Improved SVR algorithm for electronic equipment fault prediction

Abstract

The present invention relates to an electronic equipment fault prediction method of an improved SVR algorithm, which extracts sample data that can characterize the entire process of electronic equipment state degradation from an electronic equipment state monitor; performs data preprocessing on sample feature data; establishes a support vector regression (SVR) prediction model; optimizes the parameters of the SVR prediction model through a particle swarm algorithm; designs a recursive multi-step prediction method based on the SVR model improved by the particle swarm algorithm, and draws an equipment state degradation curve based on the SVR model; designs a fault warning based on a sliding time window according to the equipment state degradation curve. The fault prediction method of the present invention objectively and accurately reflects the operating state of the monitored electronic equipment, predicts the future state change trend of the equipment, and promptly provides a fault warning, providing a basis for guiding subsequent equipment fault management and maintenance work.

Description

Technical Field

[0001] The present invention relates to a device fault judgment technology, and in particular to an electronic equipment fault prediction method based on multi-step prediction and particle swarm algorithm to improve SVR algorithm. Background Art

[0002] With the widespread application of new technologies and the continuous enrichment and enhancement of functions and performance, the complexity and precision of modern large-scale equipment systems are becoming increasingly higher. In order to enable the equipment to operate stably for a long time and meet the needs of production development, the equipment is required to have a very low tolerance for failures, which conflicts with the risks that may be brought about by increasingly complex equipment systems. At present, traditional system maintenance methods such as regular inspections and post-management maintenance can no longer adapt to the development trend of modern equipment. The demand for efficient, accurate, and low-cost equipment monitoring and maintenance systems, scientific maintenance management technologies, and situation-based maintenance methods is becoming increasingly strong.

[0003] With the development of modern sensor technology and information technology, more efficient condition-based maintenance technology is becoming more and more popular. It uses advanced fault diagnosis and prediction technology to detect according to the actual operation of the equipment, and measures and counts the dynamic information of the equipment during operation to predict the trend of equipment health changes, predict equipment failures, and propose reasonable repair measures based on the actual operation of the equipment. This condition-based maintenance method, especially the key link in condition-based maintenance - fault prediction, can avoid the problem of excessive maintenance in preventive maintenance, thereby greatly reducing the number of maintenance times, reducing expenditures, and improving maintenance efficiency.

[0004] The commonly used equipment failure prediction method in the industrial field is the failure prediction based on expert knowledge and experience, such as the patent "A knowledge base-based equipment failure prediction system" (CN107966942A). However, this method relies on the domain expert knowledge base, has strong subjectivity and limitations, and has high requirements for the richness of expert knowledge and the level of expert knowledge. Therefore, the demand for data-driven failure prediction methods is increasing. The main idea is to establish a fault warning model based on the historical data of the equipment, and set the threshold according to the generated warning model, such as the patent "A machine learning-based failure prediction method" (CN108304941A). This method establishes an accurate model that conforms to the equipment mechanism and has strong versatility and adaptability. Common methods include regression models, time series models, gray models, neural network models, etc. However, considering the complex working environment of electronic equipment and the limited historical data of electronic equipment, the commonly used methods cannot meet this demand, and the support vector regression (SVR) model has the advantages of wide application range, good generalization ability, strong robustness, simple operation, etc., and is more suitable for small sample learning. Therefore, the SVR prediction model can well meet the problems of high noise interference and small amount of historical data in electronic equipment fault prediction. However, the SVR prediction model still has some shortcomings, such as the difficulty in determining the smoothing parameter C and kernel parameter σ2 in the algorithm model, the lack of explainability of the SVR prediction model prediction process and its poor prediction stability. Summary of the invention

[0005] Aiming at the problems existing in the application of support vector regression algorithm to equipment fault prediction, an electronic equipment fault prediction method based on improved SVR algorithm is proposed. It can objectively and accurately reflect the operating status of the monitored electronic equipment, predict the future status change trend of the equipment, and provide fault warning in time, thus providing a basis for guiding the subsequent equipment fault management and maintenance work.

