A precise periodic detection method, system and storage medium

Abstract

The present invention discloses an accurate periodic detection method, system and storage medium. The method includes: preprocessing the captured time series based on an outlier removal method to obtain a preprocessed time series; performing discrete wavelet transform denoising processing on the preprocessed time series to obtain denoised time series at different levels; calculating the autocorrelation function for the denoised time series at different levels to obtain candidate period values; and performing periodic verification calculation on the candidate period values in combination with historical data to obtain the final period value. The system includes: a first data acquisition module, a second data acquisition module, a third data acquisition module and a final data acquisition module. By using the present invention, the accuracy of the prediction result can be improved without reducing the prediction efficiency. As an accurate periodic detection method, system and storage medium, the present invention can be widely applied to the technical field of computer data processing.

Description

Technical Field

[0001] The present invention relates to the technical field of computer data processing, and in particular, to an accurate periodic detection method, system and storage medium. Background Art

[0002] With the explosive growth of data scale and applications, cloud computing has become a fast and effective solution. During the process of tenants using cloud computing services, a large number of time series are generated. Since there are a large number of periodic behaviors in the behaviors of tenants, the time series also contains periodic information. Periodic detection is an important task in time series processing and plays a crucial role in applications such as classification, clustering, and scheduling of time series. However, the current time series periodic prediction methods are often interfered by outliers and noise, resulting in a large number of inaccurate prediction results. Summary of the Invention

[0003] In order to solve the above technical problems, the object of the present invention is to provide an accurate periodic detection method, system and storage medium, which can improve the accuracy of prediction results without reducing the prediction efficiency.

[0004] The first technical solution adopted by the present invention is: an accurate periodic detection method, comprising the following steps:

[0005] Based on the outlier removal method, preprocess the captured time series to obtain the preprocessed time series;

[0006] Perform discrete wavelet transform denoising on the preprocessed time series to obtain denoised time series at different levels;

[0007] Calculate the autocorrelation function of the denoised time series at different levels to obtain candidate period values;

[0008] Perform periodic verification calculation on the candidate period values in combination with historical data to obtain the final period value.

[0009] Further, the step of based on the outlier removal method, preprocess the captured time series to obtain the preprocessed time series specifically includes:

[0010] Capture the original time series according to the usage information of network traffic data;

[0011] Perform data smoothing on the original time series to obtain a smoothed time series;

[0012] Perform outlier removal on the smoothed time series to obtain the preprocessed time series.

[0013] Furthermore, the step of removing outliers from the smoothed time series to obtain a preprocessed time series specifically includes:

[0014] Perform data monitoring on the smooth time series to obtain the abnormal point time series;

[0015] Classify the outlier time series to obtain the system error outlier time series and the non-system error outlier time series;

[0016] The time series of system error outlier points is processed based on the method of replacing outliers with average values ​​to obtain a first processed time series;

[0017] The time series of the non-systematic error outliers is processed based on the homogenization method to obtain a second processed time series;

[0018] The first processed time series and the second processed time series are combined to obtain a preprocessed time series.

[0019] Furthermore, the step of performing discrete wavelet transform denoising on the pre-processed time series to obtain denoised time series at different levels specifically includes:

[0020] The preprocessed time series is converted into multi-level scaling coefficients and multi-level scaling coefficients are constructed;

[0021] Zero-filling processing is performed on the multi-level scale coefficients to obtain zero-filled scale coefficients;

[0022] The zero-filled scaling coefficients are inversely transformed to the original domain to obtain denoised time series at different levels.

[0023] Furthermore, the step of calculating the autocorrelation function of the denoised time series at different levels to obtain the candidate period value specifically includes:

[0024] The autocorrelation function is used to calculate the different levels of the denoised time series at different levels, and the prediction results of multiple periodic prediction values ​​are obtained;

[0025] Make judgments based on the forecast results of the periodic forecast value;

[0026] It is determined that there are repeated values ​​in the prediction results, and the majority period value in the results is selected as the candidate period value;

[0027] It is determined that the prediction results are polarized, and the polar period values ​​of the prediction results are selected as candidate period values;

[0028] It is determined that there are abnormal period prediction values ​​in the prediction results, the abnormal period values ​​are eliminated, and the mean of the remaining prediction results is selected as the candidate period value.

