CN116127324A - Fault detection method and device for key components in uninterruptible power supply - Google Patents
Fault detection method and device for key components in uninterruptible power supply Download PDFInfo
- Publication number
- CN116127324A CN116127324A CN202310155943.0A CN202310155943A CN116127324A CN 116127324 A CN116127324 A CN 116127324A CN 202310155943 A CN202310155943 A CN 202310155943A CN 116127324 A CN116127324 A CN 116127324A
- Authority
- CN
- China
- Prior art keywords
- data
- model
- training
- fault detection
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 51
- 238000005070 sampling Methods 0.000 claims abstract description 32
- 229940050561 matrix product Drugs 0.000 claims abstract description 24
- 238000000034 method Methods 0.000 claims abstract description 23
- 238000007781 pre-processing Methods 0.000 claims abstract description 19
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 7
- 238000004458 analytical method Methods 0.000 claims description 18
- 238000004891 communication Methods 0.000 claims description 18
- 238000002955 isolation Methods 0.000 claims description 11
- 230000005540 biological transmission Effects 0.000 claims description 10
- 238000013500 data storage Methods 0.000 claims description 10
- 230000003287 optical effect Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000011156 evaluation Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 6
- 238000003062 neural network model Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 5
- 238000012795 verification Methods 0.000 claims description 5
- 238000011176 pooling Methods 0.000 claims description 4
- 238000004140 cleaning Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 238000007689 inspection Methods 0.000 claims description 3
- 238000006467 substitution reaction Methods 0.000 claims description 3
- 238000010200 validation analysis Methods 0.000 claims description 3
- 238000000638 solvent extraction Methods 0.000 claims description 2
- 238000003745 diagnosis Methods 0.000 description 9
- 238000012423 maintenance Methods 0.000 description 6
- 230000007547 defect Effects 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 3
- 230000000750 progressive effect Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 239000002253 acid Substances 0.000 description 1
- 238000013523 data management Methods 0.000 description 1
- 238000004146 energy storage Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 230000010076 replication Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J9/00—Circuit arrangements for emergency or stand-by power supply, e.g. for emergency lighting
- H02J9/04—Circuit arrangements for emergency or stand-by power supply, e.g. for emergency lighting in which the distribution system is disconnected from the normal source and connected to a standby source
- H02J9/06—Circuit arrangements for emergency or stand-by power supply, e.g. for emergency lighting in which the distribution system is disconnected from the normal source and connected to a standby source with automatic change-over, e.g. UPS systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02M—APPARATUS FOR CONVERSION BETWEEN AC AND AC, BETWEEN AC AND DC, OR BETWEEN DC AND DC, AND FOR USE WITH MAINS OR SIMILAR POWER SUPPLY SYSTEMS; CONVERSION OF DC OR AC INPUT POWER INTO SURGE OUTPUT POWER; CONTROL OR REGULATION THEREOF
- H02M7/00—Conversion of ac power input into dc power output; Conversion of dc power input into ac power output
- H02M7/02—Conversion of ac power input into dc power output without possibility of reversal
- H02M7/04—Conversion of ac power input into dc power output without possibility of reversal by static converters
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02M—APPARATUS FOR CONVERSION BETWEEN AC AND AC, BETWEEN AC AND DC, OR BETWEEN DC AND DC, AND FOR USE WITH MAINS OR SIMILAR POWER SUPPLY SYSTEMS; CONVERSION OF DC OR AC INPUT POWER INTO SURGE OUTPUT POWER; CONTROL OR REGULATION THEREOF
- H02M7/00—Conversion of ac power input into dc power output; Conversion of dc power input into ac power output
- H02M7/42—Conversion of dc power input into ac power output without possibility of reversal
- H02M7/44—Conversion of dc power input into ac power output without possibility of reversal by static converters
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Power Engineering (AREA)
- Biophysics (AREA)
- Emergency Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Business, Economics & Management (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Stand-By Power Supply Arrangements (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a fault detection method of key components in an uninterruptible power supply, which comprises the following steps: judging whether the current operation mode is set as a training stage of the model: if the training stage is in, performing data preprocessing and data set dividing by using the historical data; if the data is in the non-training stage, current data are collected, data preprocessing is carried out, and the historical data and the current data comprise operation parameter data and image data; the operation parameter data is input into a convolutional neural network model combined with incremental learning, and the image data is input into a tensor network model operated by using multidimensional feature sampling and matrix product state; judging whether the current running mode is set as the training order of the model or not again: if the model is in the training stage, training the model, and if the model meets the requirements, deploying the trained model; if the method is in the non-training stage, outputting a fault detection result, and providing a fault detection device using the method.
