CN113642779A - ResNet50 network key equipment residual life prediction method based on feature fusion - Google Patents

Abstract

The invention discloses a method for predicting the remaining life of key equipment in a ResNet50 network based on feature fusion. First, correlation analysis is performed on data information collected by various sensors in the system equipment, and feature variables with a low degree of correlation with the life are filtered to determine the impact on the life of the system. Then use the ISOMAP algorithm to fuse the highly correlated feature variables to obtain the fused feature variables, effectively integrating sensor information; for the fused feature variables, and the remaining life at the corresponding moment of the system constitute the total remaining life prediction sample set, and finally Establish a prediction model and predict the remaining life, and obtain the prediction result of the system's remaining life. The invention accurately predicts the service life of the equipment in the industrial field, and prompts the staff to perform preventive maintenance on the equipment in time.

Description

Remaining life prediction method of key equipment in ResNet50 network based on feature fusion

technical field

The invention belongs to the technical field of monitoring and maintenance of an intelligent manufacturing system, in particular to a method for predicting the remaining life of key equipment in a ResNet50 network based on feature fusion.

Background technique

Prognostic and Health Management (PHM) technology is crucial in modern industry and has been widely used in civil aviation, automotive and manufacturing. With the accumulation of time, due to the influence of internal factors and various external factors (such as wear, external impact, load, operating environment), the performance and health status of the equipment will show different degrees of degradation trend, and the accumulation will eventually Cause equipment failure, resulting in immeasurable economic losses and safety accidents. Remaining useful life (RUL) prediction is one of the most important components of Predictive Health Management (PHM) technology in modern industry, and it is defined as the length of useful life from the current moment to the end of that life. In order to improve the safety and reliability of the system and reduce the maintenance cost, it is necessary to ensure that the equipment is maintained at the best time, and the parts are replaced regularly to prevent the equipment failure caused by the accumulation of faults. By monitoring the sensor data of each component of the equipment and accurately predicting the remaining service life of the equipment at a certain time, it is possible to find an appropriate time to perform preventive maintenance on the equipment, reduce equipment downtime and save costs.

Since the prediction of the remaining service life of the system is mainly based on a large amount of monitoring data collected by multiple sensors, the overall degradation status of the system is evaluated and the system is maintained in time. A single sensor feature is not enough to accurately judge the health of a system, so multiple features are needed. However, too many features will cause information redundancy and increase the computational burden. Therefore, the features with low correlation with lifespan are filtered and the remaining multi-sensor features are fused to obtain better health indicators, improve the accuracy of prediction methods, and make preventive maintenance more timely and effective. Aiming at the problem that too many features lead to dimensionality explosion in the research of remaining life prediction, which leads to low computational efficiency, a method for predicting the remaining life of key equipment in the ResNet50 network based on feature fusion is proposed, which can not only enhance the confidence in the research on remaining life prediction , plays a key role in the advancement of the discipline and industry.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method for predicting the remaining life of key equipment in the ResNet50 network based on feature fusion, which can accurately predict the life of the equipment in the industrial field and prompt the staff to perform preventive maintenance on the equipment in time.

The technical solution adopted by the present invention is a method for predicting the remaining life of key equipment in the ResNet50 network based on feature fusion, which is specifically implemented according to the following steps:

Step 1. Perform a correlation analysis on the data information collected by each sensor in the system equipment, filter the characteristic variables with a low degree of correlation with the lifespan, and determine the characteristic variables with a high degree of correlation that affect the lifespan of the system;

Step 2, performing feature fusion on the highly correlated feature variables obtained in the step 1: using the ISOMAP algorithm to fuse the highly correlated feature variables in step 1 to obtain fused feature variables, effectively integrating sensor information;

Step 3. Build a data set construction strategy: For the fusion feature variables obtained in step 2, and the remaining life at the corresponding moment of the system constitute the total remaining life prediction sample set, and divide it into a training set and a test set;

Step 4: Establish a prediction model and predict the remaining life: use the training set of the total remaining life prediction sample set obtained in step 3 as the input of the deep residual network Resnet50 to train the prediction network, and use the trained network to predict the test samples. Obtain the prediction result of the remaining life of the system.