[0029] Furthermore, the formula of the autocorrelation function is expressed as follows:

[0030]

[0031] In the above formula, ACF k is the autocorrelation function, μ is the mean of the time series, σ 2 is the variance of the time series, X = (X 0 , x 1 ,..., X N-1 ) is the time series, N represents the length of the time series, and k represents the constant coefficient value.

[0032] Furthermore, the step of performing periodic verification calculation on the candidate period value by combining historical data to obtain the final period value specifically includes:

[0033] Select part of the historical data for discrete wavelet transform denoising processing and autocorrelation function calculation to obtain multiple candidate period values of the historical data;

[0034] Perform verification processing on multiple candidate period values of the historical data in the first processed time series to obtain the final predicted period value;

[0035] Perform weighted calculation on the historical data based on multiple candidate period values of the historical data to obtain the predicted value of future data;

[0036] Judge the length of the historical data. When it is judged that the length of the historical data ≥ 3 times the period length, predict the future data, compare the time series of the future data with the first processed time series, and verify the correctness of the candidate period value;

[0037] Calculate the Pearson coefficients between the predicted value of future data and the first processed time series respectively to obtain multiple Pearson coefficient values;

[0038] Select the maximum value of the Pearson coefficients, judge the Pearson coefficient values. When it is judged that the Pearson coefficient value is higher than the set threshold, perform verification on the correctness of the period prediction;

[0039] Select the candidate period value with the maximum Pearson coefficient as the final period value, increase the length of the historical data and continue the prediction, perform verification on the stability of the period prediction, and obtain the predicted value of long-term future data;

[0040] Compare the predicted value of long-term future data with the first processed time series, judge the stability of the period prediction. When it is judged that the gap between the time series of the two is less than the preset value, output the final period value.

[0041] The second technical solution adopted by the present invention is: a precise periodic detection system, including:

[0042] The first data acquisition module is used to preprocess the captured time series to obtain a preprocessed time series;

[0043] The second data acquisition module is used to perform discrete wavelet transform denoising processing on the preprocessed time series to obtain denoised time series at different levels;

[0044] The third data acquisition module is used to calculate the autocorrelation function of the denoised time series at different levels to obtain candidate period values;

[0045] The final data acquisition module is used to perform periodic verification calculation on the candidate period values in combination with historical data to obtain the final period value.

[0046] A computer storage medium has a computer program stored thereon. When the computer program is executed by a processor, the data prediction method is implemented.

[0047] The beneficial effects of the method and system of the present invention are as follows: The present invention uses an accurate period prediction method based on the frequency-time joint domain. Through the discrete wavelet transform (DWT) denoising processing technology, high-frequency noise is removed while the original frequency domain information is retained. Then, the candidate period values are calculated through the autocorrelation function (ACF). Finally, the candidate period values are subjected to periodic verification, which improves the accuracy of the calculation results while ensuring the calculation accuracy. Description of the Drawings

[0048] Figure 1 is a flowchart of the steps of an accurate periodic detection method of the present invention;

[0049] Figure 2 is a structural block diagram of an accurate periodic detection system of the present invention;

[0050] Figure 3 is a schematic diagram of obtaining time series outliers and fluctuations of the present invention;

[0051] Figure 4 is a calculation flowchart of the DWT algorithm of the present invention; Detailed Embodiments

[0052] The following further describes the present invention in detail with reference to the drawings and specific embodiments. For the step numbers in the following embodiments, they are only set for the convenience of explanation and illustration, and no limitation is imposed on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0053] The present invention utilizes an accurate periodic prediction method based on the frequency-time joint domain. Through discrete wavelet transform (DWT), while retaining the original frequency domain information, high-frequency noise is removed. Then, candidate periods are calculated through autocorrelation function calculation (ACF), which can improve the accuracy of the prediction result without reducing the prediction efficiency.

[0054] Referring to Figure 1 , the present invention provides an accurate periodic detection method, and the method includes the following steps:

[0055] S1. Based on the method of removing outliers, preprocess the captured time series to obtain the preprocessed time series.