Description
Technical Field
The invention relates to the technical field of power electronics, in particular to a fault detection method and device for key components in an uninterruptible power supply.
Background
In the information age, uninterruptible power supply systems (Uninterruptible Power System, UPS) mainly consist of an energy storage module (typically a lead-acid battery), a rectifier, an inverter, a control switch and the like, and utilize the battery to transmit electric energy, and convert direct current into alternating current (commercial power) through module circuits such as the rectifier, the inverter and the like, so as to transmit uninterrupted stable electric energy for important power electronic equipment. In the big data age, the UPS plays an important role in the aspects of guaranteeing information safety and the like, particularly in places such as large hospitals, the power supply object of the UPS is diagnosis and treatment instruments, operation equipment and the like, and the normal operation of the UPS is ensured.
Periodic maintenance and repair of a UPS is a specialized technical task, typically performed by specialized engineering maintenance personnel. However, manual maintenance has the limitations of large dependence on experience and knowledge of technicians, large hysteresis of maintenance and overhaul, incapability of realizing predictive operation and maintenance, large consumption of manpower, material resources, financial resources and the like. Therefore, the limitation of a manual maintenance mode can be effectively overcome by providing an intelligent fault diagnosis method and a corresponding device thereof, but the conventional general fault diagnosis method has the defects of poor mobility and self-adaptation capability among different fields, huge parameter quantity of a depth model, difficulty in realizing edge equipment deployment, real-time prediction analysis and the like.
Disclosure of Invention
The fault detection method and device for the key components in the uninterruptible power supply are high in robustness and generalization.
In one aspect, the present application provides a fault detection method for internal critical components of an uninterruptible power supply, including:
judging whether the current operation mode is set as a training stage of the model: if the training stage is in, performing data preprocessing and data set dividing by using the historical data; if the data is in the non-training stage, current data are collected, data preprocessing is carried out, and the historical data and the current data comprise operation parameter data and image data;
the operation parameter data is input into a convolutional neural network model combined with incremental learning, and the image data is input into a tensor network model operated by using multidimensional feature sampling and matrix product state;
judging whether the current running mode is set as the training order of the model or not again: if the model is in the training stage, training the model, and if the model meets the requirements, deploying the trained model; and if the training stage is in the non-training stage, outputting a fault detection result.
Further, the internal key components of the uninterruptible power supply comprise a storage battery, a rectifier, an inverter and a bypass isolation transformer, and the operation parameter data comprise a main input voltage value U im Bypass input voltage valueU ib Inverter output voltage U oi Inverter output current I oi Inverter frequency value f, load amount R, battery charging voltage U c Battery charging current I c A total of 8 operating parameter variables; the image data includes optical inspection images at 4 locations of the battery, rectifier, inverter and bypass isolation transformer.
Further, 8 operation parameter variables included in the operation parameter data in the historical data are collected, the data from the previous 30 days to the current moment are collected, and the average value of parameters is calculated for 1 time per minute per hour to form an operation parameter data set; for the optical detection images at 4 positions, images from the previous 360 days to the current moment are acquired, and 1 image is acquired every 12 hours, so that an image class data set is formed.
Further, "performing data preprocessing and partitioning of data sets using historical operating parameter data and image data" includes: performing data cleaning and data normalization operations on the operation parameter data by using missing value filling, outlier substitution and noise value smoothing, and performing contrast enhancement, denoising and image clipping operations on image data, wherein the steps of acquiring current operation parameter data and image data and performing data preprocessing comprise the following steps: the operating parameter dataset and the image dataset are divided into a training set and a validation set according to a selected scale.
Further, "collecting current data, and performing data preprocessing" includes performing normalization operation on the operation parameter data and performing image cropping operation on the image data.
