain the spectrum characteristics. According to the spectrum characteristics, identify the key frequency bands and dominant frequency components.

[0091] According to the results of spectrum analysis, an adaptive filter suitable for each modal data is designed; the frequency domain data is adaptively filtered to suppress interference and noise and enhance the information of key frequency bands; the filtered frequency domain data is inverse Fourier transformed to obtain the preprocessed time series data.

[0092] The preprocessed multimodal data is mapped to the initial state of the quantum bit; according to the goal of anomaly detection, the Hamiltonian of the problem is constructed, including energy functions and constraints; the Hamiltonian is converted into a quantum gate circuit for quantum annealing calculation.

[0093] Set the annealing parameters, including initial temperature, termination temperature, and annealing time; use the initial state of the quantum bit as the starting point of annealing, and apply the quantum gate operation corresponding to the Hamiltonian; by slowly adjusting the annealing parameters, the quantum state evolves along the path with the lowest energy and gradually approaches the target state.

[0094] After the annealing process is completed, the quantum bits are measured to obtain the results of the classical bits; based on the measurement results, the corresponding anomaly detection results are decoded and potential abnormal data points are identified; the preliminary anomaly detection results are analyzed and evaluated to provide input for subsequent steps.

[0095] Construct a deep belief network model, including multi-layer restricted Boltzmann machines and top-level classifiers; use preprocessed data of different modalities as the input of the network, and learn the high-level features of the data through unsupervised layer-by-layer pre-training; use supervised fine-tuning methods to optimize network parameters and achieve nonlinear combination and dimensionality reduction of features.

[0096] Collect historical data, including normal data and known abnormal data, to form a training set; extract the characteristics of the current light-guiding lighting state, such as light intensity, spectral distribution, etc.; combine historical data and current state characteristics to train a dynamic threshold model and learn the optimal threshold for abnormality judgment.

[0097] Compare the fused features output by the deep belief network with the output of the dynamic threshold model; mark the data points that exceed the threshold as potential outliers; apply a density clustering algorithm, such as DBSCAN, to cluster the potential outliers and identify outlier clusters.

[0098] For the abnormal points in the abnormal cluster, their time and space attributes are extracted; based on graph theory, an association network between the abnormal points is constructed, with nodes representing the abnormal points and edges representing the association strength between the abnormal points; through the association network, the spatiotemporal evolution laws and propagation patterns of the abnormal points in the abnormal cluster are explored.

[0099] Construct a recurrent neural network model, such as LSTM or GRU, to learn the temporal characteristics of abnormal patterns; use abnormal clusters and abnormal association networks as network inputs to capture the temporal dependency of abnormal evolution through sequence-to-sequence learning; train the recurrent neural network model and optimize network parameters to achieve effective representation of abnormal patterns.

[0100] On the basis of recurrent neural network, the attention mechanism is introduced to adaptively allocate the importance of different moments and different abnormal points; through the attention weight, the key abnormal features are highlighted and the noise and redundant information are suppressed; the abnormal feature vector is generated to comprehensively represent the key features of the abnormal pattern.

[0101] For the abnormal feature vectors of different modes, the quantum mutual information between them is calculated; the quantum entanglement degree between different modal anomalies is calculated through the characteristic decomposition of the quantum state density matrix; the quantum state entanglement degree matrix is ​​constructed to quantify the correlation strength between multi-modal anomalies.

[0102] The abnormal feature vector and quantum state entanglement matrix are used as the final anomaly detection results; the anomaly detection results are visualized, such as anomaly distribution diagram, association network diagram, etc.; an anomaly detection report is output, including detailed information such as anomaly type, occurrence time, and association relationship.

[0103] In this embodiment, data is collected from multiple angles and dimensions through a spectrum analyzer, an acoustic sensor, and an electromagnetic field detector, covering multiple physical properties such as visible spectrum, infrared spectrum, sound waves, and electromagnetic fields, thereby improving the comprehensiveness of data perception; multimodal data preprocessing is performed by combining wavelet transform, Fourier transform, and adaptive filtering technologies to effectively remove noise and enhance the analysis quality of data; anomaly detection is performed based on quantum annealing Hamiltonian, and the optimal quantum state is obtained through adiabatic quantum calculation to improve the accuracy and efficiency of anomaly detection; a deep belief network is used for feature nonlinear combination and dimensionality reduction to improve the fusion effect of different modal data and enhance the ability to identify anomalies; the detection threshold is dynamically adjusted based on historical data and current status, and a dynamic threshold model is constructed to ensure the flexibility and accuracy of anomaly detection; anomaly clusters in high-dimensional space are identified through density clustering, and the spatiotemporal relationship is analyzed through graph theory to enhance the correlation analysis between anomalies; a recurrent neural network is trained in combination with an attention mechanism to capture abnormal time series features, extract key abnormal features, and ensure accurate identification of abnormal patterns; multimodal abnormal correlations are quantified through quantum mutual information, a quantum state entanglement matrix is ​​generated, and the correlation between multimodal anomalies is deeply mined.

[0104] In an optional embodiment, a quantum annealing Hamiltonian is constructed, multimodal data is encoded into an initial superposition state of quantum bits as a starting point for anomaly detection, an adiabatic quantum computing method is used, and annealing parameters are gradually adjusted to make the initial superposition state evolve toward a target state to obtain an optimal quantum state. The optimal quantum state is projected to a classical state through a quantum measurement operation, and obtaining a preliminary anomaly detection result includes:

[0105] Extract features from multimodal data to obtain feature vectors, map the feature vectors to the spin configuration of quantum bits, and construct a problem Hamiltonian, where each feature vector corresponds to the spin state of a quantum bit; construct a quadratic energy function to represent the problem Hamiltonian as a combination of interaction terms and monomer terms between quantum bits; select the transverse magnetic field as the driving Hamiltonian, and use the Pauli X operator to represent each quantum bit;

[0106] Combining the problem Hamiltonian and the driving Hamiltonian, controlling the relative strength of the problem Hamiltonian and the driving Hamiltonian in combination with the time-varying coefficient, determining the total Hamiltonian, and obtaining a quantum annealing Hamiltonian;

[0107] Based on the quantum annealing Hamiltonian, the multimodal data is normalized and converted into quantum state amplitude, the quantum state amplitude is converted into a quantum superposition state using a quantum gate operation, an initial quantum state is constructed, an annealing time and an annealing scheduling function are set, the time-varying coefficient in the total Hamiltonian is adjusted, and a time-varying total Hamiltonian is obtained, the initial quantum state and the time-varying total Hamiltonian are substituted into the Schrödinger equation, the Schrödinger equation is solved, and the evolution trajectory of the quantum state over time is calculated;

[0108] During the quantum state evolution process, the system energy change rate is monitored. When the energy change rate is lower than a preset threshold, a quantum tunneling operator is introduced into the evolution trajectory to change the current quantum state; a quantum fluctuation operator is periodically applied to the evolution trajectory to introduce random phase disturbances;

[0109] The quantum bits of different modes are entangled together to obtain an entangled state, and controlled quantum gate operations are applied to the entangled state to perform parallel computing, calculate the entanglement entropy, quantify the degree of correlation between the modes, determine the quantum measurement basis based on the entangled state, perform projection measurement, project the final quantum state onto the quantum measurement basis, repeatedly perform the projection measurement, obtain the probability distribution of the measurement result, calculate the discreteness of the probability distribution, determine whether there is an abnormality according to a preset discreteness threshold, and output a preliminary abnormality detection result.