[0056] S11. Capture the original time series according to the usage information of network traffic data;

[0057] S12. Perform data smoothing on the original time series to obtain the smoothed time series;

[0058] Specifically, referring to Figure 3 , collect the original network traffic data, obtain the time series from the collected network traffic data. During the collection process of network traffic data, there will be a short-term data missing phenomenon. For example, when the network is under high-load operation, in order to ensure the transmission of traffic data, the collection of network traffic data will be temporarily stopped. The short-term missing of traffic data collection will be manifested as the missing of time series data, generating a minimum value, resulting in data fluctuations. Such outliers caused by systematic errors need to be removed through data preprocessing. Due to user usage, instantaneous traffic peaks may occur in the traffic, which is caused by the normal business of users, but continuous peaks may also be generated. Therefore, it is necessary to smooth the data while retaining the fluctuations of the traffic, and try to reduce the appearance of spikes in the fluctuations.

[0059] S13. Perform outlier removal processing on the smoothed time series to obtain the preprocessed time series.

[0060] S131. Detect the smoothed time series segments with data fluctuations in the smoothed time series to obtain the outlier time series;

[0061] Specifically, form a new sequence with the change amount of the data, detect the data through the Tukey test method, and classify the outliers. The Tukey test method calculates the upper quartile Q3 and the lower quartile Q1 of the data, and calculates the maximum estimate Q3 + k(Q3 - Q1) and the minimum estimate Q1 - k(Q3 - Q1) according to the coefficient k. In this article, the result k = 1.5 obtained by the Tukey test method is used to find the outlier sequence in the data.

[0062] S132. Classify the abnormal point time series to obtain the system error abnormal point time series and the non-system error abnormal point time series;

[0063] S133. Process the system error abnormal point time series based on the method of replacing abnormal values with the average value to obtain the first processed time series;

[0064] S134. Process the non-system error abnormal point time series based on the homogenization method to obtain the second processed time series;

[0065] S135. Combine the first processed time series and the second processed time series to obtain the preprocessed time series.

[0066] Specifically, the method of using the change amount sequence of data to find the abnormal point sequence is to facilitate the distinction of different types of abnormal points. For the minimum value caused by system error, it is reflected in the abnormal point sequence as a continuous sequence of length 2. The first value is less than the minimum value estimate, and the second value is greater than the maximum value estimate. For such abnormal points, the average value of the front and back data is used to replace the abnormal value. For other types of abnormal values, the homogenization method is used to replace the original abnormal value sequence with a uniformly spaced sequence that first increases and then decreases to smooth the data while retaining the change information of the flow, and output the preprocessed time series.

[0067] S2. Perform discrete wavelet transform denoising processing on the preprocessed time series to obtain denoised time series at different levels.

[0068] S21. Convert the preprocessed time series into multi-level scale coefficients and construct multi-level scale coefficients;

[0069] S22. Perform zero-padding processing on the multi-level scale coefficients to obtain zero-padded scale coefficients;

[0070] S23. Inverse-transform the zero-padded scale coefficients into the original domain to obtain denoised time series at different levels.

[0071] Specifically, the discrete wavelet transform DWT algorithm is used to perform denoising at different levels on the input time series to facilitate the later periodic detection of the periodic value. Perform DWT calculation on the preprocessed time series. Assume that the time series X of length N = (X 0 , X 1 ,..., X N-1 ), and calculate the wavelet coefficients w j and the scale coefficients v j

[0072]

[0073] In the above formula, j represents the level of the time series, wj represents the wavelet coefficients of the time series, v j represents the scale factor of the time series, h l represents a high-pass filter, g l represents a low-pass filter, k represents the constant coefficient value, N is the length of the time series, and N j ≡N*2 -j , v 0 =X.

[0074] For the preprocessed time series, the low-frequency component is very important, which often contains the characteristics of the time series, while the high-frequency component corresponds to the details or differences of the time series. For the task of periodic prediction, although the high-frequency component also contains a small amount of periodic information, most of it is noise. Removing the high-frequency part can increase the accuracy of ACF prediction. Therefore, the zero-filling method is used to remove the wavelet coefficients calculated at each layer, and only the proportional coefficient is used for wavelet reconstruction to obtain denoised time series at different levels and output denoised time series at different levels.

