Method and system for diagnosing health state of power equipment based on multi-modal data fusion

Abstract

The invention discloses a multi-modal data fusion power equipment health state diagnosis method and system, and belongs to the field of power equipment state monitoring, and the method comprises the steps: S100, collecting infrared thermal imaging data, vibration signals, current harmonic data and partial discharge signals of power equipment, and carrying out the time synchronization and space registration; and S200, acquiring infrared thermal imaging data, vibration signals, current harmonic data and partial discharge signals, and inputting the infrared thermal imaging data, the vibration signals, the current harmonic data and the partial discharge signals into the dynamic weight fusion model to obtain an equipment health state score and a fault type. And S300, according to the equipment health state score and the fault type, triggering grading alarm. According to the invention, timely early warning can be carried out on potential fault hidden dangers of power equipment.

Description

Technical Field

[0001] The present invention relates to the field of power equipment status monitoring, and in particular to a power equipment health status diagnosis method and system based on multimodal data fusion. Background Art

[0002] In recent years, power equipment has played a key role in all walks of life as they are an important component in maintaining production and life. In the power distribution system, switchgear is a key component for protecting and controlling equipment, ensuring the reliable operation of the power system. However, the heat generated by power equipment during operation and potential failures can lead to degraded equipment performance, increased downtime, and even fire and safety risks.

[0003] In existing technologies, single sensor diagnosis is susceptible to interference (such as infrared imaging is affected by environmental radiation), and fixed weight fusion models are difficult to adapt to the operating characteristics of different equipment such as transformers and circuit breakers. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a method and system for diagnosing the health status of power equipment using multimodal data fusion, which can provide timely warnings of potential power equipment failures and quickly locate the causes of the failures.

[0005] On the one hand, an embodiment of the present invention provides a method for diagnosing the health status of power equipment by multimodal data fusion, comprising the following steps: S100, collecting infrared thermal imaging data, vibration signals, current harmonic data and partial discharge signals of power equipment, and performing time synchronization and spatial registration; S200, obtaining infrared thermal imaging data, vibration signals, current harmonic data and partial discharge signals, inputting them into a dynamic weight fusion model, and obtaining a health status score of the equipment and a fault type; the dynamic weight fusion model comprises: a feature extraction layer for extracting features from infrared thermal imaging data, vibration signals, current harmonic data sets and partial discharge signals, and obtaining temperature gradient features, high-frequency component features of vibration signals, current harmonic data sets and partial discharge signals. The odd harmonic characteristics of harmonics and the pulse characteristics of partial discharge; the working condition perception layer is used to monitor the working condition parameters of the equipment in real time and predict the current working condition type through the LSTM network; the working condition parameters include real-time load rate, ambient temperature and humidity, and cooling wind speed; the working condition type includes at least one of steady state, overload, high temperature and mechanical wear; the attention weight allocation layer is used to calculate the dynamic weight of each mode based on the working condition type and feature correlation using a multi-head attention mechanism; the fusion decision layer is used to generate the equipment health status score and fault probability distribution through weighted summation, and finally output the equipment status level and locate the abnormal part; S300, triggering a graded alarm based on the equipment health status score and fault type.

[0006] According to some embodiments of the present invention, step S200 includes: S210, extracting the temperature gradient matrix, the high-frequency energy ratio, the THD distortion rate, and the odd harmonic amplitude ratio based on infrared thermal imaging data, vibration signals, current harmonic data, and partial discharge signals; S220, obtaining operating condition parameters; capturing the long-term dependency of the operating condition parameters based on the bidirectional LSTM, and predicting the current operating condition type; S230, calculating the dynamic weight of each mode based on an improved multi-head attention structure, the multi-head attention structure including an operating condition attention head and a feature attention head; the operating condition attention head outputs an operating condition weight vector based on the operating condition parameters, and the feature attention head outputs a feature correlation weight based on the feature vectors of each mode, and the calculation formula is: ;W is the feature correlation weight; is the working condition adaptation factor, which is generated by the BiLSTM hidden state through the fully connected layer and Sigmoid activation; Q is the query matrix, which represents the current window working condition parameters; K is the key matrix, which represents the historical window working condition sequence; is the inter-modal energy function; is the dimension of the key vector; ;in, is the BiLSTM hidden state at the current time step; It is a trainable parameter that controls the intensity of the impact of working conditions on attention; is the Sigmoid function; ;in, are the eigenvectors of different modes, the trainable matrix We is a diagonal matrix whose diagonal elements are optimized by gradient descent and initialized to obey the normal distribution with mean 0 and standard deviation 0.02; i, j are modal indices; S240, dynamically weight fusion is performed on the working condition weight vector and the feature correlation weight to obtain the final weight; S250, a weight distribution heat map is generated to show the contribution ratio of each mode in the decision; and a SHAP value interpretation report is output.

