Model training method, electrolytic capacitor remaining life prediction method and system

Abstract

The present invention discloses a model training method, an electrolytic capacitor remaining life prediction method and system, the training method comprising: obtaining a historical training data set, the historical training data set comprising multiple output parameters of an electrolytic capacitor and the remaining use time corresponding to each output parameter, the output parameter comprising output current and / or output frequency; training a neural network model based on the historical training data to obtain an electrolytic capacitor remaining life prediction model. The present invention obtains multiple output parameters of an electrolytic capacitor and a historical training data set of the remaining use time corresponding to each output parameter, and trains a neural network model based on the historical training data to obtain an electrolytic capacitor remaining life prediction model; so as to accurately predict the remaining use time of the electrolytic capacitor using the electrolytic capacitor remaining life prediction model, thereby improving the accuracy of the electrolytic capacitor remaining life prediction model.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a model training method, and a method and system for predicting the remaining life of an electrolytic capacitor. Background Art

[0002] The inverter is an ideal AC transmission device and is widely used in various fields. Among them, the high-voltage inverter (output voltage exceeds 3kV) is widely used due to its high reliability, high efficiency and good starting performance. The main circuit of the high-voltage inverter consists of an input phase-shifting transformer and a power unit. The electrolytic capacitor is an important component of the power unit, and its quality directly determines the performance of the variable frequency speed regulation system. However, the electrolytic capacitor is prone to aging and has a limited lifespan. In order to avoid the situation where the reliability of the high-voltage inverter is reduced due to the failure of the electrolytic capacitor, it is particularly important to predict the remaining life of the electrolytic capacitor in advance.

[0003] When the electrolytic capacitor ages, its ESR (equivalent series resistance) will increase and C will decrease, so the aging degree of the electrolytic capacitor can be characterized by the current values ​​of ESR and C. Traditionally, the electrolytic capacitor is removed from the inverter main circuit and measured using a handheld LCR (third generation mobile communication system) tester or a capacitor ESR meter, which is an offline measurement.

[0004] In addition, the Arrhenius equation is also a commonly used method to predict the life of electrolytic capacitors based on the failure mechanism of electrolytic capacitors. This model assumes that temperature is the only environmental stress that affects the life of electrolytic capacitors and that the failure time conforms to an exponential distribution. This model calculates the power loss P of the electrolytic capacitor. loss To determine the life of the electrolytic capacitor.

[0005] The static data of electrolytic capacitors obtained by offline measurement has certain reference value, but it is of little significance for real-time estimation of remaining life, and this method does not meet the needs of practical applications because the electrolytic capacitors are not allowed to be removed during the use of the inverter.

[0006] In addition, the power loss P based on the Arrhenius formula is loss It is related to the ripple current flowing through the capacitor. The ripple current is not a fixed value, but changes with the frequency, which makes it difficult to calculate the power loss P. lossWhen calculating the ripple current, mathematical tools are needed to perform complex spectrum transformation on the ripple current to form ripple components at each discrete frequency point. Finally, the ESR value of the electrolytic capacitor at each frequency point is calculated comprehensively to obtain the total AC ripple loss of the electrolytic capacitor, which makes this method difficult to calculate. In addition, the calculation results of this method are highly sensitive to temperature. In practical applications, it is very difficult to add a temperature sensor to the electrolytic capacitor and obtain accurate temperature data, which leads to large errors in the life prediction results. Summary of the invention

[0007] The technical problem to be solved by the present invention is to overcome the defects of the prior art in predicting the remaining life of an electrolytic capacitor, such as high calculation difficulty and inaccurate prediction results, and to provide a model training method, a method and system for predicting the remaining life of an electrolytic capacitor.

[0008] The present invention solves the above technical problems through the following technical solutions:

[0009] A first aspect of the present invention provides a model training method, the training method comprising:

[0010] Acquire a historical training data set, the historical training data set including a plurality of output parameters of the electrolytic capacitor and a remaining usage time corresponding to each output parameter, the output parameter including an output current and / or an output frequency;

[0011] Training a neural network model based on the historical training data to obtain the electrolytic capacitor remaining life prediction model;

[0012] The electrolytic capacitor remaining life prediction model takes the output parameters of the electrolytic capacitor as input and takes the predicted remaining use time of the electrolytic capacitor as output.

[0013] Preferably, after the step of obtaining a historical training data set, the model training method further comprises:

[0014] Performing outlier and null value processing on the historical training data set to obtain an outlier and null value processed historical training data set;

[0015] The historical training data set after the abnormal value and null value processing is normalized to obtain a normalized historical training data set.

[0016] Preferably, the model training method further comprises:

[0017] Filtering feature data and label data from the historical training data set to obtain a sample data set;

[0018] Extracting a test set from the sample data set according to a preset ratio;

[0019] Using the test set to test the prediction result of the electrolytic capacitor life prediction model to obtain the predicted remaining use time corresponding to the test set;

[0020] Obtaining the actual remaining usage time corresponding to the test set;

[0021] Calculating the root mean square error and the coefficient of determination of the electrolytic capacitor remaining life prediction model based on the predicted remaining use time corresponding to the test set and the actual remaining use time corresponding to the test set as the loss value of the electrolytic capacitor remaining life prediction model;

[0022] The electrolytic capacitor remaining life prediction model is optimized based on the loss value.

