CN-121997127-A - Camera operation state evaluation method and system based on multisource sensing data fusion

Abstract

The invention provides a camera operation state evaluation method and a camera operation state evaluation system based on multi-source sensing data fusion, which relate to the technical field of camera operation state evaluation and comprise the steps of acquiring multi-source sensing data; the method comprises the steps of obtaining standardized preprocessing data, calculating the association degree of each extracted feature and a state of a camera fault through mutual information entropy to form a key feature set, calculating the influence weight of each parameter in the energy conversion process through finite element simulation, constructing a layered fusion frame by combining the key feature set, voiceprint-vibration fusion features and an inter-equipment parameter association matrix to obtain each feature layer, calculating fusion subvectors of each feature layer through a weighted fusion algorithm to obtain a global fusion feature vector, inputting the global fusion feature vector into a camera state evaluation model which is trained in advance, and outputting a camera running state result. The invention comprehensively covers the operation state monitoring of the electric, mechanical, thermal and cooling systems of the camera, and provides technical support for lean operation and maintenance of the camera.

Inventors

• LUO WEI
• YUAN JINGYU
• LI HAO
• YANG WENZHE

Assignees

• 国网能源哈密煤电有限公司

Dates

Publication Date

20260508

Application Date

20260112

Claims (10)

1. 1. The camera running state evaluation method based on multi-source sensing data fusion is characterized by comprising the following steps of: Constructing a space-time alignment model based on the multi-source sensing data, unifying the data with different acquisition frequencies to the same time dimension, generating a self-adaptive filtering threshold value by combining a normal operation data feature library of the camera, filtering impulse noise, harmonic interference and environmental noise in the data, and completing dimension unification of all parameters through range normalization to obtain standardized preprocessing data; Aiming at standardized preprocessing data, a specific analysis method is adopted to extract state characteristics, wherein the specific analysis method comprises the steps of carrying out wavelet packet transformation on electric quantity data, extracting frequency band energy duty ratio and characteristic frequency amplitude, carrying out trend analysis on temperature data, extracting temperature change rate and gradient difference value, carrying out Hilbert-Huang transformation on vibration data, extracting instantaneous frequency and amplitude, carrying out Mel frequency cepstrum coefficient analysis on voiceprint data, extracting spectrum envelope and formant frequency, carrying out statistical analysis on cooling system parameters, extracting mean value and variance of the cooling system parameters, calculating the association degree of each extracted characteristic and a phase-change machine fault state through mutual information entropy after the characteristic extraction of various data, and screening the characteristic that the association degree reaches a preset threshold value to form a key characteristic set, wherein the phase-change machine fault state comprises stator winding faults and bearing bush abrasion faults; The method comprises the steps of extracting vibration characteristics and voiceprint characteristics from a key characteristic set, calculating time domain correlation and frequency domain coupling degree of the two characteristics, constructing a cross-modal characteristic mapping relation based on a coupling result to generate voiceprint-vibration fusion characteristics, constructing a physical correlation model based on a phase-modulator energy transfer chain, incorporating the electrical energy characteristics, temperature characteristics, cooling system characteristics and voiceprint-vibration fusion characteristics into the physical correlation model, calculating influence weights of all parameters in an energy conversion process through finite element simulation, generating an inter-device parameter correlation matrix based on the weights, and constructing a layered fusion frame by combining the key characteristic set, the voiceprint-vibration fusion characteristics and the inter-device parameter correlation matrix to obtain all characteristic layers, wherein all the characteristic layers are divided into an electrical characteristic layer, a temperature characteristic layer, an acoustic vibration fusion layer and a cooling characteristic layer; Calculating correlation coefficients of the fusion sub-vectors of the feature layers and the overall running state of the camera through an attention mechanism, distributing interlayer weights according to the correlation coefficients, and carrying out weighted aggregation on the fusion sub-vectors to obtain a global fusion feature vector; the global fusion feature vector is input into a pre-trained camera state evaluation model, wherein the camera state evaluation model is constructed based on a deep confidence network, and is trained until a loss function reaches a preset standard through camera normal operation data, simulation fault data and historical fault data, the global fusion feature vector is subjected to state recognition through the camera state evaluation model, and a camera operation state result is output, wherein the operation state result comprises a normal state, an abnormal state and a fault type, and the fault type comprises a stator winding fault, a bearing bush abrasion fault, a cooling system blocking fault and a bearing abnormal sound fault.