[0110] The spin configuration specifically refers to mapping the feature vector to the spin state of the qubit. Each qubit can be in a different spin direction (such as "up" or "down"), corresponding to the two basic states in the quantum system (usually denoted |0> and |1>). These spin configurations represent different quantum states of each feature in the system.

[0111] The problem Hamiltonian is an energy function used to represent the problem to be solved. It describes the interaction between the spin states of each qubit in the quantum system and the overall energy of the system. Different spin configurations of the eigenvectors will correspond to different Hamiltonians, and the optimal solution can be found by minimizing the problem Hamiltonian. Hamiltonian is usually used to describe the total energy of a physical system and is one of the most important elements in the quantum annealing process.

[0112] The quadratic energy function represents the combination of the interaction terms and the monomer terms between the qubits in the Hamiltonian of the problem. Its form is a quadratic function whose energy depends on the state (spin configuration) of the qubit, including the coupling terms between the spins and the contribution of a single spin. By constructing the quadratic energy function, the interaction of the system and the energy of the spin state can be described.

[0113] The driving Hamiltonian is a quantum system used to guide the system to evolve from an initial state to a target state. The driving Hamiltonian is usually selected as a transverse magnetic field, and acts on each quantum bit through the Pauli X operator to drive the superposition state in the quantum system. During the quantum annealing process, the driving Hamiltonian controls the evolution path of the system, helping it to gradually anneal from a high energy state to a low energy state (i.e., the optimal solution).

[0114] The Pauli X operator is a basic quantum operator acting on a quantum bit, similar to the "flip" operation in classical computing. It changes the state of a quantum bit from |0> to |1>, or from |1> to |0>, which is essentially a "NOT" operation. In the present invention, the Pauli X operator is used to represent the transverse spin change of a quantum bit, and combined with the driving Hamiltonian to help the system evolve.

[0115] Preprocess the multimodal data, such as denoising and normalization; use appropriate feature extraction methods, such as principal component analysis and independent component analysis, to extract the key features of the data; construct feature vectors from the extracted features, with each feature vector representing a data sample; map each feature vector to the spin configuration of a quantum bit, and establish a correspondence between the feature space and the Hilbert space; determine the spin state of the quantum bit, such as up-spin, down-spin, etc., based on the value of the feature vector.

[0116] According to the mapped quantum bit spin state, the problem Hamiltonian is constructed; the problem Hamiltonian consists of interaction terms and monomer terms between quantum bits, reflecting the correlation between features and the contribution of a single feature; the interaction terms can be represented by a quadratic energy function to describe the coupling strength between quantum bits; the monomer terms can be represented by the Pauli Z operator to describe the energy level difference of a single quantum bit; the transverse magnetic field is selected as the driving Hamiltonian, and the Pauli X operator is used to represent each quantum bit, introducing the quantum tunneling effect; the relative strength of the problem Hamiltonian and the driving Hamiltonian is controlled by the time-varying coefficient to construct the total Hamiltonian.

[0117] Normalize the eigenvalues ​​of multimodal data and convert them into corresponding quantum state amplitudes; use quantum gate operations, such as Hadamard gates and rotating gates, to convert the quantum state amplitudes into quantum superposition states; use the converted quantum superposition states as the initial quantum states to represent the starting point of the annealing process; set the annealing time and annealing scheduling function to control the rate and trend of quantum state evolution; and obtain the time-varying total Hamiltonian by adjusting the time-varying coefficients in the total Hamiltonian to describe the dynamics of quantum state evolution.

[0118] Substitute the initial quantum state and the time-varying total Hamiltonian into the Schrödinger equation; solve the Schrödinger equation to obtain the evolution trajectory of the quantum state over time; during the solution process, monitor the energy change rate of the system, and when the energy change rate is lower than a preset threshold, the system is considered to have entered a local minimum; at the local minimum, introduce a quantum tunneling operator into the evolution trajectory to jump out of the local minimum and continue searching for a better solution; periodically apply a quantum fluctuation operator to the evolution trajectory, introduce random phase perturbations, and increase the diversity of quantum states.

[0119] Select a suitable entanglement method to entangle quantum bits of different modes to form an entangled state; apply controlled quantum gate operations on the entangled state, perform parallel computing, and accelerate the search process; calculate the entanglement entropy of the entangled state, quantify the degree of correlation between different modes, and reveal abnormal cross-modal characteristics; based on the entangled state, determine the final quantum measurement basis, such as the computational basis, Bell basis, etc.; perform projection measurement on the final quantum state and project it onto the selected quantum measurement basis; repeat the projection measurement to obtain the probability distribution of the measurement results; calculate the discreteness of the probability distribution, that is, the frequency difference of different results; compare the discreteness with the preset threshold. If the discreteness exceeds the threshold, it is determined that an abnormality exists and output the preliminary abnormality detection result.

[0120] In this embodiment, by extracting features from multimodal data and mapping them to quantum bit spin configurations, the quantum processing capability of the data is improved and the accuracy of anomaly detection is enhanced; by constructing the problem Hamiltonian and quadratic energy function, the interaction between quantum bits is clarified, which helps to more accurately characterize the problem to be solved; by combining the driving Hamiltonian and the time-varying coefficient, the dynamic evolution of the quantum state is realized, and the ability of the system to adapt to different data characteristics is enhanced; real-time monitoring of the system energy change rate and the introduction of quantum tunneling operators can effectively overcome local minima and ensure that the system evolves towards the global optimal solution; quantum bits of different modes are entangled together to achieve parallel computing, improve computing efficiency, and provide new solutions to complex problems; through the projection measurement and probability distribution analysis of the quantum measurement basis, combined with the discreteness threshold to determine anomalies, an efficient and sensitive anomaly detection mechanism is provided; calculating the degree of correlation between the entanglement entropy quantization modes helps to deeply understand the intrinsic relationship between multimodal data and provide support for subsequent analysis.

