Time sequence data feature extraction method based on frequency GINI coefficient

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a time series data feature extraction method based on a frequency GINI coefficient, and the method comprises the following steps: collecting an input data stream, extracting feature parameters in the data stream, recording the numerical value of each feature parameter at each time point, arranging the time series data of each feature parameter, and obtaining a feature parameter set. According to the method, by introducing an automatic feature engineering method with a standard GINI coefficient as a core, the efficiency and precision of data preprocessing are improved. By means of GINI coefficient calculation and threshold setting, abnormal features in the data can be recognized and marked, and errors caused by data anomalies are reduced. In addition, dependence on expert intervention is reduced through automatic feature parameter marking and exception isolation operation, and the capacity of processing a big data set is improved. The method not only accelerates the processing flow of the feature engineering, but also improves the generalization ability of the model because the method can continuously adapt to new data input.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method for extracting time series data features based on frequency GINI coefficients. Background Art

[0002] Automated feature engineering is a technique in the field of applied artificial intelligence, especially machine learning. Its main purpose is to identify, create and optimize features in data through automated methods, thereby improving the performance of data analysis models. Feature engineering is one of the key steps in building effective machine learning models, because good features can provide more accurate data representation, making the model more accurate and efficient.

[0003] However, existing technologies are insufficient in terms of automation and real-time data processing capabilities. In particular, without the support of automated tools, the data processing process is susceptible to human bias and it is difficult to maintain processing efficiency and accuracy on dynamically changing data sets. In addition, existing technologies often fail to update feature selection and optimization criteria in real time when processing large-scale data, resulting in limited model adaptability and accuracy. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings existing in the prior art and to propose a time series data feature extraction method based on frequency GINI coefficient.

[0005] In order to achieve the above object, the present invention adopts the following technical solution, a method for extracting time series data features based on frequency GINI coefficient, comprising the following steps:

[0006] Collect input data streams, extract feature parameters from the data streams, extract feature X, feature Y, feature Z, and feature W, record the value of each feature parameter at each time point, organize the time series data of each feature parameter, and obtain a feature parameter set;

[0007] Calculate the distribution frequency of each feature parameter in the feature parameter set, count the number of occurrences of each feature parameter in each value range, calculate the frequency value of each value range, and obtain a feature distribution set;

[0008] Based on the feature distribution set, a standard GINI coefficient of each feature parameter is calculated, and for each feature parameter, the value is sorted from small to large, and the cumulative distribution frequency is calculated, and then the GINI coefficient is calculated according to the cumulative distribution frequency to generate a feature GINI coefficient set;

[0009] Set a GINI coefficient threshold range, compare the GINI coefficient value in the characteristic GINI coefficient set with the threshold range, and for each characteristic parameter, determine whether its GINI coefficient exceeds the threshold range. If so, mark the characteristic parameter as abnormal, and obtain an abnormal characteristic parameter set;

[0010] For the abnormal characteristic parameter set, data correction, alarm triggering and data point isolation operations are performed to correct the values ​​of the abnormal characteristic parameters, trigger alarm information, isolate abnormal data points, and obtain a processed characteristic set.

[0011] Preferably, the steps of acquiring the characteristic parameter set are:

[0012] Collect input data streams, extract feature parameters in the data streams, including feature X, feature Y, feature Z, and feature W, and obtain a preliminary feature parameter list;

[0013] Based on the preliminary feature parameter list, record the value of each feature parameter at each time point, organize the values ​​to form time series data, and generate a time series data set;

[0014] According to the time series data set, the formula is adopted:

[0015]

[0016] Calculate the time series weighted average of each feature parameter, where S represents the weighted average of the feature parameter set, x i is the value of the characteristic parameter at time point i, t i is the corresponding time weight, and the feature parameter set is obtained.

[0017] Preferably, the step of acquiring the feature distribution set is:

[0018] Acquire time series data of each feature parameter from the feature parameter set to obtain a time series data set of each feature;

[0019] Calculate the number of occurrences of the feature parameter within each value range to obtain the frequency distribution data of each feature;

[0020] According to the frequency distribution data of each feature, the formula is used:

[0021]

[0022] Calculate the frequency value of each value range and generate a feature distribution set; where F(x) represents the frequency value, f k is the number of occurrences in the kth value range, x k is the center value of the range, μ is the mean of the values, ∈ is a small positive number to ensure that the denominator is non-zero, and N is the total number of occurrences of the value.

