CN115034312B - Fault diagnosis method for dual neural network model satellite power system - Google Patents

Fault diagnosis method for dual neural network model satellite power system Download PDF

Info

Publication number
CN115034312B
CN115034312B CN202210673879.0A CN202210673879A CN115034312B CN 115034312 B CN115034312 B CN 115034312B CN 202210673879 A CN202210673879 A CN 202210673879A CN 115034312 B CN115034312 B CN 115034312B
Authority
CN
China
Prior art keywords
data
neural network
satellite power
network model
model
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.)
Active
Application number
CN202210673879.0A
Other languages
Chinese (zh)
Other versions
CN115034312A (en
Inventor
孔寒冰
郭小强
章仕起
王凡
吕泽毅
华长春
乔瑟夫·格莱罗
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Yanshan University
Original Assignee
Yanshan University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Yanshan University filed Critical Yanshan University
Priority to CN202210673879.0A priority Critical patent/CN115034312B/en
Publication of CN115034312A publication Critical patent/CN115034312A/en
Application granted granted Critical
Publication of CN115034312B publication Critical patent/CN115034312B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/40—Testing power supplies
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Abstract

The invention discloses a fault diagnosis method for a dual neural network model satellite power system, which comprises the following steps: s1, collecting satellite power supply system data; s2, preprocessing data of the satellite power supply system: dividing the acquired data into a multi-dimensional time sequence through a sliding time window, and processing the multi-dimensional time sequence through a trained stacked noise reduction self-encoder to remove invalid interference data; s3, satellite power system data classification: classifying the preprocessed data; s4, training a double neural network model: inputting the classified datA type I and datA type II in the S3 into A CTCN-GRU model and A TCN-A-GRU model respectively for training; s5, dual neural network model use: and (3) inputting the actually acquired data into the trained dual neural network for recognition after S2 and S3, comparing the output of the network model with the state of an actual satellite power supply system, and evaluating the network performance.