[0006] The technical solution of the present invention is: an electronic equipment fault prediction method using an improved SVR algorithm, specifically comprising the following steps:

[0007] 1) Data preparation and selection: Select several sets of relatively complete full life cycle monitoring data from the electronic equipment status data set, and select characteristic parameters that can characterize different fault types of electronic equipment and can be continuously monitored and recorded as equipment degradation characteristic variables;

[0008] 2) Preprocessing the complete life cycle monitoring data selected in step 1): the data is measured state data and time series two-dimensional data, and the two-dimensional data is divided into a training set and a test set with reference to the degradation characteristic variables, and every four adjacent data in the training set and the test set are divided into a group to prepare for the subsequent multi-step prediction;

[0009] 3) Establishing an SVR prediction model: sending the training set in step 2) to the SVR prediction model for model training, and using the test set to verify the trained model to obtain the prediction effect;

[0010] 4) The particle swarm algorithm is used to optimize the parameters of the SVR prediction model established in step 3) to obtain the optimal smoothing parameter C and kernel function coefficient σ 2 ;

[0011] 5) Based on the improved SVR model of particle swarm algorithm, a recursive multi-step prediction method was designed, and the equipment state degradation curve was drawn based on it:

[0012] The idea of ​​recursive multi-step prediction is shown in the following formula:

[0013] prediction(t)=model(obs(t-1),obs(t-2),...,obs(tm))

[0014] prediction(t+1)=model(prediction(t),obs(t-1),obs(t-2),...,obs(t-m+1))

[0015] Wherein, prediction(t) is the predicted data obtained according to the SVR prediction model at time t, model is the SVR prediction model after substituting the optimal parameters, obs(t-1), obs(t-2), ..., obs(tn) are the actual training data measured by the monitor before time t, prediction(t+1) is the predicted data obtained according to the SVR prediction model after combining recursive multi-step prediction, and every four adjacent data in step 2) are divided into a group, where m is 4, that is, the data of the four moments before time t and the predicted data at time t are used as training data to predict the data at time t+1, and so on to generate predicted data;

[0016] Draw the equipment status degradation curve based on the predicted data;

[0017] 7) According to the data degradation curve, design a fault warning based on a sliding time window.

[0018] Furthermore, in step 2), Gaussian white noise is added to the data set on the basis of dividing the data set to achieve data enhancement, which is specifically expressed by the following expression:

[0019]

[0020] Where x(n) is the actual signal in the data set, s(n) is the signal after adding noise, n is the time when the signal is sampled, and N is the total number of signal samples. The size of the input noise is set by setting the value of the signal-to-noise ratio (SNR).

[0021] Furthermore, the specific implementation steps of step 4) include:

[0022] 4.1) Population initialization: Initialize the SVR model parameters, including the penalty factor C and the kernel function parameter σ 2 , generate the initial population;

[0023] 4.2) Constructing fitness function: The fitness function f(x) is constructed using the fitting error of the SVR prediction model;

[0024] 4.3) Particle Swarm Optimization Algorithm Parameter Setting: Set the population size, maximum genetic generation number and algorithm termination condition in the algorithm;

[0025] 4.4) Calculate individual extreme values ​​and global optimal solutions: Input the SVR prediction model for training and calculate individual extreme values ​​and global optimal solutions according to the fitness function;

[0026] 4.5) Update speed and position: Update the speed and position of each particle according to the calculated individual extreme value and global optimal solution;

[0027] 4.6) Determine whether the termination condition is met: Whether the fitness value meets the requirement can be used as the termination condition of the algorithm, or the set maximum genetic generation number can be used as the termination condition; if the termination condition of the algorithm is not met, return to 4.4) and continue to execute until the termination condition of the algorithm is met;

[0028] 4.7) Get the optimal smoothing parameter C and kernel coefficient σ 2 .

[0029] Furthermore, the implementation method of step 6) is as follows: in the data degradation curve, a time frame of a specified unit length is used to check the health value on the time series, and the value corresponding to the specified time is replaced by the average value calculated in the time frame, and a threshold model is set based on the sliding time window: Where E max is the maximum value of the average health value of the equipment, k is the warning threshold coefficient, E an is the fault warning threshold, and then according to the output value of each time window and E an Compare to determine whether to alarm.