[0023] Preferably, the steps of acquiring the characteristic GINI coefficient set are:

[0024] Extracting frequency distribution data of each feature from the feature distribution set to obtain a frequency distribution data set;

[0025] For each feature parameter, sort the values ​​from small to large, calculate the cumulative distribution frequency of each feature, and generate a cumulative distribution frequency set;

[0026] Based on the cumulative distribution frequency set, the formula is adopted:

[0027]

[0028] Calculate the GINI coefficient of each feature parameter and generate a feature GINI coefficient set; where G i represents the GINI coefficient of feature i, p k represents the frequency value of the kth data point, n is the total number of data points, is the average of all frequency values, and δ is a small amount added to avoid the denominator being zero.

[0029] Preferably, the steps of acquiring the abnormal feature parameter set are:

[0030] Obtaining the GINI coefficient of each characteristic parameter from the characteristic GINI coefficient set to obtain a GINI coefficient data set;

[0031] Set the GINI coefficient threshold range, compare the GINI coefficient of each feature with the set threshold, determine whether each feature parameter exceeds the threshold, and generate a list of over-threshold features;

[0032] Based on the super-threshold feature list, according to:

[0033] E i ={x|G(x)>T low ∧G(x) <T high}

[0034] Judge each characteristic parameter to obtain an abnormal characteristic parameter set; among which E i represents the abnormal feature parameter set, G(x) is the GINI coefficient of feature parameter x, T low and T high are the lower and upper thresholds of the GINI coefficient respectively.

[0035] Preferably, the step of obtaining the processed feature set is:

[0036] Acquire the feature parameters marked as abnormal from the abnormal feature parameter set to obtain an abnormal feature parameter list;

[0037] Data correction, alarm triggering and data point isolation operations are performed on abnormal characteristic parameters to generate a preliminary corrected data set.

[0038] Preferably, the step of acquiring the processed feature set further comprises:

[0039] Integrate the preliminary correction data set using the formula:

[0040]

[0041] Re-judge the corrected feature parameter values ​​to obtain the processed feature set; f represents the processed feature set, d i is the correction data of the ith characteristic parameter, c i is the correction factor used to adjust the influence of each data point.

[0042] Compared with the prior art, the advantages and positive effects of the present invention are:

[0043] The present invention improves the efficiency and accuracy of data preprocessing by introducing an automated feature engineering method with the standard GINI coefficient as the core. By using the GINI coefficient calculation and threshold setting, abnormal features in the data can be identified and marked, reducing errors caused by data anomalies. In addition, automated feature parameter marking and abnormal isolation operations reduce dependence on expert intervention and improve the ability to process large data sets. It not only accelerates the processing flow of feature engineering, but also improves the generalization ability of the model, because it can continuously adapt to new data inputs and timely adjust and optimize feature selection criteria. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the steps of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0046] See also Figure 1 The present invention provides a technical solution, a method for extracting time series data features based on frequency GINI coefficient, comprising the following steps:

[0047] Collect input data streams, extract feature parameters in the data streams, extract feature X, feature Y, feature Z, and feature W, record the value of each feature parameter at each time point, organize the time series data of each feature parameter, and obtain a feature parameter set.

[0048] Calculate the distribution frequency of each feature parameter in the feature parameter set, count the number of occurrences of each feature parameter in each value range, calculate the frequency value of each value range, and obtain the feature distribution set.

[0049] Based on the feature distribution set, the standard GINI coefficient of each feature parameter is calculated. For each feature parameter, the values ​​are sorted from small to large, the cumulative distribution frequency is calculated, and then the GINI coefficient is calculated based on the cumulative distribution frequency to generate a feature GINI coefficient set.

[0050] Set the GINI coefficient threshold range, compare the GINI coefficient value in the feature GINI coefficient set with the threshold range, and for each feature parameter, determine whether its GINI coefficient exceeds the threshold range. If so, mark the feature parameter as abnormal and obtain the abnormal feature parameter set.

[0051] For the abnormal feature parameter set, data correction, alarm triggering and data point isolation operations are performed to correct the values ​​of the abnormal feature parameters, trigger alarm information, isolate abnormal data points, and obtain the processed feature set.