2. 2. The method for evaluating the running state of the camera based on multi-source sensing data fusion is characterized by comprising the steps of carrying out wavelet packet transformation on electric quantity data, extracting frequency band energy duty ratio and characteristic frequency amplitude, carrying out trend analysis on temperature data, extracting temperature change rate and gradient difference value, carrying out Hilbert-Huang transformation on vibration data, extracting instantaneous frequency and amplitude, carrying out mel frequency cepstrum coefficient analysis on voiceprint data, extracting spectrum envelope and formant frequency, carrying out statistical analysis on parameters of a cooling system, extracting mean value and variance of the parameters, calculating the association degree of each extracted feature and the fault state of the camera through mutual information entropy after feature extraction of various data is completed, screening the feature that the association degree reaches a preset threshold value, and forming a key feature set, wherein the fault state of the camera comprises stator winding faults and bearing bush wear faults, and the method comprises the following steps: Performing characteristic extraction on different types of data in a standardized preprocessing data matrix, performing frequency domain analysis on electric quantity data by adopting wavelet packet transformation, wherein the number of decomposition layers is 3, db4 wavelet basis is selected, frequency band energy duty ratio and characteristic frequency amplitude are extracted, performing sliding window trend analysis on temperature data, wherein the window size is set to 10min, temperature change rate and temperature gradient difference value are extracted, performing time-frequency domain analysis on vibration data by adopting Hilbert-yellow transformation, the number of decomposition layers is 5, instantaneous frequency and instantaneous amplitude are extracted, performing acoustic characteristic analysis on voiceprint data by adopting mel frequency cepstrum coefficient, the number of mel filter bank is 24, cepstrum coefficient order is 12, frequency spectrum envelope and formant parameters are extracted, the formant parameters comprise first to third formant frequencies, calculating parameter distribution characteristics by adopting a statistical analysis method on cooling system parameters, setting a time window to be 1min, and extracting mean value and variance to form an initial characteristic set; according to the initial feature set, a fault association matrix is constructed, the row dimension of the matrix is the total number of features in the initial feature set, the column dimension is the type of a phase-change machine fault state, the phase-change machine fault state comprises a stator winding fault and a bearing bush abrasion fault, and the association degree between each extracted feature and the phase-change machine fault state is calculated by adopting mutual information entropy, wherein the calculation formula is as follows: in the formula, Is characterized by And fault state Is used for the correlation degree of the number of the pieces of the data, As a feature-fault joint probability distribution, Is characterized by Is provided with a distribution of the edge probability of (c), Is in fault state Is a boundary probability distribution of (1); And setting a relevance threshold, and screening features with relevance greater than or equal to the relevance threshold to form a key feature set, wherein the relevance threshold is determined according to historical fault data of the camera and evaluation accuracy requirements.
3. 3. The method for evaluating the running state of the camera based on multi-source sensing data fusion according to claim 1 is characterized by comprising the steps of extracting vibration features and voiceprint features from a key feature set, calculating time domain correlation and frequency domain coupling degree of the two features, constructing a cross-modal feature mapping relation based on a coupling result, generating voiceprint-vibration fusion features, constructing a physical association model based on a camera energy transfer chain, incorporating the electrical energy features, temperature features, cooling system features and voiceprint-vibration fusion features into the physical association model, calculating influence weights of all parameters in an energy conversion process through finite element simulation, generating an inter-device parameter association matrix based on the weights, and constructing a layered fusion frame by combining the key feature set, the voiceprint-vibration fusion features and the inter-device parameter association matrix, wherein all feature layers are divided into an electrical feature layer, a temperature feature layer, a voiceprint fusion layer and a cooling feature