Small sample mechanical remaining life prediction method based on support vector machine model

Abstract

The present invention relates to a method for predicting the remaining life of a small sample machine based on a support vector machine model, comprising the following steps: selecting a number of measurement variables with obvious trends from the operation signal data of a mechanical product as characteristic quantities describing the mechanical degradation process, and obtaining characteristic values ​​of each characteristic quantity; preprocessing the characteristic values: using a Butterworth filter algorithm to smooth the characteristic quantities of each mechanical degradation process; mapping the characteristic values ​​to the same range by a normalization method; converting the remaining life of the machine at each time point in the existing data into the proportion of the remaining life to the total life, and normalizing it to the range of [0, 1]; dividing a plurality of training sets and validation sets into combinations, each combination corresponding to a support vector machine model; training each support vector machine model; and using each support vector machine model to perform online remaining life prediction. The method makes full use of limited mechanical operation data and can realize accurate prediction of the remaining life of the machine.

Description

Technical Field

[0001] The invention relates to a small sample mechanical remaining life prediction method based on a support vector machine model, and belongs to the field of mechanical product remaining life prediction. Background Art

[0002] In the prior art, the remaining life of a machine is defined as the length of time from the current time point until the machine completely fails. Remaining life prediction is an important part of fault diagnosis and health management (PHM). Accurate remaining life prediction can provide effective suggestions and solutions for the maintenance and replacement of the machine, thereby greatly improving the reliability and stability of the system.

[0003] The data-driven method for predicting the remaining life of machinery is the mainstream of current research. The data-driven method uses machine learning algorithms to establish a mapping relationship between monitoring data and the remaining life of machinery. Common methods include artificial neural networks (ANN), neural-fuzzy systems (NF systems) and deep learning (DL). However, the current machine learning algorithm requires a large number of high-quality sample points to train the model. When the number of samples is small or the overall quality of the samples is poor, the final prediction result has a large error. In engineering practice, it often takes several days or even months for a machine to go from normal operation to complete failure. Therefore, it is often expensive to obtain data collection of mechanical operation signals for the entire life cycle. In addition, environmental noise and mechanical manufacturing errors will cause a lot of interference to the observed signal, reducing the quality of the data. Support vector machine (SVM) is a machine learning algorithm currently suitable for small sample conditions, and has a certain degree of application in the remaining life of machinery.

[0004] Its disadvantage is that due to improper selection of model parameters and large differences in signals of different components, the prediction method based on SVM will produce overfitting and underfitting problems, reducing the prediction accuracy. Summary of the invention

[0005] The purpose of the present invention is to overcome the problems existing in the prior art and provide a small sample mechanical remaining life prediction method based on a support vector machine model, make full use of limited mechanical operation data, establish multiple SVM models, calculate the optimal model parameters through an optimization algorithm, introduce a weight automatic update algorithm to allocate the model weight, thereby achieving accurate prediction of the remaining life of the machinery.

[0006] In order to solve the above technical problems, a small sample mechanical remaining life prediction method based on a support vector machine model of the present invention comprises the following steps:

[0007] Step 1: Analyze the operation signal data of the mechanical product, select several measurement variables with obvious trends as characteristic quantities to describe the mechanical degradation process, and obtain the characteristic values ​​of each characteristic quantity;

[0008] Step 2: Preprocess the selected eigenvalues: Use the Butterworth filter algorithm to smooth the eigenvalues ​​of each mechanical degradation process to reduce random noise; Use the normalization method to map the eigenvalues ​​of the eigenvalues ​​to the same range to facilitate training the support vector machine model;

[0009] Step 3: Convert the remaining life of the machine at each time point in the existing data into the ratio of the remaining life to the total life, and normalize it to the range of [0, 1];

[0010] Step 4: Divide the data into multiple training sets and validation sets according to the size and characteristics of the existing data. Each combination corresponds to a support vector machine model.

[0011] Step 5: training each support vector machine model;

[0012] Step 6: Use each support vector machine model to perform online remaining life prediction.

