Index threshold determination method and device, equipment, storage medium and program product

Abstract

The embodiment of the invention provides an index threshold determination method and device, equipment, a storage medium and a program product. The method comprises the steps that eigenvalue weighting calculation of different statistical features is carried out on historical index values in a historical index value sequence and predicted index values in a predicted index value sequence to obtain sensitivity values; under the condition that the sensitivity value is located in a preset threshold interval, extracting a plurality of first index values meeting a preset alarm rule from a historical index value sequence, and extracting a plurality of first prediction thresholds with the same timestamps as the plurality of first index values from a prediction threshold sequence, performing fitting calculation on the plurality of first index values and the plurality of first prediction thresholds to obtain a first tolerance coefficient; and adjusting the prediction threshold sequence based on the first tolerance coefficient to obtain a target prediction threshold sequence. The scheme can adapt to trend change or jitter of different index data to carry out threshold adjustment, and provides a reliable index threshold for index early warning.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technology, and in particular, to a method, device, equipment, storage medium, and program product for determining an index threshold. Background Art

[0002] With the rapid development of the Internet industry, different Internet companies have respectively built their own large-scale business systems, and a vast amount of diverse index data is generated during the operation of these business systems. For these key indicators covering business performance, system performance, and infrastructure status, Internet companies urgently need to build an efficient monitoring and anomaly detection mechanism to ensure the high stability of service operation and improve the efficiency of operation and maintenance work.

[0003] In the related art, a dynamic threshold algorithm is used for index monitoring, which can flexibly adjust the threshold setting according to the real-time changes of index data. However, due to the diversity of trends and the difference in fluctuations between different index data, the threshold prediction results of specific dynamic threshold algorithms are relatively fixed and cannot adapt to the trend changes or jitters of different index data for threshold adjustment. As a result, the accuracy of index early warning based on the predicted threshold is relatively low, and problems such as false alarms and missed alarms are likely to occur, which need to be improved. Summary of the Invention

[0004] The embodiments of the present application provide a method, device, equipment, storage medium, and program product for determining an index threshold, which solves the problem that the threshold prediction results of the dynamic threshold algorithm in the related art are relatively fixed and cannot adapt to the trend changes or jitters of different index data for threshold adjustment, resulting in relatively low accuracy of index early warning based on the threshold, and problems such as false alarms and missed alarms are likely to occur. It can adapt to the trend changes or jitters of different index data for threshold adjustment, provide a reliable index threshold for index early warning, and improve the accuracy of early warning.

[0005] In a first aspect, the embodiments of the present application provide a method for determining an index threshold, the method comprising:

[0006] Obtain a historical index value sequence, calculate the historical index value sequence according to a set sequence prediction algorithm to obtain a predicted index value sequence and a predicted threshold sequence, the historical index value sequence includes a plurality of historical index values, and the predicted index value sequence includes a plurality of predicted index values;

[0007] Perform eigenvalue weighted calculation of different statistical characteristics on the plurality of historical index values and the plurality of predicted index values to obtain a sensitivity value, and the sensitivity value is used to adjust the tolerance level of the predicted threshold sequence;

[0008] When the sensitivity value is within a preset threshold range, extract multiple first metric values that meet the preset alarm rules from the historical metric value sequence, and extract multiple first prediction thresholds with the same timestamps as the multiple first metric values from the prediction threshold sequence. Perform fitting calculations on the multiple first metric values and the multiple first prediction thresholds to obtain a first tolerance coefficient;

[0009] Adjust the prediction threshold sequence based on the first tolerance coefficient to obtain a target prediction threshold sequence for anomaly warning of real-time metric values. The target prediction threshold sequence includes multiple metric thresholds.

[0010] In a second aspect, an embodiment of the present application further provides a metric threshold determination device, which includes:

[0011] A prediction sequence determination module configured to obtain a historical metric value sequence, calculate the historical metric value sequence according to a set sequence prediction algorithm to obtain a prediction metric value sequence and a prediction threshold sequence. The historical metric value sequence includes multiple historical metric values, and the prediction metric value sequence includes multiple prediction metric values;

[0012] A sensitivity value determination module configured to perform eigenvalue weighted calculations with different statistical characteristics on the multiple historical metric values and the multiple prediction metric values to obtain a sensitivity value, and the sensitivity value is used to adjust the tolerance degree of the prediction threshold sequence;

[0013] A tolerance coefficient determination module configured to, when the sensitivity value is within a preset threshold range, extract multiple first metric values that meet the preset alarm rules from the historical metric value sequence, and extract multiple first prediction thresholds with the same timestamps as the multiple first metric values from the prediction threshold sequence. Perform fitting calculations on the multiple first metric values and the multiple first prediction thresholds to obtain a first tolerance coefficient;

[0014] A first metric threshold determination module configured to adjust the prediction threshold sequence based on the first tolerance coefficient to obtain a target prediction threshold sequence for anomaly warning of real-time metric values. The target prediction threshold sequence includes multiple metric thresholds.

