Power fault diagnosis method and system based on multi-modal data fusion

Abstract

The invention relates to the technical field of power fault diagnosis, in particular to a multi-modal data fusion-based power fault diagnosis method and system, which comprehensively collects and analyzes electrical, environmental and mechanical parameters of power equipment through a multi-modal data fusion technology, constructs a multi-modal parameter feature matrix, and performs multi-modal data fusion on the basis of the multi-modal parameter feature matrix. Dynamic anomaly detection and personalized monitoring strategy formulation are realized, the accuracy and timeliness of fault diagnosis are improved, false alarm and missing alarm are reduced, operation and maintenance resource allocation is optimized, and stable operation and efficient power supply of a power system are ensured; the method comprises the following steps: respectively acquiring multi-modal data information of all same power equipment in a transformer area; feature extraction is carried out on the multi-modal data information of each power device to obtain a multi-modal parameter set; and carrying out data integration on all the multi-modal parameter sets which are collected at the same time to obtain a multi-modal parameter characteristic matrix which can represent the operation states of all the power equipment of the same model in the transformer area.

Description

Technical Field

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

[0002] As the scale and complexity of power systems continue to increase, the reliability and stability of power equipment have become an important concern for the power industry. Power fault diagnosis is a key link in ensuring the safe operation of power systems. Its accuracy and timeliness are directly related to the quality and efficiency of power supply.

[0003] The existing power equipment fault diagnosis method mainly monitors a single power equipment independently and diagnoses the fault type based on a fixed monitoring threshold or a fixed fault identification model. The existing method ignores the synchronization of the same equipment under the same operating conditions and cannot dynamically adjust the identification of abnormal power equipment according to the actual operation of the power equipment. For example, if the instantaneous current values ​​of all the same power equipment in the substation exceed the fixed monitoring threshold, according to the previous diagnosis method, all power equipment will be given a fault alarm. In fact, this collective instantaneous current exceeding the standard may be caused by transient events at the system level (such as a sudden increase in grid load, large motor start-up, etc.), and not every device is actually at fault. This "one-size-fits-all" diagnosis method will not only lead to an increase in the false alarm rate and waste operation and maintenance resources, but may also cover up the real fault signal and affect the stability of the substation power system. Summary of the invention

[0004] To solve the above technical problems, the present invention provides a method and system for electric power fault diagnosis based on multimodal data fusion, which comprehensively collects and analyzes the electrical, environmental and mechanical parameters of electric power equipment through multimodal data fusion technology, constructs a multimodal parameter feature matrix, realizes dynamic anomaly detection and personalized monitoring strategy formulation, improves the accuracy and timeliness of fault diagnosis, reduces false alarms and missed alarms, optimizes the allocation of operation and maintenance resources, and ensures the stable operation and efficient power supply of the power system.

[0005] In a first aspect, the present invention provides a method for diagnosing electric power faults based on multimodal data fusion, the method comprising:

[0006] Collect multi-modal data information of all the same power equipment in the substation area separately;

[0007] For each power device, feature extraction is performed to obtain a multimodal parameter set; the multimodal parameter set includes electrical parameters, environmental parameters, and mechanical parameters;

[0008] All multimodal parameter sets collected simultaneously are integrated to obtain a multimodal parameter feature matrix capable of characterizing the operating status of all power equipment of the same model in the substation; in the multimodal parameter feature matrix, multimodal parameters of the same power equipment are located in the same column; and the same modal parameters corresponding to different power equipment are located in the same row;

[0009] Inputting the multimodal parameter feature matrix into the abnormal power equipment identification model, the abnormal power equipment identification model outputs an abnormal vector of the abnormal power equipment; the abnormal vector includes the identity number of the power equipment and the degree of abnormality;

[0010] According to the abnormal vector, the abnormal power equipment monitoring strategy is generated; the abnormal power equipment monitoring strategy includes monitoring target parameters and acquisition frequency of the target parameters, and the target parameters include at least one of electrical parameters, environmental parameters and mechanical parameters;

[0011] Based on the abnormal power equipment monitoring strategy, the parameters of the abnormal power equipment are collected to obtain a time series image of the target parameters that can characterize the operating status of the abnormal power equipment;

[0012] The target parameter time series image is input into the power fault identification model, and the power fault identification model outputs the fault type of the abnormal power equipment.

[0013] Furthermore, the multimodal parameter feature matrix is:

[0014]

[0015] Among them, the multi-modal parameters of the same power equipment are located in the same column, and the same modal parameters corresponding to different power equipment are located in the same row; E1, E2, …, E N Represents the electrical parameter vector of all power equipment, H1, H2, ..., H N Represents the environmental parameter vector of all power equipment, M1, M2, ..., M N Represents the mechanical parameter vector of all electrical equipment.

[0016] Furthermore, the multimodal parameter set is:

[0017]

[0018] Among them, X i represents the multimodal parameter set, E i , H i and M i are the electrical parameters, environmental parameters and mechanical parameter vectors of the i-th power equipment respectively; where E i , H i and M i Further broken down into:

[0019]

[0020] Among them, V i ,I i ,P i ,Q i ,S i They represent the electrical parameters such as voltage, current, active power, reactive power and apparent power of the i-th power equipment respectively;

[0021]

[0022] Among them, T i ,RH i ,P atm,i They represent environmental parameters such as temperature, relative humidity and atmospheric pressure respectively;

[0023]

[0024] Among them, ω i ,x i ,v i ,a i Represent the mechanical parameters of rotation speed, displacement, velocity and acceleration respectively.

[0025] Furthermore, the method for obtaining the multimodal parameter feature matrix includes:

[0026] Clean the collected multimodal data to remove invalid or abnormal data points;

[0027] Normalize or standardize the data so that different types of measurements from different sensors can be compared on the same scale;

[0028] Then the electrical parameters, environmental parameters and mechanical parameters of each power equipment are organized in a matrix;

[0029] Construct a multimodal parameter feature matrix. Construct a multimodal parameter feature matrix through the above arrangement.

[0030] Furthermore, the method for calculating the abnormality degree includes:

[0031] For each power device i, the abnormality level AS of the power device is calculated based on the electrical parameters, environmental parameters and mechanical parameters. i ;

[0032] When there is N e Electrical parameters, N h environmental parameters and N m When the mechanical parameters are equal, the calculation formula of abnormality degree is:

[0033]

[0034] Among them, the standardized value of the jth electrical parameter of the i-th device is:

[0035]

[0036] in, represents the jth electrical parameter of the i-th device, μ e ,σ e are the mean and standard deviation of electrical parameters, respectively;

[0037] The standardized value of the jth environmental parameter of the i-th device is:

[0038]

[0039] in, represents the jth environmental parameter of the i-th device, μ h ,σ h are the mean and standard deviation of environmental parameters, respectively;

[0040] Normalized value of the jth mechanical parameter of the i-th device:

[0041]

[0042] in, represents the jth mechanical parameter of the i-th device, μ m ,σ m are the mean and standard deviation of environmental parameters, respectively;

[0043] α, β and γ are weight coefficients corresponding to electrical parameters, environmental parameters and mechanical parameters, which are used to balance the importance of different modes;

[0044] like Then the power equipment i is considered abnormal, where is the abnormality threshold.