layer, and the method comprises the steps of: Extracting vibration characteristics and voiceprint characteristics from the key characteristic set, wherein the vibration characteristics comprise instantaneous frequency and instantaneous amplitude, the voiceprint characteristics comprise spectral envelope and formant parameters, and the calculation formula of the time domain correlation of the two is as follows: in the formula, Characterised by vibrations With voiceprint features Is used to determine the covariance of (1), And The standard deviation of the two is respectively the standard deviation of the two, Time domain correlation coefficients for both; The frequency domain coupling degree is calculated as follows: in the formula, For the cross-power spectral density of both, And Respectively the power spectral density of each of them, The frequency domain coupling degree; Constructing a cross-modal feature mapping relation based on the coupling result, and generating voiceprint-vibration fusion features, wherein the vibration features comprise instantaneous frequencies and instantaneous amplitudes, and the voiceprint features comprise frequency spectrum envelopes and formant parameters; Constructing a physical association model based on an energy transfer chain of a phase modulator, and respectively constructing a mapping of electric quantity and electromagnetic torque, association of mechanical energy and heat energy, coupling of heat energy and cooling parameters through an electromagnetic induction equation, a friction loss equation and a cooling system heat exchange equation to form a closed loop conversion model of energy transfer of the phase modulator; embedding voiceprint-vibration fusion characteristics, key characteristic concentrated electric quantity characteristics, temperature characteristics and cooling system characteristics into a model according to energy conversion nodes, wherein the electric quantity characteristics correspond to electromagnetic energy links, the voiceprint-vibration fusion characteristics correspond to mechanical energy links, the temperature characteristics correspond to heat energy accumulation links, the cooling system characteristics correspond to heat energy dissipation links, each parameter value is sequentially and independently changed through finite element simulation, the variation of total energy of the system is recorded, each parameter influence weight is calculated, and then an inter-equipment parameter association matrix is generated based on a weight matrix; The method comprises the steps of combining a key feature set, generated voiceprint-vibration fusion features and an inter-equipment parameter correlation matrix to construct a layered fusion frame, extracting feature groups corresponding to diagonal elements of the correlation matrix, dividing electrical quantity features directly related to electromagnetic energy conversion, temperature features related to heat energy accumulation, voiceprint-vibration fusion features and cooling system features related to heat energy dissipation into an electrical feature layer, a temperature feature layer, a voiceprint-vibration fusion layer and a cooling feature layer respectively according to an energy transfer chain conversion sequence, carrying out secondary verification on cross-layer features through cosine similarity, determining that features in the same feature layer belong to the same energy conversion stage, and storing the features of all layers in a matrix form.
4. 4. The method for evaluating the running state of the camera based on multi-source sensing data fusion according to claim 1, wherein the method is characterized in that a weighted fusion algorithm is adopted for each feature layer to calculate fusion sub-vectors, wherein the weight of the weighted fusion algorithm is determined according to the fault contribution degree of each feature in a historical fault case, the correlation coefficient of the fusion sub-vectors of each feature layer and the overall running state of the camera is calculated through an attention mechanism, the interlayer weight is distributed according to the correlation coefficient, and the weighted aggregation is carried out on each fusion sub-vector, so that a global fusion feature vector is obtained, and the method comprises the following steps: Based on the output electrical characteristic layer, temperature characteristic layer, sound vibration fusion layer and cooling characteristic layer, respectively executing in-layer weighted fusion for each layer characteristic, wherein the method comprises the steps of analyzing and counting historical fault cases through fault mode influence, determining fault contribution coefficients of different characteristics in each layer, wherein the frequency band energy duty ratio contribution is higher than the characteristic frequency amplitude, the temperature gradient difference contribution is higher than the temperature change rate in the temperature characteristic layer, the sound print-vibration coupling characteristic contribution is higher than the single mode characteristic in the sound vibration fusion layer, the variance contribution is higher than the mean value in the cooling characteristic layer, calculating fusion subvectors of each layer by adopting a weighted summation algorithm by taking the fault contribution