[0023] A second aspect of the present invention provides a model training system, the training system comprising a first acquisition module and a training module;

[0024] The first acquisition module is used to acquire a historical training data set, the historical training data set includes multiple output parameters of the electrolytic capacitor and the remaining usage time corresponding to each output parameter, the output parameter includes output current and / or output frequency;

[0025] The training module is used to train the neural network model based on the historical training data to obtain the electrolytic capacitor remaining life prediction model;

[0026] The electrolytic capacitor remaining life prediction model takes the output parameters of the electrolytic capacitor as input and takes the predicted remaining use time of the electrolytic capacitor as output.

[0027] Preferably, the model training system further comprises a first processing module and a second processing module;

[0028] The first processing module is used to perform outlier and null value processing on the historical training data set to obtain the historical training data set after outlier and null value processing;

[0029] The second processing module is used to normalize the historical training data set after the abnormal value and null value processing to obtain a normalized historical training data set.

[0030] Preferably, the model training system further includes a screening module, an extraction module, a testing module, a second acquisition module, a calculation module and an optimization module;

[0031] The screening module is used to screen feature data and label data from the historical training data set to obtain a sample data set;

[0032] The extraction module is used to extract a test set from the sample data set according to a preset ratio;

[0033] The testing module is used to test the prediction result of the electrolytic capacitor life prediction model using the test set to obtain the predicted remaining use time corresponding to the test set;

[0034] The second acquisition module is used to obtain the actual remaining usage time corresponding to the test set;

[0035] The calculation module is used to calculate the root mean square error and the determination coefficient of the electrolytic capacitor remaining life prediction model based on the predicted remaining use time corresponding to the test set and the actual remaining use time corresponding to the test set as the loss value of the electrolytic capacitor remaining life prediction model;

[0036] The optimization module is used to optimize the electrolytic capacitor remaining life prediction model based on the loss value.

[0037] A third aspect of the present invention provides a method for predicting the remaining life of an electrolytic capacitor, the method comprising:

[0038] Acquiring real-time output parameters of the electrolytic capacitor, wherein the real-time output parameters include real-time output current and / or real-time output frequency;

[0039] The real-time output parameters are input into the electrolytic capacitor remaining life prediction model trained by the model training method described in the first aspect above to output the remaining use time of the electrolytic capacitor.

[0040] A fourth aspect of the present invention provides a system for predicting the remaining life of an electrolytic capacitor, the prediction system comprising an output parameter acquisition module and an input module;

[0041] The output parameter acquisition module is used to acquire the real-time output parameters of the electrolytic capacitor, and the real-time output parameters include real-time output current and / or real-time output frequency;

[0042] The input module is used to input the real-time output parameters into the electrolytic capacitor remaining life prediction model trained by the model training system described in the second aspect above, so as to output the remaining service life of the electrolytic capacitor.

[0043] The fifth aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the processor implements the model training method as described in the first aspect, or executes the electrolytic capacitor remaining life prediction method as described in the third aspect.

[0044] The sixth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the model training method as described in the first aspect is implemented, or the remaining life prediction method of the electrolytic capacitor as described in the third aspect is executed.

[0045] On the basis of being in accordance with the common sense in the art, the above-mentioned preferred conditions can be arbitrarily combined to obtain the preferred embodiments of the present invention.

[0046] The positive and progressive effects of the present invention are:

[0047] The present invention obtains a historical training data set of multiple output parameters of an electrolytic capacitor and the remaining use time corresponding to each output parameter, and trains a neural network model based on the historical training data to obtain an electrolytic capacitor remaining life prediction model; so that the electrolytic capacitor remaining life prediction model can be used to accurately predict the remaining use time of the electrolytic capacitor, thereby improving the accuracy of the electrolytic capacitor remaining life prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is the first flow chart of the model training method of Example 1 of the present invention.

[0049] Figure 2 This is the second flow chart of the model training method of embodiment 1 of the present invention.

[0050] Figure 3 This is the third flow chart of the model training method of embodiment 1 of the present invention.

[0051] Figure 4 This is a first comparison chart of the predicted remaining usage time and the actual remaining usage time of the model training methods of embodiments 1 and 2 of the present invention.

[0052] Figure 5 This is a second comparison chart of the predicted remaining usage time and the actual remaining usage time of the model training methods of embodiments 1 and 2 of the present invention.

[0053] Figure 6 This is a module diagram of the model training system of Example 2 of the present invention.

[0054] Figure 7 This is a schematic diagram of the structure of an electronic device according to Embodiment 3 of the present invention.

[0055] Figure 8 This is a flow chart of a method for predicting the remaining life of an electrolytic capacitor according to Embodiment 5 of the present invention.

[0056] Fig. 9 This is a module schematic diagram of the electrolytic capacitor remaining life prediction system of Example 6 of the present invention. DETAILED DESCRIPTION

[0057] The present invention is further described below by way of examples, but the present invention is not limited to the scope of the examples.