[0045] Furthermore, the method for obtaining the abnormal vector includes:

[0046] Clean the data in the generated multimodal parameter feature matrix, which contains all multimodal parameters of the device and the measured values ​​of all devices on the parameters;

[0047] The cleaned data is divided into a training set and a test set; according to the detection task of abnormal power equipment, the support vector machine model is selected as the basic model of the power equipment identification model, and the model parameters are adjusted using cross-validation; the selected model is trained using the training set data, and the model performance is evaluated on the validation set; the trained power equipment identification model is used to measure the abnormality of each device, reflecting the degree to which the device deviates from normal behavior;

[0048] The multimodal parameter feature matrix is ​​input into the power equipment identification model, and the power equipment identification model outputs an abnormal vector list, which includes the identity number of the abnormal power equipment and the degree of abnormality.

[0049] Furthermore, the method for obtaining the multimodal parameter set includes:

[0050] Preprocess the collected multimodal data information of each power device, including removing invalid data, removing outliers, calibrating data, and interpolating missing values;

[0051] Extract electrical parameter features from pre-processed multi-modal data information, including voltage / current waveform analysis, power analysis, and transient event detection;

[0052] Extract environmental parameter features from the preprocessed multimodal data information, including temperature, humidity and air pressure features;

[0053] Extract mechanical parameter features from pre-processed multi-modal data information, including vibration analysis, displacement and speed, and pattern recognition;

[0054] Normalize and standardize the extracted electrical parameter characteristics, environmental parameter characteristics and mechanical parameters;

[0055] All normalized and standardized electrical parameter characteristics, environmental parameter characteristics, and mechanical parameters are integrated to form a multimodal parameter set.

[0056] On the other hand, the present application also provides a power fault diagnosis system based on multimodal data fusion, the system comprising:

[0057] The multimodal data acquisition module is used to collect multimodal data information of all the same type of power equipment in the substation in real time or periodically, including electrical parameters, environmental parameters and mechanical parameters;

[0058] A multimodal feature extraction module is used to preprocess and extract features from the multimodal data information of each power device to form a feature vector containing multiple types of parameters to form a multimodal parameter set;

[0059] The multimodal data integration module is used to integrate the multimodal parameter sets of all power equipment into a multimodal parameter feature matrix, in which each column represents the multimodal parameter of a specific device and each row represents the performance of the same modal parameter on different devices;

[0060] An abnormal power equipment identification module is used to analyze the multimodal parameter feature matrix using a model trained by a machine learning or deep learning algorithm, identify power equipment that exhibits abnormal behavior compared to a normal operating mode, and provide an abnormal vector, including the equipment's identification number and degree of abnormality;

[0061] The abnormal power equipment monitoring strategy generation module generates a customized monitoring strategy for each abnormal power equipment based on the abnormal vector information, including the key monitoring target parameters and the corresponding acquisition frequency;

[0062] The time series image acquisition module is used to collect the target parameter time series data of abnormal power equipment according to the abnormal power equipment monitoring strategy and construct the target parameter time series image;

[0063] The power fault identification module is used to input the target parameter time series portrait into the trained power fault identification model, output the fault type of the abnormal power equipment, and determine the specific fault type of the abnormal power equipment.

[0064] In a third aspect, the present application provides an electronic device, comprising a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the transceiver, the memory, and the processor are connected via the bus, and the computer program, when executed by the processor, implements the steps of any one of the above methods.

[0065] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps in any one of the above-mentioned methods when executed by a processor.

[0066] By collecting multimodal data information of power equipment, including electrical parameters, environmental parameters and mechanical parameters, it is possible to more comprehensively evaluate the working status of the equipment and improve the accuracy of fault diagnosis; feature extraction is performed on the multimodal data of each device to form a multimodal parameter set, which is further integrated into a multimodal parameter feature matrix, making the data more structured and convenient for subsequent processing; using the abnormal power equipment identification model, dynamically identify power equipment that exhibits abnormal behavior compared with the normal operating mode, adapt to the dynamic changes of the power system, and reduce false alarms and missed alarms; generate a customized monitoring strategy for each abnormal power equipment based on the abnormal vector, including key monitoring parameters and acquisition frequency, so as to effectively allocate monitoring resources and reduce operation and maintenance costs; by collecting the time series portrait of the target parameters of the abnormal power equipment, the change trend of the equipment in the time series is captured, so that technicians can quickly locate the cause of the fault; the target parameter time series is converted into a time series image. The sequence portrait is input into the trained power fault identification model to accurately determine the specific fault type of the equipment, facilitating the adoption of targeted maintenance measures; through multimodal data analysis, it is possible to distinguish between system-level transient events and equipment faults themselves, avoiding false alarms and missed alarms caused by "one-size-fits-all" diagnosis; accurate and timely fault diagnosis can effectively prevent the occurrence of potential faults and ensure the overall stability and reliability of the power system; customized monitoring strategies enable operation and maintenance personnel to carry out equipment inspection and maintenance work in a targeted manner, reduce unnecessary inspections, and optimize resource utilization; through multimodal data fusion technology, the electrical, environmental and mechanical parameters of power equipment are comprehensively collected and analyzed, and a multimodal parameter feature matrix is ​​constructed to realize dynamic anomaly detection and personalized monitoring strategy formulation, improve the accuracy and timeliness of fault diagnosis, reduce false alarms and missed alarms, optimize the allocation of operation and maintenance resources, and ensure the stable operation and efficient power supply of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 is a flow chart of the present invention;

[0068] Figure 2 It is a flow chart of a method for obtaining a multi-modal parameter feature matrix;

[0069] Figure 3 It is a structural diagram of the power fault diagnosis system based on multimodal data fusion. DETAILED DESCRIPTION

[0070] In the description of the present application, those skilled in the art should know that the present application can be implemented as a method, an apparatus, an electronic device, and a computer-readable storage medium. Therefore, the present application can be specifically implemented in the following forms: complete hardware, complete software (including firmware, resident software, microcode, etc.), a combination of hardware and software. In addition, in some embodiments, the present application can also be implemented in the form of a computer program product in one or more computer-readable storage media, and the computer-readable storage medium contains computer program code.

