A method for identifying diesel engine data

Abstract

The present invention provides a method for identifying diesel engine data, including two parts, model construction and model application, which are carried out sequentially. The model construction includes: S1: establishing a general rule base based on general knowledge of the field, and establishing a rule base based on the field knowledge of experts; S2: reading a data table with a capacity of not less than 4G and complete field information, and using its Gaussian filtered data as training samples; S3: dividing the training samples into three types of training samples A, B, and C in combination with the criteria of the expert knowledge rule base; S4: extracting features for the three types of training samples A, B, and C respectively, and labeling the feature data; S5: importing the feature data and the label as the final training set into a random forest classifier, and obtaining A, B, and C sub-classifiers respectively; the model application part mainly uses the above steps to form a data table. The diesel engine data identification method described in the present invention has better data identification and matching accuracy for the actual data collected from high-power turbocharged diesel engines.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial data mining, and in particular relates to a diesel engine data recognition method combining an expert knowledge rule base with a machine learning classifier. Background Art

[0002] The current method of entering the database for high-power turbocharged diesel engine test data is mostly offline. Due to the large amount of data actually collected and stored, field loss is inevitable during the storage and file migration process. This problem will make the original historical data unusable, resulting in a large amount of information resources that can support data analysis and fault research. Waste. On the other hand, if such data is processed solely using expert experience, the accuracy and timeliness of the recognition of massive data cannot be guaranteed. In the existing technology, there are very few methods for industrial scenarios, especially for the recognition of real-time data collected by high-power turbocharged diesel engines, and the diesel engine data recognition method for information office is not suitable for industrial scenarios, so it is difficult to learn from. Summary of the invention

[0003] In view of this, the present invention aims to propose a diesel engine data identification method to solve the problem of missing fields in the actual data collected by a high-power turbocharged diesel engine.

[0004] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0005] A diesel engine data identification method includes two parts: model construction and model application. The model construction includes the following steps:

[0006] S1: Establish a general knowledge rule base based on the general knowledge of the field, and establish an expert knowledge rule base based on the domain knowledge of experts;

[0007] S2: Read a data table with a capacity of no less than 4G and complete field information, and use its Gaussian filtered data as training samples;

[0008] S3: Combine the criteria of the expert knowledge rule base to divide the training samples into three types of training samples: A, B, and C;

[0009] S4: Extract features for the three types of training samples A, B, and C respectively, and perform label processing on the feature data;

[0010] S5: Import the feature data and labels as the final training set into the random forest classifier, and train the random forest classifier with the labels as the classification type to obtain sub-classifiers A, B, and C respectively;

[0011] The model application part includes the following steps:

[0012] P1: Read the data table of the diesel engine to be identified with more than 5000 data entries, and then complete Gaussian smoothing and noise reduction according to the method of S2. If there are not enough data entries after the data processing is completed, the data identification process is skipped and P2 to P5 are not executed;

[0013] P2: Based on the general rule base built in step S1, identify the timestamp field and number sequence field in the imported data; complete the merging of the three types of training data sets A, B, and C according to the steps of S3;

[0014] P3: extracting features from the three types of training samples A, B, and C in step S4 to form feature data samples;

[0015] P4: Input the feature data samples of categories A, B, and C into sub-classifiers A, B, and C respectively, and sub-classifiers A, B, and C will output the label values ​​corresponding to their multiple field data respectively;

[0016] P5: Add column names to the original imported data form according to the predicted label values ​​to form a neat data table with header information.