Further, "inputting the operation parameter data into the convolutional neural network model in combination with the incremental learning" includes:
calculating the matching degree M between the original sample and the newly added sample i Weight parameter of new convolutional network = weight parameter of original convolutional network x M i Weight parameter of +New convolutional network X (1-M i ) After passing through the convolution network, the data flow obtains the output of the fault detection result through a classifier formed by three full-connection layers, and the matching degree M i The calculation formula of (2) is as follows:
wherein the convolution network is composed of 5 convolution+pooling blocks, k is the size of the feature, n represents the newly added sample, o represents the best matching sample of n, g (n i ) And g (o) i ) The ith features of n and o are denoted respectively, min (·) and max (·) are denoted respectively for minimum and maximum operations, M i The value range of (1) is located at (0, 1)]Between them.
Further, the tensor network model using multidimensional feature sampling and matrix product state operation is composed of a multidimensional sampling layer, a matrix product state network containing 4 MPS operations, a multidimensional sampling layer, a matrix product state network containing 3 MPS operations, and a product state network containing 2 MPS operations which are sequentially connected, wherein the image is divided into 8 sub-areas equally, the multidimensional sampling layer comprises average area sampling (average value of each sub-area is calculated as output), maximum area sampling (maximum value of each sub-area is selected as output), and relevant area sampling (correlation crs of the local area is calculated as output by the following formula), the correlation crs of the local area and tensor C mps The calculation formula of (2) is as follows:
where p represents the total number of pixels per sub-region, (x, y) represents the pixel location per sub-region, E (·) represents the Euclidean distance, MPS represents the matrix product state,
representing low rank tensors, l 1 ，l 2 ，...，l N Representing the introduction of the N-dimensional tensor.
Further, judging whether the trained model meets deployment requirements by adopting accuracy, precision, recall and task matching degree evaluation index TMD, if not, returning to the step of utilizing historical data to perform data preprocessing and data set division, wherein a calculation formula of the task matching degree evaluation index TMD and model deployment judgment index D are as follows:
wherein T is i And F i The method is characterized in that the method respectively represents the number of correct prediction and the number of incorrect prediction of the ith task on the verification set, n is the number of tasks, apr represents the average value of accuracy, precision and recall rate recovery, and min (·) and avg (·) represent minimum and average operation respectively.
On the other hand, the application provides a fault detection device of the inside key part of uninterrupted power supply, includes: the system comprises a data acquisition module, an upper computer operation module and a display alarm module, wherein the data acquisition module comprises an image acquisition device, an electric signal acquisition device, a data processing unit and a data transmission line, the data processing unit is used for completing preprocessing operation of image data and operation parameter data, the upper computer operation module comprises a communication interface, an operation mode judgment device, a server side trainer, a side deployment device and a data storage unit, the communication interface is used for data stream transmission and information communication with other modules, the operation mode judgment device is used for setting a model in a training stage or a non-training stage, the server side trainer is used for completing the training process of a multi-task learning model, the side end deployment device is used for realizing light edge device deployment of the model which completes training and meets deployment requirements, the operation result output by the algorithm model is stored in the data storage unit, the display alarm module comprises a communication interface, a result display unit, a fault alarm unit and a data storage unit, the communication interface is used for data transmission and information communication of different modules, the result display unit is used for carrying out visual display on the operation result of the model, the result analysis unit is used for carrying out logic judgment analysis according to the output result, and transmitting instructions to the fault alarm unit and the data alarm unit is used for recording the fault alarm unit.
The invention provides a fault detection method and device for key components in an uninterruptible power supply, which are a multi-mode and multi-task analysis method, are suitable for intelligent analysis requirements on various faults and different components under different detection environments and ensure the accuracy and reliability of results, and comprise the steps of intelligently analyzing collected operation parameter data by combining a convolution neural network model of incremental learning, effectively solving the field self-adaptation problem of the model, and using a tensor network of multidimensional feature sampling and matrix product state operation to cope with the high redundancy parameter challenges of a general deep image recognition model. In addition, an integrated system comprising a data acquisition module, an upper computer operation module and a display alarm module is provided to realize the full-flow deployment of the fault diagnosis algorithm model on the UPS. The fault detection method and the fault detection device for the key components in the uninterruptible power supply improve the automation and intelligent level of UPS fault diagnosis and simultaneously efficiently ensure the self-adaptive learning and lightweight deployment capacity of an algorithm model.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the description of the embodiments below are briefly introduced, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a fault detection method for key components in an uninterruptible power supply provided by the invention;
FIG. 2 is a schematic diagram of a fault diagnosis model of key components within the uninterruptible power supply of FIG. 1;
fig. 3 is a structural diagram of a fault detection device for key components in an uninterruptible power supply.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. It is to be understood that the described embodiments are preferred embodiments of the invention and should not be taken as excluding other embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without creative efforts, are within the protection scope of the present invention.