coefficients of each characteristic as weights, and the calculation formula is as follows: in the formula, Is the first The layer is fused with the sub-vectors, Respectively corresponding to an electric, temperature, sound and vibration fusion and cooling characteristic layer, Is the first Failure contribution coefficient of layer k feature and satisfies , Is the first The layer k-th feature vector is used to determine, Is the first Layer feature quantity; Introducing an attention mechanism to evaluate importance of the obtained fusion sub-vectors of each layer, and constructing a health degree label vector by taking historical health state data of a camera as a reference, wherein a health state corresponding label value is 1, and a fault state corresponding label value is 0; And carrying out dimension normalization processing on the obtained fusion sub-vectors of each layer, uniformly adjusting the fusion sub-vectors into 64-dimensional vectors, and carrying out weighted aggregation on the normalized fusion sub-vectors according to the obtained attention weight to generate a global fusion feature vector, wherein the global fusion feature vector comprises multi-dimensional state information of an electric, mechanical, thermal and cooling system of a camera.
5. 5. The method for evaluating the running state of the camera based on the multi-source sensing data fusion according to claim 1, wherein the global fusion feature vector is input into a pre-trained camera state evaluation model, wherein the camera state evaluation model is constructed based on a deep belief network and is trained by camera normal running data, simulated fault data and historical fault data until a loss function reaches a preset standard: Performing dimension verification on the global fusion feature vector according to the input dimension requirement of a camera state evaluation model, after confirming that the vector format is consistent with a model input interface, importing the vector format into a feature input layer of a pre-trained camera state evaluation model, wherein the camera state evaluation model is constructed based on a deep belief network, the network comprises 3 limited Boltzmann machine pre-training layers and 1 back propagation fine tuning layer, wherein the number of nodes of the pre-training layers is sequentially set to 128, 64 and 32, optimizing network initial parameters through a contrast divergence algorithm, and adjusting a weight matrix through the back propagation algorithm until the model converges; the camera state evaluation model carries out multi-layer nonlinear transformation and feature mapping on the input global fusion feature vector, then realizes state identification through a softmax classifier of an output layer, and finally outputs a camera running state result.
6. 6. A camera operation state evaluation system based on multi-source sensing data fusion, the camera operation state evaluation method based on multi-source sensing data fusion according to claim 1, comprising: The acquisition module is used for acquiring multi-source sensing data through sensors of a camera stator, a rotor, an excitation system and a cooling system, constructing a space-time alignment model based on the multi-source sensing data, unifying data with different acquisition frequencies to the same time dimension, generating a self-adaptive filtering threshold value by combining a camera normal operation data feature library, filtering impulse noise, harmonic interference and environmental noise in the data, and completing dimension unification of all parameters through range normalization to obtain standardized preprocessing data; The extraction module is used for extracting state characteristics by adopting a specific analysis method aiming at standardized preprocessing data, wherein the specific analysis method comprises the steps of carrying out wavelet packet transformation on electric quantity data, extracting frequency band energy duty ratio and characteristic frequency amplitude, carrying out trend analysis on temperature data, extracting temperature change rate and gradient difference value, carrying out Hilbert-Huang transformation on vibration data, extracting instantaneous frequency and amplitude, carrying out Mel frequency cepstrum coefficient analysis on voiceprint data, extracting spectrum envelope and formant frequency, carrying out statistical analysis on cooling system parameters, extracting mean value and variance of the parameters, calculating the association degree of each extracted characteristic and a phase adjustment machine fault state through mutual information entropy after the characteristic extraction of various data is completed, and screening the characteristic that the association degree reaches a preset threshold value to form a key characteristic set, wherein the phase adjustment machine fault state comprises stator winding faults and bearing bush abrasion faults; The system comprises a key feature set, a calculation module, a phase-modulator energy transfer chain, a layered fusion frame, an electric feature layer, a temperature feature layer, an acoustic vibration fusion layer and a cooling feature layer, wherein the key feature set is used for extracting vibration features and voiceprint features from the key feature set, calculating the time domain correlation and frequency domain coupling