[0007] According to some embodiments of the present invention, step S220 includes: S221, inputting the multi-source operating condition parameters in the sliding window into the bidirectional LSTM network, and extracting historical and future potential trend features through the forward and reverse layers of the bidirectional LSTM network respectively; and dynamically weighting the parameters of key time points in the window based on the attention mechanism; S222, calculating the probability distribution of each operating condition type through the Softmax function. If three consecutive sliding windows predict the same non-steady-state operating condition, it is determined to be a continuous abnormal operating condition; S223, calculating the KL divergence of the current operating condition data distribution and the historical benchmark in real time. If the KL divergence exceeds the threshold of 0.1, freezing the BiLSTM underlying parameters and fine-tuning the attention layer weights.

[0008] According to some embodiments of the present invention, the time synchronization and spatial registration include: using GPS or PTP protocol to time stamp the multi-sensor data; and aligning the infrared thermal image with the physical position of the device through feature point matching.

[0009] According to some embodiments of the present invention, the method for extracting temperature gradient features includes: performing super-resolution reconstruction on infrared thermal images to generate a high-precision temperature distribution map; and using a Sobel operator to calculate the temperature gradient and identify micron-level crack hot spots.

[0010] According to some embodiments of the present invention, optimizing the dynamic weight fusion model through reinforcement learning includes dynamically adjusting the decision weight of vibration data in mechanical wear faults based on historical diagnostic accuracy feedback.

[0011] According to some embodiments of the present invention, the method further includes early perception of weak faults: constructing a generative adversarial network and synthesizing abnormal samples using normal data; and jointly analyzing the implicit correlation between infrared temperature fluctuations and vibration spectra in the frequency domain.

[0012] According to some embodiments of the present invention, the method further includes: when the diagnosis result is that the connection is loose, automatically triggering the robot to perform a bolt tightening operation; after maintenance, a second infrared scan is performed to verify the temperature drop, and if it does not meet expectations, an alarm is upgraded.

[0013] According to some embodiments of the present invention, the method further includes: constructing a three-dimensional thermal field simulation model of the equipment, inputting real-time data to simulate the heat diffusion path; and locating the abnormal coordinates of the heat source through a back propagation algorithm.

[0014] Another embodiment of the present invention provides a multimodal data fusion power equipment health status diagnosis system, including: a data acquisition module for integrating an infrared thermal imager, a vibration sensor, a current transformer and a partial discharge detector; and a cloud analysis platform for executing the method described above and outputting diagnosis results.

[0015] The embodiments of the present invention achieve at least the following beneficial effects: Through multimodal data fusion and dynamic weight allocation, the embodiments of the present invention can more accurately reflect the health status of equipment and improve diagnostic accuracy. They can also adapt to the operating characteristics and operating conditions of different equipment, increasing the breadth and practicality of diagnosis. Automated and intelligent diagnostic and alarm processes reduce the burden of manual monitoring and improve management efficiency. By early sensing of subtle faults and constructing a three-dimensional thermal field simulation model of the equipment, potential faults can be promptly discovered and prevented, ensuring the safe and reliable operation of the equipment.

[0016] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments with reference to the following drawings, in which:

[0018] Figure 1 A schematic flow chart of a method according to an embodiment of the present invention;

[0019] Figure 2 This is a schematic block diagram of modules of a system according to an embodiment of the present invention.