[0058] Example 1

[0059] like Figure 1 As shown, this embodiment provides a model training method, which includes:

[0060] Step 101: Acquire a historical training data set, the historical training data set including multiple output parameters of the electrolytic capacitor and the remaining usage time corresponding to each output parameter, the output parameter including output current and / or output frequency;

[0061] In this embodiment, the historical training data set O can be sorted from far to near according to the collection time t, for example, the output current C at each time point and / or the output frequency f at each time point and the remaining usage time of the electrolytic capacitor at that time point (i.e., the remaining life of the electrolytic capacitor at that time point) L are obtained. It should be noted that the remaining life of the electrolytic capacitor is calculated by the Arrhenius formula, and the accurate value of the remaining life of the electrolytic capacitor can be obtained after adjustment.

[0062] In this embodiment, the remaining use time of the electrolytic capacitor corresponding to the output parameter is obtained by calculation. For example, the remaining use time of the current electrolytic capacitor (ie, the calendar life of the current electrolytic capacitor) CL is calculated. The specific calculation is shown in formula (1):

[0063]

[0064] Where, CL represents the calendar life of the current electrolytic capacitor, CL 0 represents the initial calendar life, Δt n Indicates the time interval between the nth sample point and the n-1th sample point.

[0065] Step 102: training a neural network model based on historical training data to obtain a remaining life prediction model for electrolytic capacitors;

[0066] In this embodiment, the electrolytic capacitor remaining life prediction model takes the output parameters of the electrolytic capacitor as input and takes the predicted remaining use time of the electrolytic capacitor as output.

[0067] It should be noted that the neural network model may be a RNN (Recurrent Neural Network) model.

[0068] In this embodiment, the LSTM (Long Short-Term Memory Network) algorithm in the recurrent neural network RNN ​​is used for model training, and appropriate parameters are selected to obtain a remaining life prediction model for an electrolytic capacitor.

[0069] This embodiment trains a neural network model based on a historical training data set to construct an electrolytic capacitor remaining life prediction model, which solves the problem of difficulty in obtaining temperature data in the remaining life assessment. While avoiding the temperature parameter, the remaining use time of the electrolytic capacitor (i.e., the electrolytic capacitor calendar life) is considered to be added as a characteristic parameter. It should be noted that the calendar life refers to the accumulation of the use time of the electrolytic capacitor from the date of production to the end of its life. This characteristic parameter can effectively guide the correct estimation of the remaining life of the electrolytic capacitor, thereby improving the value of the electrolytic capacitor remaining life prediction model in practical applications.

[0070] In one feasible solution, Figure 2 As shown, the model training method also includes:

[0071] Step 1010: Process the historical training data set for outliers and null values ​​to obtain the historical training data set after outlier and null value processing;

[0072] In this embodiment, deletion or filling and replacement processing is performed according to the proportion of abnormal values ​​and null values. Specifically, when the proportion of abnormal values ​​or null values ​​is below a preset value, deletion processing is performed; when the proportion of abnormal values ​​or null values ​​is above a preset value, filling and replacement processing is performed.

[0073] It should be noted that the preset value may be set to 5%, or may be set to other values ​​according to actual conditions, which is not specifically limited here.

[0074] Step 1011: normalize the historical training data set after outlier and null value processing to obtain a normalized historical training data set.

[0075] In this embodiment, in order to avoid the influence of dimension, the historical training data set O after outlier and null value processing is normalized to obtain the normalized historical training data set O norm , for any column of data o j , and its processing formula is shown in formula (2):

[0076]

[0077] In the formula, O ij represents the i-th value of the j-th column data in O, O j_min Indicates the jth column data o j The minimum value of O j_max Indicates the jth column data o j The maximum value of .

[0078] Step 102 specifically includes: training a neural network model based on the normalized historical training data set to obtain a remaining life prediction model for electrolytic capacitors.

[0079] In one feasible solution, Figure 3 As shown, the model training method also includes:

[0080] Step 103: Filter feature data and label data from the historical training data set to obtain a sample data set;

[0081] In this embodiment, for example, from the normalized historical training data set O norm The filtered feature data is X = {L, f, C, CL}, the filtered label data is Y = {L}, and the composed sample data set is D = {X, Y}.

[0082] It should be noted that, in this embodiment, the sample data set D can also be processed by time sequence dislocation, that is, a new feature data column X is added. t-m , that is, the value of the feature data X at m time points tm before time point t (m = 1, 2, 3...), and its corresponding label data is the remaining usage time (i.e., remaining life) L at time point t and k time points after t. That is, the data structure of the sample data set D after time sequence misalignment processing is shown in Table 1:

[0083] Table 1

[0084]

[0085] This embodiment can achieve batch data processing by performing time-series dislocation processing on the sample data set D and using the feature data of the first m time points to predict the remaining life of the next k time points. Specifically, the dislocation processing enables the recurrent neural network RNN ​​to change from the traditional method of only being able to predict the value of the next time point to being able to batch predict the values ​​of multiple time points when dealing with time series data, thereby improving the prediction efficiency.

[0086] Step 104: extract a test set from the sample data set according to a preset ratio;

[0087] In this embodiment, the test set includes feature data and label data.

[0088] It should be noted that the preset ratio is set according to actual conditions and is not specifically limited here.