degree of the two, and constructing a cross-modal feature mapping relation based on a coupling result to generate voiceprint-vibration fusion features; the aggregation module is used for calculating fusion sub-vectors of each feature layer by adopting a weighted fusion algorithm, wherein the weight of the weighted fusion algorithm is determined according to the fault contribution degree of each feature in the historical fault case; The training module is used for inputting the global fusion feature vector into a pre-trained camera state evaluation model, wherein the camera state evaluation model is constructed based on a deep confidence network and is trained to a loss function reaching a preset standard through camera normal operation data, simulation fault data and historical fault data, and carrying out state recognition on the global fusion feature vector through the camera state evaluation model to output a camera operation state result, wherein the operation state result comprises a normal state, an abnormal state and a fault type, and the fault type comprises a stator winding fault, a bearing bush abrasion fault, a cooling system blocking fault and a bearing abnormal sound fault.
7. 7. The multi-source sensory data fusion based camera operating state assessment system of claim 6, wherein the extraction module comprises: The analysis and extraction unit is used for carrying out characteristic extraction on different types of data in the standardized preprocessing data matrix, carrying out frequency domain analysis on the electric quantity data by adopting wavelet packet transformation, wherein the number of decomposition layers is 3, db4 wavelet basis is selected, and frequency band energy duty ratio and characteristic frequency amplitude are extracted; adopting sliding window trend analysis on the temperature data, wherein the window size is set to be 10min, and extracting the difference value of the temperature change rate and the temperature gradient; carrying out time-frequency domain analysis on the vibration data by adopting Hilbert-Huang transform, decomposing the vibration data into 5 layers, and extracting instantaneous frequency and instantaneous amplitude; carrying out acoustic feature analysis on voiceprint data by adopting a Mel frequency cepstrum coefficient, wherein the number of the Mel filter banks is 24, the number of the cepstrum coefficient orders is 12, and extracting spectral envelope and formant parameters, wherein the formant parameters comprise first to third formant frequencies; The construction unit is used for constructing a fault association matrix according to the initial feature set, wherein the row dimension of the matrix is the total number of features in the initial feature set, the column dimension is the type of a phase-change machine fault state, the phase-change machine fault state comprises a stator winding fault and a bearing bush abrasion fault, and the association degree of each extracted feature and the phase-change machine fault state is calculated by adopting mutual information entropy, and the calculation formula is as follows: in the formula, Is characterized by And fault state Is used for the correlation degree of the number of the pieces of the data, As a feature-fault joint probability distribution, Is characterized by Is provided with a distribution of the edge probability of (c), Is in fault state Is a boundary probability distribution of (1); the screening unit is used for setting a relevance threshold value, and screening the features with relevance greater than or equal to the relevance threshold value to form a key feature set, wherein the relevance threshold value is determined according to historical fault data of the camera and evaluation accuracy requirements.
8. 8. The multi-source sensory data fusion based camera operating state assessment system of claim 6, wherein the computing module comprises: the extraction unit is used for extracting vibration characteristics and voiceprint characteristics from the key characteristic set, wherein the vibration characteristics comprise instantaneous frequency and instantaneous amplitude, the voiceprint characteristics comprise spectrum envelope and formant parameters, and the calculation formula of the time domain correlation of the two is as follows: in the formula, Characterised by vibrations With voiceprint features Is used to determine the covariance of (1), And The standard deviation of the two is respectively the standard deviation of the two, Time domain correlation coefficients for both; The frequency domain coupling degree is calculated as follows: in the formula, For the cross-power spectral density of both, And Respectively the power spectral density of each of them, The frequency domain coupling degree; The generation unit is used for constructing a cross-modal feature mapping relation based on the coupling result and generating voiceprint-vibration fusion features, wherein the vibration features comprise instantaneous frequencies and instantaneous amplitudes, and the voiceprint features comprise frequency spectrum envelopes and formant parameters; The system comprises a building unit, a phase modulation energy transfer unit, a physical association model, a sound print-vibration fusion feature, a key feature concentrated electric quantity feature, a temperature feature and a cooling system