[0075] S3. Calculate the autocorrelation function of the denoised time series at different levels to obtain candidate period values.

[0076] S31, calculating different levels of denoised time series at different levels respectively through autocorrelation functions to obtain prediction results of multiple period prediction values;

[0077] S32, making a judgment based on the prediction result of the periodic prediction value;

[0078] S33, determining that there are repeated values ​​in the prediction result, selecting the majority period value in the result as the candidate period value;

[0079] S34, judging that the prediction results are bipolarly distributed, selecting bipolar period values ​​of the prediction results as candidate period values;

[0080] S35: If it is determined that there are abnormal period prediction values ​​in the prediction results, the abnormal period values ​​are removed, and the average of the remaining prediction results is selected as the candidate period value.

[0081] Specifically, refer to Figure 4 After obtaining the denoised time series at different levels through DWT, ACF is used to calculate the period of the denoised time series at different levels. The autocorrelation value of any series with lag values ​​can be obtained through ACF calculation. Specifically, it describes the degree of correlation between the current value of the series and its future value. The time series can contain components such as trend, seasonality, periodicity and residuals. ACF will consider all these components when looking for correlation, including direct and indirect correlation information. For a time series of length N, X = (X 0 ,X 1,...,X N-1 ) Calculate the ACF according to the following formula k :

[0082]

[0083] In the above formula, ACF k is the autocorrelation function, μ is the mean of the time series, and σ 2 is the variance of the time series. X = (X 0 , X 1 ,..., X N-1 ) is the time series, N represents the length of the time series, and k represents the constant coefficient value.

[0084] Based on the calculation result of the obtained ACF, further select the maximum value among the local peaks as the period calculated by the ACF. For the denoised time series at different levels, calculate different ACF prediction results respectively to obtain the predicted values of multiple periods. Make a judgment according to the prediction results of the period prediction values. When there are repeated values in the prediction results of the period, select the mode in the prediction results as the candidate period. When the prediction results of the period show a polarized distribution, due to errors in the DWT process, the ACF will predict long periods and short periods respectively. Select the two poles of the prediction results as the candidate periods. When there are abnormal points in the prediction results of the period, remove the abnormal points and select the mean value of the remaining prediction results as the candidate period.

[0085] S4. Combine historical data to perform periodic verification calculation on the candidate period values to obtain the final period value.

[0086] S41. Select some historical data for discrete wavelet transform denoising processing and autocorrelation function calculation to obtain multiple candidate period values of historical data;

[0087] Specifically, since the calculation process of the ACF tends to discover long periods and is prone to outliers and noise, the calculation results will also have errors due to calculation errors in the DWT process or abnormal points in the time series. Therefore, it is necessary to verify the correctness and stability of the candidate period values to achieve the purpose of accurately predicting the period. For the existing historical data, first select a small part of the data for periodic prediction, and then decide whether to use more data to predict the period or verify the stability of the period according to the prediction results. Take out some historical data and obtain multiple candidate period values of historical data through DWT and ACF.

[0088] S42. Verify the multiple candidate period values of historical data in the first processed time series to obtain the final predicted period value;

[0089] S43. Perform weighted calculation on the historical data based on multiple candidate period values of historical data to obtain the predicted value of future data;

[0090] Specifically, since the calculation of ACF may produce prediction results for multiple cycles, it is necessary to verify each candidate period value of historical data in the source data, find the most suitable candidate period value of historical data, obtain the first historical data period value, and based on the obtained first historical data period value, perform weighted processing on the first historical data period value to obtain the future data period prediction value.

[0091] S44. Judge the length of historical data. When it is judged that the length of historical data ≥ 3 times the period length, predict future data, compare the time series of future data with the first processed time series, and verify the correctness of the candidate period value.

[0092] Specifically, for a time series X of length N = (X 0 , X 1 ,..., X N-1 ), after obtaining the period T, the prediction result at time t is X t = (X t-T + X t-2T + X t-3T ) / 3. When the future data period prediction value is less than 3 periods, use the real data as the prediction data, and further wait until there is enough historical information to verify the correctness of the period. For the correctness verification of period prediction, predict the future data period prediction value based on the candidate period value, and compare the prediction result with the real data. If the prediction result is close to the real time series, it is considered that the result predicted by the candidate period value is correct.