[0089] Step 105: Use the test set to test the prediction result of the electrolytic capacitor life prediction model to obtain the predicted remaining use time corresponding to the test set;

[0090] Step 106: Obtain the actual remaining usage time corresponding to the test set;

[0091] In this embodiment, after normalizing the historical training data set, the predicted remaining usage time corresponding to the test set and the actual remaining usage time corresponding to the test set are also denormalized to obtain the predicted remaining usage time corresponding to the denormalized test set and the actual remaining usage time corresponding to the denormalized test set;

[0092] In this embodiment, the predicted remaining usage time Y corresponding to the test set pred The actual remaining usage time Y corresponding to the test set true The data denormalization process is shown in formula (3):

[0093] y' i =y max· y i +(1.0-y i )y min (3)

[0094] In the formula, y' i Represents the i-th data after denormalization, y i represents the i-th input data, y max is the maximum value of the Y sequence before normalization, y min is the minimum value of the Y sequence before normalization.

[0095] Step 107, calculating the root mean square error and determination coefficient of the electrolytic capacitor remaining life prediction model based on the predicted remaining use time corresponding to the test set and the actual remaining use time corresponding to the test set as the loss value of the electrolytic capacitor remaining life prediction model;

[0096] Step 108: Optimize the electrolytic capacitor remaining life prediction model based on the loss value.

[0097] In this embodiment, the root mean square error and the coefficient of determination are used to evaluate the performance of the electrolytic capacitor remaining life prediction model, and the calculation formulas thereof are shown in formula (4) and formula (5):

[0098]

[0099]

[0100] In the formula, RMSE represents the root mean square error, R 2 represents the coefficient of determination, N represents the number of test set samples, y' pred_i represents the denormalized predicted remaining life corresponding to the i-th test sample, y' true_i represents the true remaining life corresponding to the i-th test sample, y' true_mean Represents the mean of the true values.

[0101] It should be noted that the smaller the RMSE, the better the performance of the electrolytic capacitor remaining life prediction model. 2 The value is between 0 and 1. 2 When the value of is greater than or equal to 0.9, it is considered that the performance of the electrolytic capacitor remaining life prediction model is relatively good.

[0102] The following is an explanation with specific examples:

[0103] For example, taking a certain frequency converter as an example, historical operation monitoring data of the frequency converter is collected as a historical training data set, and the historical training data set is sorted from far to near according to the collection time t of the frequency converter operation monitoring data. The historical training data set includes the output frequency f and / or the output current C and the remaining life L of the electrolytic capacitor calculated and adjusted according to the Arrhenius formula. The data collection time interval is 2s, and there are 306853 data in total;

[0104] After analysis, the historical training data set contains abnormal data with output current of NAN (invalid data), and the proportion of the abnormal data is very small (for example, the proportion of the abnormal data is less than 0.1%), so the historical training data set with abnormal data can be directly deleted;

[0105] The current calendar life CL of the inverter is calculated according to formula (1); and the historical training data set O is normalized according to formula (2) to obtain the normalized historical training data set O norm ;

[0106] From the normalized historical training data set O norm The feature data is selected as X = {f, C, CL, L}, and the label data is selected as Y = {L} to form a sample data set D = {X, Y};

[0107] The sample data set D is processed by time series dislocation. For example, the feature data of the first 100 time points are used to predict the label data (i.e., the remaining life) of the last 9 time points. The partial data obtained when m=100 and k=9 are shown in Table 2:

[0108] Table 2

[0109]

[0110] The test set is extracted from the sample data set at a ratio of 10%, and the feature data in the test set is input into the electrolytic capacitor life prediction model to obtain the predicted remaining use time corresponding to the test set (i.e., the label data corresponding to the test set); the actual remaining use time corresponding to the test set is obtained; the predicted remaining use time Y corresponding to the test set is calculated using the above formula (3) pred The actual remaining usage time Y corresponding to the test set truePerform data denormalization and compare the predicted remaining usage time with the actual remaining usage time. Figure 4 As shown, Figure 4 In the above example, predict is the predicted remaining usage time, and true is the actual remaining usage time. Figure 4 For comparison, the current calendar life CL of the inverter in the feature data X is deleted, and the above steps from performing time series misalignment processing on the sample data set D to performing data denormalization processing are repeated to obtain the prediction result without including the current calendar life CL of the inverter. In this case, the comparison between the predicted remaining usage time and the actual remaining usage time is as follows: Figure 5 As shown;

[0111] The performance evaluation index of the electrolytic capacitor remaining life prediction model is calculated according to formulas (4) and (5), as shown in Table 3. The determination coefficient R of the electrolytic capacitor remaining life prediction model established by adding the current calendar life CL of the inverter is 2 When it is greater than or equal to 0.9, the electrolytic capacitor remaining life prediction model has a better effect, while the electrolytic capacitor remaining life prediction model established without adding the inverter's current calendar life CL has a lower R 2 It is -0.0489, and the effect of the remaining life prediction model of electrolytic capacitors is poor.

[0112] Table 3

[0113] Remaining life prediction model for electrolytic capacitors RMSE <![CDATA[R 2 ]]> Consider the current calendar life CL 217.1368 0.9091 The current calendar life CL is not taken into account 737.8993 -0.0489

[0114] This embodiment obtains a historical training data set of output current and / or output frequency and the remaining service life of the electrolytic capacitor, and trains a neural network model based on the historical training data to obtain an electrolytic capacitor remaining life prediction model; so that the electrolytic capacitor remaining life prediction model can be used to accurately predict the remaining service life of the electrolytic capacitor, thereby improving the accuracy and practicality of the electrolytic capacitor remaining life prediction model and optimizing the prediction efficiency of the electrolytic capacitor remaining life prediction model.