[0052] The steps to obtain the feature parameter set are:

[0053] Collect input data streams, extract feature parameters in the data streams, including feature X, feature Y, feature Z, and feature W, and obtain a preliminary feature parameter list;

[0054] Based on the preliminary feature parameter list, record the value of each feature parameter at each time point, organize the values ​​to form time series data, and generate a time series data set;

[0055] According to the time series data set, the formula is used:

[0056]

[0057] Calculate the time series weighted average of each feature parameter, where S represents the weighted average of the feature parameter set, x i is the value of the characteristic parameter at time point i, t i is the corresponding time weight, and the feature parameter set is obtained.

[0058] Specifically, the input data stream is collected, and the characteristic parameters in the data stream, including feature X, feature Y, feature Z and feature W, are extracted through real-time monitoring equipment. A preliminary characteristic parameter list is formed based on the real-time reading of each characteristic parameter. This process involves the parsing of various sensor inputs and preliminary data formatting. The data provided by each sensor needs to be noise filtered and preliminarily verified to ensure the accuracy and availability of the data, which includes checking the range, type and frequency of the data. The purpose is to provide clean and structured input for subsequent data processing steps and obtain a preliminarily processed characteristic parameter list.

[0059] Record the value of each characteristic parameter at each time point, and organize these values ​​to form time series data. This involves time synchronization and data integration of data collection points to ensure that all data are correctly recorded in chronological order. The organization process includes the unification of time labels and alignment of data, as well as the processing of missing data, such as filling in through interpolation or using the average of previous and subsequent data points. In addition, detection and processing of outliers are also necessary to ensure the consistency and accuracy of the time series and generate a time series data set.

[0060] formula The calculation process:

[0061] The formula is used to calculate the time-series weighted average of the feature parameter set, where S represents the weighted average of the feature parameter set, and x i is the value of the characteristic parameter at time point i, t i is the weight at that time point.

[0062] Through actual data collection, let x 1 ,x 2 ,x 3 The parameter values ​​at time points 1, 2, and 3 are 10, 20, and 30 respectively, and the time weight t 1 ,t 2 ,t 3 They are 1, 2, and 3 respectively. The calculation process is as follows:

[0063] S=10×1+20×2+30×3=10+40+90=140

[0064] The result shows that through weighted averaging, we can obtain an overall characterization value of the feature parameter set that comprehensively considers the time influence and parameter value. The value is 140, which reflects the weighted feature influence of all considered time points.

[0065] The steps to obtain the feature distribution set are:

[0066] Obtain the time series data of each feature parameter from the feature parameter set to obtain a time series data set for each feature;

[0067] Calculate the number of occurrences of the feature parameter within each value range to obtain the frequency distribution data of each feature;

[0068] According to the frequency distribution data of each feature, the formula is used:

[0069]

[0070] Calculate the frequency value of each value range and generate a feature distribution set; where F(x) represents the frequency value, f k is the number of occurrences in the kth value range, x kis the center value of the range, μ is the mean of the values, ∈ is a small positive number to ensure that the denominator is non-zero, and N is the total number of occurrences of the value.

[0071] Specifically, the time series data of each feature parameter is obtained from the feature parameter set, and these data are sorted to construct a time series data set. This step involves collecting data streams and extracting specific values ​​of feature X, feature Y, feature Z, and feature W from them. The integrity and accuracy of the data must be ensured during the data collection process. The value of each feature is indexed by a time tag to ensure that the time series of the data is completely corresponding, so that the data set can be effectively used in subsequent steps to obtain a time series data set of each feature.

[0072] Calculate the number of occurrences of feature parameters in each value range. According to the collected time series data set, use statistical methods to segment the data of each feature. Each segment represents a value range. Count the data points in each segment and calculate the number of occurrences within these value ranges. This process involves the basic frequency statistics method of statistics. Through frequency statistics, we can preliminarily observe the distribution trend of the data, and then organize the frequency distribution data to obtain the frequency distribution data of each feature.

[0073] formula The calculation process:

[0074] During the feature parameter collection process, for feature X, the number of times recorded in 10 different value ranges is:

[0075] f k ={10,15,20,25,30,35,40,45,50,55}

[0076] The center value of each value range:

[0077] x k ={1,2,3,4,5,6,7,8,9,10}

[0078] For example, the average value of all feature values:

[0079] μ=5.5

[0080] The small amount added is:

[0081] ∈=0.01

[0082] Total occurrences:

[0083] N = 365

[0084] During the calculation process, for each k, f k 、x k , μ and ∈ are put into the formula to obtain the frequency value F(xk ), the specific calculation is as follows:

[0085]

[0086] This process is continued until F(10). Specifically, the deviation between the center value and the mean value of each value range is taken into account to more accurately evaluate the influence of each value range, and the final feature distribution set will reflect the actual distribution of different feature values.