[0020] Reference numerals:

[0021] Data collection module 100, cloud analysis platform 200. DETAILED DESCRIPTION

[0022] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0023] In the description of the present invention, "several" means one or more, "more" means two or more, "greater than," "less than," and "exceed" are understood to exclude the number itself, while "above," "below," and "within" are understood to include the number itself. The use of "first" and "second" in the description is solely for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, implicitly specifying the number of the indicated technical features, or implicitly specifying the order of the indicated technical features.

[0024] The embodiments of the present invention use infrared thermal imaging technology to perform real-time online temperature monitoring of important parts of power equipment, such as: temperature monitoring of busbars and copper busbar connections in distribution cabinets (switch cabinets); temperature monitoring of cable joints in distribution cabinets and distribution boxes; temperature monitoring of power equipment such as transformers, lightning arresters, and voltage transformers; temperature monitoring of large motors, operating equipment, bodies, and mechanical connection parts; temperature monitoring of closed busbars, connectors, and plugs; and so on.

[0025] The embodiments of the present invention address the significant limitations of single sensors, poor adaptability to operating conditions, and difficulty detecting early faults in power equipment health monitoring through dynamic fusion of multimodal data and intelligent diagnosis. Traditional methods rely on single sensor data, such as infrared or vibration, which is susceptible to environmental noise and can lead to false alarms. This solution integrates four types of heterogeneous data: infrared thermal imaging, vibration signals, current harmonics, and partial discharge. Using a dynamic weight allocation model, it quantifies the contributions of different modalities, significantly improving diagnostic reliability. To address operating disturbances such as equipment load fluctuations and ambient temperature changes, an LSTM operating condition perception layer is introduced to identify operating status in real time. A multi-head attention mechanism is used to dynamically adjust the weights of each modality, ensuring high accuracy across diverse scenarios such as substations and wind power plants. High-resolution temperature gradient feature extraction and time-frequency analysis of partial discharge signals, combined with a threshold alarm strategy optimized by reinforcement learning, can identify potential defects such as insulation aging and bearing microcracks, which are difficult to detect using traditional methods, and provide early warning of faults.

[0026] Reference Figure 1 The embodiment of the present invention provides a method for diagnosing the health status of power equipment based on multimodal data fusion, comprising the following steps:

[0027] S100: Collect infrared thermal imaging data, vibration signals, current harmonic data, and partial discharge signals of power equipment, and perform time synchronization and spatial registration.

[0028] S200: Obtain infrared thermal imaging data, vibration signals, current harmonic data, and partial discharge signals, input them into a dynamic weight fusion model, and obtain a device health status score and fault type.

[0029] S300. Trigger a graded alarm based on the device health status score and fault type.

[0030] The dynamic weight fusion model of the embodiment of the present invention includes:

[0031] The feature extraction layer is used to extract features from infrared thermal imaging data, vibration signals, and partial discharge signals of current harmonic data sets to obtain temperature gradient features, high-frequency component features of vibration signals, odd harmonic features of current harmonics, and pulse features of partial discharges.

[0032] The working condition perception layer is used to monitor the working condition parameters of the equipment in real time and predict the current working condition type through the LSTM network; the working condition parameters include real-time load rate, ambient temperature and humidity, and cooling wind speed; the working condition type includes at least one of steady state, overload, high temperature, and mechanical wear.

[0033] The attention weight allocation layer is used to calculate the dynamic weight of each modality based on the working condition type and feature correlation using a multi-head attention mechanism.

[0034] The fusion decision layer is used to generate equipment health status scores and fault probability distributions through weighted summation, and ultimately output the equipment status level and locate abnormal parts.

[0035] In some embodiments, step S200 includes:

[0036] S210. Extract the temperature gradient matrix, high-frequency energy ratio, THD distortion rate, and odd harmonic amplitude ratio based on infrared thermal imaging data, vibration signals, current harmonic data, and partial discharge signals.

[0037] S220. Obtain operating condition parameters; capture the long-term dependencies of the operating condition parameters based on the bidirectional LSTM and predict the current operating condition type.