[0015] In a third aspect, an embodiment of the present application further provides a metric threshold determination device, which includes:

[0016] One or more processors;

[0017] A storage device configured to store one or more programs,

[0018] When the one or more programs are executed by the one or more processors, the one or more processors implement the metric threshold determination method described in the embodiments of the present application.

[0019] In a fourth aspect, an embodiment of the present application further provides a non-volatile storage medium storing computer-executable instructions, and the computer-executable instructions are configured to execute the metric threshold determination method described in the embodiments of the present application when executed by a computer processor.

[0020] In a fifth aspect, an embodiment of the present application further provides a computer program product, which includes a computer program. The computer program is stored in a computer-readable storage medium, and at least one processor of the device reads and executes the computer program from the computer-readable storage medium, so that the device executes the metric threshold determination method described in the embodiments of the present application.

[0021] In the embodiments of the present application, by obtaining a historical metric value sequence, calculating the historical metric value sequence according to a set sequence prediction algorithm, a predicted metric value sequence and a predicted threshold sequence are obtained. Among them, the historical metric value sequence includes multiple historical metric values, and the predicted metric value sequence includes multiple predicted metric values; eigenvalue weighted calculations with different statistical characteristics are performed on the multiple historical metric values and the multiple predicted metric values to obtain a sensitivity value; when the sensitivity value is within a preset threshold interval, multiple first metric values that meet the preset alarm rule are extracted from the historical metric value sequence, and multiple first predicted thresholds with the same timestamps as the multiple first metric values are extracted from the predicted threshold sequence. Fitting calculations are performed on the multiple first metric values and the multiple first predicted thresholds to obtain a first tolerance coefficient; the predicted threshold sequence is adjusted based on the first tolerance coefficient to obtain a target predicted threshold sequence for abnormal warning of real-time metric values. In the above solution, by calculating the historical metric value sequence according to the set sequence prediction algorithm, a predicted metric value sequence and a predicted threshold sequence that conform to the change of historical metric data can be initially determined. By performing eigenvalue weighted calculations with different statistical characteristics on the historical metric values in the historical metric value sequence and the predicted metric values in the predicted metric value sequence to obtain a sensitivity value, the correlation characteristics between the historical metric value sequence and the predicted metric value sequence can be effectively combined, and the sensitivity value for reasonably determining the tolerance degree of specifically adjusting the predicted threshold sequence can be determined. By judging the threshold interval where the sensitivity value is located, a tolerance coefficient matching it can be calculated correspondingly. Based on this first tolerance coefficient, the predicted threshold sequence can be adjusted to a target predicted threshold sequence, which can adapt to the trend changes of different metric data for threshold adjustment, provide reliable metric thresholds for metric warning, and improve warning accuracy. Description of the Drawings

[0022] Figure 1Flowchart of a method for determining an index threshold provided by an embodiment of the present application;

[0023] Figure 2 Flowchart of a method for determining an index threshold provided by an embodiment of the present application, including a process of calculating an upper limit ratio coefficient and an upper limit offset coefficient;

[0024] Figure 3 Flowchart of a method for determining an index threshold provided by an embodiment of the present application, including a process of calculating a lower limit ratio coefficient and a lower limit offset coefficient;

[0025] Figure 4 Flowchart of a method for determining an index threshold provided by an embodiment of the present application, including a process of adjusting a prediction threshold sequence based on a second tolerance coefficient;

[0026] Figure 5 Flowchart of a method for determining an index threshold provided by an embodiment of the present application, including a process of adjusting a prediction threshold sequence based on a third tolerance coefficient;

[0027] Figure 6 Flowchart of a method for determining an index threshold provided by an embodiment of the present application, including a process of calculating a sensitivity value;

[0028] Figure 7 Flowchart of a method for determining an index threshold provided by an embodiment of the present application, including a process of calculating an index eigenvalue;

[0029] Figure 8 Structural block diagram of an index threshold determination device provided by an embodiment of the present application;

[0030] Figure 9 Structural schematic diagram of an index threshold determination device provided by an embodiment of the present application. Detailed implementation manners

[0031] The following further describes the embodiments of the present application in detail with reference to the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the embodiments of the present application, rather than limiting the embodiments of the present application. In addition, it should be noted that for the sake of description, only parts related to the embodiments of the present application are shown in the drawings, rather than all structures.