[0087] The steps to obtain the characteristic GINI coefficient set are:

[0088] Extract the frequency distribution data of each feature from the feature distribution set to obtain a frequency distribution data set;

[0089] For each feature parameter, sort the values ​​from small to large, calculate the cumulative distribution frequency of each feature, and generate a cumulative distribution frequency set;

[0090] Based on the cumulative distribution frequency set, the formula is used:

[0091]

[0092] Calculate the GINI coefficient of each feature parameter and generate a feature GINI coefficient set; where G i represents the GINI coefficient of feature i, p k represents the frequency value of the kth data point, n is the total number of data points, is the average of all frequency values, and δ is a small amount added to avoid the denominator being zero.

[0093] Specifically, data is extracted from the feature parameter set to generate a time series data set. During this process, the numerical changes of each feature parameter at different time points are analyzed in detail, and these changes are recorded to construct the time series representation of the data. This method can comprehensively evaluate the volatility of the feature parameters over time, which is crucial for understanding the performance of features under different conditions. Especially when dealing with data involving time-sensitive analysis, time series analysis of feature data can more accurately capture the dynamic changes of data and provide the necessary basic information for subsequent data analysis. The core purpose of this stage is to establish a comprehensive and detailed time series data set for more in-depth statistical analysis and pattern recognition to obtain time series data sets for each feature.

[0094] Calculate the number of occurrences of the feature parameter in each value range. This process identifies and records the frequency of each value range by analyzing the distribution of each feature parameter value. This operation requires accurate classification and counting of each feature value to ensure the integrity and accuracy of the data. The generation of frequency distribution data of feature parameters involves detailed sorting and classification of the data. The accurate classification and counting of these data is to ensure that the analysis of the data in subsequent steps can reflect the actual distribution. This classification and counting process requires not only the accurate execution of the algorithm, but also a reasonable data processing strategy to avoid any deviation or error in the data processing process, so as to generate frequency distribution data for each feature.

[0095] formula:

[0096]

[0097] Formula calculation process: First, in order to calculate the GINI coefficient G of the characteristic parameter i , we must extract the frequency distribution of each feature from the dataset p k Collect data in the data, which reflects the frequency of occurrence of eigenvalues ​​in various ranges. Calculate the frequency p of each eigenvalue k With the average frequency The difference is calculated by adding a small amount δ, usually 0.01, to avoid the denominator being zero and increase the stability of the calculation. The square root and absolute value processing is applied to each frequency value to ensure that all values ​​are non-negative and take into account the size of the difference.

[0098] Next, using the weighted sum method, each frequency value p k It is weighted by its corresponding adjusted serial number 2k-n-1. This weighting method takes into account the position of the data point, increases the weight of the data points at both ends of the serial number, and emphasizes the uneven distribution. The denominator is the sum of the adjusted frequency values, which is used to standardize the GINI coefficient to ensure that the result is between 0 and 1, better reflecting the degree of inequality of the feature.

[0099] By calculating the given frequency distribution and substituting the data into the above formula, the GINI coefficient of each feature can be specifically calculated, providing a powerful tool for measuring inequality or diversity for data analysis and further statistical processing.

[0100] The steps to obtain the abnormal feature parameter set are:

[0101] Obtain the GINI coefficient of each feature parameter from the feature GINI coefficient set to obtain a GINI coefficient data set;

[0102] Set the GINI coefficient threshold range, compare the GINI coefficient of each feature with the set threshold, determine whether each feature parameter exceeds the threshold, and generate a list of over-threshold features;

[0103] Based on the super-threshold feature list, according to:

[0104] E i ={x|G(x)>T low ∧G(x) <T high}

[0105] Judge each characteristic parameter to obtain an abnormal characteristic parameter set; among which E i represents the abnormal feature parameter set, G(x) is the GINI coefficient of feature parameter x, T low and T high are the lower and upper thresholds of the GINI coefficient respectively.