[0115] Example 2

[0116] like Figure 6 As shown, this embodiment provides a model training system, which includes a first acquisition module 1 and a training module 2;

[0117] A first acquisition module 1 is used to acquire a historical training data set, the historical training data set including multiple output parameters of the electrolytic capacitor and the remaining usage time corresponding to each output parameter, the output parameter including output current and / or output frequency;

[0118] In this embodiment, the historical training data set O can be sorted from far to near according to the collection time t, for example, the output current C at each time point and / or the output frequency f at each time point and the remaining usage time of the electrolytic capacitor at that time point (i.e., the remaining life of the electrolytic capacitor at that time point) L are obtained. It should be noted that the remaining life of the electrolytic capacitor is calculated by the Arrhenius formula, and the accurate value of the remaining life of the electrolytic capacitor can be obtained after adjustment.

[0119] In this embodiment, the remaining use time of the electrolytic capacitor corresponding to the output parameter is obtained by calculation. For example, the remaining use time of the current electrolytic capacitor (ie, the calendar life of the current electrolytic capacitor) CL is calculated using formula (1) in Embodiment 1.

[0120] Training module 2, used for training a neural network model based on historical training data to obtain a remaining life prediction model for electrolytic capacitors;

[0121] In this embodiment, the electrolytic capacitor remaining life prediction model takes the output parameters of the electrolytic capacitor as input and takes the predicted remaining use time of the electrolytic capacitor as output.

[0122] It should be noted that the neural network model may be a RNN (Recurrent Neural Network) model.

[0123] In this embodiment, the LSTM algorithm in the recurrent neural network RNN ​​is used for model training, and appropriate parameters are selected to obtain a remaining life prediction model for electrolytic capacitors.

[0124] This embodiment trains a neural network model based on a historical training data set to construct an electrolytic capacitor remaining life prediction model, which solves the problem of difficulty in obtaining temperature data in the remaining life assessment. While avoiding the temperature parameter, the remaining use time of the electrolytic capacitor (i.e., the electrolytic capacitor calendar life) is considered to be added as a characteristic parameter. It should be noted that the calendar life refers to the accumulation of the use time of the electrolytic capacitor from the date of production to the end of its life. This characteristic parameter can effectively guide the correct estimation of the remaining life of the electrolytic capacitor, thereby improving the value of the electrolytic capacitor remaining life prediction model in practical applications.

[0125] In one feasible solution, Figure 6 As shown, the model training system also includes a first processing module 3 and a second processing module 4;

[0126] The first processing module 3 is used to process the outliers and null values ​​of the historical training data set to obtain the historical training data set after the outliers and null values ​​are processed;

[0127] In this embodiment, deletion or filling and replacement processing is performed according to the proportion of abnormal values ​​and null values. Specifically, when the proportion of abnormal values ​​or null values ​​is below a preset value, deletion processing is performed; when the proportion of abnormal values ​​or null values ​​is above a preset value, filling and replacement processing is performed.

[0128] It should be noted that the preset value may be set to 5%, or may be set to other values ​​according to actual conditions, which is not specifically limited here.

[0129] The second processing module 4 is used to normalize the historical training data set after the abnormal value and the null value are processed to obtain the normalized historical training data set.

[0130] In this embodiment, in order to avoid the influence of dimension, the historical training data set O after outlier and null value processing is normalized to obtain the normalized historical training data set O norm , use formula (2) in the embodiment to normalize any column of data o j .

[0131] The training module 2 is specifically used to train a neural network model based on the normalized historical training data set to obtain a remaining life prediction model for electrolytic capacitors.

[0132] In one feasible solution, Figure 6 As shown, the model training system also includes a screening module 5, an extraction module 6, a testing module 7, a second acquisition module 8, a calculation module 9 and an optimization module 10;

[0133] A screening module 5 is used to screen feature data and label data from the historical training data set to obtain a sample data set;

[0134] In this embodiment, for example, from the normalized historical training data set O norm The filtered feature data is X = {L, f, C, CL}, the filtered label data is Y = {L}, and the composed sample data set is D = {X, Y}.

[0135] It should be noted that, in this embodiment, the sample data set D can also be processed by time sequence dislocation, that is, a new feature data column X is added. t-m , that is, the value of the feature data X at m time points tm before time point t (m=1, 2, 3...), and its corresponding label data is the remaining usage time (i.e., remaining life) L at the time point t and k time points after t, that is, the data structure after the time sequence misalignment processing of the sample data set D is as shown in Table 1 in Example 1.

[0136] This embodiment can achieve batch data processing by performing time-series dislocation processing on the sample data set D and using the feature data of the first m time points to predict the remaining life of the next k time points. Specifically, the dislocation processing enables the recurrent neural network RNN ​​to change from the traditional method of only being able to predict the value of the next time point to being able to batch predict the values ​​of multiple time points when dealing with time series data, thereby improving the prediction efficiency.

[0137] Extraction module 6, used to extract a test set from the sample data set according to a preset ratio;

[0138] In this embodiment, the test set includes feature data and label data.

[0139] It should be noted that the preset ratio is set according to actual conditions and is not specifically limited here.