[0093] S45. Calculate the Pearson coefficients between the future data prediction value and the first processed time series respectively to obtain multiple Pearson coefficient values.

[0094] S46. Select the maximum value of the Pearson coefficients, judge the Pearson coefficient values, and when it is judged that the Pearson coefficient value is higher than the set threshold, perform the correctness verification of period prediction.

[0095] S47. Select the candidate period value with the maximum Pearson coefficient as the final period value, increase the length of historical data and continue to predict to perform the stability verification of period prediction and obtain the long-term future data prediction value.

[0096] Specifically, based on the obtained multiple historical data candidate period values, calculate the Pearson correlation coefficient (PCC) between the prediction result of the future time series using the candidate period value and the first processed time series, which is used to represent the correlation degree between the prediction result and the true result. The closer the calculation result is to 1, the stronger the correlation. Further, take the period corresponding to the maximum value of PCC as the predicted period value. If the result of PCC is higher than the predefined threshold, it indicates that there is a strong correlation between the prediction result and the true data, indicating that the period prediction is correct. The period can be used to predict more long-term data to verify the stability of the period, and the long-term future data period prediction value can be obtained. If the result of PCC is lower than the predefined threshold, it indicates that there is a certain difference between the prediction result and the true data, and the prediction data cannot represent the true data. Therefore, it is necessary to increase the historical data and re-predict the period value.

[0097] S48. Compare the long-term future data prediction value with the first processed time series, and judge the stability of the period prediction. If it is judged that the gap between the time series of the two is less than the preset value, output the final period value.

[0098] Specifically, for the stability of the period prediction, based on the historical data period value, continue to predict more long-term future data to obtain the long-term future data prediction value. If the prediction result can be kept close to the true time series for a long time, it indicates that the period can exist stably for a long time, and the period is considered relatively stable. The simulation experiment data of the present invention is as follows in the table:

[0099]

[0100]

[0101] Refer to Figure 2 , a precise periodic detection system, including:

[0102] The first data acquisition module is used to preprocess the captured time series to obtain the preprocessed time series;

[0103] The second data acquisition module is used to perform discrete wavelet transform denoising processing on the preprocessed time series to obtain denoised time series at different levels;

[0104] The third data acquisition module is used to calculate the autocorrelation function of the denoised time series at different levels to obtain candidate period values;

[0105] The final data acquisition module is used to perform periodic verification calculation on the candidate period values in combination with historical data to obtain the final period value.

[0106] The present invention also provides a computer storage medium, on which a computer program is stored. If the method of the present invention is implemented in the form of software functional units and sold or used as an independent product, it can be stored in this computer storage medium. Based on such an understanding, to implement all or part of the processes in the above-described embodiment methods of the present invention, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer storage medium. When the computer program is executed by a processor, the steps of the above various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer storage medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory, random access memory, electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer storage medium does not include electrical carrier signals and telecommunication signals.

[0107] The content in the above method embodiments is applicable to the system embodiments of the present invention. The functions specifically implemented in the system embodiments of the present invention are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0108] The above has specifically described the preferred embodiments of the present invention, but the present invention is not limited to the described embodiments. Those skilled in the art can also make various equivalent deformations or substitutions without departing from the spirit of the present invention, and these equivalent deformations or substitutions are all included within the scope defined by the claims of this application.

Claims

1. An accurate periodicity detection method, It is characterized in that The following steps are involved: Based on the method of removing outliers, the captured time series is preprocessed to obtain the preprocessed time series; The preprocessed time series is subjected to discrete wavelet transform denoising to obtain denoised time series at different levels; The autocorrelation function of the denoised time series at different levels is calculated to obtain the candidate period value; Combine historical data to periodically verify and calculate the candidate cycle values ​​to obtain the final cycle value; The step of calculating the autocorrelation function of the denoised time series at different levels to obtain the candidate period value specifically includes: The autocorrelation function is used to calculate the different levels of the denoised time series at different levels, and the prediction results of multiple periodic prediction values ​​are obtained; Make judgments based on the forecast results of the periodic forecast value; It is determined that there are repeated values ​​in the prediction results, and the majority period value in the results is selected as the candidate period value; It is determined that the prediction results are polarized, and the polar period values ​​of the prediction results are selected as candidate period values; It is determined that there are abnormal period prediction values ​​in the prediction results, the abnormal period values ​​are eliminated, and the mean of the remaining prediction results is selected as the candidate period value.