[0038] S230. Calculate the dynamic weight of each modality based on the improved multi-head attention structure. The multi-head attention structure includes a working condition attention head and a feature attention head. The working condition attention head outputs a working condition weight vector based on the working condition parameters, and the feature attention head outputs a feature correlation weight based on the feature vectors of each modality. The calculation formula is:

[0039] ;

[0040] Where W is the feature correlation weight; is the working condition adaptation factor, which is generated by the BiLSTM hidden state through the fully connected layer and Sigmoid activation; Q is the query matrix, which represents the current window working condition parameters; K is the key matrix, which represents the historical window working condition sequence; is the inter-modal energy function; is the dimension of the key vector;

[0041] ;

[0042] in, is the BiLSTM hidden state at the current time step; It is a trainable parameter that controls the intensity of the impact of working conditions on attention; is the Sigmoid function;

[0043] ;

[0044] in, are the eigenvectors of different modes, the trainable matrix We is a diagonal matrix whose diagonal elements are optimized by gradient descent and initialized to obey the normal distribution with mean 0 and standard deviation 0.02; i, j are the mode indices.

[0045] S240: Dynamically fuse the working condition weight vector and the feature correlation weight to obtain a final weight.

[0046] S250: Generate a weight distribution heat map to show the contribution ratio of each mode in the decision-making; output a SHAP value explanation report.

[0047] In some embodiments, step S220 of the embodiment of the present invention includes:

[0048] S221. Input the multi-source operating condition parameters in the sliding window into the bidirectional LSTM network, extract the historical and future potential trend features respectively through the forward and reverse layers of the bidirectional LSTM network; and dynamically weight the parameters of the key time points in the window based on the attention mechanism.

[0049] S222. Calculate the probability distribution of each operating condition type using the Softmax function. If three consecutive sliding windows predict the same non-steady-state operating condition, determine it as a continuous abnormal operating condition.

[0050] S223. Calculate the KL divergence of the current working condition data distribution and the historical benchmark in real time. If the KL divergence exceeds the threshold of 0.1, freeze the BiLSTM underlying parameters and fine-tune the attention layer weights.

[0051] In some embodiments, the time synchronization and spatial registration of the embodiments of the present invention include: using GPS or PTP protocol to time stamp the multi-sensor data; and aligning the infrared thermal image with the physical position of the device through feature point matching.

[0052] In some embodiments, the method for extracting temperature gradient features of an embodiment of the present invention includes: performing super-resolution reconstruction on infrared thermal images to generate a high-precision temperature distribution map; using a Sobel operator to calculate the temperature gradient and identify micron-level crack hot spots.

[0053] In some embodiments, the method of an embodiment of the present invention includes: optimizing a dynamic weight fusion model through reinforcement learning, including: dynamically adjusting the decision weight of vibration data in mechanical wear failure based on historical diagnostic accuracy feedback.

[0054] In some embodiments, the method of the embodiment of the present invention also includes early perception of weak faults: constructing a generative adversarial network and synthesizing abnormal samples using normal data; and jointly analyzing the implicit correlation between infrared temperature fluctuations and vibration spectra through frequency domain.

[0055] In some embodiments, the method of the embodiment of the present invention also includes: when the diagnosis result is that the connection is loose, automatically triggering the robot to perform the bolt tightening operation; after maintenance, a second infrared scan is performed to verify the temperature drop, and if it does not meet expectations, an alarm is upgraded.

[0056] In some embodiments, the method of the embodiment of the present invention further includes: constructing a three-dimensional thermal field simulation model of the equipment, inputting real-time data to simulate the heat diffusion path; and locating the abnormal coordinates of the heat source through a back propagation algorithm.

[0057] Reference Figure 2 Another embodiment of the present invention provides a multi-modal data fusion power equipment health status diagnosis system, including:

[0058] The data acquisition module 100 is used to integrate an infrared thermal imager, a vibration sensor, a current transformer and a partial discharge detector.

[0059] The cloud analysis platform 200 is used to execute the above-mentioned multi-modal data fusion power equipment health status diagnosis method and output the diagnosis results.

[0060] Although specific embodiments are described herein, those skilled in the art will recognize that many other modifications or alternative embodiments are also within the scope of the present disclosure. For example, any of the functions and / or processing capabilities described in conjunction with a particular device or component may be performed by any other device or component. In addition, although various exemplary implementations and architectures have been described in accordance with the embodiments of the present disclosure, those skilled in the art will recognize that many other modifications to the exemplary implementations and architectures described herein are also within the scope of the present disclosure.