[0106] Specifically, the GINI coefficient of each characteristic parameter is obtained from the characteristic GINI coefficient set. First, data collection is performed to organize the GINI coefficient of each characteristic parameter under different conditions. These data usually come from a detailed database or are obtained through real-time monitoring. The GINI coefficient of each characteristic parameter is calculated by analyzing the characteristic distribution data, which involves the central tendency and dispersion of the data. The GINI coefficient data of each characteristic parameter reflects the unequal distribution of the characteristic in the population. Through statistical methods, such as calculating the frequency of the characteristic and the integral of the frequency, the GINI coefficient data set can be accurately obtained. This process ensures the accuracy and reliability of the data and obtains the GINI coefficient data set.

[0107] Set the threshold range of the GINI coefficient. The threshold range of the GINI coefficient is determined through historical data and industry standards provided by the data analysis team. This process involves a large amount of data comparison and historical record analysis to ensure that the set threshold can reflect abnormal feature parameters while not being too sensitive to cause frequent false alarms. Compare the GINI coefficient of each feature with the set threshold, identify feature parameters that exceed the normal range through calculation and logical judgment, and generate a list of super-threshold features. This process is achieved through programming and algorithms, which ensures processing speed and accuracy and generates a list of super-threshold features.

[0108] According to the super-threshold feature list, the set GINI coefficient threshold range is used to perform abnormal detection of feature parameters, where G(x) represents the GINI coefficient of feature parameter x, T low and T high Represent the lower and upper thresholds of the GINI coefficient respectively.

[0109] Determine the value of G(x), for example, G(x) = 0.35, and set the threshold to T low= 0.25 and T high =0.40, if G(x) falls within this interval, such as 0.35>0.25 and 0.35<0.40, then x is marked as an abnormal feature. Ensure that only features that are truly beyond the normal range of variation are marked, and ensure that each feature parameter is accurately calculated to obtain an abnormal feature parameter set.

[0110] The steps to obtain the processed feature set are:

[0111] Acquire the feature parameters marked as abnormal from the abnormal feature parameter set to obtain an abnormal feature parameter list;

[0112] Perform data correction, alarm triggering and data point isolation operations on abnormal characteristic parameters to generate a preliminary corrected data set;

[0113] Integrate the preliminary correction data set using the formula:

[0114]

[0115] Re-judge the corrected feature parameter values ​​to obtain the processed feature set; f represents the processed feature set, d i is the correction data of the ith characteristic parameter, c i is the correction factor used to adjust the influence of each data point.

[0116] Specifically, the formula The calculation process:

[0117] When processing abnormal feature parameter sets, each corrected data d i Re-estimation is performed using a logistic regression model, which is represented by the above formula. In the formula, R f represents the sum of the processed feature set, d i is the correction data of the ith characteristic parameter, c i is the corresponding correction factor, which is used to adjust the influence of this data point.

[0118] There is a simplified data set, where d = [0.5, 1.2, 0.3] and the correction coefficient c = [1, -1, 0.5], then the calculation process is as follows:

[0119]

[0120]

[0121] R f =0.3655+0.3226+0.1867

[0122] R f ≈0.8748

[0123] The result shows that the weighted sum of the comprehensive corrected feature parameters is 0.8748, which represents the comprehensive characteristics of the corrected data and reflects the statistical performance of the adjusted data points in the feature set.

[0124] The characteristic parameters extracted from the abnormal characteristic parameter set are confirmed to be parameters beyond the normal range through detailed inspection. This operation ensures that only the parameters that really need attention are marked and enter the subsequent correction process. By accumulating these data, a list containing all abnormal characteristics is established. This list not only reflects the abnormal status of each parameter, but also provides a direct input source for subsequent processing.

[0125] The acquired list of abnormal characteristic parameters is used as input and processed through a series of predetermined data correction measures. Each characteristic parameter not only undergoes data correction, but also triggers corresponding alarms and data point isolation operations. These measures ensure the quality of the data and the responsiveness of the system. The corrected data set collects all these adjusted parameters, providing a basis for further data integration and analysis.