2. According to claim 1, a precise periodicity detection method, It is characterized in that The step of preprocessing the captured time series based on the outlier removal method to obtain the preprocessed time series specifically includes: Capturing raw time series based on usage information of network traffic data; Perform data smoothing on the original time series to obtain a smoothed time series; The smoothed time series is processed to remove outliers to obtain the preprocessed time series.

3. According to claim 2, a precise periodicity detection method, It is characterized in that The step of removing outliers from the smoothed time series to obtain a preprocessed time series specifically includes: Perform data monitoring on the smooth time series to obtain the abnormal point time series; Classify the outlier time series to obtain the system error outlier time series and the non-system error outlier time series; The time series of system error outlier points is processed based on the method of replacing outliers with average values ​​to obtain a first processed time series; The time series of the non-systematic error outliers is processed based on the homogenization method to obtain a second processed time series; The first processed time series and the second processed time series are combined to obtain a preprocessed time series.

4. According to claim 3, a precise periodicity detection method, It is characterized in that The step of performing discrete wavelet transform denoising on the pre-processed time series to obtain denoised time series at different levels specifically includes: The preprocessed time series is converted into multi-level scaling coefficients and multi-level scaling coefficients are constructed; Zero-filling processing is performed on the multi-level scale coefficients to obtain zero-filled scale coefficients; The zero-filled scaling coefficients are inversely transformed to the original domain to obtain denoised time series at different levels.

5. According to claim 4, a precise periodicity detection method, It is characterized in that The formula of the autocorrelation function is as follows: In the above formula, ACF k is the autocorrelation function, μ is the mean of the time series, and σ 2 is the variance of the time series. X = (X 0 , X 1 ,..., X N-1 ) is the time series, N represents the length of the time series, and k represents the value of the constant coefficient.

6. The accurate periodic detection method according to claim 5, wherein, the step of performing periodic verification calculation on the candidate period value by combining historical data to obtain the final period value specifically includes: Selecting part of the historical data for discrete wavelet transform denoising processing and autocorrelation function calculation to obtain multiple candidate period values of the historical data; Performing verification processing on multiple candidate period values of the historical data in the first processing time series to obtain the final predicted period value; Performing weighted calculation on the historical data based on multiple candidate period values of the historical data to obtain the predicted value of future data; Judging the length of the historical data. When it is judged that the length of the historical data is ≥ 3 times the period length, predicting future data, comparing the time series of the future data with the first processing time series, and verifying the correctness of the candidate period value; Calculating the Pearson coefficients between the predicted value of the future data and the first processing time series respectively to obtain multiple Pearson coefficient values; Selecting the maximum value of the Pearson coefficients, judging the Pearson coefficient values. When it is judged that the Pearson coefficient value is higher than the set threshold, verifying the correctness of the period prediction; Selecting the candidate period value with the maximum Pearson coefficient value as the final period value, increasing the length of the historical data and continuing the prediction, and performing stability verification of the period prediction to obtain the predicted value of the long-term future data; Comparing the predicted value of the long-term future data with the first processing time series, judging the stability of the period prediction. When it is judged that the gap between the time series of the two is less than the preset value, outputting the final period value.

7. An accurate periodic detection system, wherein, for executing the accurate periodic detection method according to claim 1, including the following modules: The first data acquisition module is used for preprocessing the captured time series to obtain the preprocessed time series; The second data acquisition module is used for performing discrete wavelet transform denoising processing on the preprocessed time series to obtain denoised time series at different levels; The third data acquisition module is used for calculating the autocorrelation function of the denoised time series at different levels to obtain candidate period values; The final data acquisition module is used for performing periodic verification calculation on the candidate period values by combining historical data to obtain the final period values.

8. A computer-readable storage medium, on which a computer program is stored, wherein, when the program is executed by a processor, the steps of the method according to any one of claims 1-6 are implemented.

Images