[0032] The terms "first", "second", etc. in the description and claims of this application are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of this application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are usually of the same category, and the number of objects is not limited. For example, the first object can be one or multiple. In addition, "and / or" in the description and claims means at least one of the connected objects, and the character " / ", generally represents an "or" relationship between the associated objects before and after.

[0033] The method for determining the index threshold provided by the embodiments of this application can calculate the sensitivity value by weighted calculation of the eigenvalues of different statistical characteristics for the historical index values in the historical index value sequence and the predicted index values in the predicted index value sequence, and judge the threshold interval where the sensitivity value is located, and select the corresponding first index value and the first predicted threshold for fitting calculation to obtain the first tolerance coefficient, which can effectively adjust the predicted threshold sequence based on the first tolerance coefficient, provide a reliable index threshold for index warning, and is beneficial to improving the accuracy of warning. The relevant indexes can include business indexes, performance indexes, infrastructure indexes, etc., and the embodiments of this application do not limit this. The method, device, equipment, storage medium, and program product for determining the index threshold provided by the embodiments of this application aim to solve the problem that the threshold prediction result of the specific dynamic threshold algorithm in the related technology is relatively fixed, and it cannot adapt to the trend change or jitter of different index data for threshold adjustment, resulting in low accuracy of the index warning based on this threshold, and it is easy to have problems such as false alarms and missed alarms.

[0034] In the method for determining the index threshold provided by the embodiments of this application, the execution subject of each step can be a computer device, which refers to any electronic device with data calculation, processing, and storage capabilities, such as a server, etc., and the embodiments of this application do not limit this.

[0035] Figure 1 It is a flowchart of a method for determining an index threshold provided by an embodiment of this application. As Figure 1 shown, it includes the following steps:

[0036] Step S101, obtain the historical index value sequence, calculate the historical index value sequence according to the set sequence prediction algorithm, and obtain the predicted index value sequence and the predicted threshold sequence, where the historical index value sequence includes multiple historical index values, and the predicted index value sequence includes multiple predicted index values.

[0037] Among them, the historical indicator value sequence can be a data sequence formed by arranging multiple historical indicator values collected at multiple set time nodes in chronological order. The sequence prediction algorithm can be the Prophet algorithm, the AutoARIMA (Autoregressive Integrated Moving Average) algorithm, etc. According to the change law of the historical indicator values in the historical indicator value sequence, multiple predicted indicator values corresponding to each historical indicator value at the corresponding future time node can be predicted. The multiple predicted indicator values can be arranged in chronological order to form a predicted indicator value sequence. Moreover, the predicted threshold corresponding to each predicted indicator value can be given according to the set confidence interval, and the predicted thresholds can be arranged in chronological order to form a predicted threshold sequence, which can specifically include an upper predicted threshold and a lower predicted threshold, constituting the uncertainty interval of the prediction. However, since the threshold prediction result of the sequence prediction algorithm is relatively fixed and does not involve threshold adjustment with different tolerance levels, it cannot provide relatively accurate predicted thresholds for adapting to the trend changes or jitters of different indicator data, and further threshold adjustment related to the tolerance level needs to be carried out through subsequent step processes.

[0038] Step S102: Calculate the sensitivity value by performing eigenvalue weighting calculations with different statistical characteristics on multiple historical indicator values and multiple predicted indicator values, where the sensitivity value is used to adjust the tolerance level of the predicted threshold sequence.

[0039] Among them, due to the trend diversity and fluctuation difference among different indicator data, for indicator data with obvious regularity, its fluctuation is small, and the adjustment of the corresponding predicted threshold requires a small tolerance level, and its early warning sensitivity is high, which can be used for abnormal alarm when the indicator data has slight jitter. For indicator data with no regularity, its fluctuation is large, and the adjustment of the corresponding predicted threshold requires a large tolerance level, and its early warning sensitivity is low, which can be used for abnormal alarm when the indicator data has large jitter. Thus, the historical indicator values in the historical indicator value sequence and the predicted indicator values in the predicted indicator value sequence can be used to perform eigenvalue weighting calculations with different statistical characteristics to obtain the sensitivity value for measuring the tolerance level of adjusting the predicted threshold sequence. The statistical characteristics can be the fluctuation characteristics of the historical indicator value sequence itself, the trend similarity characteristics, the absolute similarity characteristics, etc. between the historical indicator value sequence and the predicted indicator value sequence, which are used to measure the stability of the indicator data. In one embodiment, the statistical characteristics adopted can be the fluctuation characteristics and the trend similarity characteristics. Among them, the fluctuation characteristic value corresponding to the fluctuation characteristics can be obtained by calculating the standard deviation of the historical indicator values in the historical indicator value sequence, and the relevant calculation formula is as follows:

[0040]

[0041] wherein, Volatility is the volatility eigenvalue, and x i is the historical index value in the historical index value sequence, u is the average value of the historical index values in the historical index value sequence, and n is the number of elements in the historical index value sequence. The larger the volatility eigenvalue, the greater the volatility of the index data, the smaller the required sensitivity, and the greater the tolerance for adjusting the prediction threshold sequence.

[0042] The trend similarity eigenvalue corresponding to the trend similarity feature can be obtained by calculating the determination coefficient of the historical index value sequence and the predicted index value sequence. This trend similarity eigenvalue can measure the trend similarity between the historical index value sequence and the predicted index value sequence. The relevant calculation formula is as follows:

[0043]

[0044] wherein, Similarity is the trend similarity eigenvalue, and y i is the historical index value in the historical index value sequence, is the predicted index value in the predicted index value sequence, is the average value of the historical index values in the historical index value sequence, and n is the number of elements in the historical index value sequence. The larger the trend similarity eigenvalue, the closer the historical index value sequence and the predicted index value sequence are, the smaller the volatility of the index data, the higher the required sensitivity, and the smaller the tolerance for adjusting the prediction threshold sequence.

[0045] Finally, the sensitivity value can be obtained by performing weighted calculation on the volatility eigenvalue and the trend similarity eigenvalue. The relevant calculation formula is as follows:

[0046] Sensitivity = k1×(1 - Volatility) + k2×Similarity,

[0047] wherein, Sensitivity is the sensitivity value, Volatility is the volatility eigenvalue, Similarity is the trend similarity eigenvalue, and k1 and k2 are weights. For example, k1 = 0.5, k2 = 0.5, but this application does not make any limitations here.

[0048] In one embodiment, the statistical features adopted can be the volatility feature and the absolute similarity feature. Among them, the volatility eigenvalue corresponding to the volatility feature can be obtained by calculating the standard deviation and the average value of the historical index values in the historical index value sequence and dividing the standard deviation by the average value. The relevant calculation formula is as follows:

[0049]

[0050] Among them, Volatility is the volatility eigenvalue, SD is the standard deviation of the historical index values in the historical index value sequence, and u is the average value of the historical index values in the historical index value sequence. The larger the volatility eigenvalue, the greater the volatility of the index data, the smaller the required sensitivity, and the greater the tolerance for adjusting the prediction threshold sequence.

[0051] The absolute similarity eigenvalue corresponding to the absolute similarity feature can be obtained by calculating the mean absolute percentage error of the historical index value sequence and the predicted index value sequence. This absolute similarity eigenvalue can measure the degree of closeness of the absolute distance between the historical index value sequence and the predicted index value sequence. The relevant calculation formula is as follows:

[0052]

[0053] Among them, MAPE is the absolute similarity eigenvalue, y i is the historical index value in the historical index value sequence, is the predicted index value in the predicted index value sequence, and n is the number of elements in the historical index value sequence. The larger the absolute similarity eigenvalue, the less close the historical index value sequence and the predicted index value sequence are, the greater the volatility of the index data, the smaller the required sensitivity, and the greater the tolerance for adjusting the prediction threshold sequence.

[0054] Finally, the sensitivity value can be obtained by performing weighted calculation on the volatility eigenvalue and the trend similarity eigenvalue. The relevant calculation formula is as follows:

[0055] Sensitivity=k3×(1-Volatility)+k4×(1-MAPE),

[0056] Among them, Sensitivity is the sensitivity value, Volatility is the volatility eigenvalue, MAPE is the absolute similarity eigenvalue, and k3 and k4 are weights. For example, k3 = 0.5, k4 = 0.5, which is not limited in this application.

[0057] Of course, other different statistical features can also be combined for eigenvalue weighted calculation, which is not limited in this application.

[0058] Step S103: When the sensitivity value is within the preset threshold range, extract multiple first index values that meet the preset warning rules from the historical index value sequence, and extract multiple first prediction thresholds with the same timestamps as the multiple first index values from the prediction threshold sequence, and perform fitting calculation on the multiple first index values and the multiple first prediction thresholds to obtain the first tolerance coefficient.