[0061] Some aspects of the present disclosure have been described above with reference to the block diagrams and flow charts of the systems, methods, systems and / or computer program products according to the exemplary embodiments. It should be understood that the combination of one or more blocks in the block diagram and the flow chart and the blocks in the block diagram and the flow chart can be realized by executing computer executable program instructions respectively. Equally, according to some embodiments, some blocks in the block diagram and the flow chart may not need to be executed in the order shown, or may not need to be executed in full. In addition, additional components and / or operations beyond those components and / or operations shown in the blocks in the block diagram and the flow chart may be present in certain embodiments.

[0062] Therefore, the blocks in the block diagrams and flow charts support combinations of means for performing the specified functions, combinations of elements or steps for performing the specified functions, and program instruction means for performing the specified functions. It should also be understood that each block in the block diagrams and flow charts, and combinations of blocks in the block diagrams and flow charts, can be implemented by a dedicated hardware computer system that performs the specific functions, elements, or steps, or a combination of dedicated hardware and computer instructions.

[0063] The program modules, applications, etc. described herein may include one or more software components, including, for example, software objects, methods, data structures, etc. Each such software component may include computer-executable instructions that, in response to execution, cause at least a portion of the functionality described herein (e.g., one or more operations of the illustrative methods described herein) to be performed.

[0064] Software component can be encoded with any one in various programming languages.A kind of exemplary programming language can be low-level programming language, such as the assembly language associated with specific hardware architecture and / or operating system platform.Comprise that the software component of assembly language instruction may need to be converted to executable machine code by assembler before being executed by hardware architecture and / or platform.Another exemplary programming language can be a more advanced programming language, and it can be transplanted across multiple architectures.Comprise that the software component of more advanced programming language may need to be converted to intermediate representation by interpreter or compiler before execution.Other examples of programming language include but are not limited to macro language, shell or command language, job control language, script language, database query or search language or report writing language.In one or more exemplary embodiments, the software component that comprises the instruction of one in the above-mentioned programming language example can be directly executed by operating system or other software component, without first being converted into another form.

[0065] Software components can be stored as files or other data storage structures. Software components of similar types or related functions can be stored together, such as in a specific directory, folder, or library. Software components can be static (e.g., preset or fixed) or dynamic (e.g., created or modified at execution time).

[0066] The embodiments of the present invention are described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary technicians in the relevant technical field without departing from the scope of the present invention.

Claims

1. A method for diagnosing the health status of power equipment based on multimodal data fusion, characterized in that: The following steps are involved: S100: Collect infrared thermal imaging data, vibration signals, current harmonic data, and partial discharge signals of power equipment, and perform time synchronization and spatial registration; S200, obtaining infrared thermal imaging data, vibration signals, current harmonic data, and partial discharge signals, inputting them into a dynamic weight fusion model, and obtaining a device health status score and fault type; The dynamic weight fusion model includes: The feature extraction layer is used to extract features from infrared thermal imaging data, vibration signals, and partial discharge signals of current harmonic data sets to obtain temperature gradient features, high-frequency component features of vibration signals, odd harmonic features of current harmonics, and pulse features of partial discharges. The operating condition perception layer is used to monitor the operating parameters of the equipment in real time and predict the current operating condition type through the LSTM network. The operating condition parameters include real-time load rate, ambient temperature and humidity, and cooling wind speed. The operating condition type includes at least one of steady state, overload, high temperature, and mechanical wear. The attention weight allocation layer is used to calculate the dynamic weight of each modality using a multi-head attention mechanism based on the working condition type and feature relevance; The fusion decision layer is used to generate equipment health status scores and fault probability distributions through weighted summation, and ultimately output equipment status levels and locate abnormal parts; S300. Trigger a graded alarm based on the device health status score and fault type.