[0121] In an optional embodiment, during the quantum state evolution process, the system energy change rate is monitored, and when the energy change rate is lower than a preset threshold, a quantum tunneling operator is introduced into the evolution trajectory to change the current quantum state; and periodically applying a quantum fluctuation operator to the evolution trajectory to introduce a random phase perturbation further comprises:

[0122] During the quantum state evolution process, the changes in system energy and quantum state are continuously monitored, and the system energy change rate and quantum state overlap are calculated;

[0123] Based on the energy change rate and quantum state overlap of the system, the application intensity and frequency of the quantum tunneling operator and the quantum fluctuation operator are dynamically adjusted to form an adaptive quantum operator application strategy, wherein the adaptive quantum operator application strategy includes:

[0124] When the system energy change rate is lower than a first preset threshold, the application intensity of the quantum tunneling operator is increased to jump out of the local minimum; when the quantum state overlap is higher than a second preset threshold, the application frequency of the quantum fluctuation operator is increased;

[0125] According to the adaptive quantum operator application strategy, in the evolution trajectory of the quantum state, a quantum tunneling operator is applied to change the current quantum state, and a quantum fluctuation operator is applied to introduce a random phase disturbance;

[0126] According to the preset period, periodically monitor the changing trend of the system energy change rate and quantum state overlap;

[0127] When the rate of decrease of the energy change rate of the system is lower than a third preset threshold, or the amplitude of change of the quantum state overlap is lower than a fourth preset threshold, adjusting the parameters of the adaptive quantum operator application strategy, including adjusting the first preset threshold, the second preset threshold, the application intensity adjustment amplitude of the quantum tunneling operator, and the application frequency adjustment amplitude of the quantum fluctuation operator;

[0128] Repeat the steps until the preset number of iterations is reached.

[0129] The quantum tunneling operator is specifically an operator that describes the passage of a quantum state through an energy barrier. In quantum mechanics, a particle can "tunnel" through a potential barrier when the energy is insufficient to overcome it. This operator is used to introduce transitions during the evolution of a quantum state, helping the system to jump out of a local minimum and explore a wider state space.

[0130] The quantum fluctuation operator is specifically an operator used to introduce randomness and perturbation to simulate the fluctuations caused by uncertainty in the quantum system. By increasing the random phase perturbation of the system, the operator can promote the diversity of quantum states, allowing the system to explore more flexibly in the state space, thereby avoiding falling into a fixed mode or local minimum.

[0131] During the evolution of the quantum state, the changes in the system energy and quantum state are continuously monitored; the system energy change rate is calculated, that is, the amplitude of the change in the system energy per unit time; the quantum state overlap, that is, the amplitude of the inner product between the current quantum state and the initial quantum state, is calculated; the quantum state overlap reflects the trend of quantum state evolution, and the higher the overlap, the further the quantum state deviates from the initial state.

[0132] According to the system energy change rate and quantum state overlap, the application intensity and frequency of quantum tunneling operators and quantum fluctuation operators are dynamically adjusted; when the system energy change rate is lower than the first preset threshold, it is considered that the system may fall into a local minimum, and the application intensity of the quantum tunneling operator is increased to help the quantum state jump out of the local minimum and explore new solution space; when the quantum state overlap is higher than the second preset threshold, it is considered that the quantum state has significantly deviated from the initial state, and the application frequency of the quantum fluctuation operator is increased, introducing more random phase disturbances and increasing the diversity of quantum states; by dynamically adjusting the application strategy of quantum operators, the evolution of quantum states is made more efficient and comprehensive, thereby improving the accuracy of anomaly detection.

[0133] According to the adaptive quantum operator application strategy, quantum tunneling operators and quantum fluctuation operators are applied in the quantum state evolution trajectory; when the application conditions of the quantum tunneling operator are met, the quantum tunneling operator is applied to the current quantum state, additional energy is introduced, the probability amplitude distribution of the quantum state is changed, and the local minimum is jumped out; when the application conditions of the quantum fluctuation operator are met, the quantum fluctuation operator is applied to the current quantum state, random phase disturbance is introduced, the phase distribution of the quantum state is changed, and the diversity of the quantum state is increased; the specific implementation of the quantum tunneling operator and the quantum fluctuation operator can be achieved by applying additional Hamiltonian terms or gate operations.

[0134] According to preset cycles, such as every certain number of evolutionary steps, the changing trends of the system energy change rate and quantum state overlap are periodically monitored; the speed of decrease of the system energy change rate is calculated, that is, the average decrease in the system energy change rate over multiple consecutive cycles; the change range of the quantum state overlap is calculated, that is, the fluctuation range of the quantum state overlap over multiple consecutive cycles; by analyzing the changing trends, the effectiveness of the current quantum operator application strategy is evaluated, providing a basis for further adjusting the strategy.

[0135] When the rate of decrease of the system energy change rate is lower than the third preset threshold, or the change amplitude of the quantum state overlap is lower than the fourth preset threshold, it is considered that the current quantum operator application strategy may not be able to effectively improve the quantum state evolution; adjust the parameters of the adaptive quantum operator application strategy, including the first preset threshold, the second preset threshold, the application intensity adjustment amplitude of the quantum tunneling operator, and the application frequency adjustment amplitude of the quantum fluctuation operator; by adjusting the strategy parameters, the application of quantum operators is made more flexible and adaptable to the dynamic changes of quantum state evolution, thereby improving the robustness of anomaly detection.

[0136] Repeat the execution steps, continuously monitor the changes in system energy and quantum state, dynamically adjust the quantum operator application strategy, apply quantum operators to change quantum states, periodically monitor the change trend, and adjust the strategy parameters; the number of iterative executions can be determined based on the preset convergence conditions or reaching a predetermined number of iterations; through multiple iterations, the quantum state evolution gradually converges to the optimal anomaly explanation, thereby improving the accuracy of anomaly detection.

[0137] In this embodiment, the energy and quantum state changes of the system are continuously monitored to ensure dynamic control of the evolution of the quantum state and enhance the adaptability of the system; the application of quantum tunneling and fluctuation operators is dynamically adjusted based on the energy change rate and quantum state overlap, thereby improving the flexibility and effectiveness of quantum state optimization; by enhancing the application intensity of the quantum tunneling operator, it is effectively avoided to fall into the local minimum and the possibility of finding the global optimal solution is increased; the application frequency of the quantum fluctuation operator is increased, and by introducing random phase disturbances, the exploration ability of the system is improved and the diversity of states is promoted; the parameters of the adaptive strategy are periodically monitored and adjusted to ensure that the system always remains efficient and sensitive during the evolution process and the optimization process is continuously improved; the algorithm parameters are flexibly adjusted according to different preset thresholds, thereby enhancing the adaptability of the algorithm to different scenarios and data features and improving the overall performance.

[0138] S102. Input the abnormal feature vector and the quantum state entanglement matrix into the biologically inspired self-organizing neural network, generate a hierarchical representation of the fault mode through dynamic weight adjustment and synaptic growth algorithm, and construct a fault topology landscape based on the hierarchical representation through topological data analysis, determine the internal relationship and evolution path between the fault modes, and generate a multi-scale continuous coherent feature descriptor;

[0139] In this embodiment, by inputting abnormal feature vectors and quantum state entanglement matrix into the neural network, in-depth analysis of fault modes is achieved, and the accuracy of abnormality identification is improved; dynamic weight adjustment and synaptic growth algorithms are applied to optimize network learning capabilities and enhance the recognition and representation of complex fault modes; a hierarchical representation of fault modes is generated to facilitate understanding of the relationship and relative importance between different faults and provide a clear view of the fault structure; by constructing a fault topology landscape, the internal relationship and evolution path between fault modes are revealed, which helps to predict and prevent potential faults; and multi-scale continuous coherent feature descriptors are generated so that fault analysis can be performed at different levels, enhancing a comprehensive understanding of the behavior of complex systems.