[0059] Among them, the sensitivity values can be binned through a preset threshold interval, so as to facilitate the selection of different calculation methods for the tolerance coefficient, and the tolerance degrees corresponding to different calculation methods of the tolerance coefficient are different. If the sensitivity value is within the preset threshold interval, for example, from 0.3 to 0.6, it can be regarded as the sensitivity value being relatively moderate. The tolerance degree for adjusting the prediction threshold sequence needs to be moderate. The first index value that meets the preset alarm rule can be extracted from the historical index value sequence. The preset alarm rule can be that among the consecutive multiple historical index values that exceed the prediction threshold limit in the historical index values, every preset number of historical index values is regarded as the first index value. For example, every 4 or 5 historical index values are regarded as the first index value. Compared with the number of historical index values that exceed the prediction threshold limit, the number of the first index values will be less. And, the prediction index value in the prediction index value sequence with the same timestamp as the first index value can be continuously combined for fitting calculation to obtain the tolerance coefficient. After adjusting the prediction threshold based on this tolerance coefficient, the original first index value that meets the preset alarm rule is no longer regarded as an alarm object, ensuring an appropriate alarm sensitivity and being applicable to index data with a moderate degree of fluctuation. The specific fitting calculation can adopt function mapping relationships such as a linear function relationship and a quadratic function relationship. For the linear function relationship fitting, the corresponding first tolerance coefficient can include a proportionality coefficient and an offset coefficient. For the quadratic function relationship fitting, the corresponding first tolerance coefficient can include a first proportionality coefficient, a second proportionality coefficient, and an offset coefficient. The specific function mapping relationship can be selected by the developer according to the fitting accuracy requirements of the actual application scenario, and this application does not make a limitation here. If the sensitivity value is greater than the upper threshold of the preset threshold interval, it can be regarded as the sensitivity value being relatively large. The tolerance degree for adjusting the prediction threshold sequence needs to be small. The target threshold sequence can be calculated by adjusting the prediction index values in the prediction index value sequence according to the standard deviation of a preset multiple, and the tolerance coefficient can be obtained by performing fitting calculation on the target threshold sequence and the prediction index value sequence. After adjusting the prediction threshold based on this tolerance coefficient, it can follow the numerical interval distribution of statistics, ensuring a high alarm sensitivity and being applicable to index data with a small degree of fluctuation. If the sensitivity value is less than the lower threshold of the preset threshold interval, it can be regarded as the sensitivity value being relatively small. The tolerance degree for adjusting the prediction threshold sequence needs to be large. The second index value that exceeds the prediction threshold limit can be extracted from the historical index value sequence, and the prediction index value in the prediction index value sequence with the same timestamp as the second index value can be combined for fitting calculation to obtain the tolerance coefficient. After adjusting the prediction threshold based on this tolerance coefficient, the second index value no longer exceeds the limit of the prediction threshold, achieving the purpose of reducing the alarm sensitivity and being applicable to index data with a large degree of fluctuation.

[0060] Step S104: Adjust the prediction threshold sequence based on the first tolerance coefficient to obtain a target prediction threshold sequence for abnormal warning of real-time index values.

[0061] Among them, after the fitting calculation in step S103, the first tolerance coefficient corresponding to the preset fitting relationship can be obtained. In one embodiment, the fitting calculation can be performed based on a linear function relationship. Correspondingly, the first tolerance coefficient includes a proportionality coefficient and an offset coefficient. Adjusting the prediction threshold sequence based on the first tolerance coefficient to obtain the target prediction threshold sequence can be specifically:

[0062] Integrating each prediction threshold in the prediction threshold sequence based on the proportionality coefficient, the offset coefficient, and the set linear function relationship to obtain the target prediction threshold sequence. The formula for adjusting the prediction threshold sequence is as follows:

[0063] y1 = a1x + b1,

[0064] where y1 is the index threshold, x is the prediction threshold in the prediction threshold sequence, a1 is the proportionality coefficient, and b1 is the offset coefficient.