[0013] As a preferred solution of the present invention, the specific process of using the Butterworth filter algorithm to smooth the characteristic quantity of each mechanical degradation process in step 2 is: filtering the original characteristic quantity, and the formula is as follows:

[0014]

[0015] Where: n represents the filter order; ω c Represents the cutoff frequency; G0 represents the gain at zero frequency, components with frequencies higher than the cutoff frequency will be filtered, while components with frequencies lower than the cutoff frequency will be retained.

[0016] As a further preferred embodiment of the present invention, the formula for the normalization process in step 3 is as follows:

[0017] y=(y max -y min )(xx min ) / (x max -x min )+y min , where: y is the normalized result; x is the original data value; y max and min They are the upper and lower bounds of the range that needs to be normalized; x maxand x min are the upper and lower bounds of the original data value.

[0018] As a further preferred embodiment of the present invention, step 5 specifically comprises the following steps:

[0019] Step 5-1, taking the preprocessed eigenvalue as the input θ of the model, and taking the ratio of the remaining life to the total life as the output R of the model;

[0020] Step 5-2, set the optimization target of multiple indicators so that the optimization direction covers the prediction accuracy and fitting speed of the support vector machine;

[0021] Step 5-3: Use the artificial bee colony optimization algorithm to optimize the parameters of each individual support vector machine to obtain the model parameters with the best regression and prediction effects;

[0022] Step 5-4: Verify the effectiveness of each model using the test set data.

[0023] As a further preferred embodiment of the present invention, the input amount θ in step 5-1 is 1 ,θ 2 , ..., θ N ], where N is the number of features, is the jth feature quantity containing K data; output R = [r1, r2, L, r K ] / T, where r i is time point t i t is the remaining life of the machine, and T is the total life of the machine.

[0024] As a further preferred embodiment of the present invention, the optimization target of setting multiple indicators in step 5-2 has the following target parameters: In the formula, mse is the mean square error of the regression effect, assuming that the actual value in the training set and the regression result are λ and Then the calculation formula of mse is: d represents the error indicator describing the prediction effect of the validation set, and v is the corresponding weight.

[0025] As a further preferred embodiment of the present invention, step 5-3 specifically comprises the following steps:

[0026] Step 5-31: Set the initial population size D, set the maximum number of iterations Max, and randomly generate the initial solution vector. The formula is as follows: ω d =l+rand(0,1)×(ul), where ω d is the generated solution vector, 1≤d≤D, containing H elements, u and 1 are the upper and lower bounds of the solution vector respectively;

[0027] Step 5-32: Search for a new solution vector near the initial solution vector according to the following formula:

[0028] ω′ d，m =ω d，m +φ(ω d，m -ω b，m ), where ω d，m is the solution vector ω d The mth element in ω b，m is the solution vector ω b The mth element in ; 1≤m≤H; φ is a random number in the range [0, 1];

[0029] Step 5-33: Calculate the fitness fit of the new solution vector. The formula is as follows:

[0030] where fit(ω d ) is the objective function to be optimized;

[0031] Step 5-34: Assign each solution vector an acceptance probability based on the size of the fitness, and select the ideal solution based on this probability. The formula is as follows:

[0032]

[0033] Step 5-35: Repeat steps 5-31 to 5-34, and update the optimal solution in each iteration until the maximum number of iterations is reached or the optimal value does not change in several iterations, that is, the model parameters with the best regression and prediction effects are obtained.

[0034] As a further preferred embodiment of the present invention, step 6 specifically comprises the following steps:

[0035] Step 6-1, perform sliding average processing on the currently running mechanical signal in real time to reduce random noise;

[0036] Step 6-2, determine the average extreme value of each characteristic quantity according to the existing data set, and perform normalization processing on the currently running mechanical signal;

[0037] Step 6-3, calculating the Euclidean distance between each feature quantity and the corresponding feature quantity in the training sample to evaluate the similarity between the current machine and the machines in the training set;

[0038] Step 6-4: according to the similarity between the training set machine and the currently observed machine in each support vector machine model, a corresponding weight is assigned to each single model, and the current remaining life ratio is obtained by weighted average to obtain the remaining life prediction result;

[0039] Step 6-5: average the remaining life prediction results over a time period to further reduce the prediction error.