The embodiment of the invention provides a fault detection method for internal key components of an Uninterruptible Power Supply (UPS), which is used for monitoring the internal key components of the UPS in real time and outputting fault diagnosis results.
The internal key components of the uninterruptible power supply comprise a storage battery, a rectifier, an inverter and a bypass isolation transformer, and operation parameter data and image data are collected through a multi-source sensor. The operating parameter data includes a main input voltage value U im Bypass input voltage value U ib Inverter output voltage U oi Inverter output current I oi Inverter frequency value f, load amount R, battery charging voltage U c Battery charging current I c A total of 8 operating parameter variables; the image data includes optical inspection images at 4 locations of the battery, rectifier, inverter and bypass isolation transformer. The fault attributes include fault field (including appearance defects and mechanism faults), fault type (including battery faults, rectifier faults, inverter faults and bypass isolation transformer faults), fault class (divided into four classes I/II/III/IV, with progressive severity of the fault) and confidence probability (i.e. how well the model is to determine the reliability of the output result).
For 8 operating parameter variables included in the operating parameter data, data from the previous 30 days to the current moment are collected, and an average value of 1 parameter per minute is calculated every 1 hour, that is, each operating parameter variable has 720 data values, and 720×8=5760 data values are total. The optical detection images at the 4 positions are collected from the previous 360 days to the current moment, and 1 image is collected every 12 hours, namely 720 images at each position, and the total of 720×4=2880 images are obtained. The 5760 data values and 2880 images are divided into a training set and a verification set according to the ratio of 4:1. The 8 operation parameter variable values at each acquisition time are 1 group, the optical detection images at 4 positions at each acquisition time are 1 group, and fault attribute labels (including fault field, fault type and fault grade) are manually marked, and the optical detection images comprise 720 labels corresponding to operation parameters and 720 labels corresponding to images.
As shown in fig. 1, a specific process of a fault detection method for an internal key component of an uninterruptible power supply provided by an embodiment of the present invention is described as follows:
step one, starting the system to power up and starting the running program.
And step two, judging the current operation mode, and checking whether the current operation mode is set as a training stage of the model.
Step three, if the training stage is in, utilizing historical data and carrying out data preprocessing operation according to requirements: aiming at the operation parameter data, performing data cleaning by using missing value filling, outlier substitution and noise value smoothing, and performing data normalization operation; for image data, contrast enhancement, denoising, and image cropping operations are performed. Dividing the dataset (i.e., 2880 images of the image class dataset and 5760 data values of the operating parameter variable) into a training set and a validation set according to a 4:1 ratio (the ratio is a suggested value set empirically, for example, 5:1, 3:1, etc. may also be selected);
step four, if the operation parameter data is in a non-training stage (prediction stage), collecting current operation parameter data or image data, and preprocessing the data according to requirements: performing standardization operation on the operation parameter data; and performing image clipping operation on the image data.
And fifthly, the operation parameter data is input into a convolution neural network model combined with incremental learning, and the image data is input into a tensor network model calculated by using multidimensional feature sampling and matrix product state.
And step six, judging the current operation mode again, and checking whether to set the current operation mode as a training stage of the model.
Step seven, if the model is in a training stage, model training is carried out, and comprehensive analysis is carried out by adopting accuracy, precision, recall and task matching degree evaluation index TMD (Task Match Degree) provided by the embodiment of the invention so as to comprehensively measure whether the trained model meets deployment requirements, and if the model meets the deployment requirements, the trained model is deployed to a subsequent uninterruptible power supply fault detection device; returning to the step three for the model which does not meet the deployment requirement;
and step eight, outputting fault detection results if the device is in a non-training stage (prediction stage), wherein the fault detection results comprise fault fields (appearance defects or mechanism faults), fault types (including storage battery faults, rectifier faults, inverter faults or bypass isolation transformer faults), fault grades (divided into four grades of I/II/III/IV, and the severity of the faults is progressive step by step) and confidence probabilities (namely, how much the model is to grasp and judge the reliability of the output results).