[0126] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for extracting time series data features based on frequency GINI coefficient, characterized in that: The following steps are involved: Collect input data streams, extract feature parameters from the data streams, extract feature X, feature Y, feature Z, and feature W, record the value of each feature parameter at each time point, organize the time series data of each feature parameter, and obtain a feature parameter set; Calculate the distribution frequency of each feature parameter in the feature parameter set, count the number of occurrences of each feature parameter in each value range, calculate the frequency value of each value range, and obtain a feature distribution set; Based on the feature distribution set, a standard GINI coefficient of each feature parameter is calculated, and for each feature parameter, the value is sorted from small to large, and the cumulative distribution frequency is calculated, and then the GINI coefficient is calculated according to the cumulative distribution frequency to generate a feature GINI coefficient set; Set a GINI coefficient threshold range, compare the GINI coefficient value in the characteristic GINI coefficient set with the threshold range, and for each characteristic parameter, determine whether its GINI coefficient exceeds the threshold range. If so, mark the characteristic parameter as abnormal, and obtain an abnormal characteristic parameter set; For the abnormal characteristic parameter set, data correction, alarm triggering and data point isolation operations are performed to correct the values ​​of the abnormal characteristic parameters, trigger alarm information, isolate abnormal data points, and obtain a processed characteristic set.

2. The method for extracting time series data features based on frequency GINI coefficient according to claim 1, characterized in that: The steps of obtaining the characteristic parameter set are: Collect input data streams, extract feature parameters in the data streams, including feature X, feature Y, feature Z, and feature W, and obtain a preliminary feature parameter list; Based on the preliminary feature parameter list, record the value of each feature parameter at each time point, organize the values ​​to form time series data, and generate a time series data set; According to the time series data set, the formula is adopted: Calculate the time series weighted average of each feature parameter, where S represents the weighted average of the feature parameter set, x i is the value of the characteristic parameter at time point i, t i is the corresponding time weight, and the feature parameter set is obtained.

3. The method for extracting time series data features based on frequency GINI coefficient according to claim 1, characterized in that: The steps of obtaining the feature distribution set are: Acquire time series data of each feature parameter from the feature parameter set to obtain a time series data set of each feature; Calculate the number of occurrences of the feature parameter within each value range to obtain the frequency distribution data of each feature; According to the frequency distribution data of each feature, the formula is used: Calculate the frequency value of each value range and generate a feature distribution set; where F(x) represents the frequency value, f k is the number of occurrences in the kth value range, x k is the center value of the range, μ is the mean of the values, ∈ is a small positive number to ensure that the denominator is non-zero, and N is the total number of occurrences of the value.

4. The method for extracting time series data features based on frequency GINI coefficient according to claim 1, characterized in that: The steps for obtaining the characteristic GINI coefficient set are: Extracting frequency distribution data of each feature from the feature distribution set to obtain a frequency distribution data set; For each feature parameter, sort the values ​​from small to large, calculate the cumulative distribution frequency of each feature, and generate a cumulative distribution frequency set; Based on the cumulative distribution frequency set, the formula is adopted: Calculate the GINI coefficient of each feature parameter and generate a feature GINI coefficient set; where G i represents the GINI coefficient of feature i, p k represents the frequency value of the kth data point, n is the total number of data points, is the average of all frequency values, and δ is a small amount added to avoid the denominator being zero.

5. The method for extracting time series data features based on frequency GINI coefficient according to claim 1, characterized in that: The steps for obtaining the abnormal feature parameter set are: Obtaining the GINI coefficient of each characteristic parameter from the characteristic GINI coefficient set to obtain a GINI coefficient data set; Set the GINI coefficient threshold range, compare the GINI coefficient of each feature with the set threshold, determine whether each feature parameter exceeds the threshold, and generate a list of over-threshold features; Based on the super-threshold feature list, according to: E i ={x∣G(x)>T low ∧G(x)<T high } Judge each characteristic parameter to obtain an abnormal characteristic parameter set; among which E i represents the abnormal feature parameter set, G(x) is the GINI coefficient of feature parameter x, T low and T high are the lower and upper thresholds of the GINI coefficient respectively.

6. The method for extracting time series data features based on frequency GINI coefficient according to claim 1, characterized in that: The steps of obtaining the processed feature set are: Acquire the feature parameters marked as abnormal from the abnormal feature parameter set to obtain an abnormal feature parameter list; Data correction, alarm triggering and data point isolation operations are performed on abnormal characteristic parameters to generate a preliminary corrected data set.

7. The method for extracting time series data features based on frequency GINI coefficient according to claim 6, characterized in that: The step of obtaining the processed feature set further includes: Integrate the preliminary correction data set using the formula: Re-judge the corrected feature parameter values ​​to obtain the processed feature set; f represents the processed feature set, d i is the correction data of the ith characteristic parameter, c i is the correction factor used to adjust the influence of each data point.

Images