[0140] A testing module 7 is used to test the prediction result of the electrolytic capacitor life prediction model using a test set to obtain the predicted remaining use time corresponding to the test set;

[0141] The second acquisition module 8 is used to obtain the actual remaining usage time corresponding to the test set;

[0142] In this embodiment, after normalizing the historical training data set, the predicted remaining usage time corresponding to the test set and the actual remaining usage time corresponding to the test set are also denormalized to obtain the predicted remaining usage time corresponding to the denormalized test set and the actual remaining usage time corresponding to the denormalized test set;

[0143] In this embodiment, the formula (3) in Embodiment 1 is used to calculate the predicted remaining usage time Y corresponding to the test set. pred The actual remaining usage time Y corresponding to the test set true Perform data denormalization.

[0144] A calculation module 9 is used to calculate the root mean square error and the determination coefficient of the electrolytic capacitor remaining life prediction model based on the predicted remaining use time corresponding to the test set and the actual remaining use time corresponding to the test set as the loss value of the electrolytic capacitor remaining life prediction model;

[0145] The optimization module 10 is used to optimize the electrolytic capacitor remaining life prediction model based on the loss value.

[0146] In this embodiment, the root mean square error and the coefficient of determination are used to evaluate the performance of the electrolytic capacitor remaining life prediction model, and the calculation formulas thereof are shown in Formula (4) and Formula (5) in Embodiment 1.

[0147] It should be noted that the smaller the RMSE, the better the performance of the electrolytic capacitor remaining life prediction model. 2The value is between 0 and 1. 2 When the value of is greater than or equal to 0.9, it is considered that the performance of the electrolytic capacitor remaining life prediction model is relatively good.

[0148] The following is an explanation with specific examples:

[0149] For example, taking a certain frequency converter as an example, historical operation monitoring data of the frequency converter is collected as a historical training data set, and the historical training data set is sorted from far to near according to the collection time t of the frequency converter operation monitoring data. The historical training data set includes the output frequency f and / or the output current C and the remaining life L of the electrolytic capacitor calculated and adjusted according to the Arrhenius formula. The data collection time interval is 2s, and a total of 306853 data;

[0150] After analysis, the historical training data set contains abnormal data with an output current of NAN, and the proportion of the abnormal data is very small (for example, the proportion of the abnormal data is less than 0.1%), so the historical training data set with the abnormal data can be directly deleted;

[0151] The current calendar life CL of the inverter is calculated according to formula (1) in Example 1; and the historical training data set O is normalized according to formula (2) in Example 1 to obtain the normalized historical training data set O norm ;

[0152] From the normalized historical training dataset O norm The feature data is selected as X = {f, C, CL, L}, and the label data is selected as Y = {L} to form a sample data set D = {X, Y};

[0153] The sample data set D is subjected to time-series misalignment processing. For example, the feature data of the first 100 time points are used to predict the label data (i.e., the remaining life) of the next 9 time points. That is, m=100, k=9. The partial data obtained is shown in Table 2 in Example 1.

[0154] A test set is obtained by extracting the feature data in the test set from the sample data set at a ratio of 10%, inputting the feature data in the test set into the electrolytic capacitor life prediction model, and obtaining the predicted remaining use time corresponding to the test set (i.e., the label data corresponding to the test set); obtaining the actual remaining use time corresponding to the test set; and using formula (3) in Example 1 to calculate the predicted remaining use time Y corresponding to the test set. pred The actual remaining usage time Y corresponding to the test set true Perform data denormalization and compare the predicted remaining usage time with the actual remaining usage time. Figure 4 As shown, Figure 4 In the above example, predict is the predicted remaining usage time, and true is the actual remaining usage time. Figure 4 For comparison, the current calendar life CL of the inverter in the feature data X is deleted, and the above steps from performing time series misalignment processing on the sample data set D to performing data denormalization processing are repeated to obtain the prediction result without including the current calendar life CL of the inverter. In this case, the comparison between the predicted remaining usage time and the actual remaining usage time is as follows: Figure 5 As shown;

[0155] The performance evaluation index of the electrolytic capacitor remaining life prediction model is calculated according to formulas (4) and (5) in Example 1, as shown in Table 3 in Example 1, and the determination coefficient R of the electrolytic capacitor remaining life prediction model established by adding the current calendar life CL of the inverter is 2 When it is greater than or equal to 0.9, the electrolytic capacitor remaining life prediction model has a better effect, while the electrolytic capacitor remaining life prediction model established without adding the inverter's current calendar life CL has a lower R 2 It is -0.0489, and the effect of the remaining life prediction model of electrolytic capacitors is poor.

[0156] This embodiment obtains a historical training data set of output current and / or output frequency and the remaining service life of the electrolytic capacitor, and trains a neural network model based on the historical training data to obtain an electrolytic capacitor remaining life prediction model; so that the electrolytic capacitor remaining life prediction model can be used to accurately predict the remaining service life of the electrolytic capacitor, thereby improving the accuracy and practicality of the electrolytic capacitor remaining life prediction model and optimizing the prediction efficiency of the electrolytic capacitor remaining life prediction model.

[0157] Example 3

[0158] Figure 7 This is a schematic diagram of the structure of an electronic device provided in Example 3 of the present invention. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the model training method of Example 1 is implemented when the processor executes the program. Figure 7 The electronic device 30 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0159] like Figure 7 As shown, the electronic device 30 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 30 may include, but are not limited to: at least one processor 31, at least one memory 32, and a bus 33 connecting different system components (including the memory 32 and the processor 31).