Description

Fault diagnosis method for dual neural network model satellite power supply system
Technical Field
The invention relates to the technical field of fault diagnosis of a satellite power supply system, in particular to a fault diagnosis method of a double-neural network model satellite power supply system.
Background
The satellite power supply system is one of core systems of the satellite, is the only energy source of the satellite, is the heart of the satellite, limits the power and the service life of the satellite, and has great influence on the on-orbit reliable operation of the satellite due to the normal operation of the satellite power supply system, meanwhile, the probability of the fault of the satellite power supply system is the largest in all satellite systems, 80% of faults have serious influence on the whole satellite task, and 45% of faults can cause the failure of the whole satellite task. Therefore, timely finding and diagnosing the type of the satellite power supply fault has great significance for guaranteeing the reliable execution of the satellite on-orbit task.
The diagnosis of the satellite power supply system is not easy due to the complex and varied working environment of the satellite and the diversification of the fault types of the satellite. Existing satellite power system fault diagnosis techniques include several types of signal processing based, model based, digital drive based. The diagnosis method based on signal processing comprises wavelet analysis, fourier transform and other methods, which appear earlier, do not need to accurately model the system, and are simple and convenient to operate, but have limited diagnosis capability on early potential faults, limited capability on mining deep fault characteristics, and no good universality. The diagnosis method based on the model comprises a state detection method, an equivalent space method and other methods, and the technology is developed to be more perfect and successfully applied to engineering, such as Livingstone software introduced by the American national aerospace agency, and provides a fault diagnosis platform based on the model. The method does not need fault data and priori knowledge, has good fault diagnosis capability, but with the more and more complex diagnosis objects, the more and more time and labor are wasted in model building, the external interference is serious, and the transportability is poor. Data-driven based methods include expert systems, fault trees, coarse sets, fuzzy theory, bayesian, and neural network methods. The method automatically mines the system behavior mode by analyzing the historical data of the diagnosis object without accurate modeling, and the data mining is usually realized by a machine learning algorithm. The deep learning technology is an important branch of machine learning, wherein a neural network algorithm is already applied to the field of fault diagnosis, but the neural network algorithm is more limited to be used in the field of fault diagnosis of a satellite power supply system, and an attempt stage is still performed. The method does not need to accurately model the system, has stronger capacity of processing data with interference and better portability, and has wide development space.
The existing neural network-based method applied to satellite power failure diagnosis inputs data acquired by a sensor into a path of model for training, and has two defects: firstly, with the development of satellite technology and telemetry technology, the data information which can be acquired becomes huge, so that a longer training time is needed for inputting a large amount of data into one path of model to carry out model training; secondly, the running environment of the satellite is very complex, part of the acquired input and output data of the satellite power supply system is influenced by the environment and greatly fluctuates due to the running state of the satellite, the other part of the data is quantitatively changed according to the task stage of the satellite, and the two data types are not accurately judged in a dividing mode; thirdly, when the neural network model is applied to satellites in different working environments or executing different tasks, the training needs to be carried out again, and the model training time needs to be spent again.
Disclosure of Invention
The invention provides a fault diagnosis method for a dual neural network model satellite power system, which can be used for autonomously diagnosing the satellite power system when the satellite power system has a fault without transmitting telemetering data back to a ground base station, thereby realizing an efficient and reliable autonomous diagnosis process.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows: a double neural network model satellite power system fault diagnosis method comprises the following steps:
s1, collecting satellite power supply system data: a large amount of data is needed for training a neural network model, and the data is acquired through sensors distributed in a satellite power system shunt regulator, a charging regulator, a discharging regulator, a bus dereferencing module, a solar cell array and a storage battery pack;
s2, preprocessing the data of the satellite power supply system: dividing the acquired data into a multidimensional time sequence through a sliding time window, and processing the multidimensional time sequence through a trained stacked noise reduction self-encoder to remove invalid interference data;
s3, classifying the data of the satellite power system: classifying the preprocessed data, dividing data acquired by sensors connected with a shunt regulator, a charge regulator and a discharge regulator into task data sets, dividing data acquired by sensors connected with a bus dereferencing module, a solar cell array and a storage battery into data sets which are greatly influenced by satellite states and respectively marked as a data type I and a data type II, and preparing for carrying out neural network model training of different data types;
s4, training a double neural network model: building A CTCN-GRU model and A TCN-A-GRU model, and inputting the datA type I and the datA type II classified in the step S3 into the CTCN-GRU model and the TCN-A-GRU model respectively for training;
step S5, using a double neural network model: and (3) carrying out actual fault diagnosis effect test through a neural network model obtained by actual training, inputting the actually acquired data into the trained dual neural network for recognition after the step S2 and the step S3, comparing the output of the network model with the state of an actual satellite power supply system, and evaluating the network performance.
The technical scheme of the invention is further improved as follows: the formula of the encoding process of the self-encoder in step S2 is as follows:
h m ＝f(Wx m +b)；
wherein x is m Training sample set, h, representing a high dimensional space m Representing an output vector set of a hidden layer obtained by an encoder, wherein f represents an activation function from an input layer to the hidden layer, W is a weight matrix, and b is a bias vector;
the decoding process of the self-encoder is formulated as:
Figure BDA0003694124510000031
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0003694124510000032
represents h m Obtaining a data set with the same dimensionality as the input sample set through inverse transformation, wherein g represents an activation function from a hidden layer to an output layer, W' represents a weight matrix, and d is a bias vector;
the mean square error is adopted as the loss function of the self-encoder, and the expression is as follows:
Figure BDA0003694124510000033
wherein, W and W' represent weight matrix, b and d are offset vectors, m is the number of layers, x is sample data,
Figure BDA0003694124510000041
for the data with the same dimensionality as the input sample set obtained by inverse transformation of the hidden layer output vector, the number of iterations is set to minimize a loss function so as to obtain the output h of the encoder m (ii) a The stacked self-encoder is formed by stacking a plurality of self-encoders.
The technical scheme of the invention is further improved as follows: in the step S3, 90% of the datA type I is divided into A training set I and 10% is divided into A testing set I, 90% of the datA type II is divided into A training set II and 10% is divided into A testing set II, and the built CTCN-GRU model and the built TCN-A-GRU model are subjected to fault diagnosis training respectively by utilizing the training set I and the training set II.
The technical scheme of the invention is further improved as follows: the CTCN-GRU model building process comprises the following steps: the TCN comprises a causal convolutional layer, an expansion convolutional layer and a residual connection structure, wherein the convolutional layer is added in front of the causal convolutional layer, and input data is subjected to feature extraction processing through the convolutional layer, the causal convolutional layer, the expansion convolutional layer and the residual connection structure, wherein the operation formula of the expansion convolution is as follows:
Figure BDA0003694124510000042
wherein x is an input sequence; f is a filter; d is the coefficient of expansion; k is the convolution kernel size; s-di ensures that only convolution operations can be performed on past inputs;
after feature extraction is carried out on the CNN layer, the CNN layer is input into a causal convolutional layer to be subjected to time-constrained data processing, the obtained data is input into an expansion convolutional layer to expand the receptive field, the TCN layer is connected with the TCN layer by using residual errors, a Dropout layer is used for regularizing a network in each layer, and the data after feature extraction is carried out on the TCN layer is input into a GRU to be further processed; wherein, the calculation formula of the output state of the GRU is as follows:
Figure BDA0003694124510000043
Figure BDA0003694124510000044
Figure BDA0003694124510000045
Figure BDA0003694124510000046
wherein x is t For input at time t, h t Output or state at time t, h t-1 At time t-1, w is the weight, σ is the activation function sigmoid (. Alpha.), tanh (. Alpha.) is the activation function, z is the activation function t 、r t 、g t Is an intermediate variable.
The technical scheme of the invention is further improved as follows: the TCN-A-GRU model building process comprises the following steps: the TCN comprises a causal convolutional layer, an expansion convolutional layer and a residual connection structure, data are further input into a GRU for processing after being processed by the TCN, and an attention mechanism is introduced, wherein a calculation formula of the attention mechanism is as follows:
Figure BDA0003694124510000051
wherein, a i Assigning attention weights; h is a total of i Outputting for a hidden layer; c is the attention mechanism output and i is the number of hidden layers.
The technical scheme of the invention is further improved as follows: and the datA in the test set I and the test set II are respectively input into the trained CTCN-GRU model and the trained TCN-A-GRU model for fault diagnosis.
Due to the adoption of the technical scheme, the invention has the technical progress that:
the invention provides two new deep learning frames and A new diagnostic mode of A dual neural network model, the collected datA is divided into two datA sets according to the characteristics of the collected datA, the fault diagnosis is carried out by adopting A dual model combining an improved time convolution network and A gated recursion unit mixed model (CTCN-GRU model) with A time convolution network and an attention machine gated recursion unit mixed model (TCN-A-GRU), the input quantity of single model datA is reduced, the training time of one diagnostic model can be shortened, the efficiency is improved, and the diagnostic processing with more pertinence can be carried out on different characteristic datA types, and the diagnostic accuracy is improved; meanwhile, when the diagnosis mode is applied to another satellite with the same kind of task data and different operation environment data with the trained satellite or a satellite with the same kind of operation environment data and different task data, only the retraining of the diagnosis model with changed input data can be carried out, and the model training time is saved.
Drawings
FIG. 1 is a schematic diagram of a data classification-based dual-model satellite power failure diagnosis mode model of the invention;
FIG. 2 is a schematic diagram of the CTCN-GRU model of the invention;
FIG. 3 is A schematic diagram of the TCN-A-GRU model of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the following examples:
as shown in fig. 1, a dual neural network model satellite power system fault diagnosis method includes the following steps:
s1, collecting satellite power supply system data: a large amount of data is needed for training a neural network model, and the data is collected by sensors distributed on a satellite power supply system shunt regulator, a charge regulator, a discharge regulator, a bus value taking module, a solar cell array and a storage battery pack;
s2, preprocessing the data of the satellite power supply system: dividing the acquired data into a multi-dimensional time sequence through a sliding time window, and processing the multi-dimensional time sequence through a trained stacked noise reduction self-encoder to remove invalid interference data; the coding process of the self-encoder is expressed as:
h m ＝f(Wx m +b)；
wherein x is m Training sample set, h, representing a high dimensional space m Representing an output vector set of a hidden layer obtained through an encoder, wherein f represents an activation function from an input layer to the hidden layer, W is a weight matrix, and b is a bias vector;
the decoding process of the self-encoder is formulated as:
Figure BDA0003694124510000061
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0003694124510000062
represents h m Obtaining a data set with the same dimensionality as the input sample set through inverse transformation, wherein g represents an activation function from a hidden layer to an output layer, W' represents a weight matrix, and d is a bias vector;
the mean square error is used as the loss function of the self-encoder, and the expression is as follows:
Figure BDA0003694124510000063
wherein, W and W' represent weight matrix, b and d are offset vectors, m is layer number, x is sample data,
Figure BDA0003694124510000064
for the data with the same dimensionality as the input sample set obtained by inverse transformation of the hidden layer output vector, the number of iterations is set to minimize a loss function so as to obtain the output h of the encoder m (ii) a The stacked self-encoder is formed by stacking a plurality of self-encoders.
S3, classifying the data of the satellite power system: classifying the preprocessed data, dividing data collected by sensors connected with a shunt regulator, a charge regulator and a discharge regulator into task data sets, dividing data collected by sensors connected with a bus value taking module, a solar cell array and a storage battery into data sets which are greatly influenced by satellite states, respectively recording the data sets as a data type I and a data type II, and preparing for carrying out neural network model training of different data types;
s4, training a double neural network model: the dual neural network model is A hybrid model of an improved time convolution neural network and A gated recursion unit, namely CTCN-GRU for short, and A hybrid model of the time convolution neural network and an attention control recursion unit, namely TCN-A-GRU for short;
as shown in FIG. 2, the CTCN-GRU model building process is as follows: the TCN comprises a causal convolutional layer, an expansion convolutional layer and a residual connection structure, wherein the convolutional layer (CNN) is added before the causal convolutional layer, the sensing field of the model is enlarged while the characteristic that the output which does not influence the time t of the TCN is only related to the time t from the previous layer and the convolution of the earlier elements is not influenced, and the input data is subjected to feature extraction processing through the convolutional layer, the causal convolutional layer, the expansion convolutional layer and the residual connection layer, wherein the operation formula of the expansion convolution is as follows:
Figure BDA0003694124510000071
wherein x is an input sequence; f is a filter; d is the coefficient of expansion; k is the convolution kernel size; s-di ensures that only convolution operations can be performed on past inputs;
after the characteristics of the CNN layers are extracted, the CNN layers are input into a causal convolutional layer to be subjected to time-constrained data processing, the obtained data are input into an expansion convolutional layer to expand the receptive field, the TCN layers are connected with one another by using residual errors, a Dropout layer is used for regularizing a network in each layer, the data subjected to characteristic extraction by the TCN are input into a GRU to be further processed, and finally a group of data is output; wherein, the calculation formula of the output state of the GRU is as follows:
Figure BDA0003694124510000072
Figure BDA0003694124510000081
Figure BDA0003694124510000082
Figure BDA0003694124510000083
wherein x is t For input at time t, h t Output or state at time t, h t-1 At time t-1, w is the weight, σ is the activation function sigmoid (. Alpha.), tanh (. Alpha.) is the activation function, z is the activation function t 、r t 、g t Is an intermediate variable.
As shown in FIG. 3, the TCN-A-GRU model building process is as follows: the TCN comprises a causal convolutional layer, an expansion convolutional layer and a residual error connection structure, data are further input into the GRU for processing after being processed by the TCN, and an attention mechanism is introduced, wherein a calculation formula of the attention mechanism is as follows:
Figure BDA0003694124510000084
wherein, a i Assigning attention weights; h is i Outputting for a hidden layer; c is attention mechanism output, i is the number of hidden layers; the attention mechanism has excellent self-regulation capability, the introduction of the attention mechanism can help to strengthen the contribution of important features to fault diagnosis, the diagnosis error caused by the data characteristics of more external interference and larger fluctuation existing in the data type II is avoided, and a more accurate diagnosis result is obtained by utilizing the self-regulation.
And dividing 90% of the datA type I into A training set I and 10% into A testing set I, dividing 90% of the datA type II into A training set II and 10% into A testing set II, and performing fault diagnosis training on the built CTCN-GRU model and the built TCN-A-GRU model respectively by using the training set I and the training set II.
Step S5, using a double neural network model: and (3) carrying out actual fault diagnosis effect test through A neural network model obtained by actual training, respectively inputting datA in the test set I and the test set II into the trained CTCN-GRU model and TCN-A-GRU model for recognition, comparing the output of the network model with the state of an actual satellite power supply system, and evaluating the network performance.

Claims (6)

1. A double neural network model satellite power system fault diagnosis method is characterized by comprising the following steps: the method comprises the following steps:
s1, collecting satellite power supply system data: a large amount of data is needed for training a neural network model, and the data is acquired through sensors distributed in a satellite power system shunt regulator, a charging regulator, a discharging regulator, a bus dereferencing module, a solar cell array and a storage battery pack;
s2, preprocessing the data of the satellite power supply system: dividing the acquired data into a multi-dimensional time sequence through a sliding time window, and processing the multi-dimensional time sequence through a trained stacked noise reduction self-encoder to remove invalid interference data;
s3, classifying the data of the satellite power system: classifying the preprocessed data, dividing data collected by sensors connected with a shunt regulator, a charge regulator and a discharge regulator into task data sets, dividing data collected by sensors connected with a bus value taking module, a solar cell array and a storage battery into data sets which are greatly influenced by satellite states, respectively recording the data sets as a data type I and a data type II, and preparing for carrying out neural network model training of different data types;
s4, training a double neural network model: building A CTCN-GRU model and A TCN-A-GRU model, and inputting the datA type I and the datA type II classified in the step S3 into the CTCN-GRU model and the TCN-A-GRU model respectively for training;
step S5, using a dual neural network model: and (3) testing the actual fault diagnosis effect through a neural network model obtained through actual training, inputting the actually acquired data into the trained dual neural network for recognition after the steps S2 and S3 are carried out, comparing the output of the network model with the state of an actual satellite power supply system, and evaluating the network performance.
2. The dual neural network model satellite power system fault diagnosis method of claim 1, wherein: the formula of the encoding process of the self-encoder in step S2 is as follows:
h m ＝f(Wx m +b)；
wherein x is m Training sample set, h, representing a high dimensional space m Representing an output vector set of a hidden layer obtained by an encoder, wherein f represents an activation function from an input layer to the hidden layer, W is a weight matrix, and b is a bias vector;
the decoding process of the self-encoder is formulated as:
Figure FDA0003694124500000021
wherein the content of the first and second substances,
Figure FDA0003694124500000022
represents h m Obtaining a data set with the same dimensionality as the input sample set through inverse transformation, wherein g represents an activation function from a hidden layer to an output layer, W' represents a weight matrix, and d is a bias vector;
the mean square error is adopted as the loss function of the self-encoder, and the expression is as follows:
Figure FDA0003694124500000023
wherein, W and W' represent weight matrix, b and d are offset vectors, m is the number of layers, x is sample data,
Figure FDA0003694124500000024
for the data with the same dimensionality as the input sample set obtained by inverse transformation of the hidden layer output vector, the number of iterations is set to minimize a loss function so as to obtain the output h of the encoder m (ii) a The stacked self-encoder is formed by stacking a plurality of self-encoders.
3. The dual neural network model satellite power system fault diagnosis method of claim 1, wherein: in the step S3, 90% of the datA type I is divided into A training set I and 10% of the datA type I is divided into A testing set I, 90% of the datA type II is divided into A training set II and 10% of the datA type II is divided into A testing set II, and the built CTCN-GRU model and the built TCN-A-GRU model are subjected to fault diagnosis training respectively by utilizing the training set I and the training set II.
4. The dual neural network model satellite power system fault diagnosis method of claim 3, wherein: the CTCN-GRU model building process comprises the following steps: the TCN comprises a causal convolutional layer, an expansion convolutional layer and a residual connection structure, wherein the convolutional layer is added before the causal convolutional layer, the input data is subjected to feature extraction processing through the convolutional layer, the causal convolutional layer, the expansion convolutional layer and the residual connection structure, and the expansion convolution has the operational formula as follows:
Figure FDA0003694124500000025
wherein x is an input sequence; f is a filter; d is the coefficient of expansion; k is the convolution kernel size; s-di ensures that only convolution operations can be performed on past inputs;
after feature extraction is carried out on the CNN layer, the CNN layer is input into a causal convolutional layer for time-constrained data processing, the obtained data is input into an expansion convolutional layer for expanding a receptive field, the TCN layers are connected with one another by using residual errors, a Dropout layer is used in each layer for regularizing a network, and the data after feature extraction is carried out on the TCN layer is input into a GRU for further processing; wherein, the calculation formula of the output state of the GRU is as follows:
Figure FDA0003694124500000031
Figure FDA0003694124500000032
Figure FDA0003694124500000033
Figure FDA0003694124500000034
wherein x is t For input at time t, h t Output or state at time t, h t-1 At time t-1, w is the weight, σ is the activation function sigmoid (. Alpha.), tanh (. Alpha.) is the activation function, z is the activation function t 、r t 、g t Is an intermediate variable.
5. The dual neural network model satellite power system fault diagnosis method of claim 3, wherein: the TCN-A-GRU model building process comprises the following steps: the TCN comprises a causal convolutional layer, an expansion convolutional layer and a residual connection structure, data are further input into a GRU for processing after being processed by the TCN, and an attention mechanism is introduced, wherein a calculation formula of the attention mechanism is as follows:
Figure FDA0003694124500000035
wherein, a i Assigning attention weights; h is i Outputting for a hidden layer; c is the attention mechanism output and i is the number of hidden layers.
6. The dual neural network model satellite power system fault diagnosis method of claim 3, wherein: and the datA in the test set I and the test set II are respectively input into A trained CTCN-GRU model and A trained TCN-A-GRU model for fault diagnosis.
CN202210673879.0A 2022-06-14 2022-06-14 Fault diagnosis method for dual neural network model satellite power system Active CN115034312B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210673879.0A CN115034312B (en) 2022-06-14 2022-06-14 Fault diagnosis method for dual neural network model satellite power system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210673879.0A CN115034312B (en) 2022-06-14 2022-06-14 Fault diagnosis method for dual neural network model satellite power system

Publications (2)

Publication Number Publication Date
CN115034312A CN115034312A (en) 2022-09-09
CN115034312B true CN115034312B (en) 2023-01-06

Family

ID=83125625

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210673879.0A Active CN115034312B (en) 2022-06-14 2022-06-14 Fault diagnosis method for dual neural network model satellite power system

Country Status (1)

Country Link
CN (1) CN115034312B (en)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110555340A (en) * 2018-05-31 2019-12-10 北京深鉴智能科技有限公司 neural network computing method and system and corresponding dual neural network implementation
CN113076973A (en) * 2021-03-09 2021-07-06 山东师范大学 Sequence anomaly detection method and system based on time convolution network
AU2021102629A4 (en) * 2021-04-22 2021-07-08 Huazhong University Of Science And Technology Lithium battery thermal runaway multi-level detection and early warning system and judgment method
CN113342476A (en) * 2021-07-05 2021-09-03 中山大学 Equipment cluster cross-domain fault prediction method and system based on transfer learning
CN113642754A (en) * 2021-04-23 2021-11-12 北京工业大学 Complex industrial process fault prediction method based on RF noise reduction self-coding information reconstruction and time convolution network
CN113673774A (en) * 2021-08-25 2021-11-19 北京航空航天大学 Aero-engine remaining life prediction method based on self-encoder and time sequence convolution network
CN113705424A (en) * 2021-08-25 2021-11-26 浙江工业大学 Performance equipment fault diagnosis model construction method based on time convolution noise reduction network

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110555340A (en) * 2018-05-31 2019-12-10 北京深鉴智能科技有限公司 neural network computing method and system and corresponding dual neural network implementation
CN113076973A (en) * 2021-03-09 2021-07-06 山东师范大学 Sequence anomaly detection method and system based on time convolution network
AU2021102629A4 (en) * 2021-04-22 2021-07-08 Huazhong University Of Science And Technology Lithium battery thermal runaway multi-level detection and early warning system and judgment method
CN113344024A (en) * 2021-04-22 2021-09-03 华中科技大学 Lithium ion battery thermal runaway grading early warning method and early warning system
CN113642754A (en) * 2021-04-23 2021-11-12 北京工业大学 Complex industrial process fault prediction method based on RF noise reduction self-coding information reconstruction and time convolution network
CN113342476A (en) * 2021-07-05 2021-09-03 中山大学 Equipment cluster cross-domain fault prediction method and system based on transfer learning
CN113673774A (en) * 2021-08-25 2021-11-19 北京航空航天大学 Aero-engine remaining life prediction method based on self-encoder and time sequence convolution network
CN113705424A (en) * 2021-08-25 2021-11-26 浙江工业大学 Performance equipment fault diagnosis model construction method based on time convolution noise reduction network

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
《AME-TCN:Attention Mechanism Enhanced Temporal Convolutional Network for Fault Diagnosis in Industrial Processes》;Jiyang Zhang等;《2021 Global Reliability and Prognostics and Health Management》;20211231;第1-6页 *
《GEO卫星电源系统故障仿真及诊断研究》;高永成;《中国优秀硕士学位论文全文数据库 信息科技辑》;20210115(第1期);第1-99页 *
《基于CNN－GRU网络的轴承故障检测算法》;张伟等;《工业仪表与自动化装置》;20210630(第6期);第88-91页 *
《基于DAE和TCN的复杂工业过程故障预测》;高学金等;《仪器仪表学报》;20210630;第42卷(第6期);第140-151页 *

Also Published As

Publication number Publication date
CN115034312A (en) 2022-09-09

Similar Documents

Publication Publication Date Title
CN110361176B (en) Intelligent fault diagnosis method based on multitask feature sharing neural network
CN109555566B (en) Steam turbine rotor fault diagnosis method based on LSTM
CN111504676B (en) Equipment fault diagnosis method, device and system based on multi-source monitoring data fusion
CN112580263B (en) Turbofan engine residual service life prediction method based on space-time feature fusion
CN113777496A (en) Lithium ion battery residual life prediction method based on time convolution neural network
CN111275164A (en) Underwater robot propulsion system fault diagnosis method
CN110657984A (en) Planetary gearbox fault diagnosis method based on reinforced capsule network
CN112132102B (en) Intelligent fault diagnosis method combining deep neural network with artificial bee colony optimization
CN113850161A (en) Flywheel fault identification method based on LSTM deep noise reduction self-encoder
CN113705396A (en) Motor fault diagnosis method, system and equipment
CN115618732A (en) Nuclear reactor digital twin key parameter autonomous optimization data inversion method
CN114357372A (en) Aircraft fault diagnosis model generation method based on multi-sensor data driving
Ma et al. A collaborative central domain adaptation approach with multi-order graph embedding for bearing fault diagnosis under few-shot samples
CN115034312B (en) Fault diagnosis method for dual neural network model satellite power system
CN112763215A (en) Multi-working-condition online fault diagnosis method based on modular federal deep learning
CN116774086B (en) Lithium battery health state estimation method based on multi-sensor data fusion
US20230350402A1 (en) Multi-task learning based rul predication method under sensor fault condition
CN112598186B (en) Improved LSTM-MLP-based small generator fault prediction method
CN115452376A (en) Bearing fault diagnosis method based on improved lightweight deep convolution neural network
CN114997214A (en) Fault diagnosis method and device for residual error intensive network
Katser et al. Data pre-processing methods for NPP equipment diagnostics algorithms: an overview
Xu et al. A novel scheme for multivariate statistical fault detection with application to the Tennessee Eastman process.
CN117688496B (en) Abnormality diagnosis method, system and equipment for satellite telemetry multidimensional time sequence data
CN114118140B (en) Multi-view intelligent fault diagnosis method and system for explosion-proof motor bearing
CN112133941B (en) Rapid fault diagnosis method for locomotive proton exchange membrane fuel cell system

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
GR01 Patent grant
GR01 Patent grant