[0071] The above-mentioned computer-readable storage medium may adopt any combination of one or more computer-readable storage media. Computer-readable storage media include: electrical, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or devices, or any combination of the above. More specific examples of computer-readable storage media include: portable computer disks, hard disks, random access memories, read-only memories, erasable programmable read-only memories, flash memories, optical fibers, optical disc read-only memories, optical storage devices, magnetic storage devices, or any combination of the above. In the present application, computer-readable storage media can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device, or device.

[0072] The acquisition, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of national laws.

[0073] The present application describes the provided methods, devices, and electronic devices through flowcharts and / or block diagrams.

[0074] It should be understood that each box in the flowchart and / or block diagram and the combination of boxes in the flowchart and / or block diagram can be implemented by computer-readable program instructions. These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine, and these computer-readable program instructions are executed by a computer or other programmable data processing device to produce a device that implements the functions / operations specified by the boxes in the flowchart and / or block diagram.

[0075] These computer-readable program instructions may also be stored in a computer-readable storage medium that enables a computer or other programmable data processing device to work in a specific manner. In this way, the instructions stored in the computer-readable storage medium produce an instruction device product including functions / operations specified in the blocks in the flowchart and / or block diagram.

[0076] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device, so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby enabling the instructions executed on the computer or other programmable data processing apparatus to provide a process for implementing the functions / operations specified in the blocks in the flowchart and / or block diagram.

[0077] The present application is described below in conjunction with the drawings in the present application.

[0078] Embodiment 1: Figure 1 to Figure 2 As shown, the power fault diagnosis method based on multimodal data fusion of the present invention specifically includes the following steps:

[0079] S1. Collect multi-modal data information of all identical power equipment in the substation area;

[0080] Step S1 requires deploying appropriate sensors and data acquisition devices on all identical power equipment in the substation area to capture various key information during equipment operation; the details are as follows:

[0081] Electrical parameters: such as voltage, current, power factor and harmonic content, which are measured by current transformers, voltage transformers, power meters and other equipment installed in the circuit;

[0082] Environmental parameters: including temperature, humidity and air pressure, etc., obtained through temperature and humidity sensors, pressure sensors, etc.;

[0083] Mechanical parameters: such as vibration frequency, displacement and rotation speed, which are monitored by vibration sensors, accelerometers, tachometers and other equipment;

[0084] In order to ensure the accuracy and real-time performance of data, the sensors used should have high precision, good stability and anti-interference ability;

[0085] Considering the security and reliability of data transmission, wired or wireless methods are used for data transmission, such as industrial Ethernet, fieldbus technology or wireless sensor networks;

[0086] For each power equipment, sensors and data acquisition devices will continuously collect data and store it locally or upload it to a central monitoring system, ensuring that a detailed data set reflecting the operating status of the power equipment can be obtained to facilitate further data processing and fault diagnosis.

[0087] By collecting data on electrical parameters, environmental parameters and mechanical parameters, the operating status of power equipment can be comprehensively evaluated from multiple angles, and whether there are potential problems with the equipment can be accurately determined; using sensors with high precision, good stability and anti-interference ability to ensure high quality of collected data, high-quality data can help improve the accuracy of fault diagnosis and facilitate subsequent analysis; real-time data collection and transmission enable the monitoring system to promptly detect and respond to abnormal conditions of power equipment, and improve the ability to prevent accidents before they occur; by adopting wired or wireless data transmission methods, appropriate communication means can be selected according to actual conditions; considering the security of data transmission, encryption and other security measures are used to prevent data leakage or tampering during transmission to protect the security of the power system; corresponding sensors and data acquisition devices are installed on newly added equipment to adapt to the expansion or changes of the power system; by reducing false alarm rates, avoiding unnecessary maintenance work and timely discovering real problems, the operation and maintenance costs are reduced; by continuously monitoring and comparing multimodal data, the dynamic adjustment of fault identification models is supported to facilitate adaptation to the actual operation of power equipment; improving the accuracy and timeliness of fault diagnosis, effectively reducing false alarm rates, and improving the overall stability of the power system.

[0088] S2. Extracting features from the multimodal data information of each electric power device to obtain a multimodal parameter set; the multimodal parameter set includes electrical parameters, environmental parameters, and mechanical parameters;

[0089] Step S2 involves extracting features from the multimodal data information of each power device to obtain a multimodal parameter set; specifically, as follows:

[0090] Preprocess the collected multimodal data information of each power device, including: removing invalid data and outliers to ensure data quality; calibrating the collected data to ensure that the data from different sensors are synchronized in time and in consistent units; filling in missing data through interpolation;

[0091] The process of processing the acquired data and extracting useful information includes: time domain analysis, calculating statistical quantities such as average, maximum, minimum and root mean square; frequency domain analysis, using Fourier transform to analyze the frequency components of the signal; time-frequency analysis, using wavelet transform and other methods to analyze the time-frequency characteristics of the signal; filtering, applying bandpass filters to remove noise interference;

[0092] After that, the electrical parameters are feature extracted, including: voltage / current waveform analysis, using Fourier transform to analyze the spectrum characteristics of voltage and current and extract harmonic content; power analysis, calculating the changing trends of active power, reactive power and apparent power, as well as the fluctuation of power factor; transient event detection, capturing transient phenomena such as voltage drop, voltage rise, and instantaneous interruption through short-time Fourier transform or wavelet transform;

[0093] Environmental parameter feature extraction, including: temperature and humidity features, analyzing the time series data of temperature and humidity, extracting the changing trends and periodic changes of temperature and humidity, etc.; air pressure features, considering the trend of air pressure changing with time and weather;

[0094] Mechanical parameter feature extraction, including: vibration analysis, using fast Fourier transform or wavelet transform to analyze vibration signals and extract the frequency components, amplitude and phase information of vibration; displacement and speed, using time domain statistical indicators such as mean, standard deviation and peak value to describe the distribution characteristics of displacement and speed; pattern recognition, using principal component analysis and independent component analysis to identify pattern features in mechanical parameters;

[0095] Select the features of the extracted multimodal data information of the power equipment. The selection methods include: correlation analysis, which removes redundant features by calculating the correlation coefficients between the features; importance assessment, which uses machine learning methods to assess the importance of features; feature dimensionality reduction, which uses principal component analysis or linear discriminant analysis to reduce feature dimensions;

[0096] Normalize or standardize the extracted features to ensure that they are compared on the same scale;

[0097] Integrate all extracted features to form a multimodal parameter set;

[0098] The multimodal parameter set is:

[0099]

[0100] Among them, X i represents the multimodal parameter set, E i , H i and M i are the electrical parameters, environmental parameters and mechanical parameter vectors of the i-th power equipment respectively; where E i , H i and M i Further broken down into:

[0101]

[0102] Among them, V i ,I i ,P i ,Q i ,S i They represent the electrical parameters such as voltage, current, active power, reactive power and apparent power of the i-th power equipment respectively;

[0103]

[0104] Among them, Ti ,RH i ,P atm,i They represent environmental parameters such as temperature, relative humidity and atmospheric pressure respectively;

[0105]

[0106] Among them, ω i ,x i ,v i ,a i Represent the mechanical parameters of rotation speed, displacement, velocity and acceleration respectively.