[0140] In an optional embodiment, the abnormal feature vector and the quantum state entanglement matrix are input into a biologically inspired self-organizing neural network, and a hierarchical representation of the fault mode is generated through dynamic weight adjustment and synaptic growth algorithm. Based on the hierarchical representation, a fault topology landscape is constructed through topological data analysis, and the internal relationship and evolution path between the fault modes are determined. The generation of a multi-scale continuous coherent feature descriptor includes:

[0141] Constructing the biologically inspired self-organizing neural network, including an input layer, a hidden layer, and an output layer, each layer consisting of a plurality of neurons, and establishing random initial connection weights between neurons;

[0142] The abnormal feature vector and the quantum state entanglement degree matrix are transmitted into the biologically inspired self-organizing neural network to trigger neuron activation, and the connection weights are dynamically adjusted according to the activity levels of adjacent neurons;

[0143] During the network training process, the activity of neurons is monitored, and synaptic growth and pruning are completed based on the synaptic growth algorithm. Through repeated iterative training, a multi-level structure is self-organized to generate a hierarchical representation of the fault mode.

[0144] Based on the hierarchical representation of the failure mode, each neuron in the biologically inspired self-organizing neural network is regarded as a data point in a high-dimensional space, and the position is determined according to the connection weight of the neuron;

[0145] A connection relationship is established between the data points using the k-nearest neighbor method to form a simple complex representing the topological structure of the data, topological data analysis is performed on the simple complex, a fault topological landscape is constructed, the connection relationship and transition paths of different regions in the fault topological landscape are analyzed, the internal relationship and evolution path between fault modes are determined, the persistence homology of the simple complex is calculated, the appearance and disappearance of topological features are recorded to form a persistence graph, and based on the persistence graph, the topological features of the fault mode are described to generate a multi-scale persistence homology feature descriptor.

[0146] Design a neural network structure including an input layer, a hidden layer, and an output layer; determine the number of neurons in each layer according to the characteristics of the system to be diagnosed; establish random initial connection weights between neurons to represent the interaction strength between neurons; the initial connection weights can be generated randomly or set based on prior knowledge.

[0147] The abnormal feature vector and quantum state entanglement matrix obtained in the anomaly detection stage are input into the biologically inspired self-organizing neural network; the abnormal feature vector represents the characteristic representation of the fault mode, and the quantum state entanglement matrix represents the correlation between different modes; the input data triggers the activation of neurons in the neural network, and the output of the neurons is calculated according to the activation function of the neurons.

[0148] The connection weights between neurons are dynamically adjusted according to the activation of neurons. A competitive learning mechanism is used to enhance or weaken the connection weights according to the activity levels of adjacent neurons. The connection weights between neurons with high activity levels are enhanced, while the connection weights between neurons with low activity levels are suppressed. By dynamically adjusting the connection weights, the neural network can adaptively learn the patterns and features of the input data.

[0149] During the network training process, the activity level of each neuron is continuously monitored; the connection structure between neurons is dynamically adjusted according to the synaptic growth algorithm; for neurons with high activity and large connection weights, the synaptic growth between them and adjacent neurons is promoted; for neurons with low activity and small connection weights, the synaptic growth between them and adjacent neurons is inhibited, and synaptic pruning is performed; through synaptic growth and pruning, the structure of the neural network can self-organize and adapt to the characteristics of the input data.

[0150] Through repeated iterative training, the biologically inspired self-organizing neural network converges to a stable state; in the converged state, the neural network forms a multi-layer structure, and neurons at different levels represent different abstract levels of fault modes; the bottom-level neurons represent specific fault characteristics, and the high-level neurons represent more abstract and generalized fault modes; by analyzing the hierarchical structure of the neural network, the hierarchical representation of the fault mode can be obtained, revealing the internal organization and correlation of the fault mode.

[0151] Each neuron in a biologically inspired self-organizing neural network is regarded as a data point in a high-dimensional space. The position of the data point in the high-dimensional space is determined according to the connection weights between neurons. The k-nearest neighbor method is used to establish a connection relationship between the data points to form a simple complex representing the topological structure of the data. Topological data analysis is performed on the simple complex to calculate its homology group and persistent homology. The fault topological landscape is constructed by analyzing the generators of the homology group and the persistent intervals of the persistent homology.

[0152] In the fault topology landscape, different connected components correspond to different fault modes; the connection relationship and transition path between connected components are analyzed to reveal the internal relationship and evolution path between fault modes; the dynamic evolution process of fault modes is explored by analyzing the appearance, merging and disappearance of connected components; key topological events are identified, such as the birth, death and merging of connected components, to reveal the key nodes of fault mode evolution.

[0153] Calculate the persistence homology of simple complexes and record the appearance and disappearance of topological features at different scales; generate a persistence graph of persistence homology to show the persistence of topological features with scale changes; based on the persistence graph, extract multi-scale topological features, such as persistence interval, life cycle, etc.; combine the extracted topological features into a multi-scale persistence homology feature descriptor to comprehensively characterize the topological properties of the fault mode; use the multi-scale persistence homology feature descriptor to achieve effective representation and comparison of fault modes.

[0154] In this embodiment, the constructed self-organizing neural network realizes effective processing of abnormal characteristics and quantum states through the input layer, hidden layer and output layer, thereby improving the ability of fault mode recognition; by monitoring the activity level of neurons and dynamically adjusting the connection weights, the adaptability and learning ability of the network to complex data are enhanced; the synaptic growth algorithm is used to form a multi-level structure, which provides a clear and systematic perspective for the representation of fault modes, facilitating analysis and understanding; the relationship between data points is established based on the neuron connection weights, and the fault topology landscape is constructed through topological analysis to reveal the complex relationship between fault modes; the continuous coherent calculation and generation of continuous graphs help capture the topological characteristics of fault modes, providing a more detailed and dynamic fault analysis method.

[0155] In an optional embodiment, the synapse growth algorithm includes:

[0156] Continuously monitoring the activity of each neuron in the biologically inspired self-organizing neural network, and quantifying the activity by recording the number of activations of each neuron within a predetermined time window;

[0157] Set high activity threshold and inactive threshold to determine the state of neurons;

[0158] Taking neurons whose activity exceeds the high-activity threshold as high-activity neurons, randomly selecting positions in the surrounding space of the high-activity neurons to generate new neurons, establishing initial connections between the new neurons and the high-activity neurons, and establishing probabilistic connections with neighboring neurons of the high-activity neurons;

[0159] Neurons with activity levels lower than the inactive threshold are regarded as inactive neurons, and the inactive neurons are marked as being in a state to be removed. When an inactive neuron is marked as being in a state to be removed multiple times within a plurality of consecutive predetermined time windows, the inactive neuron is removed from the biologically inspired self-organizing neural network, and the connection weights with neighboring neurons are proportionally distributed to surrounding neurons;

[0160] Update the connection matrix of the biologically inspired self-organizing neural network according to the newly added and removed neurons, and recalculate the positional relationship of all neurons in the biologically inspired self-organizing neural network;

[0161] Based on the overall activity status, the high activity threshold and the inactivity threshold are periodically adjusted.