[0065] In one embodiment, the fitting calculation can be performed based on a quadratic function relationship. Correspondingly, the first tolerance coefficient can include a first proportionality coefficient, a second proportionality coefficient, and an offset coefficient. Adjusting the prediction threshold sequence based on the first tolerance coefficient to obtain the target prediction threshold sequence can be specifically:

[0066] Integrating each prediction threshold in the prediction threshold sequence based on the first proportionality coefficient, the second proportionality coefficient, the offset coefficient, and the set quadratic function relationship to obtain the target prediction threshold sequence. The formula for adjusting the prediction threshold sequence is as follows:

[0067] y2 = a2x 2 + b2x + c1,

[0068] where y2 is the index threshold, x is the prediction threshold in the prediction threshold sequence, a2 is the first proportionality coefficient, b2 is the second proportionality coefficient, and c1 is the offset coefficient.

[0069] Thus, after adjusting the prediction threshold sequence, the target prediction threshold sequence can be obtained. For real-time index values that exceed the index threshold limit in the target prediction threshold sequence, abnormal warnings can be issued to trigger relevant background notifications to achieve the purpose of monitoring the index.

[0070] As described above, by obtaining the historical indicator value sequence, calculating the historical indicator value sequence according to the set sequence prediction algorithm, a predicted indicator value sequence and a predicted threshold sequence are obtained, where the historical indicator value sequence includes multiple historical indicator values, and the predicted indicator value sequence includes multiple predicted indicator values; performing eigenvalue weighted calculation with different statistical characteristics on multiple historical indicator values and multiple predicted indicator values to obtain a sensitivity value; when the sensitivity value is within a preset threshold interval, extracting multiple first indicator values that meet the preset alarm rule from the historical indicator value sequence, and extracting multiple first predicted thresholds with the same timestamps as the multiple first indicator values from the predicted threshold sequence, performing fitting calculation on the multiple first indicator values and the multiple first predicted thresholds to obtain a first tolerance coefficient; adjusting the predicted threshold sequence based on the first tolerance coefficient to obtain a target predicted threshold sequence for abnormal warning of real-time indicator values. In the above solution, by calculating the historical indicator value sequence according to the set sequence prediction algorithm, a predicted indicator value sequence and a predicted threshold sequence that conform to the change of historical indicator data can be initially determined. By performing eigenvalue weighted calculation with different statistical characteristics on the historical indicator values in the historical indicator value sequence and the predicted indicator values in the predicted indicator value sequence to obtain a sensitivity value, the correlation characteristics between the historical indicator value sequence and the predicted indicator value sequence can be effectively combined, and the sensitivity value for reasonably determining the tolerance degree of specifically adjusting the predicted threshold sequence can be determined. By judging the threshold interval where the sensitivity value is located, a tolerance coefficient matching it can be calculated correspondingly. Based on this first tolerance coefficient, the predicted threshold sequence can be adjusted to a target predicted threshold sequence, which can adapt to the trend changes of different indicator data for threshold adjustment, provide a reliable indicator threshold for indicator warning, and improve the warning accuracy.

[0071] Figure 2 FIG. is a flowchart of a method for determining an indicator threshold including a process of calculating an upper limit ratio coefficient and an upper limit offset coefficient provided by an embodiment of the present application. As Figure 2 shown, it includes the following steps:

[0072] Step S201, obtain a historical indicator value sequence, calculate the historical indicator value sequence according to the set sequence prediction algorithm, to obtain a predicted indicator value sequence and a predicted threshold sequence, where the historical indicator value sequence includes multiple historical indicator values, and the predicted indicator value sequence includes multiple predicted indicator values.

[0073] Step S202, perform eigenvalue weighted calculation with different statistical characteristics on multiple historical indicator values and multiple predicted indicator values to obtain a sensitivity value, where the sensitivity value is used to adjust the tolerance degree of the predicted threshold sequence.

[0074] Step S203: When the sensitivity value is within the preset threshold range, extract multiple first metric values that meet the preset alarm rule from the historical metric value sequence, and extract multiple first prediction thresholds with the same timestamps as the multiple first metric values from the prediction threshold sequence, where the multiple first prediction thresholds include multiple upper limit prediction thresholds.

[0075] Among them, the upper limit prediction threshold can be the maximum predicted value obtained by prediction based on the sequence prediction algorithm. If the historical metric value is greater than the upper limit prediction threshold with the same corresponding timestamp, it can be regarded as a metric value that may trigger an abnormal warning in the future. Correspondingly, the preset alarm rule can be that among the consecutive multiple historical metric values greater than the upper limit prediction threshold in the historical metric values, every preset number of historical metric values is regarded as a first metric value.

[0076] Step S204: Calculate the standard deviation of the multiple historical metric values.

[0077] Step S205: Perform a first ratio calculation on each first metric value and the upper limit prediction threshold with the same timestamp to obtain multiple first result values, and determine the maximum value among the multiple first result values as the upper limit ratio coefficient.