[0107] By collecting data on electrical parameters, environmental parameters and mechanical parameters, the operating status of power equipment can be comprehensively evaluated from multiple angles, and whether there are potential problems with the equipment can be accurately determined; by performing detailed preprocessing and feature extraction on the multimodal data of each device, it is ensured that the extracted features can effectively characterize the operating status of the equipment and improve the accuracy of fault diagnosis; dynamic adjustment of abnormal power equipment identification is performed according to the actual operating conditions of the power equipment, avoiding a "one-size-fits-all" diagnosis method and reducing the risk of false alarms and missed alarms; by generating monitoring strategies for abnormal power equipment, a customized monitoring solution is provided for each abnormal power equipment, including the target parameters and acquisition frequency for monitoring, so as to effectively allocate monitoring resources and reduce operation and maintenance costs; through multimodal data analysis, It can distinguish between system-level transient events and equipment failures, avoiding false alarms and missed alarms caused by "one-size-fits-all" diagnosis; accurate and timely fault diagnosis can effectively prevent the occurrence of potential faults and ensure the overall stability and reliability of the power system; customized monitoring strategies allow operation and maintenance personnel to conduct equipment inspection and maintenance work more targeted, reduce unnecessary inspections, and optimize resource utilization; by providing rich data support, it is convenient for operation and maintenance personnel to make more informed decisions; by integrating electrical, environmental and mechanical multimodal data, it improves the accuracy and timeliness of power fault diagnosis, reduces false alarms and missed alarms, comprehensively evaluates the status of power equipment, supports dynamic adjustment of monitoring strategies, optimizes resource allocation, and ensures the stable operation of the power system and efficient and reliable power supply.

[0108] S3, integrating all multimodal parameter sets collected simultaneously to obtain a multimodal parameter feature matrix capable of characterizing the operating status of all power equipment of the same model in the substation;

[0109] In step S3, all collected multimodal parameter sets are integrated to form a multimodal parameter feature matrix; specifically, it includes:

[0110] S31. First, clean the collected multimodal data to remove invalid or abnormal data points to ensure data quality;

[0111] S32, normalizing or standardizing the data so that different types of measurements from different sensors can be compared on the same scale;

[0112] S33, then organizing the electrical parameters, environmental parameters and mechanical parameters of each power device into a matrix; each column represents a specific power device, including all multi-modal parameters of the power device, and each row represents a specific modal parameter, such as voltage, temperature or vibration intensity;

[0113] Construct a multimodal parameter feature matrix. Through the above arrangement, a multimodal parameter feature matrix is ​​constructed, in which each column corresponds to a specific power equipment in the substation area, and the column contains all the electrical, environmental and mechanical parameters of this equipment. Each row represents a specific type of parameter, and this row shows the measured values ​​of all equipment on this parameter.

[0114] The multimodal parameter feature matrix is:

[0115]

[0116] Among them, the multi-modal parameters of the same power equipment are located in the same column, and the same modal parameters corresponding to different power equipment are located in the same row; E1, E2, …, E N Represents the electrical parameter vector of all power equipment, H1, H2, ..., H N Represents the environmental parameter vector of all power equipment, M1, M2, ..., M N Represents the mechanical parameter vector of all electrical equipment.

[0117] By collecting data on electrical parameters, environmental parameters, and mechanical parameters, the operating status of power equipment is comprehensively evaluated from multiple perspectives, and it is accurately determined whether there are potential problems with the equipment. By performing detailed preprocessing and feature extraction on the multimodal data of each device, it is ensured that the extracted features can effectively characterize the operating status of the equipment and improve the accuracy of fault diagnosis. Dynamic adjustment of abnormal power equipment identification is performed according to the actual operating conditions of the power equipment, avoiding a "one-size-fits-all" diagnosis method and reducing the risk of false alarms and missed alarms. By generating monitoring strategies for abnormal power equipment, a customized monitoring solution is provided for each abnormal power equipment, including the target parameters for monitoring and the acquisition frequency, so as to effectively allocate monitoring resources and reduce operation and maintenance costs. Through multimodal data analysis, it is possible to distinguish between system-level transient events and equipment failures themselves, avoiding Avoid the problem of false alarms and missed alarms caused by "one-size-fits-all" diagnosis; accurate and timely fault diagnosis can effectively prevent the occurrence of potential faults and ensure the overall stability and reliability of the power system; customized monitoring strategies allow operation and maintenance personnel to conduct equipment inspection and maintenance work more targeted, reduce unnecessary inspections, and optimize resource utilization; by integrating electrical, environmental and mechanical multimodal data, improve the accuracy and timeliness of power fault diagnosis, and reduce false alarms and missed alarms; comprehensively evaluate the status of power equipment, support dynamic adjustment of monitoring strategies, optimize resource allocation, and ensure the stable operation of the power system and efficient and reliable power supply; by constructing a multimodal parameter feature matrix, a comprehensive evaluation of the equipment operating status can be achieved, dynamic monitoring strategy adjustment can be supported, false alarms and missed alarms can be reduced, resource allocation can be optimized, and stable operation of the power system can be ensured.

[0118] S4, inputting the multimodal parameter feature matrix into an abnormal power equipment identification model, and the abnormal power equipment identification model outputs an abnormal vector of the abnormal power equipment; the abnormal vector includes the identity number of the power equipment and the degree of abnormality;

[0119] Step S4 is to input the multimodal parameter feature matrix into the abnormal power equipment identification model to identify which power equipment exhibits abnormal behavior; specifically as follows:

[0120] First, build a power equipment identification model, including:

[0121] In the multimodal parameter feature matrix generated in step S3, the multimodal parameters of the same power device are located in the same column, including all the multimodal parameters of the device; the same modal parameters corresponding to different power devices are located in the same row, showing the measured values ​​of all devices on the parameter;

[0122] Clean the acquired data and check the data quality to ensure there are no abnormal or invalid data points;

[0123] Divide the data into a training set and a test set. The training set is used to train the model, and the test set is used to test the model performance.