[0017] Furthermore, the specific method of establishing the expert knowledge rule base in S1 is: pre-setting three data classification rule conditions in the knowledge rule base, which are:

[0018] ① The data that meets the following conditions are classified as Class A data, so that Class A data includes the required speed, actual speed, and required gear rod:

[0019] X(i)>=800, X(i) refers to the data sequence of any variable in the A-type data;

[0020] The above rules are generally applicable to high-power turbocharged diesel engines with normal operation, but they do not rule out the occurrence of a very small number of random abnormal data. The filter operator introduced by the following formula can eliminate this random factor:

[0021] ∑sgnx(X(i)-800)>0.99*n, n is the length of the differential sequence, and 0.99 is the filter operator;

[0022] ② The following conditions are met and classified as Class C data, so that Class C data includes atmospheric pressure, intake temperature, water temperature, and voltage value:

[0023] X(i)<110, X(i) refers to the data sequence formed by any variable in the C-type data;

[0024] The above rules are generally applicable to high-power turbocharged diesel engines with normal operation, but they do not rule out the occurrence of a very small number of random abnormal data. The filter operator introduced by the following formula can eliminate this random factor:

[0025] ∑sgnx(110-X(i))>0.99*n, n is the length of the differential sequence, and 0.99 is the filter operator;

[0026] ③ Data that do not meet ① and ② are classified as Class B data, and Class B data includes oil pressure, intake pressure, and exhaust temperature.

[0027] Furthermore, the knowledge rule base of S1 includes at least two general rule conditions. One is that the encoding format of the timestamp string must comply with the international encoding method to be determined as a timestamp field. If there is no field that meets the conditions, it will not be output; the other is that the numbering sequence is stored in the form of an integer variable, and the variable values ​​are arranged in ascending order, and the ascending increment is 1. This rule also introduces a filter operator to smooth random noise:

[0028] ∑sgnx(Diff_X(i))>0.99*n

[0029] Among them, Diff_X is the difference sequence formed by the number sequence, n is the difference sequence length, and 0.99 is the filter operator. Only when the above conditions are met can it be determined as a number sequence field. If there is no field that meets the conditions, it will not be output.

[0030] Furthermore, S2 specifically includes: for each column of the imported data with a capacity higher than 4G, a Gaussian filter with a specified template size of 3×3 and a standard deviation of 0.8 is used to perform smoothing and noise reduction on each sequence.

[0031] Furthermore, the S3 specifically includes: dividing the data according to the expert knowledge rule base built into step S1, and the rules can ensure that the required speed, actual speed, and required gear rod are used as training sample A; the oil pressure, intake pressure, and exhaust temperature are used as training sample B; and the intake temperature, atmospheric pressure, water temperature, 15V voltage, and 24V voltage are used as training sample C.

[0032] Furthermore, the specific method of step S4 is:

[0033] S41: Determine a feature set, and for each column of data samples of training sample A, calculate the maximum value, minimum value, average value, signal-to-noise ratio, power spectrum density, first-order main frequency, and root mean square for each 5000 data sequence as a unit time window data Xa; the above 7-dimensional features form a training sample, and the column name of the data sequence is used as its label value;

[0034] S42: According to the determined feature set, for each column of data samples of training sample B, every 5000 data sequences are taken as one unit time window data Xb to calculate the average value, maximum value, kurtosis, skewness, kurtosis, median value, minimum value, and first-order difference; the above 8-dimensional features form one training sample, and the column name of the data sequence is taken as its label value;

[0035] S43: According to the determined feature set, for each column of data samples of the training sample C, every 5000 data sequences are taken as a unit time window data Xc to calculate their mean, standard deviation, median, first-order difference value, and minimum value; the above five-dimensional features form a training sample, and the column name of the data sequence is used as its label value.

[0036] Compared with the prior art, the diesel engine data identification method of the present invention has the following advantages:

[0037] (1) The diesel engine data recognition method described in the present invention has better data recognition and matching accuracy for the actual data collected from high-power turbocharged diesel engines.

[0038] (2) The diesel engine data identification method described in the present invention, with the support of the expert knowledge rule base, automatically solves the problems of missing historical data fields and missing identifiers, thereby improving work efficiency and reducing labor time costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0040] Figure 1 It is a flow chart of a diesel engine data identification method combining an expert knowledge rule base and a machine learning classifier in an embodiment of the present invention;

[0041] Figure 2 The specific discrimination rules of the expert knowledge rule base in the embodiment of the present invention;

[0042] Figure 3 The specific discrimination rules of the general rule base in the embodiment of the present invention;

[0043] Figure 4 It is a feature set used by different types of data sets in the embodiments of the present invention;

[0044] Figure 5 Schematic diagram of a method for generating feature extraction samples in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0046] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and the like are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", and the like may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0047] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood by specific circumstances.