[0162] During the operation of the biologically inspired self-organizing neural network, the activity of each neuron is continuously monitored; a predetermined time window, such as 1 second or 10 seconds, is set to count the number of neuron activations; within each time window, the number of activations of each neuron is recorded, that is, the number of times the output of the neuron exceeds the activation threshold; the activity of the neuron is quantified based on the number of activations, and the activity can be expressed as the ratio of the number of activations to the length of the time window.

[0163] Two activity thresholds, namely the high activity threshold and the inactive threshold, are set to judge the state of neurons; the high activity threshold represents the lower limit of the activity of neurons in a highly active state, such as 0.8 or 0.9; the inactive threshold represents the upper limit of the activity of neurons in an inactive state, such as 0.1 or 0.2; the activity threshold can be adjusted according to specific problems and data characteristics to adapt to different application scenarios.

[0164] Neurons whose activity exceeds the high-activity threshold are called high-activity neurons. New neurons are generated by randomly selecting positions in the space around high-activity neurons. The newly generated neurons establish initial connections with high-activity neurons, and the connection weights are set to random values. The new neurons also establish probabilistic connections with neighboring neurons of high-activity neurons, and the connection probability is inversely proportional to the distance. By generating new neurons, the network's expressive power is increased, which helps capture new patterns and features in the data.

[0165] Neurons whose activity is lower than the inactive threshold are called inactive neurons; inactive neurons are marked as to-be-removed, indicating that the neuron may need to be removed; a number of consecutive time windows, such as 3 or 5, is set to determine the long-term inactivity of neurons; if a neuron is marked as to-be-removed multiple times in multiple consecutive time windows, it will be removed from the network; when removing inactive neurons, the connection weights between it and neighboring neurons are proportionally distributed to the surrounding neurons; by removing inactive neurons, the network structure is simplified and the impact of redundancy and noise is reduced.

[0166] According to the newly added and removed neurons, the connection matrix of the biologically inspired self-organizing neural network is updated; the connection matrix represents the connection relationship and weights between neurons, and needs to be adjusted accordingly according to the changes in neurons; the positional relationship of all neurons in the high-dimensional space is recalculated, considering the impact of the newly added and removed neurons on the network topology; the updated connection matrix and positional relationship reflect the dynamic changes of the network structure, which helps to maintain the adaptability of the network.

[0167] According to the activity status of the entire network, the high activity threshold and the inactivity threshold are adjusted periodically; if the overall activity of the network is high, the high activity threshold can be appropriately increased to limit the generation rate of new neurons; if the overall activity of the network is low, the inactivity threshold can be appropriately lowered to slow down the removal rate of inactive neurons; by dynamically adjusting the activity threshold, the network growth and pruning process can be synchronized with the characteristics and changes of the data.

[0168] In this embodiment, by continuously monitoring the number of activations of each neuron and quantifying the activity, the network's responsiveness to state changes is enhanced; setting high activity and inactive thresholds helps to quickly identify and activate important neurons, while effectively removing inefficient neurons and optimizing the network structure; generating new neurons based on highly active neurons promotes network expansion and self-optimization, and improves learning and adaptability; when removing inactive neurons, the connection weights are allocated proportionally to ensure the continuity of information transmission and the overall stability of the network; regularly adjusting the activity threshold ensures that the network adapts to different data characteristics and dynamic environments, and improves overall performance and flexibility.

[0169] S103. Input the multi-scale continuous homology feature descriptors into the pre-built fuzzy cognitive graph reasoning system, model the causal relationship between concept nodes, generate a fault scenario model including concept node activation values ​​and relationship weight matrix, diagnose the fault based on the fault scenario model using nonlinear fuzzy reasoning algorithm, realize online learning in combination with adaptive resonance theory, update the fault diagnosis knowledge base, obtain the diagnosis results, and generate a fault diagnosis report including fault type, fault location and fault severity.

[0170] In this embodiment, a fuzzy cognitive graph reasoning system is used to transform multi-scale continuous coherent feature descriptors into fault scenario models, thereby improving the understanding of complex fault relationships. A nonlinear fuzzy reasoning algorithm is used for fault diagnosis, thereby improving the accuracy and flexibility of diagnosis and adapting to different fault characteristics. Adaptive resonance theory is combined to implement online learning, thereby ensuring that the fault diagnosis knowledge base is continuously updated to adapt to new data and fault modes. A detailed fault diagnosis report is generated, covering the fault type, location, and severity, providing clear decision support information for subsequent processing and maintenance.

[0171] In an optional embodiment, the multi-scale continuous homology feature descriptor is input into a pre-built fuzzy cognitive graph reasoning system, and the causal relationship between concept nodes is modeled to generate a fault scenario model including concept node activation values ​​and a relationship weight matrix. Based on the fault scenario model, the fault is diagnosed using a nonlinear fuzzy reasoning algorithm, and online learning is realized in combination with the adaptive resonance theory. The fault diagnosis knowledge base is updated to obtain the diagnosis result, and a fault diagnosis report including the fault type, fault location and fault severity is generated, including:

[0172] Construct a fuzzy cognitive graph reasoning system, which includes multiple concept nodes and inter-node links. The concept nodes represent fault characteristics and fault states, and the inter-node links represent the causal impact of faults.

[0173] Mapping the multi-scale continuous homology feature descriptors to the concept nodes of the fuzzy cognitive graph reasoning system and assigning initial activation values ​​to the corresponding concept nodes; analyzing the mutual influence between the concept nodes, determining the connection strength and influence direction between the nodes, and establishing a causal relationship model;

[0174] Based on the initial activation value of the concept node and the causal relationship model, a fault scenario model is generated, wherein the fault scenario model includes a concept node activation value vector and a relationship weight matrix;

[0175] The fault scenario model is analyzed using a nonlinear fuzzy inference algorithm, wherein the analysis includes initializing a system state, applying a relationship weight matrix to calculate mutual influences between nodes, and updating node activation values ​​using a nonlinear conversion function, and repeatedly iterating the analysis until the fault scenario model reaches a stable state, thereby obtaining a final concept node activation value;

[0176] Based on the final concept node activation value, diagnose the fault, determine the fault type, location and severity, and obtain a preliminary fault result;

[0177] Combining adaptive resonance theory to realize online learning, compare the similarity between the preliminary fault result and the known fault mode, update the fault diagnosis knowledge base according to the similarity, update the existing fault mode information when the similarity is higher than the preset threshold, and add the preliminary fault result as a new fault mode to the fault diagnosis knowledge base when the similarity is lower than the preset threshold;

[0178] Based on the preliminary fault diagnosis result and the updated fault diagnosis knowledge base, a final diagnosis result is obtained, and a fault diagnosis report is generated according to the final diagnosis result. The fault diagnosis report includes the fault type, fault location and fault severity.