[0124] Based on the result of feature selection in step S2, the feature set to be finally used is further confirmed using methods such as correlation analysis and importance assessment;

[0125] According to the task of detecting abnormal power equipment, select the appropriate model, such as support vector machine, clustering algorithm, isolation forest or autoencoder; use cross-validation and other techniques to adjust model parameters to achieve the best performance;

[0126] Use the training set data to train the selected model; evaluate the model performance on the validation set and adjust the model until you are satisfied;

[0127] The trained power equipment identification model is used to measure the abnormality of each device, reflecting the degree to which the device deviates from normal behavior; the threshold of the abnormality level is set according to business needs to determine whether the device is abnormal;

[0128] The multimodal parameter feature matrix is ​​input into the power equipment identification model, and the power equipment identification model outputs an abnormal vector list, which includes the identity number and abnormal degree of the abnormal power equipment; the abnormal vector list is used to formulate a monitoring strategy for abnormal equipment;

[0129] Adjust the power equipment identification model according to the evaluation results and retrain the power equipment identification model if necessary;

[0130] The method for calculating the abnormality degree in the abnormal vector of the abnormal power equipment includes:

[0131] For each power device, the abnormal power device identification model outputs an abnormality degree. The abnormality degree is a numerical value that indicates the degree to which the device deviates from normal behavior. The higher the score, the more likely the device is to be abnormal.

[0132] A threshold is set based on historical data or expert experience to distinguish between normal and abnormal devices; the threshold is designed to be fixed or dynamically adjusted according to different operating conditions;

[0133] Input the real-time or multi-period multimodal parameter feature matrix into the trained model to obtain the abnormality level of each device; when the abnormality level of a device exceeds the set threshold, the device is judged as an abnormal device and its abnormality level is recorded;

[0134] For each device that is judged to be abnormal, an abnormality vector is output, which includes the device's identification number and the degree of abnormality;

[0135] For each electrical device i, the abnormality level AS of the electrical device is calculated based on the electrical parameters, environmental parameters and mechanical parameters. i ;

[0136] When there is N e Electrical parameters, N h environmental parameters and N m When the mechanical parameters are equal, the calculation formula of abnormality degree is:

[0137]

[0138] Among them, the standardized value of the jth electrical parameter of the i-th device is:

[0139]

[0140] in, represents the jth electrical parameter of the i-th device, μ e ,σ e are the mean and standard deviation of electrical parameters, respectively;

[0141] The standardized value of the jth environmental parameter of the i-th device is:

[0142]

[0143] in, represents the jth environmental parameter of the i-th device, μ h ,σ h are the mean and standard deviation of environmental parameters, respectively;

[0144] Normalized value of the jth mechanical parameter of the i-th device:

[0145]

[0146] in, represents the jth mechanical parameter of the i-th device, μ m ,σ m are the mean and standard deviation of environmental parameters, respectively;

[0147] α, β and γ are weight coefficients corresponding to electrical parameters, environmental parameters and mechanical parameters, which are used to balance the importance of different modes;

[0148] like Then the power equipment i is considered abnormal, where is the abnormality threshold.

[0149] By adopting a multimodal parameter feature matrix, the state changes of power equipment can be captured from multiple angles, thereby improving the accuracy of anomaly detection. Data cleaning ensures the quality of data input into the power equipment identification model, avoiding the impact of noise or erroneous data on model training. Methods such as correlation analysis and importance assessment are used to further confirm the feature set that is ultimately used, ensuring that the power equipment identification model can focus on the most relevant features, thereby improving the efficiency and accuracy of the power equipment identification model. The power equipment identification model parameters are adjusted through cross-validation, ensuring that the power equipment identification model not only performs well on the training set, but also has good generalization capabilities on unseen data. The degree of anomaly can quantify the degree to which the equipment deviates from the normal operating state, making it easier for operation and maintenance personnel to quickly identify potential problems and take corresponding measures. The threshold is adjusted based on actual needs and operating conditions. Flexible settings, which can be fixed or dynamically adjusted, help adapt to different scenarios; choose a variety of models for anomaly detection, such as support vector machines, clustering algorithms, isolation forests or autoencoders, etc., making the selection of power equipment identification models more flexible and diverse; by continuously monitoring the multimodal parameter feature matrix in real time or multiple time periods, abnormal situations can be discovered in time to prevent failures; the output anomaly vector contains the equipment's identity number and degree of abnormality, which makes it convenient for maintenance personnel to quickly locate the problem equipment and understand its degree of abnormality; the power equipment identification model can effectively identify abnormal power equipment, provide quantitative assessment of the degree of abnormality, and help the maintenance team take action faster and reduce losses caused by equipment failures. At the same time, due to the adjustability and scalability of the power equipment identification model, it can adapt to the ever-changing environment and technological advances.

[0150] S5. Generate the abnormal power equipment monitoring strategy according to the abnormal vector; the abnormal power equipment monitoring strategy includes monitoring target parameters and acquisition frequency of the target parameters, and the target parameters include at least one of electrical parameters, environmental parameters and mechanical parameters;

[0151] Step S5 is a process of generating a monitoring strategy for abnormal power equipment based on the abnormal vector; the details are as follows:

[0152] Determine which devices need special attention based on the degree of abnormality in the abnormal vector; analyze the multimodal parameter feature matrix to find out the specific parameters that cause the score abnormality in order to determine the focus of monitoring;

[0153] According to the degree of abnormality in the abnormality vector and the relevant data in the multimodal parameter feature matrix, the parameters that need to be focused on are determined; the monitoring indicators include electrical parameters, environmental parameters and mechanical parameters; when the abnormality is mainly caused by electrical parameters, the electrical parameters should become the key monitoring target; if the environmental parameters change greatly, they should also be listed as monitoring targets;

[0154] Adjust the monitoring frequency according to the degree of abnormality and the importance of the equipment; for equipment with a high degree of abnormality or critical equipment, increase the collection frequency to monitor the equipment status more frequently; for equipment with a low score or non-critical equipment, keep the collection frequency low to save resources;

[0155] Based on the above analysis, a specific monitoring strategy is generated for each abnormal power equipment. The monitoring strategy is presented in the form of a document. The monitoring strategy document contains: equipment identification number, a list of target parameters to be monitored, and the collection frequency of each target parameter;

[0156] By analyzing the degree of abnormality and the multimodal parameter feature matrix, we can specifically determine which parameters to monitor, making monitoring more accurate and effective; adjust the monitoring frequency according to the importance and abnormality of the equipment, reasonably allocate monitoring resources, avoid unnecessary data collection, and reduce data processing costs; by carefully analyzing abnormal equipment and determining monitoring priorities, we can effectively reduce the false alarm rate and avoid waste of resources caused by excessive monitoring; through detailed monitoring of abnormal parameters, potential faults can be identified faster and more accurately, and the accuracy of fault detection can be improved; the monitoring strategy can be flexibly adjusted according to changes in equipment status to adapt to different operating conditions and environmental changes; by generating monitoring strategy documents, a clear operating guide is provided for operation and maintenance personnel, so that they can take quick action and reduce fault response time; by generating this abnormal power equipment monitoring strategy, the accuracy and timeliness of power fault diagnosis can be effectively improved, while resource allocation can also be optimized, false alarm rate can be reduced, and the overall stability and efficiency of the power system can be improved.