[0048] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0049] A diesel engine data identification method, such as Figures 1 to 5 As shown in the figure, it includes two parts: model construction and model application. Model construction includes the following steps:

[0050] S1: Establish a general knowledge rule base based on the general knowledge of the field, and establish an expert knowledge rule base based on the domain knowledge of experts;

[0051] S2: Read a data table with a capacity of no less than 4G and complete field information, and use its Gaussian filtered data as training samples;

[0052] S3: Combine the criteria of the expert knowledge rule base to divide the training samples into three types of training samples: A, B, and C;

[0053] S4: Extract features for the three types of training samples A, B, and C respectively, and perform label processing on the feature data;

[0054] S5: Import the feature data and labels as the final training set into the random forest classifier, and train the random forest classifier with the labels as the classification type to obtain sub-classifiers A, B, and C respectively;

[0055] Expert knowledge plays an important role in performance evaluation, especially when the data relationships are complex and the data distribution is similar, it can provide guidance with reference significance with the help of experts' experience and insight. In the expert knowledge system, general knowledge / common sense knowledge can also be considered as a simple type of expert knowledge. Figure 2 As shown, the expert knowledge in this embodiment is a threshold reference range obtained by using the actual variation range of multiple parameters of a high-power turbocharged diesel engine, including:

[0056] 1) The speed range of the normal operating condition of a high-power turbocharged diesel engine is 800-2300r / min;

[0057] 2) The normal range of oil pressure of high-power turbocharged diesel engine is 250~800kpa;

[0058] 3) The normal range of exhaust temperature of high-power turbocharged diesel engines is not higher than 750°C;

[0059] 4) The normal range of intake air temperature of a high-power turbocharged diesel engine is not higher than 110°C;

[0060] 5) The normal range of water temperature of high-power turbocharged diesel engines is not higher than 100°C;

[0061] 6) The voltage fluctuation range of high-power turbocharged diesel engines shall not exceed 15%;

[0062] 7) The atmospheric pressure of the test site is generally not higher than 101kpa;

[0063] With the help of the above expert input, the expert knowledge rule base established in S1 is as shown in the attached Figure 2 As shown, the three classification rule conditions used in this embodiment are the same as the expert knowledge rule base conditions described in S1 of this patent, namely:

[0064] ① The data that meets the following conditions are classified as Class A data, so that Class A data includes the required speed, actual speed, and required gear rod:

[0065] ∑sgnx(X(i)-800)>0.99*n, n is the length of the differential sequence, and 0.99 is the filtering operator.

[0066] ② The following conditions are met and classified as Class C data, so that Class C data includes atmospheric pressure, intake temperature, water temperature, and voltage value:

[0067] ∑sgnx(110-X(i))>0.99*n, n is the length of the differential sequence, and 0.99 is the filtering operator.

[0068] ③ Data that do not meet ① and ② are classified as Class B data, and Class B data includes oil pressure, intake pressure, and exhaust temperature;

[0069] Furthermore, the general rule base of S1 is as shown in the attached Figure 3 As shown, it includes at least two general rule conditions. First, the timestamp string encoding format must comply with the international encoding method to be determined as a timestamp field. If there is no field that meets the conditions, it will not be output. Second, the numbering sequence is stored in the form of an integer variable, and the variable values ​​are arranged in ascending order, with an ascending increment of 1. This rule also introduces a filter operator to smooth random noise:

[0070] ∑sgnx(Diff_X(i))>0.99*n

[0071] Among them, Diff_X is the difference sequence formed by the number sequence, n is the difference sequence length, and 0.99 is the filter operator. Only when the above conditions are met can it be determined as a number sequence field. If there is no field that meets the conditions, it will not be output.