[0179] Identify key concepts in the field of fault diagnosis, such as fault characteristics, fault status, etc., as concept nodes of the fuzzy cognitive map; analyze the causal relationship and influence degree between concept nodes, and determine the links between nodes and their directions; use fuzzy linguistic variables to describe the influence intensity between nodes, such as "strong", "medium", "weak", etc., to indicate the degree of causal relationship; construct a fuzzy cognitive map containing multiple concept nodes and links between nodes to form a knowledge representation of fault diagnosis.

[0180] The multi-scale persistent coherent feature descriptors extracted from the biologically inspired self-organizing neural network are mapped to the concept nodes of the fuzzy cognitive map; each concept node corresponds to one or more persistent coherent features, indicating the fault feature or state represented by the node; according to the value of the persistent coherent feature, the initial activation value is assigned to the corresponding concept node, indicating the initial strength of the feature or state.

[0181] Analyze the mutual influence and causal relationship between conceptual nodes, determine the connection strength and influence direction of the links between nodes; use fuzzy rules to describe the causal relationship between nodes, such as "if the activation value of node A is high, the activation value of node B will also increase"; adjust and optimize fuzzy rules based on domain knowledge and expert experience so that they can accurately reflect the causal relationship of the fault; form a complete causal relationship model to represent the mutual influence and propagation between fault characteristics and states.

[0182] Based on the initial activation value of the concept node and the causal relationship model, a model representing the specific fault scenario is generated; the fault scenario model includes two key components: the concept node activation value vector and the relationship weight matrix; the concept node activation value vector represents the activation degree of each node in the current fault scenario, reflecting the strength of the fault characteristics and status; the relationship weight matrix represents the weight of the link between nodes, reflecting the influence of the causal relationship.

[0183] The fault scenario model is analyzed using a nonlinear fuzzy reasoning algorithm to simulate the evolution and propagation of the fault; the system state is initialized, and the concept node activation value vector is used as the starting point of the reasoning process; the relationship weight matrix is ​​applied to calculate the comprehensive effect of each node on other nodes and update the node activation value; a nonlinear transformation function, such as an S-type function or a hyperbolic tangent function, is used to perform a nonlinear transformation on the node activation value to simulate the nonlinear characteristics of the fault; the reasoning process is repeated until the fault scenario model reaches a stable state and the final concept node activation value is obtained.

[0184] Based on the final conceptual node activation value, a preliminary fault diagnosis is performed; nodes with higher activation values ​​are analyzed to determine possible fault types and locations; the severity of the fault is estimated based on the relative size of the node activation values; and preliminary fault diagnosis results are generated, including the fault type, location, and severity.

[0185] Adaptive resonance theory is used to realize online learning and updating of fault diagnosis knowledge; the similarity between the preliminary fault diagnosis results and the known fault modes is compared, and their distance or correlation in the feature space is calculated; if the similarity is higher than the preset threshold, it means that the preliminary diagnosis result matches the known fault mode, and the information of the existing fault mode, such as fault frequency, severity, etc., is updated; if the similarity is lower than the preset threshold, it means that the preliminary diagnosis result is a new fault mode, which is added to the fault diagnosis knowledge base to expand the coverage of the knowledge base; through adaptive resonance learning, the fault diagnosis system can continuously accumulate and update knowledge and adapt to new fault modes.

[0186] Generate final fault diagnosis results by comprehensively considering preliminary fault diagnosis results and updated fault diagnosis knowledge base; verify and correct preliminary diagnosis results, refer to information in knowledge base to improve diagnosis accuracy and reliability; generate detailed fault diagnosis report, including fault type, location, severity, possible causes and recommended measures; provide diagnosis report to maintenance personnel and decision makers to support fault handling and prevention.

[0187] In this embodiment, by constructing a fuzzy cognitive graph reasoning system, fault characteristics and states are clearly represented, and the systematic nature of fault analysis is enhanced; the links between nodes represent causal effects, which helps to understand the interactions and impact directions between faults and improve the depth of fault diagnosis; the nonlinear fuzzy reasoning algorithm is used to achieve stable state analysis of the fault scenario model to ensure accurate conceptual node activation values; based on the final activation value, the fault type, location and severity can be accurately determined to improve the reliability of fault diagnosis; online learning is carried out in combination with adaptive resonance theory, and the fault diagnosis knowledge base is dynamically updated to ensure that the system adapts to new fault modes; a detailed fault diagnosis report is generated, providing comprehensive information including fault type, location and severity to support subsequent decision-making and maintenance.

[0188] Figure 2 FIG. 1 is a schematic diagram of a structural diagram of a uniform light guide lighting fault diagnosis system for cultural tourism based on artificial intelligence according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0189] The first unit is used to deploy a multimodal perception network, collect multimodal data, process the multimodal data through a quantum annealing multimodal anomaly detection algorithm, detect data points exceeding a preset threshold in a high-dimensional feature space, and output an abnormal feature vector and a quantum state entanglement degree matrix;

[0190] The second unit is used to input the abnormal feature vector and the quantum state entanglement matrix into the biologically inspired self-organizing neural network, generate a hierarchical representation of the fault mode through dynamic weight adjustment and synaptic growth algorithm, and build a fault topology landscape based on the hierarchical representation through topological data analysis, determine the internal relationship and evolution path between the fault modes, and generate a multi-scale continuous coherent feature descriptor;

[0191] The third unit is used to input the multi-scale continuous homology feature descriptors into the pre-built fuzzy cognitive graph reasoning system, model the causal relationship between concept nodes, generate a fault scenario model including concept node activation values ​​and relationship weight matrix, and diagnose the fault based on the fault scenario model using nonlinear fuzzy reasoning algorithm, combine adaptive resonance theory to realize online learning, update the fault diagnosis knowledge base, obtain the diagnosis results, and generate a fault diagnosis report including fault type, fault location and fault severity.

[0192] According to a third aspect of the embodiments of the present invention,

[0193] An electronic device is provided, comprising:

[0194] processor;

[0195] a memory for storing processor-executable instructions;

[0196] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0197] A fourth aspect of the embodiments of the present invention is:

[0198] 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.

[0199] 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.