The calculation formula of the task matching degree evaluation index TMD and the model deployment judgment index D are as follows:
wherein T is i And F i Respectively representing the number of correct prediction and the number of incorrect prediction of the ith task on the verification set, wherein n is the number of tasks (in the embodiment of the invention, n=2, including an analysis model for operation parameter data and an analysis model for image data), apr represents the average value of accuracy, precision and recall, and min (·) and avg (·) represent minimum and average operation respectively.
As shown in fig. 2, which shows the principle of a convolutional neural network model incorporating incremental learning and a tensor network model using multidimensional feature sampling and matrix product state operations.
On the one hand, for the operation parameter data, a convolution neural network model combined with incremental learning is input, and the matching degree M between an original sample and a newly added sample is calculated first i And according to M i Determining how much of the weight parameters of the original convolutional network can be used to obtain the new convolutional network by means of parameter replication, i.e. the weight parameters of the new convolutional network = weight parameters of the original convolutional network x M i Weight parameter of +New convolutional network X (1-M i ) The method comprises the steps of carrying out a first treatment on the surface of the The convolution network consists of 5 convolution+pooling blocks, and the settings of the size kernel_size, the step size stride, the number of filters num_filters, and the pooling size and step size stride_stride of the convolution kernel are given in the figure; after passing through the convolution network, the data stream can obtain the output of the fault detection result through a classifier formed by three full-connection layers. Wherein the matching degree M i Wherein k is the size of the feature, n represents the newly added sample, o represents the best matching sample of n, g (n i ) And g (o) i ) The ith features of n and o are denoted respectively, min (·) and max (·) are denoted respectively for minimum and maximum operations, M i The value range of (1) is located at (0, 1)]Between them.
On the other hand, for detected image data, input to a tensor network model using multidimensional feature sampling and matrix product state operation, it is made up of a multidimensional sampling layer and matrix product state network, wherein the image is divided equally into 8 sub-areas, the multidimensional sampling layer includes average area sampling (average value of each sub-area is calculated as output), maximum area sampling (maximum value of each sub-area is selected as output), and correlation area sampling (correlation crs (correlation of region sampling) of the local area is calculated as output by the following formula, where p is the total number of pixels of each sub-area, (x, y) represents the pixel position of each sub-area, and E (·) represents the euclidean distance). The calculation formula of the correlation crs of the local area is as follows:
the matrix product state MPS (matrix product state) is a tensor network state that approximates a higher order tensor by a network of a series of contracted lower order tensors, α N N tensor node representing MPS operation, specifically, one index is l 1 ，l 2 ，...，l N N-dimensional tensors pass through low rank tensors
Approximation is performed to obtain approximated tensor C mps The method comprises the following steps: />
The tensor network model using multidimensional feature sampling and matrix product state operation proposed by the embodiment is composed of a multidimensional sampling layer, a matrix product state network containing 4 MPS operations, a multidimensional sampling layer, a matrix product state network containing 3 MPS operations and a product state network containing 2 MPS operations which are sequentially connected, and finally, fault detection results (including fault field, fault type, fault level and confidence probability) are output.
As shown in fig. 3, the embodiment of the invention provides a fault detection device for internal key components of an uninterruptible power supply based on incremental learning and tensor network, which comprises a data acquisition module, an upper computer operation module and a display alarm module.
The data acquisition module comprises an image acquisition device, an electric signal acquisition device, a data processing unit and a data transmission line. The image collector is used for collecting optical detection images of the storage battery, the rectifier, the inverter and the bypass isolation transformer at 4 positions. The electric signal acquisition equipment is used for acquiring operation parameter data of the storage battery, the rectifier, the inverter and the bypass isolation transformer, and comprises a main input voltage value U im Bypass input voltage value U ib Inverter output voltage U oi Inverter output current I oi Inverter frequency value f, load amount R, battery charging voltage U c Battery charging current I c Electrical data for a total of 8 parameters. The data processing unit is used for completing preprocessing operation of the image data and the operation parameter data. The data transmission line is used for transmitting data streams and communicating information with other modules.
The upper computer operation module comprises a communication interface, an operation mode judging device, a server-side trainer, a side-side deployment device and a data storage unit. The communication interface is used for data stream transmission and information communication with other modules. The running mode judger is used for setting the model in a training phase or a non-training phase. The server side trainer is used for completing the training process of the multi-task learning model, the side end deployment device realizes light-weight edge end equipment deployment on the model which completes the training and meets the deployment requirement, and the operation result output by the algorithm model is stored in the data storage unit, so that the data management and the call at any time are convenient.