feature, wherein the physical association model is built based on an energy transfer chain of the phase modulation, the mapping of electric quantity and electromagnetic torque, the association of mechanical energy and heat energy, the coupling of heat energy and cooling parameters are respectively built through an electromagnetic induction equation, a friction loss equation and a cooling system heat exchange equation to form a closed loop conversion model of the phase modulation energy transfer; The construction determining unit is used for constructing a layered fusion frame by combining the key feature set, the generated voiceprint-vibration fusion features and the parameter incidence matrix among devices, extracting feature groups corresponding to diagonal elements of the incidence matrix, dividing electric quantity features directly related to electromagnetic energy conversion, temperature features related to heat energy accumulation, voiceprint-vibration fusion features and cooling system features related to heat energy dissipation into an electric feature layer, a temperature feature layer, a voiceprint-vibration fusion layer and a cooling feature layer respectively according to an energy transfer chain conversion sequence, performing secondary verification on cross-layer features through cosine similarity, determining that features in the same feature layer belong to the same energy conversion stage, and storing the features of all layers in a matrix form.
9. 9. The multi-source sensory data fusion based camera operating state assessment system of claim 6, wherein the aggregation module comprises: The fusion unit is used for respectively executing in-layer weighted fusion for each layer of characteristics based on the output electrical characteristic layer, the temperature characteristic layer, the sound vibration fusion layer and the cooling characteristic layer, wherein the method comprises the steps of analyzing and counting historical fault cases through fault mode influence, determining fault contribution coefficients of different characteristics in each layer, wherein the frequency energy duty contribution is higher than the characteristic frequency amplitude in the electrical characteristic layer, the temperature gradient difference contribution is higher than the temperature change rate in the temperature characteristic layer, the sound print-vibration coupling characteristic contribution is higher than the single mode characteristic in the sound vibration fusion layer, the variance contribution is higher than the mean value in the cooling characteristic layer, calculating fusion subvectors of each layer by adopting a weighted summation algorithm by taking the fault contribution coefficients of each characteristic as weights, and the calculation formula is as follows: in the formula, Is the first The layer is fused with the sub-vectors, Respectively corresponding to an electric, temperature, sound and vibration fusion and cooling characteristic layer, Is the first Failure contribution coefficient of layer k feature and satisfies , Is the first The layer k-th feature vector is used to determine, Is the first Layer feature quantity; The evaluation unit is used for introducing an attention mechanism to evaluate the importance of the obtained fusion sub-vectors of each layer, and constructing a health degree label vector by taking historical health state data of the camera as a reference, wherein the health state corresponding label value is 1, and the fault state corresponding label value is 0; The generation unit is used for carrying out dimension normalization processing on the obtained fusion sub-vectors of each layer, uniformly adjusting the dimension normalization processing to 64-dimension vectors, carrying out weighted aggregation on the normalized fusion sub-vectors according to the obtained attention weight, and generating a global fusion feature vector, wherein the global fusion feature vector comprises multi-dimension state information of an electric, mechanical, thermal and cooling system of a camera.
10. 10. The multi-source sensory data fusion based camera operating state assessment system of claim 6, wherein the training module comprises: The verification unit is used for carrying out dimension verification on the global fusion feature vector according to the input dimension requirement of the camera state evaluation model, leading the feature vector into a feature input layer of the camera state evaluation model which is trained in advance after confirming that the vector format is consistent with a model input interface, wherein the camera state evaluation model is constructed based on a deep belief network, the network comprises 3 limited Boltzmann machine pre-training layers and1 back propagation fine tuning layer, the node number of the pre-training layers is sequentially set to 128, 64 and 32, the initial parameters of the network are optimized through a contrast divergence algorithm, and then a weight matrix is adjusted through the back propagation algorithm until the model converges; The output unit is used for carrying out multi-layer nonlinear transformation and feature mapping on the input global fusion feature vector by the camera state evaluation model, realizing state identification by a softmax classifier of an output layer and finally outputting a camera running state result.