2. The method for diagnosing the health status of electric power equipment based on multimodal data fusion according to claim 1, characterized in that: The step S200 includes: S210, extracting the temperature gradient matrix, high-frequency energy ratio, THD distortion rate, and odd harmonic amplitude ratio based on infrared thermal imaging data, vibration signals, current harmonic data, and partial discharge signals; S220, obtaining operating condition parameters; capturing long-term dependencies of the operating condition parameters based on the bidirectional LSTM, and predicting the current operating condition type; S230. Calculate the dynamic weight of each modality based on an improved multi-head attention structure, wherein the multi-head attention structure includes a working condition attention head and a feature attention head; the working condition attention head outputs a working condition weight vector based on the working condition parameters, and the feature attention head outputs a feature correlation weight based on the feature vectors of each modality. The calculation formula is: ； Where W is the feature correlation weight; is the working condition adaptation factor, which is generated by the BiLSTM hidden state through the fully connected layer and Sigmoid activation; Q is the query matrix, which represents the current window working condition parameters; K is the key matrix, which represents the historical window working condition sequence; is the inter-modal energy function; is the dimension of the key vector; ； in, is the BiLSTM hidden state at the current time step; It is a trainable parameter that controls the intensity of the impact of working conditions on attention; is the Sigmoid function; ； in, are the eigenvectors of different modes, the trainable matrix We is a diagonal matrix whose diagonal elements are optimized by gradient descent and initialized to obey the normal distribution with mean 0 and standard deviation 0.02; i, j are the mode indices; S240, dynamically fusion the working condition weight vector and the feature correlation weight to obtain a final weight; S250: Generate a weight distribution heat map to show the contribution ratio of each mode in the decision-making; output a SHAP value explanation report.

3. The method for diagnosing the health status of electric power equipment based on multimodal data fusion according to claim 2, characterized in that: The step S220 includes: S221. Input the multi-source operating condition parameters in the sliding window into a bidirectional LSTM network, extract historical and future potential trend features through the forward and reverse layers of the bidirectional LSTM network respectively; and dynamically weight the parameters of key time points in the window based on the attention mechanism; S222. Calculate the probability distribution of each operating condition type using the Softmax function. If three consecutive sliding windows predict the same non-steady-state operating condition, determine it as a continuous abnormal operating condition. S223. Calculate the KL divergence of the current working condition data distribution and the historical benchmark in real time. If the KL divergence exceeds the threshold of 0.1, freeze the BiLSTM underlying parameters and fine-tune the attention layer weights.

4. The method for diagnosing the health status of electric power equipment based on multimodal data fusion according to claim 1, characterized in that: The time synchronization and spatial registration include: Use GPS or PTP protocol to time-stamp multi-sensor data; align infrared thermal images with the physical location of the device through feature point matching.

5. The method for diagnosing the health status of electric power equipment based on multimodal data fusion according to claim 1, characterized in that: The method for extracting the temperature gradient feature includes: The infrared thermal image is reconstructed with super-resolution to generate a high-precision temperature distribution map; the Sobel operator is used to calculate the temperature gradient and identify micron-level crack hot spots.

6. The method for diagnosing the health status of electric power equipment based on multimodal data fusion according to claim 1, characterized in that: The method includes optimizing the dynamic weight fusion model through reinforcement learning, including: Based on historical diagnostic accuracy feedback, the decision weight of vibration data in mechanical wear failure is dynamically adjusted.

7. The method for diagnosing the health status of electric power equipment based on multimodal data fusion according to claim 1, characterized in that: The method also includes early detection of subtle faults: Build a generative adversarial network to synthesize abnormal samples using normal data; The implicit correlation between infrared temperature fluctuations and vibration spectra is analyzed jointly in the frequency domain.

8. The method for diagnosing the health status of electric power equipment based on multimodal data fusion according to claim 1, characterized in that: The method further comprises: When the diagnosis result is that the connection is loose, the robot is automatically triggered to perform the bolt tightening operation; After maintenance, a second infrared scan is performed to verify the temperature drop. If the temperature drop does not meet expectations, an alarm is upgraded.

9. The method for diagnosing the health status of electric power equipment based on multimodal data fusion according to claim 1, characterized in that: The method further comprises: Build a three-dimensional thermal field simulation model of the equipment and input real-time data to simulate the heat diffusion path; The abnormal coordinates of heat sources are located using the back propagation algorithm.

10. A multimodal data fusion power equipment health status diagnosis system, characterized by: include: Data acquisition module for integrating infrared thermal imager, vibration sensor, current transformer and partial discharge detector; A cloud analysis platform is used to execute the method according to any one of claims 1 to 9 and output a diagnosis result.

Images