[0078] Among them, in this embodiment, a linear function relationship is used for fitting calculation, and the upper limit ratio coefficient and the upper limit offset coefficient need to be calculated correspondingly. The first ratio calculation can be to divide each first metric value by the upper limit prediction threshold with the same timestamp to obtain multiple first result values, or to divide the subtraction result of each first metric value and the standard deviation by the upper limit prediction threshold with the same timestamp to obtain multiple first result values. This application does not make a limitation here. The first result value can be used to represent the ratio value of the first metric value relative to the upper limit prediction threshold with the same timestamp. By determining the maximum value among the multiple first result values as the upper limit ratio coefficient, the product of the calculated upper limit ratio coefficient and each upper limit prediction threshold can cover the corresponding first metric value with the same timestamp. Optionally, a maximum ratio coefficient can be set. If the calculated upper limit ratio coefficient exceeds the maximum ratio coefficient, the upper limit ratio coefficient is adjusted to be consistent with the maximum ratio coefficient to prevent extreme values in the first metric values from affecting the accuracy of the upper limit ratio coefficient.

[0079] Step S206: Perform a first integration calculation on each first metric value and the upper limit prediction threshold with the same timestamp based on the standard deviation and the upper limit ratio coefficient to obtain multiple second result values, and determine the maximum value among the multiple second result values as the upper limit offset coefficient.

[0080] Among them, the first integration calculation can be to obtain multiple second result values by inverse calculation based on the linear function relationship. Specifically, the relevant formula for the first integration calculation is as follows:

[0081] b' = y + k'Δb - a'*x,

[0082] Wherein, b' is the second result value, y is the first index value, Δb is the standard deviation, a' is the upper limit proportionality coefficient, x is the upper limit prediction threshold with the same timestamp as the first index value, and k' is a preset multiple. For example, k' = 1, which is not limited in this application. The standard deviation Δb can be used to compensate for data fluctuations, making the calculated second result value more stable. The second result value can be used to represent the offset value of the first index value relative to the upper limit prediction threshold with the same timestamp. By determining the maximum value among multiple second result values as the upper limit offset coefficient, the product result of the upper limit proportionality coefficient and each upper limit prediction threshold, plus the upper limit offset coefficient, can cover the corresponding first index value with the same timestamp. Optionally, a maximum offset coefficient can be set. If the calculated upper limit offset coefficient exceeds the maximum offset coefficient, the upper limit offset coefficient is adjusted to be consistent with the maximum offset coefficient to prevent extreme values in the first index value from affecting the accuracy of the upper limit offset coefficient.

[0083] Step S207: Adjust the prediction threshold sequence based on the upper limit proportionality coefficient and the upper limit offset coefficient to obtain the target prediction threshold sequence for real-time anomaly warning of index values.

[0084] As described above, by calculating the first ratio and the first integration calculation to determine the upper limit proportionality coefficient and the upper limit offset coefficient, multiple first index values that meet the preset warning rules can be effectively matched, and the tolerance adjustment range of the preset threshold sequence can be determined, making the index threshold more in line with the actual fluctuations of the index data.

[0085] Figure 3 This is a flowchart of a method for determining an index threshold including a process of calculating a lower limit proportionality coefficient and a lower limit offset coefficient provided by an embodiment of the present application. As Figure 3 shown, it includes the following steps:

[0086] Step S301: Obtain a historical index value sequence, and calculate the historical index value sequence according to the set sequence prediction algorithm to obtain a predicted index value sequence and a prediction threshold sequence, where the historical index value sequence includes multiple historical index values, and the predicted index value sequence includes multiple predicted index values.

[0087] Step S302: Perform eigenvalue weighted calculation with different statistical characteristics on multiple historical index values and multiple predicted index values to obtain a sensitivity value, where the sensitivity value is used to adjust the tolerance degree of the prediction threshold sequence.

[0088] Step S303: When the sensitivity value is within the preset threshold range, extract multiple first metric values that meet the preset alarm rule from the historical metric value sequence, and extract multiple first prediction thresholds with the same timestamps as the multiple first metric values from the prediction threshold sequence, where the multiple first prediction thresholds include multiple lower prediction thresholds.

[0089] Among them, the lower prediction threshold can be the minimum predicted value obtained by prediction based on the sequence prediction algorithm. If the historical metric value is less than the lower prediction threshold with the same timestamp, it can be regarded as a metric value that may trigger an anomaly warning in the future. Correspondingly, the preset alarm rule can be that among multiple consecutive historical metric values in the historical metric values that are less than the lower prediction threshold, every preset number of historical metric values is regarded as a first metric value.