The present invention is also characterized in that,

Step 1 is as follows:

其中，

为系统中i时刻第n个传感器的监测数据信息，i＝1,2,...,t，该系统的剩余寿命Y表示为Y＝{yⁱ}，其中，yⁱ为系统对应i时刻的剩余寿命；Step 1.1. Number the data information collected by the sensors in the system equipment, n is the serial number of the sensor, x _n is the monitoring data information of the nth sensor, and the monitoring data X is expressed as

in,

is the monitoring data information of the nth sensor at time i in the system, i=1,2,...,t, the remaining life Y of the system is expressed as Y={y ⁱ }, where y ⁱ is the time corresponding to the system i remaining life;

Step 1.2, use correlation analysis to calculate the correlation coefficient

和

分别为监测数据信息x_n和剩余寿命Y的标准差；Among them, Cov(x _n , Y) is the covariance between the monitoring data information x _n and the remaining life Y,

and

are the standard deviation of the monitoring data information x _n and the remaining life Y respectively;

具体如下：Step 1.3. Calculate the Pearson correlation coefficient between the monitoring data information x _n and the remaining life Y

details as follows:

范围：

Correlation coefficient

scope:

Retain the highly correlated feature variables whose results exceed 0.8.

Step 2 is as follows:

Step 2.1. Select a neighborhood and construct a neighborhood graph G: set the nearest neighbor parameter to k, and construct a sample set D={x ₁ ,x ₂ ,...,x _m }, m is the number of sample points in the sample set D of the original d-dimensional space, and two data points x _h and x _j are arbitrarily selected in the sample set D={x ₁ ,x ₂ ,...,x _m }, where , h, j are the numbers of the two data points selected in the sample set D, h, j∈1,2,3...m, the distance between the first data point x _h and the neighbor parameter k is set as the Euclidean distance , when another data point x _j is one of the k points of the nearest neighbor parameters of x _h , it means that they are adjacent, and the neighborhood graph G has edges, that is, the neighborhood graph G is constructed;

Step 2.2. Calculate the shortest path distance matrix: If there are edge connections between two points x _h and x _j on the neighborhood graph G, define the shortest path between the sample points x _h and x _j as d _G (x _h , x _j ) =d(x _h , x _j ); otherwise d _G (x _h , x _j )=∞, where d(x _h , x _j ) is the Euclidean distance between data points x _h and x _j ;

For l∈1,2,3...m, we have:

d _G (x _h ,x _j )=min{d _G (x _h ,x _j ),d _G (x _h ,x _l )+d _G (x _l ,x _j )} (3)

Among them, x _l is a new sample point in the sample set D, d _G (x _h ,x _l ) represents the shortest path between the sample points x _h and x _l , d _G (x _l ,x _j ) represents the sample point x The shortest path between _l and x _j , d _G (x _h ,x _j ) represents the shortest path between sample points x _h and x _j ;

其中，m是d维原始空间的样本集D中样本点的个数，h，j，l是样本集D中选取的三个数据点的编号，h，j，l∈1,2,3...m；Based on this, the shortest path distance matrix can be determined as

Among them, m is the number of sample points in the sample set D of the d-dimensional original space, h, j, l are the numbers of the three data points selected in the sample set D, h, j, l ∈ 1, 2, 3. ..m;

Step 2.3. Construct a new d'-dimensional sample space by reducing the dimension of the d-dimensional original space, and keep the distance between the d'-dimensional new sample space and the d-dimensional original space unchanged, and obtain the inner product matrix B of the samples after dimensionality reduction: Let B=Z ^T Z∈R ^m×m , where d is the dimension of the original space, d' is the dimension of the new sample space, Z is the matrix representation of the d-dimensional original space embedded in the d'-dimensional new sample space, and R is set of real numbers;

After constructing the new sample space, d'<d, and the Euclidean distance of any two samples in the new sample space of dimension d' is equal to the distance in the original space of dimension d, namely:

||z _h -z _j ||=dist _hj (4)

Among them, z _h , z _j represent the Euclidean distance of the sample points x _h , x _j in the d'-dimensional new sample space, respectively, and dist _hj represents the distance between the sample points x _h and x _j in the d-dimensional original space;

Expand Equation (4) and square both sides of the equal sign to get:

b_hh与b_jj以此类推，则有：

b _hh and b _jj and so on, there are:

Among them, b _hj is the element of the hth row and j column in the inner product matrix B, b _hh and b _jj and so on, dist _hj represents the distance between the sample points x _h and x _j in the d-dimensional original space, dist _h , dist _·j and dist _· and so on;

The inner product matrix B can be obtained by obtaining the elements in the inner product matrix B such as b _hj , b _hh , b _jj ;

Step 2.4, construct a d-dimensional embedding for the shortest path distance matrix D _G , and finally obtain the matrix representation Z∈R ^d′×m in which the samples are embedded in the d'-dimensional new sample space through the d-dimensional original space;

The inner product matrix B obtained in step 2.3 obtains the eigenvalue matrix and the eigenvector matrix through eigendecomposition, namely:

B=VΛV ^T (8)

Among them, Λ=diag(λ ₁ ,λ ₂ ,...,λ _d ) is a diagonal matrix composed of eigenvalues, λ ₁ ≥λ ₂ ≥...≥λ _d , λ ₁ ,λ ₂ ,... , λ _d is the eigenvalue obtained by decomposition, and V is the corresponding eigenvector matrix.