[0160] The bus 33 includes a data bus, an address bus, and a control bus.

[0161] The memory 32 may include a volatile memory, such as a random access memory (RAM) 321 and / or a cache memory 322 , and may further include a read-only memory (ROM) 323 .

[0162] The memory 32 may also include a program / utility 325 having a set (at least one) of program modules 324, such program modules 324 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0163] The processor 31 executes various functional applications and data processing by running the computer program stored in the memory 32, such as the model training method of Example 1 of the present invention.

[0164] The electronic device 30 may also communicate with one or more external devices 34 (e.g., keyboards, pointing devices, etc.). Such communication may be performed via an input / output (I / O) interface 35. Furthermore, the model generating device 30 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 36. Figure 7 As shown, the network adapter 36 communicates with other modules of the model-generated device 30 via the bus 33. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the model-generated device 30, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.

[0165] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0166] Example 4

[0167] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the model training method provided in Embodiment 1 is implemented.

[0168] The readable storage medium may include but is not limited to: a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device or any suitable combination of the above.

[0169] In a possible implementation manner, the present invention can also be implemented in the form of a program product, which includes a program code. When the program product runs on a terminal device, the program code is used to enable the terminal device to execute the model training method described in Example 1.

[0170] Among them, the program code for executing the present invention can be written in any combination of one or more programming languages, and the program code can be executed completely on the user device, partially on the user device, as an independent software package, partially on the user device and partially on a remote device, or completely on the remote device.

[0171] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that this is only for illustration and the protection scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but these changes and modifications all fall within the protection scope of the present invention.

[0172] Example 5

[0173] like Figure 8 As shown, this embodiment provides a method for predicting the remaining life of an electrolytic capacitor, the prediction method comprising:

[0174] Step 201, obtaining a real-time output parameter of an electrolytic capacitor, the real-time output parameter including a real-time output current and / or a real-time output frequency;

[0175] Step 202: input the real-time output parameters into the electrolytic capacitor remaining life prediction model trained by the model training method of Example 1 to output the remaining use time of the electrolytic capacitor.

[0176] While avoiding temperature parameters, this embodiment only considers the real-time output current and / or real-time output frequency in the acquired real-time output parameters to be input into the trained electrolytic capacitor remaining life prediction model, so as to achieve accurate prediction of the remaining use time of the electrolytic capacitor at batch time points, and effectively reduce the calculation difficulty while reducing the input parameters of the electrolytic capacitor remaining life prediction model, thereby improving the accuracy of predicting the remaining use time.

[0177] Example 6

[0178] like Fig. 9 As shown, this embodiment provides a system for predicting the remaining life of an electrolytic capacitor, and the prediction system includes an output parameter acquisition module 61 and an input module 62;

[0179] An output parameter acquisition module 61 is used to acquire real-time output parameters of the electrolytic capacitor, where the real-time output parameters include real-time output current and / or real-time output frequency;

[0180] The input module 62 is used to input the real-time output parameters into the electrolytic capacitor remaining life prediction model trained by the model training system of Example 2 to output the remaining use time of the electrolytic capacitor.

[0181] While avoiding temperature parameters, this embodiment only considers the real-time output current and / or real-time output frequency in the acquired real-time output parameters to be input into the trained electrolytic capacitor remaining life prediction model, so as to achieve accurate prediction of the remaining use time of the electrolytic capacitor at batch time points, and effectively reduce the calculation difficulty while reducing the input parameters of the electrolytic capacitor remaining life prediction model, thereby improving the accuracy of predicting the remaining use time.

[0182] Example 7

[0183] A schematic diagram of the structure of an electronic device provided in Embodiment 7 of the present invention. The schematic diagram of the structure of the electronic device in this embodiment is similar to Figure 7 The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the method for predicting the remaining life of an electrolytic capacitor of Embodiment 5 is implemented. Figure 7 The electronic device 30 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0184] like Figure 7 As shown, the electronic device 30 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 30 may include, but are not limited to: at least one processor 31, at least one memory 32, and a bus 33 connecting different system components (including the memory 32 and the processor 31).

[0185] The bus 33 includes a data bus, an address bus, and a control bus.

[0186] The memory 32 may include a volatile memory, such as a random access memory (RAM) 321 and / or a cache memory 322 , and may further include a read-only memory (ROM) 323 .

[0187] The memory 32 may also include a program / utility 325 having a set (at least one) of program modules 324, such program modules 324 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0188] The processor 31 executes various functional applications and data processing by running the computer program stored in the memory 32, such as the electrolytic capacitor remaining life prediction method of embodiment 5 of the present invention.

[0189] The electronic device 30 may also communicate with one or more external devices 34 (e.g., keyboards, pointing devices, etc.). Such communication may be performed via an input / output (I / O) interface 35. Furthermore, the model generating device 30 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 36. Figure 7 As shown, the network adapter 36 communicates with other modules of the model-generated device 30 via the bus 33. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the model-generated device 30, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.

[0190] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0191] Example 8

[0192] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method for predicting the remaining life of an electrolytic capacitor provided in Embodiment 5 is implemented.