[0072] The S2 specifically includes: for each column of the imported data with a capacity higher than 4G, a Gaussian filter with a specified template size of 3×3 and a standard deviation of 0.8 is used to perform smoothing and noise reduction on each sequence.

[0073] The S3 specifically includes: dividing the data according to the expert knowledge rule base built into S1, and the rules can ensure that the required speed, actual speed, and required gear rod are used as training sample A; the oil pressure, intake pressure, and exhaust temperature are used as training sample B; the intake temperature, atmospheric pressure, water temperature, 15V voltage, and 24V voltage are used as training sample C.

[0074] The S4 specifically includes: S41: according to the attached Figure 4 The feature set determined in the above is used to calculate the maximum value, minimum value, average value, signal-to-noise ratio, power spectrum density, first-order main frequency, and root mean square of each column data sample of training sample A as 5000 data sequences as a unit time window data Xa; the above 7-dimensional features form a training sample, and the column name of the data sequence is used as its label value. S42: According to the attached Figure 4 The feature set determined in the above is used to calculate the average value, maximum value, kurtosis, skewness, kurtosis, median value, minimum value and first-order difference of each column data sample of training sample B as a unit time window data Xb; the above 8-dimensional features form a training sample, and the column name of the data sequence is used as its label value. S43: According to the attached Figure 4 The feature set determined in the above step is used to calculate the mean, standard deviation, median, first-order difference and minimum value of each column data sample of the training sample C as a unit time window data Xc. The above five-dimensional features form a training sample, and the column name of the data sequence is used as its label value. The form of the feature samples formed by the three types of data A, B and C in this step conforms to the attached Figure 5 .

[0075] The model application part of the diesel engine data identification method is as shown in the attached Figure 1 As shown, it includes the following steps:

[0076] P1: Read the data table of the diesel engine to be identified with more than 5000 data entries, and then complete Gaussian smoothing and noise reduction according to the method of S2. If there are not enough data entries after the data processing is completed, the data identification process is skipped and P2 to P5 are not executed;

[0077] Data cleaning & filtering mainly uses Python scripts to batch integrate a large amount of historical data collected from high-power turbocharged diesel engines, and complete the removal of outliers and noise reduction processing of the data. The Gaussian filter used in noise reduction processing is a linear smoothing filter that has a good inhibitory effect on noise that obeys the normal distribution. This processing can significantly improve the accuracy of the machine learning sub-classifier model, and the more reasonable the configuration of the template and standard deviation parameters and the larger the data volume, the higher the matching accuracy of the sub-classifier.

[0078] P2: Based on the built-in S1 Figure 3 The general rule base shown identifies the timestamp field and the number sequence field in the imported data; secondly, the three types of training data sets A, B, and C are merged according to the step S3;

[0079] P3: According to the steps described in S4, feature extraction is performed on the three types of training samples A, B, and C to form the following Figure 5 The feature data samples shown;

[0080] P4: Input the feature data samples of categories A, B, and C into sub-classifiers A, B, and C respectively, and sub-classifiers A, B, and C will output the label values ​​(i.e., column names) corresponding to their multiple field data respectively;

[0081] P5: Add column names to the original imported data form according to the predicted label values ​​(i.e. column names) to form a neat data table with header information.

[0082] The P4 specifically includes: P41: feeding each column of Class A prediction input samples into sub-classifier A, and recording the output label values ​​in sequence; P42: feeding each column of Class B prediction input samples into sub-classifier B, and recording the output label values ​​in sequence; P43: feeding each column of Class C prediction input samples into sub-classifier C, and recording the output label values ​​in sequence.

[0083] The P5 specifically includes: restoring the various label values ​​output by the sub-classifiers A, B, and C to the original data headers respectively, so that each column of data in the data table to be identified has a complete header, and the overall process of data matching has been completed.