计算Z的结果后可得到矩阵表示为：

此时的结果

便是融合后的特征变量；Then the matrix representation Z of the d-dimensional original space embedded in the d'-dimensional new sample space is calculated as:

After calculating the result of Z, the matrix can be obtained as:

result at this time

is the feature variable after fusion;

为特征值构成的对角矩阵，特征值

V_*为对应的特征向量矩阵；

为d'维新样本空间中i时刻第m'个样本点的融合特征变量，m'是d'维新样本空间中样本点的个数。where, there are d ^* non-zero eigenvalues,

is a diagonal matrix of eigenvalues, eigenvalues

V _* is the corresponding eigenvector matrix;

is the fusion feature variable of the m'th sample point at time i in the d'-dimensional sample space, where m' is the number of sample points in the d'-dimensional sample space.

Step 3 is as follows:

和系统剩余寿命Y＝{yⁱ}重构为剩余寿命预测样本集

分别为重构的剩余寿命预测样本集中的特征变量，并按照7:3的比例划分为训练集和验证集，其中，yⁱ为系统对应i时刻的剩余寿命；Step 3.1. Construct a sample set: combine the fused feature variables obtained in step 2

and the remaining life of the system Y={y ⁱ } is reconstructed into the remaining life prediction sample set

are the feature variables in the reconstructed remaining life prediction sample set, and are divided into training set and validation set according to the ratio of 7:3, where y ⁱ is the remaining life of the system corresponding to time i;

Step 3.2, construct sub-training set and sub-test set: divide the training set divided in step 3.1 into sub-training set and sub-test set according to the ratio of 7:3 again;

Step 3.3. Construct the total training set and the total test set: Combine the feature variables in the sub-training set obtained in step 3.2 into the total training set in the form of column stacking. The feature variables in the set are merged into the total test set in the form of column stacking, and finally the total remaining life prediction sample set consisting of the total training set, the total test set and the validation set obtained in step 3.1 is formed.

Step 4 is as follows:

Step 4.1. Build a 50-layer deep residual network ResNet50: The deep residual network ResNet50 consists of three parts, namely the convolution pooling part, the four residual block parts, and the pooling and flattening part. There are two residual block parts. The basic structure is the identity residual block and the convolution residual block;

Step 4.2. Feature extraction: The training set of the total remaining life prediction sample set obtained in step 3 is used as the input of the deep residual network Resnet50, and the one-dimensional convolution kernel is used to extract the features of the training set, and the extracted convolution features are processed. max pooling to obtain pooled features;

Step 4.3. Feature extraction of residual block: The pooled feature is used as the input of the residual block. The residual block of the deep residual network Resnet50 consists of the following four parts: a convolution residual block and two identity residual blocks , one convolutional residual block and three identity residual blocks, one convolutional residual block and five identity residual blocks, one convolutional residual block and two identity residual blocks, the above four residual blocks The difference blocks are sequentially connected to form the core part of the residual network;

Step 4.4, establish a prediction model: integrate the output of the residual block through the average pooling layer, and then obtain the prediction model through the flattening layer;

Step 4.5. Obtain the remaining life prediction result: Input the test set into the prediction model to obtain the remaining life prediction result. Each time a prediction is performed, the output prediction result is obtained, and the error between the prediction result and the real data of the remaining life collected by the system is reversed. It is propagated to the deep residual network Resnet50, and the parameters in the network are adjusted. When the error of the network is less than 0.05, it means that the network has converged. The trained network is used to predict the test samples, and finally the prediction result of the remaining life of the system is obtained. , to achieve accurate prediction of the remaining life of the system.

The beneficial effect of the present invention is that, a method for predicting the remaining life of key equipment in the ResNet50 network based on feature fusion, eliminates the low correlation degree features affecting the life of the system equipment through the method of correlation analysis, and uses the ISOMAP algorithm to fuse the high correlation degree features, Effectively integrate sensor information and eliminate coupling between sensors. Combining the ISOMAP algorithm and the deep residual network Resnet50, the fused sensor feature information is used as the input of the deep residual network Resnet50 to train the prediction network, learn the hidden layer relationship between each feature, and build the remaining life prediction model. Prediction of lifespan. The accuracy and efficiency of the proposed method in remaining life prediction are verified by experimental simulation.

Description of drawings

Fig. 1 is the overall flow chart of the residual life prediction method of ResNet50 network key equipment based on feature fusion of the present invention;

Fig. 2 is a graph showing the results of correlation analysis of sensor characteristic variables in the present invention;

Fig. 3 adopts ISOMAP algorithm to carry out feature fusion flow chart in the present invention;

Fig. 4 is the overall structure diagram of the deep residual network ResNet50 in the present invention;

5 is a flowchart of the prediction process of the deep residual network ResNet50 in the present invention;

FIG. 6 is a graph of the prediction result of the present invention.

Detailed ways

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

A method for predicting the remaining life of key equipment in the ResNet50 network based on feature fusion of the present invention, the flowchart is shown in Figure 1, and is specifically implemented according to the following steps:

Step 1. Retain highly correlated feature variables that affect the operating time of system equipment: perform correlation analysis on the data information collected by each sensor in the system equipment, filter feature variables that are less correlated with lifespan, and determine high correlations that affect system lifespan characteristic variable;

With reference to Figures 2 to 5, step 1 is as follows:

其中，

为系统中i时刻第n个传感器的监测数据信息，i＝1,2,...,t，该系统的剩余寿命Y表示为Y＝{yⁱ}，其中，yⁱ为系统对应i时刻的剩余寿命；Step 1.1. Number the data information collected by the sensors in the system equipment, n is the serial number of the sensor, x _n is the monitoring data information of the nth sensor, and the monitoring data X is expressed as

in,

is the monitoring data information of the nth sensor at time i in the system, i=1,2,...,t, the remaining life Y of the system is expressed as Y={y ⁱ }, where y ⁱ is the time corresponding to the system i remaining life;

Step 1.2, use correlation analysis to calculate the correlation coefficient

和

分别为监测数据信息x_n和剩余寿命Y的标准差；Among them, Cov(x _n , Y) is the covariance between the monitoring data information x _n and the remaining life Y,

and

are the standard deviation of the monitoring data information x _n and the remaining life Y respectively;

具体如下：Step 1.3. Calculate the Pearson correlation coefficient between the monitoring data information x _n and the remaining life Y

details as follows:

范围：

Correlation coefficient

scope:

Retain the highly correlated feature variables whose results exceed 0.8.

Step 2, performing feature fusion on the highly correlated feature variables obtained in the step 1: using the ISOMAP algorithm to fuse the highly correlated feature variables in step 1 to obtain fused feature variables, effectively integrating sensor information;

Step 2 is as follows:

Step 2.1. Select a neighborhood and construct a neighborhood graph G: set the nearest neighbor parameter to k, and construct a sample set D={x ₁ ,x ₂ ,...,x _m }, m is the number of sample points in the sample set D of the original d-dimensional space, and two data points x _h and x _j are arbitrarily selected in the sample set D={x ₁ ,x ₂ ,...,x _m }, where , h, j are the numbers of the two data points selected in the sample set D, h, j∈1,2,3...m, the distance between the first data point x _h and the neighbor parameter k is set as the Euclidean distance , when another data point x _j is one of the k points of the nearest neighbor parameters of x _h , it means that they are adjacent, and the neighborhood graph G has edges, that is, the neighborhood graph G is constructed;

Step 2.2. Calculate the shortest path distance matrix: If there are edge connections between two points x _h and x _j on the neighborhood graph G, define the shortest path between the sample points x _h and x _j as d _G (x _h , x _j ) =d(x _h , x _j ); otherwise d _G (x _h , x _j )=∞, where d(x _h , x _j ) is the Euclidean distance between data points x _h and x _j ;

For l∈1,2,3...m, we have:

d _G (x _h ,x _j )=min{d _G (x _h ,x _j ),d _G (x _h ,x _l )+d _G (x _l ,x _j )} (3)

Among them, x _l is a new sample point in the sample set D, d _G (x _h ,x _l ) represents the shortest path between the sample points x _h and x _l , d _G (x _l ,x _j ) represents the sample point x The shortest path between _l and x _j , d _G (x _h ,x _j ) represents the shortest path between sample points x _h and x _j ;

其中，m是d维原始空间的样本集D中样本点的个数，h，j，l是样本集D中选取的三个数据点的编号，h，j，l∈1,2,3...m；Based on this, the shortest path distance matrix can be determined as

Among them, m is the number of sample points in the sample set D of the d-dimensional original space, h, j, l are the numbers of the three data points selected in the sample set D, h, j, l ∈ 1, 2, 3. ..m;

Step 2.3. Construct a new d'-dimensional sample space by reducing the dimension of the d-dimensional original space, and keep the distance between the d'-dimensional new sample space and the d-dimensional original space unchanged, and obtain the inner product matrix B of the samples after dimensionality reduction: Let B=Z ^T Z∈R ^m×m , where d is the dimension of the original space, d' is the dimension of the new sample space, Z is the matrix representation of the d-dimensional original space embedded in the d'-dimensional new sample space, and R is set of real numbers;

After constructing the new sample space, d'<d, and the Euclidean distance of any two samples in the new sample space of dimension d' is equal to the distance in the original space of dimension d, namely:

||z _h -z _j ||=dist _hj (4)

Among them, z _h , z _j represent the Euclidean distance of the sample points x _h , x _j in the d'-dimensional new sample space, respectively, and dist _hj represents the distance between the sample points x _h and x _j in the d-dimensional original space;

Expand Equation (4) and square both sides of the equal sign to get:

b_hh与b_jj以此类推，则有：

b _hh and b _jj and so on, there are:

Among them, b _hj is the element of the hth row and j column in the inner product matrix B, b _hh and b _jj and so on, dist _hj represents the distance between the sample points x _h and x _j in the d-dimensional original space, dist _h , dist _·j and dist _· and so on;

The inner product matrix B can be obtained by obtaining the elements in the inner product matrix B such as b _hj , b _hh , b _jj ;

Step 2.4, construct a d-dimensional embedding for the shortest path distance matrix D _G , and finally obtain the matrix representation Z∈R ^d′×m in which the samples are embedded in the d'-dimensional new sample space through the d-dimensional original space;

The inner product matrix B obtained in step 2.3 obtains the eigenvalue matrix and the eigenvector matrix through eigendecomposition, namely:

B=VΛV ^T (8)

Among them, Λ=diag(λ ₁ ,λ ₂ ,...,λ _d ) is a diagonal matrix composed of eigenvalues, λ ₁ ≥λ ₂ ≥...≥λ _d , λ ₁ ,λ ₂ ,... , λ _d is the eigenvalue obtained by decomposition, and V is the corresponding eigenvector matrix.

计算Z的结果后可得到矩阵表示为：

此时的结果

便是融合后的特征变量；Then the matrix representation Z of the d-dimensional original space embedded in the d'-dimensional new sample space is calculated as:

After calculating the result of Z, the matrix can be obtained as:

result at this time

is the feature variable after fusion;

为特征值构成的对角矩阵，特征值

V_*为对应的特征向量矩阵；

为d'维新样本空间中i时刻第m'个样本点的融合特征变量，m'是d'维新样本空间中样本点的个数。where, there are d ^* non-zero eigenvalues,

is a diagonal matrix of eigenvalues, eigenvalues

V _* is the corresponding eigenvector matrix;

is the fusion feature variable of the m'th sample point at time i in the d'-dimensional sample space, where m' is the number of sample points in the d'-dimensional sample space.

Step 3. Build a data set construction strategy: For the fusion feature variables obtained in step 2, and the remaining life at the corresponding moment of the system constitute the total remaining life prediction sample set, and divide it into a training set and a test set;

Step 3 is as follows:

和系统剩余寿命Y＝{yⁱ}重构为剩余寿命预测样本集

分别为重构的剩余寿命预测样本集中的特征变量，并按照7:3的比例划分为训练集和验证集，其中，yⁱ为系统对应i时刻的剩余寿命；Step 3.1. Construct a sample set: combine the fused feature variables obtained in step 2

and the remaining life of the system Y={y ⁱ } is reconstructed into the remaining life prediction sample set

are the feature variables in the reconstructed remaining life prediction sample set, and are divided into training set and validation set according to the ratio of 7:3, where y ⁱ is the remaining life of the system corresponding to time i;

Step 3.2, construct sub-training set and sub-test set: divide the training set divided in step 3.1 into sub-training set and sub-test set according to the ratio of 7:3 again;

Step 3.3. Construct the total training set and the total test set: Combine the feature variables in the sub-training set obtained in step 3.2 into the total training set in the form of column stacking. The feature variables in the set are merged into the total test set in the form of column stacking, and finally the total remaining life prediction sample set consisting of the total training set, the total test set and the validation set obtained in step 3.1 is formed.

Step 4: Establish a prediction model and predict the remaining life: use the training set of the total remaining life prediction sample set obtained in step 3 as the input of the deep residual network Resnet50 to train the prediction network, and use the trained network to predict the test samples. Obtain the prediction result of the remaining life of the system, and promptly prompt the staff to perform preventive maintenance on the overall equipment.

Step 4 is as follows:

Step 4.1. Build a 50-layer deep residual network ResNet50: The deep residual network ResNet50 consists of three parts, namely the convolution pooling part, the four residual block parts, and the pooling and flattening part. There are two residual block parts. The basic structure is the identity residual block and the convolution residual block respectively; the overall structure is shown in Figure 4. The difference between the two residual blocks is whether there is a convolution layer at the shortcut connection, which is used to solve the problem of dimension mismatch.

Step 4.2. Feature extraction: The training set of the total remaining life prediction sample set obtained in step 3 is used as the input of the deep residual network Resnet50, and the one-dimensional convolution kernel is used to extract the features of the training set, and the extracted convolution features are processed. max pooling to obtain pooled features;

Step 4.3. Feature extraction of residual block: The pooled feature is used as the input of the residual block. The residual block of the deep residual network Resnet50 consists of the following four parts: a convolution residual block and two identity residual blocks , one convolutional residual block and three identity residual blocks, one convolutional residual block and five identity residual blocks, one convolutional residual block and two identity residual blocks, the above four residual blocks The difference blocks are sequentially connected to form the core part of the residual network. With the deepening of the number of network layers, the residual results are approached to 0 through the unique shortcut connection of the residual blocks to reduce the loss of input;

Step 4.4, establish a prediction model: integrate the output of the residual block through the average pooling layer, and then obtain the prediction model through the flattening layer;

Step 4.5. Obtain the remaining life prediction result: Input the test set into the prediction model to obtain the remaining life prediction result. The specific life prediction process is shown in Figure 5. Each time a prediction is made to obtain the output prediction result, the error between the prediction result and the real data of the remaining life collected by the system is back-propagated to the deep residual network Resnet50, and the parameters in the network are adjusted. When the error is less than 0.05, it means that the network has converged, and the trained network is used to predict the test sample, and finally the prediction result of the remaining life of the system is obtained, so as to realize the accurate prediction of the remaining life of the system.

Example

超过0.8的4个特征变量。将前200个寿命周期数据作为总训练集，将后76个寿命周期数据作为总测试集。基于以上数据，采用本发明的方法进行多变量剩余寿命预测，为了更加清晰地描述实验结果，将仿真结果可视化，结果见图6。通过观察图6可知，所提模型的预测结果与实际数据误差较小，效果较好。This experiment takes the turbine engine system as the research object, and collects the sample data of 21 sensors under four failure modes. Taking the data set of one failure mode as an example (100 engines have a total of 20,631 sample data), randomly selected 17th engine data (276 life cycle data in total). After correlation analysis and ISOMAP feature fusion, the correlation coefficient is retained

4 feature variables over 0.8. Take the first 200 life cycle data as the total training set and the last 76 life cycle data as the total test set. Based on the above data, the method of the present invention is used for multivariate remaining life prediction. In order to describe the experimental results more clearly, the simulation results are visualized, and the results are shown in Figure 6 . By observing Figure 6, it can be seen that the prediction result of the proposed model has a small error with the actual data, and the effect is better.

Claims (5)

1. A ResNet50 network key equipment residual life prediction method based on feature fusion is characterized by comprising the following steps:

step 1, carrying out correlation analysis on data information acquired by each sensor in system equipment, filtering characteristic variables with low correlation degree with service life, and determining high correlation degree characteristic variables influencing the service life of a system;

step 2, carrying out feature fusion on the high-association feature variables obtained in the step 1: fusing the high-association characteristic variables in the step 1 by using an ISOMAP algorithm to obtain fused characteristic variables and effectively integrate sensor information;

step 3, constructing a data set construction strategy: forming a total sample set of residual life prediction by aiming at the fusion characteristic variables obtained in the step 2 and the residual life at the corresponding moment of the system, and dividing the total sample set into a training set and a testing set;

and 4, step 4: establishing a prediction model and predicting the residual life: and (3) training the prediction network by taking the training set of the total residual life prediction sample set obtained in the step (3) as the input of the depth residual error network Resnet50, and predicting the test sample by using the trained network to obtain the residual life prediction result of the system.

2. The method for predicting the residual life of the ResNet50 network key equipment based on feature fusion as claimed in claim 1, wherein the step 1 is as follows:

step 1.1, numbering data information acquired by a sensor in system equipment, wherein n represents a serial number of the sensor, and x represents a serial number of the sensor_nFor the monitoring data information of the nth sensor, the monitoring data X is expressed as

Wherein,

for the monitoring data information of the nth sensor at the time i in the system, i is 1,2ⁱIn which yⁱThe residual life of the system corresponding to the i moment;

step 1.2, calculating correlation coefficient by using correlation analysis

Wherein, Cov (x)_nY) is monitoring data information x_nAnd the covariance between the remaining lifetime Y,

and

respectively monitoring data information x_nAnd standard deviation of residual life Y;

step 1.3, calculating monitoring data information x_nPearson correlation coefficient rho of residual life Y_xnYThe method comprises the following steps:

correlation coefficient ρ_xnYThe range is as follows:

and keeping the high-association characteristic variable with the result exceeding more than 0.8.

3. The method for predicting the residual life of the ResNet50 network key equipment based on feature fusion as claimed in claim 2, wherein the step 2 is as follows:

step 2.1, selecting neighborhoodAnd constructing a neighborhood graph G: setting a neighbor parameter as k, and constructing a sample set D ═ x by using the high-relevancy feature variable retained in the step 1.3₁,x₂,...,x_mM is the number of sample points in a sample set D of the original D-dimensional space, where { x ═ x }₁,x₂,...,x_mArbitrarily choose some two data points x_hAnd x_jWherein h, j is the number of two data points selected in the sample set D, h, j belongs to 1,2,3_hSetting the distance from the adjacent parameter k as Euclidean distance, and setting another data point x_jIs x_hWhen one of the nearest neighbor parameters k points is found, the nearest neighbor parameters are shown to be adjacent, and the neighborhood graph G has edges, namely the neighborhood graph G is constructed;

step 2.2, calculating a shortest path distance matrix: if two points x on neighborhood graph G_h、x_jDefining sample point x when there is an edge join_hAnd x_jThe shortest path between is d_G(x_h,x_j)＝d(x_h,x_j) (ii) a Otherwise d_G(x_h,x_j) Infinity, wherein d (x)_h,x_j) Is a data point x_hAnd x_jThe Euclidean distance of;

for l ∈ 1,2,3.. m, there are:

d_G(x_h,x_j)＝min{d_G(x_h,x_j),d_G(x_h,x_l)+d_G(x_l,x_j)} (3)

wherein x is_lIs a new sample point in the sample set D, D_G(x_h,x_l) Represents a sample point x_hAnd x_lShortest path between, d_G(x_l,x_j) Represents a sample point x_lAnd x_jShortest path between, d_G(x_h,x_j) Represents a sample point x_hAnd x_jThe shortest path therebetween;

based on this, the shortest path distance matrix can be determined as

Wherein m is the number of sample points in a sample set D of the D-dimensional original space, h, j and l are the numbers of three data points selected in the sample set D, and h, j and l belong to 1,2 and 3.. m;

step 2.3, constructing a d 'dimension new sample space by reducing the dimension of the d dimension original space, keeping the distance between the d' dimension new sample space and the d dimension original space unchanged, and solving an inner product matrix B of the reduced samples: let B be Z^TZ∈R^m×mWherein d is the dimension of the original space, d 'is the dimension of the new sample space, Z is the matrix representation of the d-dimension original space embedded with the d' -dimension new sample space, and R is a real number set;

after the new sample space is constructed, d '< d, and the Euclidean distance of any two samples in the d' dimension new sample space is equal to the distance in the d dimension original space, namely:

||z_h-z_j||＝dist_hj (4)

wherein z is_h,z_jRespectively representing sample points x_h,x_jEuclidean distance, dist, in d' Vietnamese sample space_hjRepresents a sample point x_hAnd x_jDistance in d-dimensional original space;

equation (4) is expanded and squared across the equality number:

b_hhand b_jjBy analogy, the following are:

wherein, b_hjIs the element of h row and j column in inner product matrix B, B_hhAnd b_jjBy analogy, dist_hjRepresents a sample point x_hAnd x_jDistance in d-dimension original space, dist_h·、dist_·jAnd dist_··And so on;

by finding b_hj，b_hh，b_jjThe elements in the inner product matrix B are equal to obtain an inner product matrix B;

step 2.4, distance matrix D for shortest path_GConstructing d-dimension embedding, and finally obtaining a matrix representation Z belonging to the sample and embedding the d' dimension new sample space in the d-dimension original space^d′×m；

And (3) obtaining an eigenvalue matrix and an eigenvector matrix through characteristic decomposition by using the inner product matrix B obtained in the step 2.3, namely:

B＝VΛV^T (8)

wherein Λ ═ diag (λ)₁,λ₂,...,λ_d) Diagonal matrices formed for eigenvalues, λ₁≥λ₂≥...≥λ_d，λ₁,λ₂,...,λ_dFor decomposing the obtained eigenvalues, V is the corresponding eigenvector matrix.

The formula for embedding the matrix representation Z of the d-dimensional original space into the d' -dimensional new sample space is:

the result of calculating Z can be expressed as a matrix:

the result at this time

Is the feature variable after fusion;

wherein, there is d^*A non-zero eigenvalue, Λ_*＝diag(λ₁,λ₂,...,λ_d*) For the formation of characteristic valuesA diagonal matrix of formed eigenvalues λ₁≥λ₂≥...≥λ_d*，V_*Is a corresponding feature vector matrix;

and (3) the fusion characteristic variable of the m 'th sample point at the i moment in the d' dimension new sample space, wherein m 'is the number of the sample points in the d' dimension new sample space.

4. The method for predicting the residual life of the ResNet50 network key equipment based on feature fusion as claimed in claim 3, wherein the step 3 is as follows:

step 3.1, constructing a sample set: fusing the characteristic variables obtained in the step 2

And the remaining life Y of the system is { Y ═ YⁱReconstructing into a residual life prediction sample set

yⁱ,

Predicting characteristic variables in the sample set for the reconstructed residual life respectively, and dividing the sample set into a training set and a verification set according to a ratio of 7:3, wherein yⁱThe residual life of the system corresponding to the i moment;

step 3.2, constructing a sub-training set and a sub-testing set: dividing the training set divided in the step 3.1 into a sub-training set and a sub-testing set according to the proportion of 7: 3;

step 3.3, constructing a total training set and a total testing set: and (3) merging the characteristic variables in the sub-training sets obtained in the step (3.2) into a total training set in a column stacking mode, and simultaneously merging the characteristic variables in the sub-testing sets into a total testing set in a column stacking mode corresponding to the merging sequence of the sub-training sets in the step (3.2), so as to finally form a residual life prediction total sample set consisting of the total training set, the total testing set and the verification set obtained in the step (3.1).

5. The method for predicting the residual life of the ResNet50 network key equipment based on feature fusion as claimed in claim 4, wherein the step 4 is as follows:

step 4.1, constructing a 50-layer depth residual error network ResNet 50: the depth residual error network ResNet50 is composed of 3 parts, namely a convolution pooling part, 4 residual error block parts and a pooling flattening part, wherein the residual error block parts have two basic structures, namely an identical residual error block and a convolution residual error block;

step 4.2, feature extraction: taking the training set of the total sample set of the residual life prediction obtained in the step 3 as the input of a depth residual error network Resnet50, performing feature extraction on the training set by adopting a one-dimensional convolution kernel, and performing maximum pooling on the extracted convolution features to obtain pooled features;

and 4.3, extracting the characteristics of the residual block: taking the pooled features as an input of a residual block, the residual block of the deep residual network Resnet50 is composed of the following four parts: the system comprises a convolution residual block, two identity residual blocks, a convolution residual block, three identity residual blocks, a convolution residual block, five identity residual blocks, a convolution residual block and two identity residual blocks, wherein the four residual blocks are connected in sequence to form a residual network core part;

step 4.4, establishing a prediction model: integrating the output of the residual block through an average pooling layer, and obtaining a prediction model through a flattening layer;

and 4.5, obtaining a residual life prediction result: inputting the test set into a prediction model to obtain a residual life prediction result, obtaining an output prediction result by predicting once, reversely transmitting an error between the prediction result and residual life real data acquired by the system to a depth residual error network Resnet50, adjusting and optimizing each parameter in the network, when the error of the network is less than 0.05, indicating the network convergence, predicting a test sample by using a trained network, finally obtaining a system residual life prediction result, and realizing accurate prediction of the residual life of the system.

Images