The display alarm module comprises a communication interface, a result display unit, a result analysis unit, a fault alarm unit and a data storage unit. The communication interface is used for data transmission and information communication of different modules, and the result display unit is used for visually displaying operation results of the model, wherein the operation results comprise a fault field (appearance defect or mechanism fault), fault types (including storage battery fault, rectifier fault, inverter fault or bypass isolation transformer fault), fault grades FL (divided into four grades of I/II/III/IV, the severity of the fault is progressive step by step) and confidence probability prob (namely, how much the model holds to judge the reliability of the output result); the result analysis unit carries out logic judgment and analysis according to the output result, transmits an instruction to the fault alarm unit, and further executes different fault alarms DA according to the instruction, wherein R/Y/B/G respectively represents red, yellow, blue and green indicator light alarms, the severity is gradually reduced in a color indication mode, and meanwhile, the fault field and the fault type are represented by the indicator light alarms at different positions; the related result analysis and fault alarm records can be stored in the data storage unit, so that the result data can be conveniently taken and analyzed.
According to the fault detection method for the internal key components of the uninterruptible power supply, provided by the embodiment of the invention, a multi-task learning model is designed for different mode data of operation parameters and images to finish intelligent fault detection tasks. Specifically, aiming at the operation parameter data, a convolutional neural network model combined with incremental learning is designed to conduct fault attribute prediction so as to effectively solve the self-adaptive learning problem of the model when a new fault class is introduced; aiming at image data, a tensor network based on multidimensional feature sampling and matrix product state operation is provided for image-based fault identification, so that the problems of high redundancy and large-scale parameters of a general depth image identification model are effectively overcome. The training set data is utilized to complete the training process of a convolutional neural network model combined with incremental learning and a tensor network operated by using multidimensional feature sampling and matrix product state, and the verification set data is utilized to complete performance evaluation of the two models, so that the model meeting the performance requirements is deployed, wherein the model comprises three conventional indexes of accuracy, precision and recall rate, and further comprises a task matching degree evaluation index (TMD) provided by the embodiment of the application.
From the above, the embodiment of the invention constructs a novel multi-mode multi-task analysis method to adapt to intelligent analysis requirements on various faults and different components under different detection environments and ensure the accuracy and reliability of results, and comprises the steps of designing a convolution neural network model combined with incremental learning to carry out intelligent analysis on collected operation parameter data, so that the field self-adaption problem of the model is effectively solved; tensor networks using multidimensional feature sampling and matrix product state operations are designed to address the high redundancy parameter challenges of general depth image recognition models. The field self-adaption and accurate and rapid fault diagnosis are realized, the automation and intelligent level of the fault diagnosis of the uninterruptible power supply is improved, meanwhile, the strong robustness and the high generalization of the fault detection model are ensured, and the self-adaption learning and the light deployment capacity of the algorithm model are ensured.
The foregoing description of the embodiments and description is presented to illustrate the scope of the invention, but is not to be construed as limiting the scope of the invention. Modifications, equivalents, and other improvements to the embodiments of the invention or portions of the features disclosed herein, as may occur to persons skilled in the art upon use of the teachings of the invention or the embodiments described above, in combination with the common general knowledge, the knowledge of one skilled in the art, and/or the prior art, may be made by any appropriate analysis, reasoning, or limited experimentation.
Claims (9)
1. A method for detecting faults of critical components in an uninterruptible power supply, the method comprising:
judging whether the current operation mode is set as a training stage of the model: if the training stage is in, performing data preprocessing and data set dividing by using the historical data; if the data is in the non-training stage, collecting current data, and carrying out data preprocessing, wherein the historical data and the current data comprise operation parameter data and image data;
inputting the operation parameter data into a convolution neural network model combined with incremental learning, and inputting the image data into a tensor network model operated by using multidimensional feature sampling and matrix product state;
judging whether the current running mode is set as the training order of the model or not again: if the model is in the training stage, training the model, and if the model meets the requirements, deploying the trained model; and if the training stage is in the non-training stage, outputting a fault detection result.