[0157] S6. Collect parameters of abnormal power equipment based on abnormal power equipment monitoring strategy to obtain a target parameter time series image that can characterize the operating state of the abnormal power equipment;

[0158] Step S6 is responsible for collecting parameters according to the abnormal power equipment monitoring strategy and building a time series image of the target parameters; specifically includes:

[0159] According to the abnormal power equipment monitoring strategy generated in step S5, the abnormal power equipment to be monitored and its corresponding monitoring target parameter list and acquisition frequency are clearly defined; the monitoring target parameters include: electrical parameters, environmental parameters and mechanical parameters;

[0160] For each power equipment marked as abnormal, data collection is started according to the parameter list in the monitoring strategy document. Data collection follows the established time interval or frequency. According to the set collection frequency, data points of electrical parameters, environmental parameters and mechanical parameters are regularly obtained from the equipment to ensure that the data obtained is representative and sufficient for subsequent analysis. The collected data needs to be stored in a structured database for subsequent processing and analysis.

[0161] Preprocess the collected data, such as removing noise, filling missing values, standardizing or normalizing, to ensure that the data quality meets the requirements for further analysis;

[0162] Use preprocessed data to build a time series portrait that can reflect the operating status of the equipment; the time series portrait can intuitively show the trend of abnormal power equipment changing over time;

[0163] Time series portrait of target parameters, including: time series graph, which is used to show the changing trend of each parameter over time; trend line, which represents the changing trend of parameters over time through fitting curve; statistical chart, such as box plot or histogram, which shows the data distribution;

[0164] The characteristics of the target parameter time series portrait include: the image should cover data over a period of time to reflect the changing trend of the equipment operation status; contain information on multiple parameters to facilitate observation of equipment status from multiple dimensions; the same parameters of different power equipment can be compared on the same scale to facilitate discovery of differences;

[0165] Analyze the constructed target parameter time series profile to extract key features that are helpful for fault identification; extract features through time series analysis methods, including: spectrum analysis, used to identify anomalies at specific frequencies; wavelet transform, used to capture patterns within local time windows; statistical analysis, used to calculate statistics such as mean and standard deviation to evaluate parameter fluctuations;

[0166] Conduct in-depth analysis of the constructed target parameter time series profile to find abnormal patterns or behaviors; the analysis results can indicate which parameters are abnormal and the degree of abnormality;

[0167] The collected data, the constructed target parameter time series portrait and the analysis results are organized into a report, which includes: equipment identification number, specific manifestations of abnormal parameters, quantitative indicators of the degree of abnormality and fault type hypothesis.

[0168] By clarifying the list of monitoring target parameters and the acquisition frequency, abnormal power equipment can be monitored in a targeted manner to avoid unnecessary extensive monitoring of all equipment, thereby improving efficiency and reducing costs; electrical parameters, environmental parameters, and mechanical parameters are collected regularly to ensure the timeliness and representativeness of the data, providing a high-quality data basis for subsequent analysis; the data preprocessing process can effectively remove noise, fill in missing values, and standardize or normalize data to ensure that the data quality meets the requirements of further analysis and improve the accuracy of data analysis; by constructing a time series portrait containing multiple parameters, the equipment status can be observed from multiple angles, which helps to fully understand the operating status of the equipment; using visualization methods such as time series graphs, trend lines, and statistical charts, the changing trend of the equipment operating status over time is intuitively displayed, allowing technicians to quickly identify abnormal patterns; using multiple methods such as spectrum analysis, wavelet transform, and statistical analysis A method is used to effectively extract key features and help identify potential failure modes, thereby achieving early warning; in-depth analysis of abnormal modes or behaviors can accurately determine which parameters are abnormal and the degree of abnormality, providing a scientific basis for subsequent maintenance and repair work; organizing them into a report format not only facilitates communication between internal teams, but also facilitates reporting to management to ensure the accuracy and consistency of information transmission; through the systematic monitoring, collection, and analysis process, a closed-loop solution is formed, which helps to systematically identify and solve abnormal problems of power equipment; through effective monitoring and maintenance of abnormal equipment, the reliability and stability of the power system are improved, and downtime and economic losses caused by equipment failures are reduced; the target parameter time series portrait provides a structured and systematic framework for the monitoring of power equipment, improves monitoring efficiency and the accuracy of fault identification, and enhances the reliability and economy of the entire power system.

[0169] S7, inputting the target parameter time series image into the power fault identification model, and the power fault identification model outputs the fault type of the abnormal power equipment;

[0170] Step S7 uses the target parameter time series image constructed in the previous step S6 to further identify the specific fault type of the abnormal power equipment; the details are as follows:

[0171] First, a power fault identification model is constructed, including:

[0172] Collect historical fault data and normal operation data of multi-modal parameter time series portraits, and clean the collected data to remove noise and outliers to ensure data quality;

[0173] Divide the data set into training set, validation set and test set to ensure that the fault types in each data set are evenly distributed and avoid data bias;

[0174] Extract key features from the time series portrait of the target parameters, including statistical features of the time series, such as mean, variance and maximum value; frequency domain features, such as Fourier transform and wavelet transform; morphological features, such as peaks and troughs;

[0175] Consider using advanced features, such as periodic features and trend features extracted from time series data;

[0176] Use feature selection methods such as recursive feature elimination (RFE), model-based feature selection, or correlation analysis to identify the most valuable features;

[0177] Choose appropriate models based on the characteristics of the power fault identification task and the characteristics of the data, including traditional machine learning models such as support vector machines, decision trees, random forests or gradient boosting trees, deep learning models such as convolutional neural networks, recurrent neural networks or long short-term memory networks, etc.

[0178] Use the training set data to train the selected model; perform cross-validation to evaluate the performance of the power fault identification model, and adjust hyperparameters to optimize the power fault identification model;

[0179] The performance of the power fault identification model is evaluated using the validation set and the test set;

[0180] Adjust the power fault identification model structure or hyperparameters, such as learning rate, number of layers or number of nodes, based on the evaluation results; use grid search or random search to find the best parameter combination;

[0181] Deploy the trained power fault identification model to a production environment for real-time or batch processing of newly collected data;

[0182] Monitor the performance of the power fault identification model to ensure that it remains effective over time; regularly retrain the power fault identification model with new data to adapt to changes in the power system;

[0183] After the power fault recognition model is built, the target parameter time series image obtained from step S6 is converted into a format suitable for input into the power fault recognition model; for example, the image is converted into a numerical feature vector, or the image is directly used as the input of the convolutional neural network; when a deep learning model is used, the image needs to be preprocessed, such as normalization or resizing.