Description

Camera operation state evaluation method and system based on multisource sensing data fusion Technical Field The invention relates to the technical field of camera operation state evaluation, in particular to a camera operation state evaluation method and system based on multi-source sensing data fusion. Background The camera serves as a key device in the power system, and plays a significant role in maintaining the stable operation of the power system. The reactive power system can transmit or absorb reactive power to a system, effectively improve the power factor, reduce the loss in a network, and has remarkable effects of adjusting the network voltage and improving the electric energy quality. In a long-distance power transmission line, a camera can maintain the stability of the power grid voltage by adjusting the excitation current of the camera according to the load condition of the power grid. When the power grid is loaded and heavy, the power grid is over-excited to run, the lagged reactive current component in the power transmission line is reduced, so that the voltage drop of the line is reduced, and when the power transmission line is lightly loaded, the power grid is underexcited to run, the lagged reactive current is absorbed, and the voltage of the power grid is prevented from rising. The operation state of the camera is accurately estimated, and the method has great significance in guaranteeing the safety, stability and economic operation of the power system. If the camera fails, local power grid voltage fluctuation can be caused, and even large-area power failure accidents are caused when the voltage fluctuation is serious, so that huge losses are brought to social production and life of people. For example, in some electricity peak periods, if the camera cannot normally provide reactive power support, the voltage of the power grid may be suddenly lowered, the normal operation of various electric equipment is affected, the factory production line is forced to be stopped, and the traffic signal lamp is out of order, so that great inconvenience is brought to society. Therefore, the running state of the camera can be mastered timely and accurately, potential fault hidden dangers can be found in advance, and corresponding measures can be taken, so that the method is important for ensuring reliable power supply of the power system. Disclosure of Invention The invention aims to provide a camera running state evaluation method, a camera running state evaluation system, camera running state evaluation equipment and a camera running state evaluation readable storage medium based on multi-source sensing data fusion, so as to solve the problems. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: In a first aspect, the present application provides a method for evaluating an operation state of a camera based on multi-source sensing data fusion, including: Constructing a space-time alignment model based on the multi-source sensing data, unifying the data with different acquisition frequencies to the same time dimension, generating a self-adaptive filtering threshold value by combining a normal operation data feature library of the camera, filtering impulse noise, harmonic interference and environmental noise in the data, and completing dimension unification of all parameters through range normalization to obtain standardized preprocessing data; Aiming at standardized preprocessing data, a specific analysis method is adopted to extract state characteristics, wherein the specific analysis method comprises the steps of carrying out wavelet packet transformation on electric quantity data, extracting frequency band energy duty ratio and characteristic frequency amplitude, carrying out trend analysis on temperature data, extracting temperature change rate and gradient difference value, carrying out Hilbert-Huang transformation on vibration data, extracting instantaneous frequency and amplitude, carrying out Mel frequency cepstrum coefficient analysis on voiceprint data, extracting spectrum envelope and formant frequency, carrying out statistical analysis on cooling system parameters, extracting mean value and variance of the cooling system parameters, calculating the association degree of each extracted characteristic and a phase-change machine fault state through mutual information entropy after the characteristic extraction of various data, and screening the characteristic that the association degree reaches a preset threshold value to form a key characteristic set, wherein the phase-change machine fault state comprises stator winding faults and bearing bush abrasion faults; The method comprises the steps of extracting vibration characteristics and voiceprint characteristics from a key characteristic set, calculating time domain correlation and frequency domain coupling degree of the two characteristics, constructing a cross-modal characteristic mapping relation