2. The method for detecting faults of internal critical components of an uninterruptible power supply according to claim 1, wherein the internal critical components of the uninterruptible power supply comprise a storage battery, a rectifier, an inverter and a bypass isolation transformer, and the operation parameter data comprise a main input voltage value U im Bypass input voltage value U ib Inverter output voltage U oi Inverter output current I oi Inverter frequency value f, load amount R, battery charging voltage U c Battery charging current I c A total of 8 operating parameter variables; the image data includes optical inspection images at 4 locations of the battery, rectifier, inverter and bypass isolation transformer.
3. The fault detection method for internal critical components of an uninterruptible power supply according to claim 2, wherein for 8 operation parameter variables included in the operation parameter data in the history data, data from the previous 30 days to the current moment are collected, and an average value of parameters is calculated 1 time per minute per hour to form an operation parameter data set; for the optical detection images at the 4 positions, the images from the previous 360 days to the current moment are acquired, and 1 image is acquired every 12 hours to form an image class data set.
4. The fault detection method for internal critical components of an uninterruptible power supply as claimed in claim 3, wherein said performing data preprocessing and partitioning of the data set using the historical operating parameter data and the image data comprises: performing data cleaning and data normalization operations on the operation parameter data by using missing value filling, outlier substitution and noise value smoothing, and performing contrast enhancement, denoising and image clipping operations on the image data, wherein the steps of collecting current operation parameter data and image data and performing data preprocessing include: the operating parameter dataset and the image dataset are divided into a training set and a validation set according to a selected scale.
5. The method for detecting the fault of the internal key component of the uninterruptible power supply according to claim 2, wherein the step of collecting current data and performing data preprocessing includes the step of performing normalization operation on the operation parameter data and performing image clipping operation on the image data.
6. The fault detection method for critical components within an uninterruptible power supply of claim 2, wherein inputting the operating parameter data in combination with an incrementally learned convolutional neural network model comprises:
calculating the matching degree M between the original sample and the newly added sample i Weight parameter of new convolutional network = weight parameter of original convolutional network x M i Weight parameter of +New convolutional network X (1-M i ) After passing through the convolution network, the data flow obtains the output of the fault detection result through a classifier formed by three full-connection layers, and the matching degree M i The calculation formula of (2) is as follows:
wherein the convolution network consists of 5 convolution+pooling blocks, k is the size of the feature, n represents the newly added sample, o represents the best matching sample of n, g (n) i ) And g (o) i ) The ith features of n and o are denoted respectively, min (·) and max (·) are denoted respectively for minimum and maximum operations, M i The value range of (1) is located at (0, 1)]Between them.
7. The fault detection method of claim 6, wherein the tensor network model using multidimensional feature sampling and matrix product state operation is composed of a multidimensional sampling layer, a matrix product state network containing 4 MPS operations, a multidimensional sampling layer, a matrix product state network containing 3 MPS operations, and a product state network containing 2 MPS operations sequentially connected, wherein the image is divided into 8 sub-regions equally, the multidimensional sampling layer includes average region sampling, maximum region sampling, and correlation crs and tensor C of the correlation region mps The calculation formula of (2) is as follows:
where p represents the total number of pixels per sub-region, (x, y) represents the pixel location per sub-region, E (·) represents the Euclidean distance, MPS represents the matrix product state,
representing low rank tensors, l 1 ,l 2 ,…,l N Representing the introduction of the N-dimensional tensor.
8. The fault detection method for internal critical components of uninterruptible power supply according to claim 7, wherein accuracy, precision, recall and task matching degree evaluation index TMD are adopted to determine whether the trained model meets deployment requirements, if not, the method returns to the step of using historical data to perform data preprocessing and data set division, wherein a calculation formula of the task matching degree evaluation index TMD and model deployment determination index D are as follows:
wherein T is i And F i The method is characterized in that the method respectively represents the number of correct prediction and the number of incorrect prediction of the ith task on the verification set, n is the number of tasks, apr represents the average value of the accuracy, the precision and the recall rate, and min (·) and avg (·) represent the minimum and average operation respectively.