[0040] As a further preferred embodiment of the present invention, the calculation formula of the Euclidean distance in step 6-3 is as follows:

[0041] Among them, θ' and θ are the characteristic value of the current measured machine and the characteristic value of a machine in the training set, respectively.

[0042] As a further preferred embodiment of the present invention, the calculation formula of the remaining life prediction result in step 6-4 is as follows:

[0043] Among them, w i is the weight of the i-th support vector machine, r i is the prediction result of the i-th support vector machine.

[0044] Compared with the prior art, the present invention has achieved the following beneficial effects: The small sample mechanical remaining life prediction method based on support vector machine of the present invention can more effectively utilize limited data samples to establish a support vector machine model, which is suitable for engineering applications where data samples are difficult to obtain. Compared with other machine learning algorithms that require large-scale and high-quality data, it has a wider range of applicability. Multiple indicators are introduced to construct optimization targets, and the optimal model parameters are obtained through the artificial bee colony optimization algorithm to improve the prediction accuracy of each support vector machine model and enhance the versatility of the model. At the same time, a weight calculation method based on Euclidean distance is introduced. According to the similarity between the feature quantity and the training sample, the corresponding weight of each support vector machine is obtained, which reduces the occurrence of overfitting and underfitting phenomena, further improves the prediction efficiency and stability, and has important engineering value. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The accompanying drawings are only provided for reference and explanation and are not intended to limit the present invention.

[0046] Figure 1 The present invention is a flowchart of an embodiment of the present invention.

[0047] Figure 2 It is a flow chart of the artificial bee colony algorithm in the present invention.

[0048] Figure 3 The online prediction effect of a single support vector machine on the remaining life of machinery.

[0049] Figure 4 The online prediction effect of weighted support vector machine on the remaining life of machinery. DETAILED DESCRIPTION

[0050] like Figure 1 As shown, the small sample mechanical remaining life prediction method based on the support vector machine model of the present invention includes the following steps:

[0051] Step 1: Analyze the operation signal data of the mechanical product, select several measurement variables with obvious trends as characteristic quantities to describe the mechanical degradation process, and obtain the characteristic values ​​of each characteristic quantity;

[0052] Step 2: Preprocess the selected eigenvalues: Use the Butterworth filter algorithm to smooth the eigenvalues ​​of each mechanical degradation process to reduce random noise; Use the normalization method to map the eigenvalues ​​of the eigenvalues ​​to the same range to facilitate training the support vector machine model;

[0053] Step 3: Convert the remaining life of the machine at each time point in the existing data into the ratio of the remaining life to the total life, and normalize it to the range of [0, 1];

[0054] Step 4: Divide the data into multiple training sets and validation sets according to the size and characteristics of the existing data. Each combination corresponds to a support vector machine model.

[0055] Step 5: training each support vector machine model;

[0056] Step 6: Use each support vector machine model to perform online remaining life prediction.

[0057] The specific process of using the Butterworth filter algorithm to smooth the characteristic quantity of each mechanical degradation process described in step 2 is: filtering the original characteristic quantity, and the formula is as follows:

[0058]

[0059] Where: n represents the filter order; ω c Represents the cutoff frequency; G0 represents the gain at zero frequency, components with frequencies higher than the cutoff frequency will be filtered, while components with frequencies lower than the cutoff frequency will be retained.

[0060] The formula for normalization in step 3 is as follows:

[0061] y=(y max -y min )(xx min ) / (x max -x min )+y min (2)

[0062] Where: y is the normalized result; x is the original data value; y max and min They are the upper and lower bounds of the range that needs to be normalized; x max and x min are the upper and lower bounds of the original data value.

[0063] Step 5 specifically includes the following steps:

[0064] Step 5-1: Use the preprocessed eigenvalues ​​as the model input θ, θ = [θ 1 ,θ 2 , ..., θ N ], where N is the number of features, is the j-th feature quantity containing K data;

[0065] The ratio of remaining life to total life is also used as the output of the model, R = [r1, r2, L, r K ] / T, where r i is time point t i t is the remaining life of the machine, and T is the total life of the machine.

[0066] Step 5-2: Set the optimization target of multiple indicators so that the optimization direction covers the prediction accuracy and fitting speed of the support vector machine; the target parameters are as follows:

[0067]

[0068] In the formula, mse is the mean square error of the regression effect, assuming that the actual value in the training set and the regression result are λ and Then the calculation formula of mse is:

[0069]

[0070] In formula (3), d represents the error index describing the prediction effect of the validation set, and v is the corresponding weight, which represents the influence of each index on the optimization result. In this method, three error indexes d are selected to evaluate the prediction effect of the validation set. Assume that the actual value and the prediction result in the validation set are γ and The calculation formulas for the three indicators are as follows:

[0071] d1: Root mean square error (RMSE):

[0072]

[0073] d2: Mean relative error (MRE):

[0074]

[0075] d3: Convergence:

[0076]

[0077] in,

[0078]

[0079]

[0080] In formulas (7) to (9), t FPT is the initial time point, t i represents the i-th time point; M(t i ) is t i The relative error between the true value and the predicted value at a given time point.

[0081] Step 5-3: Use the artificial bee colony optimization algorithm to optimize the parameters of each individual support vector machine to obtain the model parameters with the best regression and prediction effects, which specifically includes the following steps:

[0082] Step 5-31: Figure 2 As shown, set the initial population size D, set the maximum number of iterations Max, and randomly generate the initial solution vector. The formula is as follows:

[0083] ω d =l+rand(0,1)×(ul) (10)

[0084] Among them, ω d is the generated solution vector, 1≤d≤D, containing H elements, u and l are the upper and lower bounds of the solution vector respectively.

[0085] Step 5-32: Search for a new solution vector near the initial solution vector according to formula (11):

[0086] ω' d,m =ω d,m +φ(ω d,m -ω b,m ) (11)

[0087] Among them, ω d,m is the solution vector ω d The mth (1≤m≤H) element in b,m is the solution vector ω b The mth (1≤m≤H) element in ; φ is a random number in the range [0, 1];

[0088] Step 5-33: Calculate the fitness fit of the new solution vector. The formula is as follows:

[0089]

[0090] Where f(ω d ) is the objective function of optimization, that is, formula (3);

[0091] Step 5-34: Assign each solution vector an acceptance probability based on the size of the fitness, and select the ideal solution based on this probability. The formula is as follows:

[0092]

[0093] Step 5-35: Repeat steps 5-31 to 5-34, and update the optimal solution in each iteration until the maximum number of iterations is reached or the optimal value does not change in several iterations, that is, the model parameters with the best regression and prediction effects are obtained.

[0094] Step 5-4: Verify the effectiveness of each model using the test set data.

[0095] Step 6 specifically includes the following steps:

[0096] Step 6-1, perform sliding average processing on the currently running mechanical signal in real time to reduce random noise;

[0097] Step 6-2, determine the average extreme value of each characteristic quantity according to the existing data set, and perform normalization processing on the currently running mechanical signal;

[0098] Step 6-3: Calculate the Euclidean distance between each feature value and the corresponding feature value in the training sample to evaluate the similarity between the current machine and the machines in the training set; the calculation formula of the Euclidean distance is as follows:

[0099]

[0100] Among them, θ' and θ are the characteristic value of the current measured machine and the characteristic value of a machine in the training set, respectively. The characteristic value can reflect the similarity between the changes.

[0101] Step 6-4: According to the similarity between the training set machinery and the currently observed machinery in each support vector machine model, a corresponding weight is assigned to each single model, and the current remaining life ratio is obtained by weighted average to obtain the remaining life prediction result. The calculation formula is as follows:

[0102]

[0103] Among them, w i is the weight of the i-th support vector machine, r i is the prediction result of the i-th support vector machine.

[0104] Step 6-5: average the remaining life prediction results over a time period to further reduce the prediction error.

[0105] Next, the remaining life prediction for the CMAPSS turbine engine data set includes the following steps:

[0106] Step 1: Select appropriate eigenvalues:

[0107] CMAPSS engine data contains 14 indicators to describe the health status of the engine, including: fan inlet temperature, compressor outlet temperature, turbine outlet temperature, fan inlet pressure, bypass pipe pressure, relative blade speed, core speed, engine pressure ratio, etc. Among them, the change trend of pressure ratio, compressor outlet static pressure and relative blade speed is more obvious, which can better reflect the degradation process of the engine, so this method selects the above three variables as characteristic values.

[0108] Steps 2 and 3: According to the frequency characteristics of different eigenvalues, set the appropriate filter order and cutoff frequency, filter the eigenvalues ​​according to formula (1), and normalize the eigenvalues ​​according to formula (2).

[0109] Step 4: Divide the training set and validation set: To simulate the small sample condition, only 5 groups of samples from the 100 samples in the training set are selected as known data, and the other samples are used as unknown data. 3 groups are selected from the 5 groups of samples as training sets, and the remaining 2 groups are used as validation sets, so a total of 10 support vector machine models can be trained.

[0110] Step 5: Train each support vector machine model:

[0111] Step 5-1, determine the input and output: Based on the preprocessed eigenvalues, construct a 3×K input matrix, where K is the total number of time points for the three groups of samples. Since the engine is in a stable operation state for an initial period of time in the CMAPSS data set, when constructing the output, the values ​​of the remaining life greater than 125 are unified to 125, and then the remaining life ratio is calculated accordingly.

[0112] Step 5-2: Set the multi-index optimization target according to the desired optimization effect: In engineering practice, overestimation of the remaining life will cause more serious consequences than underestimation, so it is necessary to avoid overestimation as much as possible. When constructing the objective function, a higher weight can be assigned to the index of fit so that the prediction results converge as quickly as possible.

[0113] Step 5-3, use the artificial bee colony algorithm to optimize the model parameters. For the CMAPSS engine model, the radial basis kernel function is selected when training the support vector machine, so the parameters that need to be determined in the model include the penalty coefficient C and the radial basis kernel function width g. Set the population size to 20, the maximum number of iterations to 50, and stop the optimization when the optimal value does not change within 10 iterations. Optimize the parameters according to formula (10-13), and finally get the optimal parameters of 10 support vector machine models. The results are shown in Table 1;

[0114] Table 1. Support vector machine model optimization parameters

[0115]

[0116]

[0117] Step 5-4: Verify the model effectiveness based on the test set: The regression effect of each model and the test effect on the validation set are shown in Table 2.

[0118] Table 2. Support vector machine model effects

[0119]

[0120] Step 6: Use each support vector machine model to predict the remaining life:

[0121] Step 6-1, 2, perform sliding average on the measured eigenvalues ​​to reduce their random noise. According to the data in the training set, the average maximum values ​​of the three eigenvectors are 2388.2587, 8.5309, and 554.3199, respectively. The three eigenvalues ​​are normalized according to the three average maximum values ​​to construct the input.

[0122] In step 6-3, the Euclidean distance between the feature value and the feature value in the training set is calculated to evaluate the similarity. Tables 3 and 4 show the Euclidean distance between the three feature quantities and the feature values ​​corresponding to each engine individual in the training set at different time points.

[0123] Table 3. Euclidean distance of eigenvalues ​​after 20 cycles

[0124]

[0125] Table 4. Euclidean distance of eigenvalues ​​under 150-cycle periodicity loop

[0126]

[0127] Step 6-4: Determine the weight of each support vector machine model. Taking the time point 20 cycles: 150 cycles as an example, the reciprocal of the Euclidean distance of each eigenvalue can reflect the similarity. Therefore, the similarity between the current test engine and engines 1-5 can be written as:

[0128] 20 cycles:

[0129]

[0130]

[0131]

[0132]

[0133]

[0134] 150 cycles:

[0135]

[0136]

[0137]

[0138]

[0139]

[0140] Each support vector machine model contains data from three engines as training samples, so the applicability of each support vector machine model to the current test engine can be determined by adding the similarities s, and the weight of each support vector machine can be determined by the proportion of the applicability. Therefore, under 20 cycles, the applicability and weight of the model are shown in Table 5:

[0141] Table 5. Applicability and weight of each model after running 20 cycles

[0142]

[0143]

[0144] Under 150 cycles, the applicability and weight of the model are shown in Table 6:

[0145] Table 6. Applicability and weight of each model after running 150 cycles

[0146]

[0147] The online prediction effect of a single support vector machine on the remaining life of an engine is as follows: Figure 3 As shown in the figure, the online prediction effect of the weighted support vector machine on the remaining life of the engine is as follows Figure 4 As shown. Among them:

[0148] The remaining life ratios predicted by the 10 support vector machine models when running for 20 cycles are: 0.9870, 0.9431, 0.9852, 0.9193, 0.9998, 1.0009, 0.9263, 0.9535, 0.9060, 1.0533. According to formula (13), the weighted remaining life ratio result is: 0.9696.

[0149] The remaining life ratios predicted by the 10 support vector machines when running for 150 cycles are: 0.7986, 0.6720, 0.8327, 0.5157, 0.6356, 0.5680, 0.6677, 0.8355, 0.6606, 0.5564. According to formula (13), the weighted remaining life ratio result is: 0.6794.

[0150] Step 6-5: In order to further reduce the prediction error, the prediction results are averaged. First, the remaining life ratio is converted into the actual life value, and the remaining life of the five most recently predicted time points is averaged to obtain the final prediction value at the current time point. The remaining life under 20 cycles and 150 cycles can be calculated as follows:

[0151]

[0152]

[0153] The actual remaining life at the two time points is 125 and 102, which shows that the predicted results are very close to the actual values.

[0154] Table 7 shows the error assessment of the remaining life prediction values ​​at all time points.

[0155] Table 7. Remaining life prediction error

[0156]

[0157] The above description is only a preferred embodiment of the present invention, and does not limit the scope of patent protection of the present invention. In addition to the above embodiments, the present invention may also have other implementations. Any technical solution formed by equivalent replacement or equivalent transformation falls within the protection scope required by the present invention. The technical features not described in the present invention can be realized by or using existing technologies, and will not be repeated here.

Claims

1. A small sample mechanical remaining life prediction method based on a support vector machine model, characterized in that: The steps include: Step 1: Analyze the operation signal data of the mechanical product, select several measurement variables with obvious trends as characteristic quantities to describe the mechanical degradation process, and obtain the characteristic values ​​of each characteristic quantity; Step 2: Preprocess the selected eigenvalues: Use the Butterworth filter algorithm to smooth the eigenvalues ​​of each mechanical degradation process to reduce random noise; Use the normalization method to map the eigenvalues ​​of the eigenvalues ​​to the same range to facilitate training the support vector machine model; Step 3: Convert the remaining life of the machine at each time point in the existing data into the ratio of the remaining life to the total life, and normalize it to the range of [0, 1]; Step 4: According to the scale and characteristics of the existing data, multiple combinations of training sets and validation sets are divided, and each combination corresponds to a support vector machine model; in order to simulate the small sample condition, only 5 groups of samples in the 100 samples in the training set are selected as known data, and the other samples are used as unknown data; 3 groups are selected from the 5 groups of samples as training sets, and the remaining 2 groups are used as validation sets, so a total of 10 support vector machine models can be trained; Step 5: training each support vector machine model; Step 6: Use each support vector machine model to perform online remaining life prediction; Step 5 specifically includes the following steps: Step 5-1, taking the preprocessed eigenvalue as the input θ of the model, and taking the ratio of the remaining life to the total life as the output R of the model; Step 5-2, set the optimization target of multiple indicators so that the optimization direction covers the prediction accuracy and fitting speed of the support vector machine; Step 5-3: Use the artificial bee colony optimization algorithm to optimize the parameters of each individual support vector machine to obtain the model parameters with the best regression and prediction effects; Step 5-4: Verify the effectiveness of each model using the test set data; Step 6 specifically includes the following steps: Step 6-1, perform sliding average processing on the currently running mechanical signal in real time to reduce random noise; Step 6-2, determine the average extreme value of each characteristic quantity according to the existing data set, and perform normalization processing on the currently running mechanical signal; Step 6-3, calculate the Euclidean distance between each feature value and the corresponding feature value in the training sample to evaluate the similarity between the current machine and the machines in the training set; the calculation formula of the Euclidean distance is as follows: Among them, θ j ' and θ j They are the characteristic value of the current measured machine and the characteristic value of a machine in the training set respectively; Step 6-4: according to the similarity between the training set machine and the currently observed machine in each support vector machine model, a corresponding weight is assigned to each single model, and the current remaining life ratio is obtained by weighted average to obtain the remaining life prediction result; Step 6-5: average the remaining life prediction results over a time period to further reduce the prediction error.

2. The method for predicting the remaining service life of a small sample machine based on a support vector machine model according to claim 1 is characterized in that: The specific process of using the Butterworth filter algorithm to smooth the characteristic quantity of each mechanical degradation process described in step 2 is: filtering the original characteristic quantity, and the formula is as follows: Where: n represents the filter order; ω c Represents the cutoff frequency; G0 represents the gain at zero frequency, components with frequencies higher than the cutoff frequency will be filtered, while components with frequencies lower than the cutoff frequency will be retained.

3. The method for predicting the remaining service life of a small sample machine based on a support vector machine model according to claim 1 is characterized in that: The formula for normalization in step 2 is as follows: and=(and max -and min )(xx min ) / (x max -x min )+y min , Where: y is the normalized result; x is the eigenvalue of the feature quantity; y max and min They are the upper and lower bounds of the range that needs to be normalized; x max and x min are the upper and lower bounds of the eigenvalues ​​of the feature quantity.

4. The method for predicting the remaining service life of a small sample machine based on a support vector machine model according to claim 1, characterized in that: The input quantity θ in step 5-1 is [θ 1 ,θ 2 , ..., θ N ], where N is the number of features, θ j is the jth feature quantity, 1≤j≤N; output quantity R=[r1,r2,...,r K ] / T, where r i is time point t i The remaining life of the machine is 1≤i≤K, and T is the total life of the machine.

5. The method for predicting the remaining service life of a small sample machine based on a support vector machine model according to claim 1, characterized in that: In step 5-2, set the optimization target of multiple indicators. The target parameters are as follows: In the formula, mse is the mean square error of the regression effect, and the actual value and regression result of the p-th predicted data point in the training set are set to be λ p and Then the calculation formula of mse is: d i represents the i-th error indicator describing the prediction effect of the validation set, v i is the corresponding weight of the i-th error indicator.

6. The method for predicting the remaining service life of a small sample machine based on a support vector machine model according to claim 1, characterized in that: Step 5-3 specifically includes the following steps: Step 5-31: Set the initial population size D, set the maximum number of iterations Max, and randomly generate the initial solution vector. The formula is as follows: ω d =l+rand(0,1)×(ul), where ω d is the generated solution vector, 1≤d≤D, containing H elements, u and l are the upper and lower bounds of the solution vector respectively; Step 5-32: Search for a new solution vector near the initial solution vector according to the following formula: ω' d,m =ω d,m +φ(ω d,m -ω b,m ), where ω d,m is the solution vector ω d The mth element in ω b,m is the solution vector ω b The mth element in , 1≤m≤H; φ is a random number in the range [0, 1]; Step 5-33: Calculate the fitness fit of the new solution vector. The formula is as follows: Where f(ω d ) is the objective function to be optimized; Step 5-34: Assign each solution vector an acceptance probability based on the size of the fitness, and select the ideal solution based on this probability. The formula is as follows: Step 5-35: Repeat steps 5-31 to 5-34, and update the optimal solution in each iteration until the maximum number of iterations is reached or the optimal value does not change in several iterations, that is, the model parameters with the best regression and prediction effects are obtained.

7. The method for predicting the remaining service life of a small sample machine based on a support vector machine model according to claim 1 is characterized in that: The calculation formula for the remaining life prediction result described in step 6-4 is as follows: Among them, w i is the weight of the i-th support vector machine, r i is the prediction result of the ith support vector machine.

Images