[0193] The readable storage medium may include but is not limited to: a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device or any suitable combination of the above.

[0194] In a possible implementation manner, the present invention can also be implemented in the form of a program product, which includes a program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the electrolytic capacitor remaining life prediction method described in Example 5.

[0195] Among them, the program code for executing the present invention can be written in any combination of one or more programming languages, and the program code can be executed completely on the user device, partially on the user device, as an independent software package, partially on the user device and partially on a remote device, or completely on the remote device.

[0196] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that this is only for illustration and the protection scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but these changes and modifications all fall within the protection scope of the present invention.

Claims

1. A model training method, It is characterized in that The training method comprises: Acquire a historical training data set, the historical training data set including multiple output parameters of the electrolytic capacitor and the remaining use time corresponding to each output parameter, the output parameter including the output current and / or output frequency after avoiding the temperature parameter; Training a neural network model based on the historical training data to obtain the electrolytic capacitor remaining life prediction model; Filtering feature data and label data from the historical training data set to obtain a sample data set; Performing time sequence misalignment processing on the sample data set; the sample data set is used to train the electrolytic capacitor remaining life prediction model; The electrolytic capacitor remaining life prediction model takes the output parameters of the electrolytic capacitor as input and takes the predicted remaining use time of the electrolytic capacitor as output.

2. The model training method according to claim 1, It is characterized in that After the step of obtaining a historical training data set, the model training method further includes: Performing outlier and null value processing on the historical training data set to obtain an outlier and null value processed historical training data set; The historical training data set after the abnormal value and null value processing is normalized to obtain a normalized historical training data set.

3. The model training method according to claim 1, It is characterized in that The model training method also includes: Extracting a test set from the sample data set according to a preset ratio; Using the test set to test the prediction result of the electrolytic capacitor life prediction model to obtain the predicted remaining use time corresponding to the test set; Obtaining the actual remaining usage time corresponding to the test set; Calculating the root mean square error and the coefficient of determination of the electrolytic capacitor remaining life prediction model based on the predicted remaining use time corresponding to the test set and the actual remaining use time corresponding to the test set as the loss value of the electrolytic capacitor remaining life prediction model; The electrolytic capacitor remaining life prediction model is optimized based on the loss value.

4. A model training system, It is characterized in that The training system includes a first acquisition module, a training module and a screening module; The first acquisition module is used to acquire a historical training data set, the historical training data set includes multiple output parameters of the electrolytic capacitor and the remaining use time corresponding to each output parameter, the output parameter includes the output current and / or output frequency after avoiding the temperature parameter; The training module is used to train the neural network model based on the historical training data to obtain the electrolytic capacitor remaining life prediction model; The screening module is used to screen feature data and label data from the historical training data set to obtain a sample data set; and is also used to perform time series misalignment processing on the sample data set; The sample data set is used to train the electrolytic capacitor remaining life prediction model; The electrolytic capacitor remaining life prediction model takes the output parameters of the electrolytic capacitor as input and takes the predicted remaining use time of the electrolytic capacitor as output.

5. The model training system as claimed in claim 4, It is characterized in that The model training system also includes a first processing module and a second processing module; The first processing module is used to perform outlier and null value processing on the historical training data set to obtain the historical training data set after outlier and null value processing; The second processing module is used to normalize the historical training data set after the abnormal value and null value processing to obtain a normalized historical training data set.

6. The model training system according to claim 4, It is characterized in that The model training system also includes an extraction module, a testing module, a second acquisition module, a calculation module and an optimization module; The extraction module is used to extract a test set from the sample data set according to a preset ratio; The testing module is used to test the prediction result of the electrolytic capacitor life prediction model using the test set to obtain the predicted remaining use time corresponding to the test set; The second acquisition module is used to obtain the actual remaining usage time corresponding to the test set; The calculation module is used to calculate the root mean square error and the determination coefficient of the electrolytic capacitor remaining life prediction model based on the predicted remaining use time corresponding to the test set and the actual remaining use time corresponding to the test set as the loss value of the electrolytic capacitor remaining life prediction model; The optimization module is used to optimize the electrolytic capacitor remaining life prediction model based on the loss value.

7. A method for predicting the remaining life of an electrolytic capacitor. It is characterized in that The prediction method comprises: Acquiring real-time output parameters of the electrolytic capacitor, wherein the real-time output parameters include real-time output current and / or real-time output frequency; The real-time output parameters are input into an electrolytic capacitor remaining life prediction model trained using the model training method described in any one of claims 1 to 3 to output the remaining use time of the electrolytic capacitor.

8. A system for predicting the remaining life of electrolytic capacitors. It is characterized in that The prediction system includes an output parameter acquisition module and an input module; The output parameter acquisition module is used to acquire the real-time output parameters of the electrolytic capacitor, and the real-time output parameters include real-time output current and / or real-time output frequency; The input module is used to input the real-time output parameters into the electrolytic capacitor remaining life prediction model trained by the model training system as described in any one of claims 4-6 to output the remaining service life of the electrolytic capacitor.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, it implements the model training method described in any one of claims 1 to 3, or executes the remaining life prediction method of the electrolytic capacitor described in claim 4.

10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, it implements the model training method described in any one of claims 1 to 3, or executes the remaining life prediction method of an electrolytic capacitor described in claim 4.

Images