9. A fault detection device for critical components inside an uninterruptible power supply, the fault detection device comprising: the system comprises a data acquisition module, an upper computer operation module and a display alarm module, wherein the data acquisition module comprises an image acquisition device, an electric signal acquisition device, a data processing unit and a data transmission line, the data processing unit is used for completing preprocessing operation of image data and operation parameter data, the upper computer operation module comprises a communication interface, an operation mode judgment device, a server side trainer, a side deployment device and a data storage unit, the communication interface is used for data stream transmission and information communication with other modules, the operation mode judgment device is used for setting a model in a training stage or a non-training stage, the server side trainer is used for completing a training process of a multi-task learning model, the side deployment device is used for realizing light-weighted edge device deployment of the model which completes training and meets deployment requirements, the operation result output by an algorithm model is stored in the data storage unit, the display alarm module comprises a communication interface, a result display unit, a result analysis unit, a fault alarm unit and a data storage unit, the communication interface is used for data transmission and information communication of different modules, the result display unit is used for visually displaying the operation result of the model, the result display unit is used for completing the training process of the multi-task learning model, the training process is met, the side deployment device is used for completing the training process of the model, the light-weighted edge end deployment device meets the deployment requirements, the operation result and meets the deployment requirements, the operation result is output by the algorithm model is recorded by the algorithm model, and the result is stored by the fault analysis unit.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310155943.0A CN116127324A (en) | 2023-02-13 | 2023-02-13 | Fault detection method and device for key components in uninterruptible power supply |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310155943.0A CN116127324A (en) | 2023-02-13 | 2023-02-13 | Fault detection method and device for key components in uninterruptible power supply |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116127324A true CN116127324A (en) | 2023-05-16 |
Family
ID=86297359
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310155943.0A Pending CN116127324A (en) | 2023-02-13 | 2023-02-13 | Fault detection method and device for key components in uninterruptible power supply |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116127324A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118014051A (en) * | 2024-04-08 | 2024-05-10 | 传申弘安智能(深圳)有限公司 | Multi-mode large model increment learning method and device and computer equipment |
-
2023
- 2023-02-13 CN CN202310155943.0A patent/CN116127324A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118014051A (en) * | 2024-04-08 | 2024-05-10 | 传申弘安智能(深圳)有限公司 | Multi-mode large model increment learning method and device and computer equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112904220B (en) | UPS (uninterrupted Power supply) health prediction method and system based on digital twinning and machine learning, electronic equipment and storable medium | |
| CN112859822B (en) | Equipment health analysis and fault diagnosis method and system based on artificial intelligence | |
| CN112418451A (en) | Transformer fault diagnosis positioning system based on digital twinning | |
| CN113156917B (en) | Power grid equipment fault diagnosis method and system based on artificial intelligence | |
| CN110929918A (en) | 10kV feeder line fault prediction method based on CNN and LightGBM | |
| CN113339204B (en) | Wind driven generator fault identification method based on hybrid neural network | |
| CN116932523B (en) | Platform for integrating and supervising third party environment detection mechanism | |
| CN117032165A (en) | Industrial equipment fault diagnosis method | |
| CN112613494B (en) | Power line monitoring abnormality identification method and system based on deep countermeasure network | |
| CN117829296A (en) | Charging pile fault diagnosis model based on large language model | |
| CN117520924A (en) | Island photovoltaic operation and maintenance fault cause analysis method and system based on multi-mode data | |
| CN116127324A (en) | Fault detection method and device for key components in uninterruptible power supply | |
| CN111143835A (en) | Non-invasive protection method for business logic of electric power metering system based on machine learning | |
| CN118150943B (en) | Distribution network ground fault positioning method, system, electronic equipment and storage medium | |
| CN118096131B (en) | Operation and maintenance inspection method based on electric power scene model | |
| CN117708720B (en) | Equipment fault diagnosis system based on knowledge graph | |
| Gao et al. | Fault prediction in electric power communication network based on improved DenseNet | |
| CN116542509A (en) | Campus logistics task management method and device | |
| CN109886292B (en) | Abnormal reason diagnosis method based on abnormal association graph | |
| CN113419140B (en) | Distribution network line fault diagnosis method and device considering human-computer-object cooperation | |
| CN115556099B (en) | Sustainable learning industrial robot fault diagnosis system and method | |
| CN118363354B (en) | Intelligent manufacturing optimization system based on digital twinning | |
| CN115035328B (en) | Converter image increment automatic machine learning system and building training method thereof | |
| CN114993678A (en) | Three-phase asynchronous motor bearing life prediction method | |
| CN118940832A (en) | Power equipment defect knowledge extraction system and method based on transfer learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |