CN113837433A - Method for predicting remaining service life of radio frequency system of active digital array radar equipment - Google Patents
Method for predicting remaining service life of radio frequency system of active digital array radar equipment Download PDFInfo
- Publication number
- CN113837433A CN113837433A CN202110927111.7A CN202110927111A CN113837433A CN 113837433 A CN113837433 A CN 113837433A CN 202110927111 A CN202110927111 A CN 202110927111A CN 113837433 A CN113837433 A CN 113837433A
- Authority
- CN
- China
- Prior art keywords
- layer
- radio frequency
- array radar
- radar equipment
- data
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
Abstract
The invention discloses a method for predicting the remaining service life of a radio frequency system of active digital array radar equipment, which relates to the technical field of active digital array radar equipment, in particular to a method for predicting the remaining service life of the radio frequency system of the active digital array radar equipment, and comprises the following steps: s1, selecting partial data of 21 working state signals of 100 digital transceiver modules in the radio frequency transceiver subsystem of the active digital array radar equipment under different working conditions as a training set, and performing data preprocessing on the original working state signals of the training set; s2, carrying out normalization processing on the preprocessed data and inputting the data into a training model; and S3, constructing a neural network model object with a sequential structure. The method for predicting the residual service life of the radio frequency transceiving subsystem not only can greatly reduce the economic cost of maintenance and repair, but also has great practical value in the aspect of enhancing the maintenance and repair guarantee capability of active digital array radar equipment in a battlefield environment.
Description
Technical Field
The invention relates to the technical field of active digital array radar equipment, in particular to a method for predicting the residual service life of a radio frequency system of the active digital array radar equipment.
Background
Modern active digital array radar equipment has been developed into complex systems involving mechanics, electronics, controls, computers, and the like, which are complex in structure, high in performance requirements, and highly integrated in technology.
At present, the maintenance and repair mode of the radio frequency transceiving subsystem of the active digital array radar equipment mainly comprises the following steps: 1. maintaining regularly; 2. after maintenance; 3. the method of maintaining according to the situation includes the following steps:
and (3) regular maintenance: the economy and the pertinence are poor, and the task type maintenance is only carried out at regular intervals;
after maintenance: firstly, the method mainly comprises the steps of directly replacing a field replaceable unit DAM module when the economic benefit is low and the risk coefficient is high; however, in practical circumstances, the possibility of damage to certain components can cause a chain reaction, leading to other failures;
and (4) visual maintenance: the working performance and the working state of the radar equipment at the moment are judged mainly depending on the experience mode of maintenance personnel, so that the maintenance behavior is determined;
the above-mentioned artificial experience is inefficient, has poor reliability and is easy to form 'wrong judgment' and 'missed judgment', which not only affects the service life of the weaponry, but also causes irreparable serious consequences even in actual military operations.
The traditional maintenance technology is not enough to apply data, and the equipment system is lack of relevant and reliable learning ability, and can not accurately judge the future service life of the system with a complex structure, and the main disadvantages are that: the learning ability is low or no due to the fact that an artificial intelligence algorithm is not fully utilized; the prediction program is simple and lacks reliability; the prediction level is relatively low, and the result cannot be reasonably explained; the data can not be fully and reasonably utilized.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a method for predicting the residual service life of a radio frequency system of active digital array radar equipment, and solves the problems in the prior art.
In order to achieve the purpose, the invention is realized by the following technical scheme: the method for predicting the remaining service life of the radio frequency system of the active digital array radar equipment comprises the following steps:
s1, selecting partial data of 21 working state signals of 100 digital transceiver modules in the radio frequency transceiver subsystem of the active digital array radar equipment under different working conditions as a training set, and performing data preprocessing on the original working state signals of the training set;
s2, carrying out normalization processing on the preprocessed data and inputting the data into a training model;
s3, constructing a neural network model object with a sequential structure;
s4, adding a first long-short term memory loop layer for the model object in the step S3;
s5, adding a convolution layer to the model in the step S3;
s6, adding a second long-short term memory cycle layer for the model in the step S3;
s7, preprocessing data of the non-life working state signals of different working conditions in the test set and carrying out normalization processing, and acquiring trend quantitative health indexes of the test set by combining deep features mined in the step S4 and the trend quantitative health index model in the step S5;
and S8, fitting the performance degradation trend of the digital transceiver module by utilizing a polynomial curve, and predicting the RUL of the digital transceiver module.
Optionally, in the first long-short term memory cycle layer in step S4, a time step, a characteristic length of each time step, a number of nerve cell units, a number of Dropout, or a return sequence value is set, and the renormalized characteristic data is used as an input of the long-short term memory cycle network.
Optionally, the convolutional layer in step S5 sets the number of convolutional layers, the number of filters, and the size of convolutional size, uses the data calculated and output by the long and short term memory loop layer as the input of the convolutional layer, and traverses the entire input data sequence by using the convolutional layer and the pooling layer to extract the local information of the feature signal and mine the deep features.
Optionally, in the second long-short term memory loop layer in step S6, a trending quantitative health indicator is constructed by using the advantage that the memory cells in the long-short term memory loop layer have long-short term memory over the time series data, and a trending quantitative health indicator model is established; the R-square decision coefficient was used as a regression accuracy evaluation criterion.
Optionally, in step S4, the process of constructing the memory cell units in the long-short term memory cycle layer is as follows: the hidden layer of the long-short term memory cycle layer introduces a set of memory units, LSTM memory units, which include three gate controllers: an input gate i, a forgetting gate f and an output gate o, which allow the network to learn when to forget the history information and when to update the memory unit with the new information; the memory cell unit of the long-short term memory cycle layer controls the long-term dependence information in the flow capture sequence of the information in the time sequence through the action of three gates, and effectively processes the sequence data; the method comprises the following steps of calculating an update state:
s401, calculating an output value ft of a forgetting stage, forgetting the influence of control history information on the state value of the current memory cell unit, and controlling what information is discarded from the cell state of the memory cell unit, wherein the calculation formula (1) is as follows:
ft=σ(Wxfxt+Whfht-1+bf) (1);
s402, a calculation and memory stage, wherein the calculation and memory stage comprises two parts:
in part, the generation of temporary memory cell unitsBefore the memory cell ct is updated, a temporary memory cell is generatedThe method is characterized in that a deep feature xt input at the current moment t and a hidden layer unit output health index ht-1 at the previous moment t-1 jointly and respectively act with respective weight matrixes in a linear combination mode and then pass through the value of an activation function tanh, and a calculation formula (2) is as follows:
and the second part calculates the value it of the input stage, the input gate controls the influence of the current deep characteristic data input on the state value of the memory unit, and the calculation formula (3) is as follows:
it=σ(Wxixt+Whiht-1+bi) (3)
s403, calculating the current memory cell state value ct, updating the current memory cell state, and calculating formula (4) as follows:
in the formula: expressing the dot-by-dot product, and according to the formula (4), the memory cell unit state updating depends on the unit value ct-1 at the last moment and the current candidate memory unit value, and the forgetting gate and the input gate respectively regulate the two parts;
s404, calculating a value ot of an output stage, controlling the output of the state value of the memory cell unit by an output gate, outputting a memory cell unit value ht of the long-term and short-term memory cycle layer by the output gate based on the current updated cell state of the memory cell unit, and calculating formulas (5) and (6) as follows:
Ot=σ(Wxoxt+Whoht-1+bo) (5)
ht=Ot*tanh(Ct) (6)
in the above expressions (1) to (4), Wxc, Wxi, Wxf, and Wxo are weight values between the input layer xt and the hidden layer ht at time t, respectively.
Optionally, in step S5, local abstract information of the data is automatically extracted by using convolution operation, local link, and weight sharing characteristics of the convolutional neural network to mine deep features.
Optionally, the implementation process of the specific convolution calculation in step S5 is as follows:
inputting the working state data of the active digital array radar equipment radio frequency receiving and transmitting subsystem after normalization into a convolutional layer, wherein the specific operation of the convolutional layer is as shown in formula (7):
in the formula: ki l(j′)Is the jth' weight of the ith convolution kernel of the ith layer,for the j th convolved local region r in the l th layer, representing convolution operation, W is the width of convolution kernel, and y is output to each convolution by using modified linear unit activation functionl(i,j)And (3) carrying out nonlinear transformation, specifically expressed as formula (8):
al(i,j)=f(yl(i,j))=max{0,yl(i,j)} (8)
in the above formula (8): y isl(i,j)For convolutional layer output values, f () is the activation function ReLU, al(i,j)Is yl(i,j)An activation value after an activation function (ReLU);
a pooling layer: and (3) performing dimensionality reduction operation by adopting maximum pooling to minimize network parameters and obtain deep features, wherein the mathematical description of the maximum pooling is shown as a formula (9):
in the above formula: a isl(i,j)Activation value output for the t-th neuron of the ith feature map of the l-th layer, V is the pooling region width, pl(i,j)Is the output value through the pooling layer.
Optionally, in step S6, a second long-short term memory loop layer is added to the model, and a trend quantitative health index model is established for output; using regression accuracy evaluation criteria, the specific accuracy calculation is shown in equation (10):
the invention provides a method for predicting the residual service life of a radio frequency system of active digital array radar equipment, which has the following beneficial effects:
the radio frequency transceiver subsystem is one of the most important subsystems of the active digital array radar equipment, and plays an irreplaceable role in keeping the sensitivity of the whole machine and improving the detection efficiency; the residual service life prediction of the radio frequency transceiving subsystem not only can greatly reduce the economic cost of maintenance and repair, but also has great practical value in the aspect of enhancing the maintenance and repair guarantee capability of active digital array radar equipment in a battlefield environment.
The method for predicting the remaining service life of the active digital array radar equipment radio frequency system has the advantage of solving the problem that the remaining service life of the active digital array radar equipment radio frequency transceiver subsystem is difficult to predict under two modes of performance degradation gradual failure and sudden failure. Firstly, preprocessing a radio frequency transceiving subsystem of active digital array radar equipment, then normalizing working state signals of each sensor obtained by preprocessing, and then carrying out data time sequence segmentation; and takes it as the input of LSTM; the output value of the LSTM utilizes the characteristics of convolution operation, weight sharing and the like of the CNN, local abstract information of data is automatically extracted to mine deep features, and the problem that a traditional feature extraction method depends on expert experience too much is avoided. Then inputting the deep features into an LSTM network, constructing a trend quantitative health index, performing polynomial curve fitting, and using a decision coefficient R square to give model fitting accuracy, and finally predicting future failure time to obtain the prediction of the RUL of the radio frequency transceiver subsystem of the active digital array radar equipment; experimental results show that the trend quantitative health index constructed by the method has good monotonous trend under two fault modes, and the prediction result can be better close to a real life value.
Drawings
FIG. 1 is a schematic diagram of a flow structure of a calculation formula according to the present invention;
FIG. 2 is a schematic diagram of the formula structure of the present invention;
FIG. 3 is a schematic diagram of a prediction model according to the present invention;
FIG. 4 is a schematic diagram of the training process of the present invention;
FIG. 5 is a schematic diagram illustrating comparison between model accuracy evaluation of a training set and a validation set according to the present invention;
FIG. 6 is a schematic diagram showing the comparison of the mean absolute error between the training set and the validation set according to the present invention;
FIG. 7 is a schematic diagram of a comparison structure of Loss function of Loss of Loss between the training set and the verification set according to the present invention;
FIG. 8 is a diagram of the software operating entity of the present invention;
FIG. 9 is a schematic diagram of a residual life prediction fit curve during use of the RF transceiver subsystem of the present invention;
FIG. 10 is a schematic diagram of the calculation of equation (1) according to the present invention;
FIG. 11 is a schematic diagram of the calculation of equation (2) according to the present invention;
FIG. 12 is a schematic diagram of the calculation of equation (4) according to the present invention;
FIG. 13 is a schematic diagram of the calculation formulas (5) and (6) according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments.
Referring to fig. 1 to 13, the present invention provides a technical solution: the method for predicting the remaining service life of the radio frequency system of the active digital array radar equipment comprises the following steps:
s1, selecting partial data of 21 working state signals of 100 digital transceiver modules in the radio frequency transceiver subsystem of the active digital array radar equipment under different working conditions as a training set, and performing data preprocessing on the original working state signals of the training set;
s2, normalizing the preprocessed data and inputting the normalized data into a training model, wherein the input feature dimension N is 25, and the input feature dimension N comprises 21 sensor data features, 3 setting features and 1 normalized life cycle feature;
s3, constructing a neural network model object with a sequential structure;
s4, adding a first long-short term memory cycle layer for the model object in the step S3, setting time steps, the characteristic length of each time step, the number of nerve cell units, the number of Dropout or the value of a return sequence, and using the normalized characteristic data as the input of a long-short term memory cycle network;
s5, adding a convolutional layer for the model in the step S3, setting the number of convolutional layers, the number of filters and the size of convolutional dimension, taking the value data calculated and output by the long-short term memory cyclic layer as the input of the convolutional layer, traversing the whole input data sequence by using the convolutional layer and the pooling layer to extract the local information of the characteristic signal and mine deep layer characteristics;
s6, adding a second long-short term memory cycle layer for the model in the step S3, constructing a trend quantitative health index by utilizing the advantage that the memory units in the long-short term memory cycle layer have long-short term memory on time series data, and establishing a trend quantitative health index model; using the R-party decision coefficient as a regression accuracy evaluation standard;
s7, preprocessing data of the non-life working state signals of different working conditions in the test set and carrying out normalization processing, and acquiring trend quantitative health indexes of the test set by combining deep features mined in the step S4 and the trend quantitative health index model in the step S5;
and S8, fitting the performance degradation trend of the digital transceiver module by utilizing a polynomial curve, and predicting the RUL of the digital transceiver module.
Optionally, in step S4, the process of constructing the memory cell units in the long-short term memory cycle layer is as follows: the hidden layer of the long-short term memory cycle layer introduces a set of memory units, LSTM memory units, which include three gate controllers: an input gate i, a forgetting gate f and an output gate o, which allow the network to learn when to forget the history information and when to update the memory unit with the new information; the memory cell unit of the long-short term memory cycle layer controls the long-term dependence information in the flow capture sequence of the information in the time sequence through the action of three gates, and effectively processes the sequence data; the method comprises the following steps of calculating an update state:
s401, calculating an output value ft of a forgetting stage, forgetting the influence of control history information on the state value of the current memory cell unit, and controlling what information is discarded from the cell state of the memory cell unit, wherein the calculation formula (1) is as follows:
ft=σ(Wxfxt+Whfht-1+bf) (1);
s402, a calculation and memory stage, wherein the calculation and memory stage comprises two parts:
in part, the generation of temporary memory cell unitsBefore the memory cell ct is updated, a temporary memory cell is generatedThe method is characterized in that a deep feature xt input at the current moment t and a hidden layer unit output health index ht-1 at the previous moment t-1 jointly and respectively act with respective weight matrixes in a linear combination mode and then pass through the value of an activation function tanh, and a calculation formula (2) is as follows:
and the second part calculates the value it of the input stage, the input gate controls the influence of the current deep characteristic data input on the state value of the memory unit, and the calculation formula (3) is as follows:
it=σ(Wxixt+Whiht-1+bi) (3)
s403, calculating the current memory cell state value ct, updating the current memory cell state, and calculating formula (4) as follows:
in the formula: expressing the dot-by-dot product, and according to the formula (4), the memory cell unit state updating depends on the unit value ct-1 at the last moment and the current candidate memory unit value, and the forgetting gate and the input gate respectively regulate the two parts;
s404, calculating a value ot of an output stage, controlling the output of the state value of the memory cell unit by an output gate, outputting a memory cell unit value ht of the long-term and short-term memory cycle layer by the output gate based on the current updated cell state of the memory cell unit, and calculating formulas (5) and (6) as follows:
Ot=σ(Wxoxt+Whoht-1+bo) (5)
ht=Ot*tanh(Ct) (6)
in the above expressions (1) to (4), Wxc, Wxi, Wxf, and Wxo are weight values between the input layer xt and the hidden layer ht at time t, respectively.
In the invention, in step S5, local abstract information of data is automatically extracted by using convolution operation, local link and weight sharing characteristics of the convolutional neural network to mine deep features.
In the present invention, the implementation process of the specific convolution calculation in step S5 is:
inputting the working state data of the active digital array radar equipment radio frequency receiving and transmitting subsystem after normalization into a convolutional layer, wherein the specific operation of the convolutional layer is as shown in formula (7):
in the formula: ki l(j′)Is the jth' weight of the ith convolution kernel of the ith layer,for the j th convolved local region r in the l th layer, representing convolution operation, W is the width of convolution kernel, and y is output to each convolution by using modified linear unit activation functionl(i,j)And (3) carrying out nonlinear transformation, specifically expressed as formula (8):
al(i,j)=f(yl(i,j))=max{0,yl(i,j)} (8)
in the above formula (8): y isl(i,j)For convolutional layer output values, f () is the activation function ReLU, al(i,j)Is yl(i,j)An activation value after an activation function (ReLU);
a pooling layer: and (3) performing dimensionality reduction operation by adopting maximum pooling to minimize network parameters and obtain deep features, wherein the mathematical description of the maximum pooling is shown as a formula (9):
in the above formula: a isl(i,j)Activation value output for the t-th neuron of the ith feature map of the l-th layer, V is the pooling region width, pl(i,j) Is the output value through the pooling layer.
In the invention, in step S6, a second long-short term memory loop layer is added to the model to establish a trend quantitative health index model for output; using regression accuracy evaluation criteria, the specific accuracy calculation is shown in equation (10):
the method for predicting the remaining service life of the active digital array radar equipment radio frequency system comprises the steps of preprocessing a radio frequency receiving and transmitting subsystem of the active digital array radar equipment, normalizing working state signals of all sensors obtained through preprocessing, and then performing data time sequence segmentation. And takes it as the input of LSTM; the output value of the LSTM utilizes the characteristics of convolution operation, weight sharing and the like of the CNN, and the local abstract information of the data is automatically extracted to mine deep features, so that the problem that the traditional feature extraction method excessively depends on expert experience is avoided; then inputting the deep features into an LSTM network, constructing a trend quantitative health index, performing polynomial curve fitting, and determining the precision of model fitting by using a decision coefficient R square, and finally predicting the future failure time to obtain the prediction of the RUL of the radio frequency transceiver subsystem of the active digital array radar equipment
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be able to cover the technical scope of the present invention and the equivalent alternatives or modifications according to the technical solution and the inventive concept of the present invention within the technical scope of the present invention.
Claims (8)
1. The method for predicting the remaining service life of the radio frequency system of the active digital array radar equipment comprises the following steps:
s1, selecting partial data of 21 working state signals of 100 digital transceiver modules in the radio frequency transceiver subsystem of the active digital array radar equipment under different working conditions as a training set, and performing data preprocessing on the original working state signals of the training set;
s2, carrying out normalization processing on the preprocessed data and inputting the data into a training model;
s3, constructing a neural network model object with a sequential structure;
s4, adding a first long-short term memory loop layer for the model object in the step S3;
s5, adding a convolution layer to the model in the step S3;
s6, adding a second long-short term memory cycle layer for the model in the step S3;
s7, preprocessing data of the non-life working state signals of different working conditions in the test set and carrying out normalization processing, and acquiring trend quantitative health indexes of the test set by combining deep features mined in the step S4 and the trend quantitative health index model in the step S5;
and S8, fitting the performance degradation trend of the digital transceiver module by utilizing a polynomial curve, and predicting the RUL of the digital transceiver module.
2. The active digital array radar equipment radio frequency system remaining useful life prediction method of claim 1, wherein: in the first long-short term memory cycle layer in step S4, the time step, the characteristic length of each time step, the number of nerve cell units, the number of Dropout, or the value of the return sequence is set, and the renormalized characteristic data is used as the input of the long-short term memory cycle network.
3. The active digital array radar equipment radio frequency system remaining useful life prediction method of claim 1, wherein: in the convolutional layer in step S5, the number of convolutional layers, the number of filters, and the size of convolutional size are set, the value data calculated and output by the long and short term memory loop layer is used as the input of the convolutional layer, and the convolutional layer and the pooling layer are used to traverse the whole input data sequence to extract the local information of the feature signal and to mine the deep features.
4. The active digital array radar equipment radio frequency system remaining service life prediction method of claim 3, wherein: in the second long-short term memory cycle layer in the step S6, a trend quantitative health index is constructed by using the advantage that the memory cells in the long-short term memory cycle layer have long-short term memory for the time series data, and a trend quantitative health index model is established; the R-square decision coefficient was used as a regression accuracy evaluation criterion.
5. The method of claim 2, wherein in step S4, the memory cell units in the long-short term memory cycle layer are constructed by: the hidden layer of the long-short term memory cycle layer introduces a set of memory units, LSTM memory units, which include three gate controllers: an input gate i, a forgetting gate f and an output gate o, which allow the network to learn when to forget the history information and when to update the memory unit with the new information; the memory cell unit of the long-short term memory cycle layer controls the long-term dependence information in the flow capture sequence of the information in the time sequence through the action of three gates, and effectively processes the sequence data; the method comprises the following steps of calculating an update state:
s401, calculating an output value ft of a forgetting stage, forgetting the influence of control history information on the state value of the current memory cell unit, and controlling what information is discarded from the cell state of the memory cell unit, wherein the calculation formula (1) is as follows:
ft=σ(Wxfxt+Whfht-1+bf) (1);
s402, a calculation and memory stage, wherein the calculation and memory stage comprises two parts:
in part, the generation of temporary memory cell unitsBefore the memory cell ct is updated, a temporary memory cell is generated The method is characterized in that a deep feature xt input at the current moment t and a hidden layer unit output health index ht-1 at the previous moment t-1 jointly and respectively act with respective weight matrixes in a linear combination mode and then pass through the value of an activation function tanh, and a calculation formula (2) is as follows:
and the second part calculates the value it of the input stage, the input gate controls the influence of the current deep characteristic data input on the state value of the memory unit, and the calculation formula (3) is as follows:
it=σ(Wxixt+Whiht-1+bi) (3)
s403, calculating the current memory cell state value ct, updating the current memory cell state, and calculating formula (4) as follows:
in the formula: expressing the dot-by-dot product, and according to the formula (4), the memory cell unit state updating depends on the unit value ct-1 at the last moment and the current candidate memory unit value, and the forgetting gate and the input gate respectively regulate the two parts;
s404, calculating a value ot of an output stage, controlling the output of the state value of the memory cell unit by an output gate, outputting a memory cell unit value ht of the long-term and short-term memory cycle layer by the output gate based on the current updated cell state of the memory cell unit, and calculating formulas (5) and (6) as follows:
Ot=σ(Wxoxt+Whoht-1+bo) (5)
ht=Ot*tanh(Ct) (6)
in the above expressions (1) to (4), Wxc, Wxi, Wxf, and Wxo are weight values between the input layer xt and the hidden layer ht at time t, respectively.
6. The active digital array radar equipment radio frequency system remaining useful life prediction method of claim 1, wherein: in step S5, local abstract information of the data is automatically extracted by using convolution operation, local link, and weight sharing characteristics of the convolutional neural network to mine deep features.
7. The active digital array radar equipment radio frequency system remaining service life prediction method according to claim 3, wherein the specific convolution calculation in the step S5 is implemented as follows:
inputting the working state data of the active digital array radar equipment radio frequency receiving and transmitting subsystem after normalization into a convolutional layer, wherein the specific operation of the convolutional layer is as shown in formula (7):
in the formula: ki l(j′)Is the jth' weight of the ith convolution kernel of the ith layer,for the j th convolved local region r in the l th layer, representing convolution operation, W is the width of convolution kernel, and y is output to each convolution by using modified linear unit activation functionl(i,j)And (3) carrying out nonlinear transformation, specifically expressed as formula (8):
al(i,j)=f(yl(i,j))=max{0,yl(i,j)} (8)
in the above formula (8): y isl(i,j)For convolutional layer output values, f () is the activation function ReLU, al(i,j)Is yl(i,j)An activation value after an activation function (ReLU);
a pooling layer: and (3) performing dimensionality reduction operation by adopting maximum pooling to minimize network parameters and obtain deep features, wherein the mathematical description of the maximum pooling is shown as a formula (9):
in the above formula: a isl(i,j)Activation value output for the t-th neuron of the ith feature map of the l-th layer, V is the pooling region width, pl(i,j)Is the output value through the pooling layer.
8. The active digital array radar equipment radio frequency system remaining service life prediction method of claim 4, wherein: in step S6, a second long-short term memory loop layer is added to the model, and a trend quantitative health index model is established for output; using regression accuracy evaluation criteria, the specific accuracy calculation is shown in equation (10):
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110927111.7A CN113837433A (en) | 2021-08-12 | 2021-08-12 | Method for predicting remaining service life of radio frequency system of active digital array radar equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110927111.7A CN113837433A (en) | 2021-08-12 | 2021-08-12 | Method for predicting remaining service life of radio frequency system of active digital array radar equipment |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113837433A true CN113837433A (en) | 2021-12-24 |
Family
ID=78960471
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110927111.7A Pending CN113837433A (en) | 2021-08-12 | 2021-08-12 | Method for predicting remaining service life of radio frequency system of active digital array radar equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113837433A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117272118A (en) * | 2023-11-17 | 2023-12-22 | 成都天奥集团有限公司 | T/R component health state prediction method, system, equipment and medium |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109409567A (en) * | 2018-09-17 | 2019-03-01 | 西安交通大学 | Complex device method for predicting residual useful life based on the double-deck shot and long term memory network |
CN109726524A (en) * | 2019-03-01 | 2019-05-07 | 哈尔滨理工大学 | A kind of rolling bearing remaining life prediction technique based on CNN and LSTM |
US20190235484A1 (en) * | 2018-01-31 | 2019-08-01 | Hitachi, Ltd. | Deep learning architecture for maintenance predictions with multiple modes |
CN110377984A (en) * | 2019-07-02 | 2019-10-25 | 中国科学院深圳先进技术研究院 | A kind of industrial equipment remaining useful life prediction technique, system and electronic equipment |
CN110472800A (en) * | 2019-08-23 | 2019-11-19 | 山东浪潮通软信息科技有限公司 | A kind of machine tool method for predicting residual useful life based on LSTM+CNN |
CN111687689A (en) * | 2020-06-23 | 2020-09-22 | 重庆大学 | Cutter wear state prediction method and device based on LSTM and CNN |
CN113139278A (en) * | 2021-04-08 | 2021-07-20 | 上海工程技术大学 | Deep learning-based bearing residual life prediction method and system |
-
2021
- 2021-08-12 CN CN202110927111.7A patent/CN113837433A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190235484A1 (en) * | 2018-01-31 | 2019-08-01 | Hitachi, Ltd. | Deep learning architecture for maintenance predictions with multiple modes |
CN109409567A (en) * | 2018-09-17 | 2019-03-01 | 西安交通大学 | Complex device method for predicting residual useful life based on the double-deck shot and long term memory network |
CN109726524A (en) * | 2019-03-01 | 2019-05-07 | 哈尔滨理工大学 | A kind of rolling bearing remaining life prediction technique based on CNN and LSTM |
CN110377984A (en) * | 2019-07-02 | 2019-10-25 | 中国科学院深圳先进技术研究院 | A kind of industrial equipment remaining useful life prediction technique, system and electronic equipment |
CN110472800A (en) * | 2019-08-23 | 2019-11-19 | 山东浪潮通软信息科技有限公司 | A kind of machine tool method for predicting residual useful life based on LSTM+CNN |
CN111687689A (en) * | 2020-06-23 | 2020-09-22 | 重庆大学 | Cutter wear state prediction method and device based on LSTM and CNN |
CN113139278A (en) * | 2021-04-08 | 2021-07-20 | 上海工程技术大学 | Deep learning-based bearing residual life prediction method and system |
Non-Patent Citations (1)
Title |
---|
何彦 等: "基于长短时记忆卷积神经网络的刀具磨损在线监测模型", 中国机械工程, vol. 31, no. 16, pages 2 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117272118A (en) * | 2023-11-17 | 2023-12-22 | 成都天奥集团有限公司 | T/R component health state prediction method, system, equipment and medium |
CN117272118B (en) * | 2023-11-17 | 2024-02-13 | 成都天奥集团有限公司 | T/R component health state prediction method, system, equipment and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109726524B (en) | CNN and LSTM-based rolling bearing residual service life prediction method | |
CN103974311B (en) | Based on the Condition Monitoring Data throat floater detection method for improving Gaussian process regression model | |
CN110221225B (en) | Spacecraft lithium ion battery cycle life prediction method | |
CN111274737A (en) | Method and system for predicting remaining service life of mechanical equipment | |
CN110288046B (en) | Fault prediction method based on wavelet neural network and hidden Markov model | |
CN116757534A (en) | Intelligent refrigerator reliability analysis method based on neural training network | |
CN111625516A (en) | Method and device for detecting data state, computer equipment and storage medium | |
US20220172069A1 (en) | Method For Updating Model Of Facility Monitoring System | |
CN111461463A (en) | Short-term load prediction method, system and equipment based on TCN-BP | |
CN115455746B (en) | Nuclear power device operation monitoring data anomaly detection and correction integrated method | |
CN114266201B (en) | Self-attention elevator trapping prediction method based on deep learning | |
CN113837433A (en) | Method for predicting remaining service life of radio frequency system of active digital array radar equipment | |
Wang et al. | A remaining useful life prediction model based on hybrid long-short sequences for engines | |
Xu et al. | High-accuracy health prediction of sensor systems using improved relevant vector-machine ensemble regression | |
CN107038450A (en) | Unmanned plane policing system based on deep learning | |
CN109657907B (en) | Quality control method and device for geographical national condition monitoring data and terminal equipment | |
CN116665798A (en) | Air pollution trend early warning method and related device | |
Dudukcu et al. | Unmanned aerial vehicles (UAVs) battery power anomaly detection using temporal convolutional network with simple moving average algorithm | |
CN112016240B (en) | Prediction method for residual stable service life of incomplete degradation equipment with similar evidence | |
Zou et al. | Fault diagnosis of shield machine based on SOM-BP neural network fusion | |
CN116136926A (en) | Underwater robot health management system based on Bayesian network | |
CN112560252A (en) | Prediction method for residual life of aircraft engine | |
Xiaozhi et al. | Remaining Useful Life Prognostics Based on Deep Combined Temporal Bidirectional Convolutional Network | |
AlDulaimi et al. | The Noisy Multipath Parallel Hybrid Model for Remaining Useful Life Estimation (NMPM) | |
Lskaafi | Fault Diagnosis and Failure Prognostics of Lithium-ion Battery based on Least Squares Support Vector Machine and Memory Particle Filter Framework |
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 |