[0030] The beneficial effects of the present invention are as follows: the electronic equipment fault prediction method of the present invention improves the SVR algorithm, 1) adopts a data-driven fault prediction method, gets rid of the disadvantages of the difficulty in obtaining the knowledge and experience of experts in determining weights and the strong subjectivity, and predicts the deterioration of the equipment status by analyzing and mining the potential laws of historical data. The present invention makes full use of the equipment status data, and the constructed fault prediction method can relatively objectively, accurately and scientifically evaluate the health status of industrial equipment; 2) Especially for the problem that the working environment of electronic equipment is complex and the required historical data is relatively small, the method of the present invention is more suitable for small sample learning than the common data-driven fault prediction method-neural network. Compared with the time series model, the method of the present invention has stronger noise resistance and better robustness. Therefore, the fault prediction method of the present invention can better meet the needs of electronic equipment fault prediction; 3) The particle swarm algorithm is used to optimize the parameters of the SVR prediction model, so that the SVR prediction model can obtain parameters that are more suitable for the data set used, thereby improving the prediction effect of the SVR prediction model; 4) A recursive multi-step prediction method is designed based on the SVR model improved by the particle swarm algorithm, so that the SVR model training and prediction are more interpretable, and its prediction results are more stable; 5) Compared with traditional fault warnings, which only have two states, healthy and faulty, the threshold model set by the method of the present invention can perform hierarchical management of equipment status according to different K value selections, and inform the status of the equipment in advance, which is conducive to users taking measures in advance and making adjustments to extend the operating life of the equipment, reduce the failure rate, and save costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a calculation flow chart of the electronic equipment fault prediction method based on multi-step prediction and particle swarm algorithm to improve SVR algorithm of the present invention;

[0032] Figure 2a This is the effect diagram of the SVR prediction model under voltage data;

[0033] Figure 2b This is the effect diagram of the SVR prediction model under temperature data;

[0034] Figure 3 It is a flow chart of parameter optimization of the SVR prediction model by using the particle swarm algorithm in the method of the present invention;

[0035] Figure 4a This is a comparison diagram of the voltage data of the method of the present invention with other prediction methods;

[0036] Figure 4b This is a comparison diagram of the temperature data of the method of the present invention with other prediction methods;

[0037] Figure 5 It is a flow chart of a fault warning based on a sliding time window of an example of the present invention;

[0038] Figure 6 This is a fault warning result diagram based on a sliding time window of an example of the present invention. DETAILED DESCRIPTION

[0039] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0040] The calculation process of the electronic equipment fault prediction method based on multi-step prediction and particle swarm algorithm improved SVR algorithm described in the present invention is as follows: Figure 1 As shown, it mainly includes the following parts:

[0041] S1. First, select several sets of relatively complete full life cycle monitoring data from the electronic equipment status data set, and select characteristic parameters that can characterize different fault types of electronic equipment and can be continuously monitored and recorded as equipment degradation characteristic variables. Different industrial equipment can select their own parameters to be monitored according to their own equipment characteristic types, mainly including current, temperature, voltage, etc.

[0042] S2, perform data preprocessing on the selected data set;

[0043] The present invention chooses to process data on a python platform. The complete equipment status data of the entire life cycle collected by the status monitor is input into a two-dimensional array through the python platform, which is respectively the measured status data and the time series, and the two-dimensional array is divided into a training set and a test set according to the experimental needs with reference to the degradation characteristic variable, and the first 80% of a set of data reflecting the entire life cycle of the equipment is used as the training set, the measured status data training set is defined as Y_train, and the time training set is defined as X_train. The last 20% of the data reflecting the entire life cycle of the equipment is used as the test set, the measured data test set is defined as Y_test, and the time test set is defined as X_test. On this basis, the data of the training set and the test set are divided into a group according to the principle of four adjacent data, for example, the first to fourth data in the two-dimensional array are used as the first group, the second to fifth data are used as the second group, and so on, to complete the data division.

[0044] In order to better test the robustness of the model and more realistically reflect the actual working environment of the equipment, Gaussian white noise is added to the data set on the basis of dividing the data set to achieve data enhancement. The specific operation is expressed by the following expression:

[0045]

[0046] Where x(n) is the actual signal in the data set, s(n) is the signal after adding noise, n is the time of signal sampling, and N is the total number of signal samples. The input noise size is set by setting the value of the signal-to-noise ratio (SNR). The present invention selects a signal-to-noise ratio of 30.

[0047] S3. Set parameters for the SVR prediction model, including the kernel function type, smoothing parameter C and kernel function coefficient σ of the SVR model. 2 ;

[0048] The present invention uses the scikit-learn library in the python platform to establish the SVR prediction model required for the experiment, and the established model is in the form of SVR (kernel, C, gamma). The present invention sets the initial SVR model parameters as svr1 = SVR (kernel = 'rbf', C = 20, gamma = 0.0001), where kernel is the kernel function type, C is the smoothing parameter, and gamma is the kernel function coefficient. Commonly used kernel functions are shown in Table 1.

[0049] Table 1

[0050]

[0051] Among them, the 'rbf' kernel function, i.e. the Gaussian kernel function, has the best model fitting effect, so the rbf kernel function is selected. It can be concluded that

[0052] Therefore, we only need to adjust the size of gamma to change the kernel function coefficient.

[0053] After the SVR prediction model is established, the training sets Y_train and X_train in the selected divided temperature data set and voltage data set (such as shown in Table 2) are input into svr.fit(X_train, Y_train) to train the model. After the SVR prediction model training is completed, the time data of the last 20% of the data set is used as X_test and input into svr.predict(test_X) to generate the prediction data. Finally, the predicted data is compared with the actual data to determine the prediction effect of the SVR model. The prediction effect diagram of the SVR prediction model under two different data sets is shown in the figure below. Figure 2a , 2b As shown. Figure 2a It can be seen intuitively that the SVR model has a very poor prediction effect under the voltage data set. Therefore, the problem of parameter selection in the SVR prediction model needs to be solved urgently.

[0054] Table 2

[0055]

[0056] At the same time, the present invention selects the mean square error (MSE) and the determination coefficient (R 2 ) is used to quantitatively measure the effect of the SVR prediction model.

[0057] The mean square error is the average of the sum of squares of the deviations between each data and the true value, so the average is calculated by subtracting the sum of squares of the estimated value from the true value, that is:

[0058] In the present invention, the estimated value is the device status data fitted by the SVR model, and the fitted data f(x i ) and the real data y i Substituting it into the formula, we can get its mean square error. The smaller the mean square error, the stronger the model's ability to fit the experimental data, but it cannot be 0. When it is 0, it means that the model completely fits the conditions we listed. At this time, the model is in an overfitting state, and the model effect is poor in actual application.

[0059] The coefficient of determination indicates the degree of correlation between the independent variable Y and the independent variable X. 2 The formula is:

[0060] Where SSE is the residual sum of squares SSE = ∑(y i -f(x i )) 2 , SST is the total sum of squares

[0061] The size of the coefficient of determination determines the closeness of the correlation. 2 When it is closer to 1, it means that the reference value of the related equation is higher, the fitting effect is better, the degree of explanation of the independent variable on the dependent variable is higher, the percentage of changes caused by the independent variable in the total change is higher, and the observation points are denser near the generated curve; on the contrary, when it is closer to 0, it means that the reference value is lower, the fitting effect is worse, the degree of explanation of the independent variable on the dependent variable is lower, the percentage of changes caused by the independent variable in the total change is lower, and the observation points are sparser near the generated curve. The MSE and R of the SVR prediction model in the voltage data set and the temperature data set are 2 As shown in Table 3.

[0062] Table 3

[0063]

[0064] It can be seen from the table that the SVR model has completely different prediction effects in different data sets under the same parameters, and it can hardly achieve prediction effects in the voltage data set. Therefore, the reasonable selection of parameters of the SVR prediction model has become an urgent problem to be solved by the SVR model. The present invention selects the particle swarm algorithm to improve the SVR prediction model, optimizes the smoothing parameter C of the SVR, and the kernel function parameter σ 2 .

[0065] S4. Optimize the parameters of the SVR prediction model by using the particle swarm algorithm. The specific steps include:

[0066] 1) Population initialization: Initialize the SVR model parameters, including the penalty factor C, the kernel function parameter σ2, and generate the initial population;

[0067] 2) Constructing the fitness function: The fitness function f(x) is constructed using the fitting error of the SVR prediction model;

[0068] 3) Particle swarm algorithm parameter setting: set the population size, maximum genetic generation and algorithm termination condition in the algorithm;

[0069] 4) Calculate individual extreme values ​​and global optimal solutions: Input the SVR prediction model for training and calculate individual extreme values ​​and global optimal solutions based on the fitness function;

[0070] 5) Update speed and position: Update the speed and position of each particle according to the calculated individual extreme value and global optimal solution;

[0071] 6) Determine whether the termination condition is met: The termination condition of the algorithm can be whether the fitness value reaches the requirement, or the maximum genetic generation number can be set as the termination condition. If the termination condition of the algorithm is not met, return to 4) and continue to execute until the termination condition of the algorithm is met;

[0072] 7) The optimal smoothing parameter C and kernel coefficient σ are obtained 2 .

[0073] The specific flow chart is as follows Figure 3 shown.

[0074] In the python platform, the present invention chooses to use the scikit-opt library to establish the particle swarm algorithm, and sets the parameters of the particle swarm algorithm to pso=PSO(func=phere, dim=2, pop=5, max_iter=1000, lb=[1,0.000001], ub=[100,100]). Among them, the minimum value of the sum of the difference between the data fitted by SVR and the real data is set as the fitness function. In the particle swarm algorithm parameter setting, the present invention sets the population size to 5, the maximum number of iterations to 1000, and the upper and lower bounds of the two parameters to 1-100 and 0.000001-100. After the iteration is completed, the optimal C and gamma values ​​are obtained.

[0075] S5. Based on the improved SVR model of particle swarm algorithm, a recursive multi-step prediction method is designed, and the equipment state degradation curve is drawn based on it. The specific steps include:

[0076] After substituting the optimal parameters into the SVR prediction model, the idea of ​​recursive multi-step prediction is adopted. The data predicted by the SVR model at time t is combined with the data actually measured before time t as training data. The original SVR model is used to obtain the predicted data at time t+1, and so on. The specific prediction idea is shown in the following formula:

[0077] prediction(t)=model(obs(t-1),obs(t-2),...,obs(tm))

[0078] prediction(t+1)=model(prediction(t),obs(t-1),obs(t-2),...,obs(t-m+1))

[0079] Among them, prediction(t) is the predicted data obtained according to the SVR prediction model at time t, model is the SVR prediction model after substituting the optimal parameters, obs(t-1), obs(t-2),..., obs(tn) are the actual training data measured by the monitor before time t, and prediction(t+1) is the predicted data obtained according to the SVR prediction model after combining recursive multi-step prediction. In the present invention, m is set to 4, that is, the data of the 4 moments before time t and the predicted data at time t are used as training data to predict the data at time t+1, and so on to generate predicted data. Finally, according to the predicted data, the equipment status degradation curve is drawn. After many experiments, the present invention selects the 4 most representative experiments, and its degradation curve is as follows Figure 4a , 4b shown.

[0080] pass Figure 4a, 4b It can be seen that the SVR model has better prediction effect than the BP neural network under small sample learning, and the prediction effect of the improved SVR prediction model (Multi-PSO-SVR) is greatly improved compared with the initial SVR prediction model. Compared with the SVR prediction model (PSO-SVR) improved by the particle swarm algorithm without multi-step prediction, its prediction effect and stability are also greatly improved. Table 4 MSE and R under voltage data set 2 Comparison,Table 5 MSE and R under temperature data set 2 The comparison quantitatively shows the prediction effect of the improved SVR prediction model compared with the initial SVR prediction model.

[0081] Table 4

[0082]

[0083] Table 5

[0084]

[0085] It can be seen from Table 4 and Table 5 that the prediction effect of the SVR prediction model improved by multi-step prediction and particle swarm algorithm is greatly improved compared with the single SVR prediction model under different data sets, and the prediction effect of the SVR prediction model improved by multi-step prediction and particle swarm algorithm is more stable. Therefore, the SVR prediction model improved by multi-step prediction and particle swarm algorithm selected by the present invention can indeed show good prediction performance in small sample learning of electronic devices.

[0086] S6. According to the data degradation curve, design a fault warning based on a sliding time window. The specific steps include:

[0087] Use a time frame of specified unit length to check the health value on the time series, and use the average value calculated in the time frame to replace the value corresponding to the specified time. Set the threshold model based on the sliding time window:

[0088] Where E max is the maximum value of the average health value of the equipment, k is the warning threshold coefficient, E an is the fault warning threshold. Then, according to the output value of each time window and E an Compare to determine whether to alarm. The flowchart of fault warning is as follows Figure 5 shown.

[0089] The present invention takes the degradation curve generated by the SVR prediction model based on multi-step prediction and particle swarm algorithm under the temperature data set as an example. The alarm is set when the temperature is greater than 35°C, that is, the threshold is set to 35, the window width is set to 4, and the sliding size is 1 each time, ensuring that the number of windows represents the time size. The average value in the time window is used instead of the corresponding value of the specified time to avoid the occurrence of false alarms due to external interference. The alarm result is as follows Figure 6 As shown. Figure 6 It can be seen that the alarm time window is 137, which is consistent with the time node when the temperature is higher than 35° C. in the actual curve. Therefore, the fault warning design based on the sliding time window set by the present invention is accurate and effective.

[0090] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. An electronic equipment fault prediction method based on an improved SVR algorithm, characterized in that: The specific steps include: 1) Data preparation and selection: Select several sets of complete life cycle monitoring data from the electronic equipment status data set, and select characteristic parameters that can characterize different fault types of electronic equipment and can be continuously monitored and recorded as equipment degradation characteristic variables; 2) Perform data preprocessing on the complete life cycle monitoring data selected in step 1): the data is measured state data and time series two-dimensional data, and the two-dimensional data is divided into training set and test set with reference to degradation characteristic variables, and every four adjacent data in the training set and test set are divided into a group to prepare for subsequent multi-step prediction; 3) Establishing the SVR prediction model: Send the training set in step 2) to the SVR prediction model for model training, and use the test set to verify the trained model to obtain the prediction effect; 4) Use the particle swarm algorithm to optimize the parameters of the SVR prediction model established in step 3) and obtain the optimal smoothing parameters C And the kernel function coefficients σ 2 ; 5) A recursive multi-step prediction method was designed based on the SVR model optimized by the particle swarm algorithm, and the equipment state degradation curve was drawn based on it: The idea of ​​recursive multi-step prediction is shown in the following formula: in, prediction ( t )for t The prediction data obtained from the SVR prediction model at all times, model is the SVR prediction model after substituting the optimal parameters, for t The actual training data measured by the monitor at the moment, prediction ( t +1) is the predicted data obtained by combining the recursive multi-step prediction according to the SVR prediction model. Every four adjacent data in step 2) are divided into a group. Here m is 4, that is, through t The data of the 4 moments before the moment and t The prediction data at the moment is used as training data to predict the t +1 time point data, and generate forecast data by analogy; Draw the equipment status degradation curve based on the predicted data; 6) According to the data degradation curve, design a fault warning based on a sliding time window.

2. The electronic equipment fault prediction method using the improved SVR algorithm according to claim 1 is characterized in that: In step 2), Gaussian white noise is added to the data set on the basis of dividing the data set to achieve data enhancement, which is specifically expressed by the following expression: ， in x ( n ) is the actual signal in the data set, s ( n ) is the signal after adding noise, n is the time when the signal is sampled, N is the total number of signal samples; the input noise level is set by setting the value of the signal-to-noise ratio (SNR).

3. The electronic equipment fault prediction method using the improved SVR algorithm according to claim 1 or 2, characterized in that: The specific implementation steps of step 4) include: 4.1) Population initialization: Initialize SVR model parameters, including penalty factors C , kernel function parameters σ 2 , generate the initial population; 4.2) Constructing fitness function: Fitness function f ( x ) is constructed using the fitting error of the SVR prediction model; 4.3) Particle Swarm Algorithm Parameter Setting: Set the population size, maximum genetic generation number and algorithm termination condition in the algorithm; 4.4) Calculate individual extreme values ​​and global optimal solutions: Input the SVR prediction model for training and calculate individual extreme values ​​and global optimal solutions based on the fitness function; 4.5) Update speed and position: Update the speed and position of each particle according to the calculated individual extreme value and global optimal solution; 4.6) Determine whether the termination condition is met: whether the fitness value meets the requirement is used as the termination condition of the algorithm, or the set maximum genetic generation number is used as the termination condition; if the algorithm termination condition is not met, return to 4.4) and continue to execute until the algorithm termination condition is met; 4.7) Get the optimal smoothing parameter C and the kernel coefficient σ 2 .

4. The electronic equipment fault prediction method using the improved SVR algorithm according to claim 3 is characterized in that: The implementation method of step 6) is as follows: in the data degradation curve, a time frame of a specified unit length is used to check the health value on the time series, and the value corresponding to the specified time is replaced by the average value calculated in the time frame, and a threshold model is set based on the sliding time window: , in E max is the maximum value of the average health value of the equipment, k is the warning threshold coefficient, E an is the fault warning threshold, and then the output value of each time window is compared with E an Compare to determine whether to alarm.

Images