[0084] As attached Figure 3 As shown, the general rules used in this embodiment are the same as the general rule base conditions described in S1 of this patent, namely:

[0085] ① The encoding format of the timestamp string must comply with the international encoding method to be determined as a timestamp field;

[0086] ② The numbering sequence is stored in the form of an integer variable, and the variable values ​​are arranged in ascending order, with an ascending increment of 1. This rule also introduces a filter operator to smooth random noise:

[0087] ∑sgnx(Diff_X(i))>0.99*n

[0088] Among them, Diff_X is the difference sequence formed by the numbering sequence, n is the length of the difference sequence, and 0.99 is the filter operator.

[0089] Basic statistical features and frequency domain features are effective derivative (dimensionality increase) methods in the field of industrial data mining. Identifying high-weight features is a prerequisite for accurate classification. Random forest is an algorithm that integrates multiple trees through the idea of ​​ensemble learning. Its basic unit is the decision tree, and its essence belongs to a major branch of machine learning - ensemble learning method. Practice shows that random forest is still quite advantageous in terms of accuracy. The training of the random forest classification model is mainly to obtain feature samples with different label values ​​in the same data set as training samples. After training with a certain amount of sample data, an effective random forest classifier is obtained. Therefore, the random forest classifier used in this embodiment is used as a subclassifier.

[0090] The following is a detailed description of the implementation method of physical sign extraction and classification model training in this embodiment:

[0091] 1. Statistical (frequency domain) feature extraction, mainly including:

[0092] 1) Sliding window processing: In order to extract features with certain statistical significance, a certain type of historical data (such as Class A) after data cleaning is segmented by sliding windows, that is, every 5,000 data are used as a basic data set for feature extraction, and there is no data overlap between adjacent basic data sets. Obviously, historical data with a length of 100,000 can be segmented into 20 groups of basic data sets;

[0093] 2) Feature sample generation: Solve the corresponding feature set for any column of a single group (such as Class A) of basic data sets. The corresponding relationship between data set type and feature set is as follows: Figure 5 As shown. Figure 4 As shown, the solved features are arranged in sequence to form a feature sample of a column of data. The larger the amount of original data, the more feature samples there are;

[0094] 2. The training of the random forest sub-classifier model mainly includes:

[0095] 1) Sample feature calculation: Combine the feature sample Feature and the original data column name Name corresponding to the feature sample into a training sample (Feature, Name);

[0096] 2) Train the random forest classifier, apply the grid search method to adjust the hyperparameters, and finally determine the optimal random forest model.

[0097] In summary, the method provided by the present invention has better data recognition and matching accuracy for the actual data collected from high-power turbocharged diesel engines; although the present invention has been exemplified as above with a preferred embodiment, this embodiment does not limit the present invention, and any person skilled in the art can make some modifications and improvements without departing from the method flow and scope of the present invention. Therefore, the scope of protection of the present invention shall be based on the definition of the claims.