[0200] 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 method for diagnosing uniform light guide lighting faults in cultural tourism based on artificial intelligence, characterized in that: include: Deploy a multimodal perception network to collect multimodal data of the cultural tourism uniform light-guiding lighting system, process the multimodal data through a quantum annealing multimodal anomaly detection algorithm, detect data points exceeding a preset threshold in a high-dimensional feature space, and output anomaly feature vectors and a quantum state entanglement degree matrix; The abnormal feature vector and quantum state entanglement matrix are input into a biologically inspired self-organizing neural network, and a hierarchical representation of the fault mode is generated through dynamic weight adjustment and synaptic growth algorithm. Based on the hierarchical representation, a fault topology landscape is constructed through topological data analysis to determine the internal relationship and evolution path between the fault modes, and generate a multi-scale continuous coherent feature descriptor; The multi-scale continuous homology feature descriptors are input into the pre-built fuzzy cognitive graph reasoning system, and the causal relationship between concept nodes is modeled to generate a fault scenario model including concept node activation values ​​and relationship weight matrix. Based on the fault scenario model, the nonlinear fuzzy reasoning algorithm is used to diagnose the fault, and the adaptive resonance theory is combined to realize online learning, update the fault diagnosis knowledge base, obtain the diagnosis results, and generate a fault diagnosis report including fault type, fault location and fault severity. Deploy a multimodal perception network to collect multimodal data of the cultural tourism uniform light-guiding lighting system, process the multimodal data through the quantum annealing multimodal anomaly detection algorithm, detect data points exceeding the preset threshold in the high-dimensional feature space, and output the abnormal feature vector and quantum state entanglement degree matrix including: The multimodal sensing network includes a spectrum analyzer, an acoustic sensor and an electromagnetic field detector, wherein the spectrum analyzer is used to collect visible spectrum and near infrared spectrum data of the light guide illumination, the acoustic sensor is used to collect ultrasonic signals and infrasonic signals, and the electromagnetic field detector is used to measure the electromagnetic field intensity and distribution; Generate a multimodal data stream through the multimodal perception network, perform wavelet transform denoising, Fourier transform spectrum analysis and adaptive filtering on the multimodal data stream to obtain pre-processed multimodal data; Construct a quantum annealing Hamiltonian, encode multimodal data into the initial superposition state of quantum bits as the starting point for anomaly detection, use adiabatic quantum computing method, gradually adjust annealing parameters to make the initial superposition state evolve toward the target state, and obtain the optimal quantum state. Through quantum measurement operations, project the optimal quantum state to the classical state to obtain preliminary anomaly detection results. Based on the preliminary anomaly detection results, multimodal data fusion is performed, and the features of different modes are nonlinearly combined and dimensionally reduced using a deep belief network to obtain fused features; according to historical data and the current light guide lighting state, the anomaly detection threshold is adaptively adjusted, a dynamic threshold model is constructed, the fused features are compared with the dynamic threshold, potential anomalies are marked, and a density clustering algorithm is applied to analyze the distribution of the potential anomalies in the high-dimensional feature space and identify anomaly clusters; an anomaly association network is constructed based on graph theory to mine the spatiotemporal relationship between anomalies in the anomaly clusters; Based on the abnormal cluster and the abnormal association network, a recurrent neural network model is trained to capture the temporal characteristics of the abnormal pattern, and key abnormal features are extracted from the recurrent neural network model in combination with an attention mechanism to generate an abnormal feature vector; The quantum mutual information between different modal anomalies is calculated, and the quantum state entanglement matrix is ​​constructed by quantifying the correlation strength of multi-modal anomalies, and the anomaly feature vector and the quantum state entanglement matrix are output.

2. The method according to claim 1, characterized in that Construct a quantum annealing Hamiltonian, encode the multimodal data into the initial superposition state of quantum bits as the starting point for anomaly detection, use the adiabatic quantum computing method, gradually adjust the annealing parameters to make the initial superposition state evolve toward the target state, and obtain the optimal quantum state. Through quantum measurement operations, project the optimal quantum state to the classical state, and obtain preliminary anomaly detection results including: Extract features from multimodal data to obtain feature vectors, map the feature vectors to the spin configuration of quantum bits, and construct a problem Hamiltonian, where each feature vector corresponds to the spin state of a quantum bit; construct a quadratic energy function to represent the problem Hamiltonian as a combination of interaction terms and monomer terms between quantum bits; select the transverse magnetic field as the driving Hamiltonian, and use the Pauli X operator to represent each quantum bit; Combining the problem Hamiltonian and the driving Hamiltonian, controlling the relative strength of the problem Hamiltonian and the driving Hamiltonian in combination with the time-varying coefficient, determining the total Hamiltonian, and obtaining a quantum annealing Hamiltonian; Based on the quantum annealing Hamiltonian, the multimodal data is normalized and converted into quantum state amplitude, the quantum state amplitude is converted into a quantum superposition state using a quantum gate operation, an initial quantum state is constructed, an annealing time and an annealing scheduling function are set, the time-varying coefficient in the total Hamiltonian is adjusted, and a time-varying total Hamiltonian is obtained, the initial quantum state and the time-varying total Hamiltonian are substituted into the Schrödinger equation, the Schrödinger equation is solved, and the evolution trajectory of the quantum state over time is calculated; During the quantum state evolution process, the system energy change rate is monitored. When the energy change rate is lower than a preset threshold, a quantum tunneling operator is introduced into the evolution trajectory to change the current quantum state; a quantum fluctuation operator is periodically applied to the evolution trajectory to introduce random phase disturbances; The quantum bits of different modes are entangled together to obtain an entangled state, and controlled quantum gate operations are applied to the entangled state to perform parallel computing, calculate the entanglement entropy, quantify the degree of correlation between the modes, determine the quantum measurement basis based on the entangled state, perform projection measurement, project the final quantum state onto the quantum measurement basis, repeatedly perform the projection measurement, obtain the probability distribution of the measurement result, calculate the discreteness of the probability distribution, determine whether there is an abnormality according to a preset discreteness threshold, and output a preliminary abnormality detection result.

3. The method according to claim 2, characterized in that During the quantum state evolution process, the system energy change rate is monitored, and when the energy change rate is lower than a preset threshold, a quantum tunneling operator is introduced into the evolution trajectory to change the current quantum state; and periodically applying a quantum fluctuation operator to the evolution trajectory to introduce a random phase disturbance also includes: During the quantum state evolution process, the changes in system energy and quantum state are continuously monitored, and the system energy change rate and quantum state overlap are calculated; Based on the energy change rate and quantum state overlap of the system, the application intensity and frequency of the quantum tunneling operator and the quantum fluctuation operator are dynamically adjusted to form an adaptive quantum operator application strategy, wherein the adaptive quantum operator application strategy includes: When the system energy change rate is lower than a first preset threshold, the application intensity of the quantum tunneling operator is increased to jump out of the local minimum; when the quantum state overlap is higher than a second preset threshold, the application frequency of the quantum fluctuation operator is increased; According to the adaptive quantum operator application strategy, in the evolution trajectory of the quantum state, a quantum tunneling operator is applied to change the current quantum state, and a quantum fluctuation operator is applied to introduce a random phase disturbance; According to the preset period, periodically monitor the changing trend of the system energy change rate and quantum state overlap; When the rate of decrease of the energy change rate of the system is lower than a third preset threshold, or the amplitude of change of the quantum state overlap is lower than a fourth preset threshold, adjusting the parameters of the adaptive quantum operator application strategy, including adjusting the first preset threshold, the second preset threshold, the application intensity adjustment amplitude of the quantum tunneling operator, and the application frequency adjustment amplitude of the quantum fluctuation operator; Repeat the steps until the preset number of iterations is reached.