[0184] The processed data is input into a trained power fault identification model, and the power fault identification model outputs the fault type of the abnormal power equipment, including one or more fault categories and their confidence scores;

[0185] According to the fault type results output by the power fault identification model, corresponding fault handling and maintenance decisions are made for the equipment.

[0186] By collecting historical fault data and normal operation data of multimodal parameter time series portraits and cleaning them, the quality and reliability of the data are ensured; the reasonable division of the data set reduces the risk of overfitting and ensures that the power fault identification model has good generalization ability; the statistical characteristics, frequency domain characteristics and morphological characteristics of the time series are used to capture information of different dimensions and improve the prediction accuracy of the power fault identification model; the feature selection method is used to reduce irrelevant or redundant features, reduce the complexity of the power fault identification model, and improve the efficiency of the power fault identification model; select a suitable model according to the task characteristics and data characteristics, whether it is a traditional machine learning model or a deep learning model, to flexibly respond to different data sizes and complexities; use cross-validation to evaluate the performance of the power fault identification model, and optimize the performance of the power fault identification model by adjusting hyperparameters to ensure the generalization ability of the power fault identification model on unknown data; use the validation set and test set to evaluate the performance of the power fault identification model, and objectively understand The effectiveness of the model; by regularly retraining the power fault identification model with new data, it adapts to the dynamic changes of the power system and ensures the timeliness of the power fault identification model; continuously monitors the performance of the power fault identification model and adjusts it as needed to ensure the long-term effectiveness of the power fault identification model; deploys the trained power fault identification model to the production environment to realize automated fault detection and improve the response speed and efficiency of the power system; by converting the target parameter time series portrait into a format suitable for input, the power fault identification model can be easily integrated into the existing monitoring system; the fault type and its confidence score output by the power fault identification model provide clear guidance, which is convenient for quickly locating the fault and taking appropriate maintenance actions; through comprehensive data preparation, rich feature extraction and selection, flexible model selection, meticulous training and evaluation, continuous improvement and maintenance, and easy-to-integrate deployment, effective identification and rapid response of power faults are achieved, improving the stability and safety of the power system.

[0187] Embodiment 2: Figure 3 As shown, the electric power fault diagnosis system based on multimodal data fusion of the present invention specifically includes the following modules;

[0188] The multimodal data acquisition module is used to collect multimodal data information of all the same type of power equipment in the substation in real time or periodically, including electrical parameters, environmental parameters and mechanical parameters;

[0189] A multimodal feature extraction module is used to preprocess and extract features from the multimodal data information of each power device to form a feature vector containing multiple types of parameters to form a multimodal parameter set;

[0190] The multimodal data integration module is used to integrate the multimodal parameter sets of all power equipment into a multimodal parameter feature matrix, in which each column represents the multimodal parameter of a specific device and each row represents the performance of the same modal parameter on different devices;

[0191] An abnormal power equipment identification module is used to analyze the multimodal parameter feature matrix using a model trained by a machine learning or deep learning algorithm, identify power equipment that exhibits abnormal behavior compared to a normal operating mode, and provide an abnormal vector, including the equipment's identification number and degree of abnormality;

[0192] The abnormal power equipment monitoring strategy generation module generates a customized monitoring strategy for each abnormal power equipment based on the abnormal vector information, including the key monitoring target parameters and the corresponding acquisition frequency;

[0193] The time series image acquisition module is used to collect the target parameter time series data of abnormal power equipment according to the abnormal power equipment monitoring strategy and construct the target parameter time series image;

[0194] The power fault identification module is used to input the target parameter time series portrait into the trained power fault identification model to determine the specific fault type of the abnormal power equipment.

[0195] The multimodal data acquisition module collects information on electrical parameters, environmental parameters, and mechanical parameters of power equipment, which facilitates a comprehensive assessment of the status of the equipment and the impact of environmental factors; the multimodal feature extraction module forms a feature vector containing multiple types of parameters by preprocessing and extracting features from various modal data, thereby improving the accuracy of fault diagnosis; the abnormal power equipment identification module uses a model trained with machine learning or deep learning algorithms to identify power equipment that exhibits abnormal behavior compared to normal operating modes, which can adapt to the dynamic changes of the power system and reduce false alarms and missed alarms; the abnormal power equipment monitoring strategy generation module generates a customized monitoring strategy for each abnormal power equipment based on the abnormal vector, including key monitoring parameters and acquisition frequency, which facilitates the effective allocation of monitoring resources and reduces operation and maintenance costs; the time series portrait acquisition module The module can capture the changing trend of abnormal power equipment in the time series, facilitate the diagnosis of problems caused by transient events, and help technicians quickly locate the cause of the fault; the power fault identification module can accurately determine the specific fault type of the equipment according to the target parameter time series portrait, so as to facilitate the adoption of targeted maintenance measures; through multimodal data analysis, it can distinguish between system-level transient events and equipment failures themselves, avoiding false alarms and missed alarms caused by "one-size-fits-all" diagnosis; through more accurate and timely fault diagnosis, it can effectively prevent the occurrence of potential faults and improve the overall stability and reliability of the power system; the power fault diagnosis system based on multimodal data fusion not only improves the accuracy and timeliness of power fault diagnosis, reduces the risk of false alarms and missed alarms, but also facilitates the optimization of the power system's operation and maintenance strategy to ensure the quality and efficiency of power supply.

[0196] The various variations and specific embodiments of the electric power fault diagnosis method based on multimodal data fusion in Example 1 are also applicable to the electric power fault diagnosis system based on multimodal data fusion in this embodiment. Through the above detailed description of the electric power fault diagnosis method based on multimodal data fusion, those skilled in the art can clearly know the implementation method of the electric power fault diagnosis system based on multimodal data fusion in this embodiment, so for the sake of brevity of the specification, it will not be described in detail here.

[0197] In addition, the present application also provides an electronic device, including a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor. The transceiver, the memory, and the processor are respectively connected via a bus. When the computer program is executed by the processor, each process of the above-mentioned method for controlling output data is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.