[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A diesel engine data identification method, characterized in that: It includes two parts: model construction and model application. The model construction includes the following steps: S1: Establish a general knowledge rule base based on the general knowledge of the field, and establish an expert knowledge rule base based on the domain knowledge of experts; The general rule base includes at least two general rule conditions. First, the timestamp string encoding format must comply with the international encoding method to be determined as a timestamp field. If there is no field that meets the conditions, it will not be output. Second, the numbering sequence is stored in the form of an integer variable, and the variable values ​​are arranged in ascending order, with an ascending increment of 1. This rule also introduces a filter operator to smooth random noise: ∑sgnx(Diff_X(i))>0.99*n Wherein, Diff_X is the difference sequence formed by the numbering sequence, n is the difference sequence length, and 0.99 is the filter operator; only when the conditions are met can it be determined as a numbering sequence field, and if there is no field that meets the conditions, it will not be output; S2: Read a data table with a capacity of no less than 4G and complete field information, and use its Gaussian filtered data as training samples; S3: Divide the training samples into three types of training samples A, B, and C according to the criteria of the expert knowledge rule base; specifically, the following steps are performed: divide the data according to the expert knowledge rule base built in step S1, and use the required speed, actual speed, and required gear rod as training sample A; use the oil pressure, intake pressure, and exhaust temperature as training sample B; and use the intake temperature, atmospheric pressure, water temperature, 15V voltage, and 24V voltage as training sample C; S4: Extract features for the three types of training samples A, B, and C respectively, and perform label processing on the feature data; S5: Import the feature data and labels as the final training set into the random forest classifier, and train the random forest classifier with the labels as the classification type to obtain sub-classifiers A, B, and C respectively; The model application part includes the following steps: P1: Read the data table of the diesel engine to be identified with more than 5000 data entries, and then complete Gaussian smoothing and noise reduction according to the method of S2. If there are not enough data entries after the data processing is completed, the data identification process is skipped and P2 to P5 are not executed; P2: Based on the general rule base built in step S1, identify the timestamp field and number sequence field in the imported data; complete the merging of the three types of training data sets A, B, and C according to the steps of S3; P3: extracting features from the three types of training samples A, B, and C in step S4 to form feature data samples; P4: Input the feature data samples of categories A, B, and C into sub-classifiers A, B, and C respectively, and sub-classifiers A, B, and C will output the label values ​​corresponding to their multiple field data respectively; P5: Add column names to the original imported data form according to the predicted label values ​​to form a neat data table with header information.

2. A diesel engine data identification method according to claim 1, characterized in that: The specific method of establishing the expert knowledge rule base in S1 is: pre-setting three data classification rule conditions in the knowledge rule base, which are: ① The data that meets the following conditions are classified as Class A data, so that Class A data includes the required speed, actual speed, and required gear rod: X(i)>=800, X(i) refers to the data sequence of any variable in the A-type data; The above rules are generally applicable to high-power turbocharged diesel engines with normal operation, but they do not rule out the occurrence of a very small number of random abnormal data. The filter operator introduced by the following formula can eliminate this random factor: ∑sgnx(X(i)-800)>0.99*n, n is the length of the differential sequence, and 0.99 is the filter operator; ② The following conditions are met and classified as Class C data, so that Class C data includes atmospheric pressure, intake temperature, water temperature, and voltage value: X(i)<110, X(i) refers to the data sequence formed by any variable in the C-type data; The above rules are generally applicable to high-power turbocharged diesel engines with normal operation, but they do not rule out the occurrence of a very small number of random abnormal data. The filter operator introduced by the following formula can eliminate this random factor: ∑sgnx(110-X(i))>0.99*n, n is the difference sequence length, 0.99 is the filter operator; ③ Data that do not meet ① and ② are classified as Class B data, and Class B data includes oil pressure, intake pressure, and exhaust temperature.

3. A diesel engine data identification method according to claim 1, characterized in that: The S2 specifically includes: for each column of the imported data with a capacity higher than 4G, a Gaussian filter with a specified template size of 3×3 and a standard deviation of 0.8 is used to perform smoothing and noise reduction on each sequence.

4. A diesel engine data identification method according to claim 1, characterized in that: The specific method of step S4 is: S41: Determine a feature set, and for each column of data samples of training sample A, calculate the maximum value, minimum value, average value, signal-to-noise ratio, power spectrum density, first-order main frequency, and root mean square for each 5000 data sequence as a unit time window data Xa; the above 7-dimensional features form a training sample, and the column name of the data sequence is used as its label value; S42: According to the determined feature set, for each column of data samples of training sample B, every 5000 data sequences are taken as one unit time window data Xb to calculate the average value, maximum value, kurtosis, skewness, kurtosis, median value, minimum value, and first-order difference; the above 8-dimensional features form one training sample, and the column name of the data sequence is taken as its label value; S43: According to the determined feature set, for each column of data samples of the training sample C, every 5000 data sequences are taken as a unit time window data Xc to calculate their mean, standard deviation, median, first-order difference value, and minimum value; the above five-dimensional features form a training sample, and the column name of the data sequence is used as its label value.

Images