4. The method according to claim 1, characterized in that The abnormal feature vector and quantum state entanglement matrix are input into the biologically inspired self-organizing neural network. Through dynamic weight adjustment and synaptic growth algorithm, a hierarchical representation of the fault mode is generated. Based on the hierarchical representation, through topological data analysis, a fault topological landscape is constructed to determine the internal relationship and evolution path between the fault modes. The multi-scale continuous coherent feature descriptors are generated, including: Constructing the biologically inspired self-organizing neural network, including an input layer, a hidden layer, and an output layer, each layer consisting of a plurality of neurons, and establishing random initial connection weights between neurons; The abnormal feature vector and the quantum state entanglement degree matrix are transmitted into the biologically inspired self-organizing neural network to trigger neuron activation, and the connection weights are dynamically adjusted according to the activity levels of adjacent neurons; During the network training process, the activity of neurons is monitored, and synaptic growth and pruning are completed based on the synaptic growth algorithm. Through repeated iterative training, a multi-level structure is self-organized to generate a hierarchical representation of the fault mode. Based on the hierarchical representation of the failure mode, each neuron in the biologically inspired self-organizing neural network is regarded as a data point in a high-dimensional space, and the position is determined according to the connection weight of the neuron; A connection relationship is established between the data points using the k-nearest neighbor method to form a simple complex representing the topological structure of the data, topological data analysis is performed on the simple complex, a fault topological landscape is constructed, the connection relationship and transition paths of different regions in the fault topological landscape are analyzed, the internal relationship and evolution path between fault modes are determined, the persistence homology of the simple complex is calculated, the appearance and disappearance of topological features are recorded to form a persistence graph, and based on the persistence graph, the topological features of the fault mode are described to generate a multi-scale persistence homology feature descriptor.

5. The method according to claim 4, characterized in that The synapse growth algorithm includes: Continuously monitoring the activity of each neuron in the biologically inspired self-organizing neural network, and quantifying the activity by recording the number of activations of each neuron within a predetermined time window; Set high activity threshold and inactive threshold to determine the state of neurons; Taking neurons whose activity exceeds the high-activity threshold as high-activity neurons, randomly selecting positions in the surrounding space of the high-activity neurons to generate new neurons, establishing initial connections between the new neurons and the high-activity neurons, and establishing probabilistic connections with neighboring neurons of the high-activity neurons; Neurons with activity levels lower than the inactive threshold are regarded as inactive neurons, and the inactive neurons are marked as being in a state to be removed. When an inactive neuron is marked as being in a state to be removed multiple times within a plurality of consecutive predetermined time windows, the inactive neuron is removed from the biologically inspired self-organizing neural network, and the connection weights with neighboring neurons are proportionally distributed to surrounding neurons; Update the connection matrix of the biologically inspired self-organizing neural network according to the newly added and removed neurons, and recalculate the positional relationship of all neurons in the biologically inspired self-organizing neural network; Based on the overall activity status, the high activity threshold and the inactivity threshold are periodically adjusted.

6. The method according to claim 1, characterized in that The multi-scale continuous homology feature descriptors are input into the pre-built fuzzy cognitive graph reasoning system, and the causal relationship between concept nodes is modeled to generate a fault scenario model containing concept node activation values ​​and relationship weight matrices. Based on the fault scenario model, the nonlinear fuzzy reasoning algorithm is used to diagnose the fault, and the adaptive resonance theory is combined to realize online learning, update the fault diagnosis knowledge base, obtain the diagnosis results, and generate a fault diagnosis report including fault type, fault location and fault severity. Construct a fuzzy cognitive graph reasoning system, which includes multiple concept nodes and inter-node links. The concept nodes represent fault characteristics and fault states, and the inter-node links represent the causal impact of faults. Mapping the multi-scale continuous homology feature descriptors to the concept nodes of the fuzzy cognitive graph reasoning system and assigning initial activation values ​​to the corresponding concept nodes; analyzing the mutual influence between the concept nodes, determining the connection strength and influence direction between the nodes, and establishing a causal relationship model; Based on the initial activation value of the concept node and the causal relationship model, a fault scenario model is generated, wherein the fault scenario model includes a concept node activation value vector and a relationship weight matrix; The fault scenario model is analyzed using a nonlinear fuzzy inference algorithm, wherein the analysis includes initializing a system state, applying a relationship weight matrix to calculate mutual influences between nodes, and updating node activation values ​​using a nonlinear conversion function, and repeatedly iterating the analysis until the fault scenario model reaches a stable state, thereby obtaining a final concept node activation value; Based on the final concept node activation value, diagnose the fault, determine the fault type, location and severity, and obtain a preliminary fault result; Combining adaptive resonance theory to realize online learning, compare the similarity between the preliminary fault result and the known fault mode, update the fault diagnosis knowledge base according to the similarity, update the existing fault mode information when the similarity is higher than the preset threshold, and add the preliminary fault result as a new fault mode to the fault diagnosis knowledge base when the similarity is lower than the preset threshold; Based on the preliminary fault result and the updated fault diagnosis knowledge base, a final diagnosis result is obtained, and a fault diagnosis report is generated according to the final diagnosis result. The fault diagnosis report includes the fault type, fault location and fault severity.

7. A uniform light guide lighting fault diagnosis system for cultural tourism based on artificial intelligence, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The first unit is used to deploy a multimodal perception network, collect multimodal data, process the multimodal data through a quantum annealing multimodal anomaly detection algorithm, detect data points exceeding a preset threshold in a high-dimensional feature space, and output an abnormal feature vector and a quantum state entanglement degree matrix; The second unit is used to input the abnormal feature vector and the quantum state entanglement matrix into the biologically inspired self-organizing neural network, generate a hierarchical representation of the fault mode through dynamic weight adjustment and synaptic growth algorithm, and build a fault topology landscape based on the hierarchical representation through topological data analysis, determine the internal relationship and evolution path between the fault modes, and generate a multi-scale continuous coherent feature descriptor; The third unit is used to input the multi-scale continuous homology feature descriptors into the pre-built fuzzy cognitive graph reasoning system, model the causal relationship between concept nodes, generate a fault scenario model including concept node activation values ​​and relationship weight matrix, and diagnose the fault based on the fault scenario model using nonlinear fuzzy reasoning algorithm, combine adaptive resonance theory to realize online learning, update the fault diagnosis knowledge base, obtain the diagnosis results, and generate a fault diagnosis report including fault type, fault location and fault severity.

8. 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 according to any one of claims 1 to 6.

9. 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 6 is implemented.

Images