[0198] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A power fault diagnosis method based on multimodal data fusion, characterized in that: The method comprises: Collect multi-modal data information of all the same power equipment in the substation area separately; For each power device, feature extraction is performed to obtain a multimodal parameter set; the multimodal parameter set includes electrical parameters, environmental parameters, and mechanical parameters; All multimodal parameter sets collected simultaneously are integrated to obtain a multimodal parameter feature matrix that can characterize the operating status of all power equipment of the same model in the substation area; Inputting the multimodal parameter feature matrix into the abnormal power equipment identification model, the abnormal power equipment identification model outputs an abnormal vector of the abnormal power equipment; the abnormal vector includes the identity number of the power equipment and the degree of abnormality; According to the abnormal vector, the abnormal power equipment monitoring strategy is generated; the abnormal power equipment monitoring strategy includes monitoring target parameters and acquisition frequency of the target parameters, and the target parameters include at least one of electrical parameters, environmental parameters and mechanical parameters; Based on the abnormal power equipment monitoring strategy, the parameters of the abnormal power equipment are collected to obtain a time series image of the target parameters that can characterize the operating status of the abnormal power equipment; The target parameter time series image is input into the power fault identification model, and the power fault identification model outputs the fault type of the abnormal power equipment.

2. The power fault diagnosis method based on multimodal data fusion according to claim 1, characterized in that: The multimodal parameter feature matrix is: ； Among them, the multi-modal parameters of the same power equipment are located in the same column, and the same modal parameters corresponding to different power equipment are located in the same row; Represents the electrical parameter vector of all power equipment, represents the environmental parameter vector of all power equipment, Represents the mechanical parameter vector of all electrical equipment.

3. The power fault diagnosis method based on multimodal data fusion according to claim 2, characterized in that: The multimodal parameter set is: ； in, represents a set of multimodal parameters, and They are The electrical parameters, environmental parameters and mechanical parameters vectors of the power equipment; and Further broken down into: ； in, Respectively represent Electrical parameters such as voltage, current, active power, reactive power and apparent power of each power equipment; ； in, They represent environmental parameters such as temperature, relative humidity and atmospheric pressure respectively; ； in, Represent the mechanical parameters of rotation speed, displacement, velocity and acceleration respectively.

4. The power fault diagnosis method based on multimodal data fusion according to claim 1, characterized in that: The method for obtaining the multimodal parameter feature matrix includes: Clean the collected multimodal data to remove invalid or abnormal data points; Normalize or standardize the data so that different types of measurements from different sensors can be compared on the same scale; Organize the electrical parameters, environmental parameters and mechanical parameters of each power equipment in a matrix; Construct a multimodal parameter feature matrix. Through the above arrangement, a multimodal parameter feature matrix is ​​constructed.

5. The power fault diagnosis method based on multimodal data fusion according to claim 1, characterized in that: The method for calculating the abnormality degree includes: For each electrical equipment , calculate the abnormality of power equipment based on electrical parameters, environmental parameters and mechanical parameters ; When there is Electrical parameters, Environmental parameters and When the mechanical parameters are equal, the calculation formula of abnormality degree is: ； Among them, The first device The standardized values ​​of the electrical parameters are: ； in, Indicates The first device Electrical parameters, are the mean and standard deviation of electrical parameters, respectively; No. The first device The standardized values ​​of the environmental parameters are: ； in, Indicates The first device Environmental parameters, are the mean and standard deviation of environmental parameters, respectively; No. The first device Standardized values ​​of mechanical parameters: ； in, Indicates The first device Mechanical parameters, are the mean and standard deviation of environmental parameters, respectively; and are weight coefficients corresponding to electrical parameters, environmental parameters, and mechanical parameters, used to balance the importance of different modes; like , then the power equipment is considered abnormal, among which is the abnormality threshold.

6. The power fault diagnosis method based on multimodal data fusion according to claim 1, characterized in that: The method for obtaining the exception vector includes: Clean the data in the generated multimodal parameter feature matrix, which contains all multimodal parameters of the device and the measured values ​​of all devices on the parameters; The cleaned data is divided into a training set and a test set; according to the detection task of abnormal power equipment, the support vector machine model is selected as the basic model of the power equipment identification model, and the model parameters are adjusted using cross-validation; the selected model is trained using the training set data, and the model performance is evaluated on the validation set; the trained power equipment identification model is used to measure the abnormality of each device, reflecting the degree to which the device deviates from normal behavior; The multimodal parameter feature matrix is ​​input into the power equipment identification model, and the power equipment identification model outputs an abnormal vector list, which includes the identity number of the abnormal power equipment and the degree of abnormality.

7. The power fault diagnosis method based on multimodal data fusion according to claim 1, characterized in that: The method for obtaining the multimodal parameter set includes: Preprocess the collected multimodal data information of each power device, including removing invalid data, removing outliers, calibrating data, and interpolating missing values; Extract electrical parameter features from pre-processed multi-modal data information, including voltage / current waveform analysis, power analysis, and transient event detection; Extract environmental parameter features from the preprocessed multimodal data information, including temperature, humidity and air pressure features; Extract mechanical parameter features from pre-processed multi-modal data information, including: vibration analysis, displacement and speed, and pattern recognition; Normalize and standardize the extracted electrical parameter characteristics, environmental parameter characteristics and mechanical parameters; All normalized and standardized electrical parameter characteristics, environmental parameter characteristics, and mechanical parameters are integrated to form a multimodal parameter set.

8. A power fault diagnosis system based on multimodal data fusion, characterized in that: The system comprises: The multimodal data acquisition module is used to collect multimodal data information of all the same type of power equipment in the substation in real time or periodically, including electrical parameters, environmental parameters and mechanical parameters; A multimodal feature extraction module is used to preprocess and extract features from the multimodal data information of each power device to form a feature vector containing multiple types of parameters to form a multimodal parameter set; The multimodal data integration module is used to integrate the multimodal parameter sets of all power equipment into a multimodal parameter feature matrix, in which each column represents the multimodal parameter of a specific device and each row represents the performance of the same modal parameter on different devices; An abnormal power equipment identification module is used to analyze the multimodal parameter feature matrix using a model trained by a machine learning or deep learning algorithm, identify power equipment that exhibits abnormal behavior compared to a normal operating mode, and provide an abnormal vector, including the equipment's identification number and degree of abnormality; The abnormal power equipment monitoring strategy generation module generates a customized monitoring strategy for each abnormal power equipment based on the abnormal vector information, including the key monitoring target parameters and the corresponding acquisition frequency; The time series image acquisition module is used to collect the target parameter time series data of abnormal power equipment according to the abnormal power equipment monitoring strategy and construct the target parameter time series image; The power fault identification module is used to input the target parameter time series portrait into the trained power fault identification model, output the fault type of the abnormal power equipment, and determine the specific fault type of the abnormal power equipment.

9. An electronic device for power fault diagnosis based on multimodal data fusion, comprising a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the transceiver, the memory, and the processor are connected via the bus, and characterized in that: When the computer program is executed by the processor, the steps in the method according to any one of claims 1 to 7 are implemented.

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

Images