[0090] Step S304: Calculate the standard deviation of the multiple historical metric values.

[0091] Step S305: Perform a second ratio calculation on each first metric value and the lower prediction threshold with the same timestamp based on the standard deviation to obtain multiple third result values, and determine the minimum value among the multiple third result values as the lower ratio coefficient.

[0092] Among them, in this embodiment, a linear function relationship is used for fitting calculation, and the lower ratio coefficient and the lower offset coefficient need to be calculated correspondingly. The second ratio calculation can be to divide the subtraction result of each first metric value and the standard deviation by the lower prediction threshold with the same timestamp to obtain multiple third result values. The standard deviation can be used to compensate for data fluctuations, making the calculated third result values more stable. It can also be to divide the subtraction result of each first metric value and a preset multiple of the standard deviation by the lower prediction threshold with the same timestamp to obtain multiple third result values. This application does not make a limitation here. The third result value can be used to represent the proportional value of the first metric value relative to the lower prediction threshold with the same timestamp. By determining the minimum value among the multiple third result values as the lower ratio coefficient, the product of the calculated lower ratio coefficient and each lower prediction threshold can cover the corresponding first metric value with the same timestamp. Optionally, a minimum ratio coefficient can be set. If the calculated lower ratio coefficient is less than the minimum ratio coefficient, the lower ratio coefficient is adjusted to be consistent with the minimum ratio coefficient to prevent extreme values in the first metric values from affecting the accuracy of the lower ratio coefficient.

[0093] Step S306: Perform a second integration calculation on each first metric value and the lower prediction threshold with the same timestamp based on the lower ratio coefficient to obtain multiple fourth result values, and determine the minimum value among the multiple fourth result values as the lower offset coefficient.

[0094] Among them, the second integration calculation can be used to inversely obtain multiple fourth result values based on a linear function relationship. Specifically, the relevant formula for the second integration calculation is as follows:

[0095] b″ = y - a″ * x + c,

[0096] where b″ is the second result value, y is the first index value, a″ is the lower limit proportionality coefficient, x is the lower limit prediction threshold with the same timestamp as the first index value, and c is a compensation term. For example, c = 0, which is not limited in this application. The fourth result value can be used to represent the offset value of the first index value relative to the lower limit prediction threshold with the same timestamp. By determining the minimum value among multiple fourth result values as the lower limit offset coefficient, the product result of the lower limit proportionality coefficient and each lower limit prediction threshold, plus the lower limit offset coefficient, can cover the corresponding first index value with the same timestamp. Optionally, a minimum offset coefficient can be set. If the calculated lower limit offset coefficient is less than this minimum offset coefficient, the lower limit offset coefficient is adjusted to be the same as this minimum offset coefficient to prevent extreme values in the first index value from affecting the accuracy of the lower limit offset coefficient.

[0097] Step S307: Adjust the prediction threshold sequence based on the lower limit proportionality coefficient and the lower limit offset coefficient to obtain a target prediction threshold sequence for real-time anomaly warning of index values.

[0098] As described above, by determining the lower limit proportionality coefficient and the lower limit offset coefficient through the second ratio calculation and the second integration calculation, multiple first index values that meet the preset warning rules can be effectively matched, and the tolerance adjustment range of the preset threshold sequence can be determined, making the index threshold more in line with the actual fluctuations of the index data.

[0099] Figure 4 It is a flowchart of a method for determining an index threshold provided by an embodiment of the present application, which includes a process of adjusting a prediction threshold sequence based on a second tolerance coefficient. As Figure 4 shown, it includes the following steps:

[0100] Step S401: Obtain a historical index value sequence, and calculate the historical index value sequence according to the set sequence prediction algorithm to obtain a predicted index value sequence and a prediction threshold sequence, where the historical index value sequence includes multiple historical index values, and the predicted index value sequence includes multiple predicted index values.

[0101] Step S402: Perform eigenvalue weighted calculation with different statistical characteristics on multiple historical index values and multiple predicted index values to obtain a sensitivity value, where the sensitivity value is used to adjust the tolerance degree of the prediction threshold sequence.

[0102] Step S403: When the sensitivity value is within the preset threshold range, extract multiple first metric values that meet the preset alarm rules from the historical metric value sequence, and extract multiple first prediction thresholds with the same timestamps as the multiple first metric values from the prediction threshold sequence. Perform fitting calculations on the multiple first metric values and the multiple first prediction thresholds to obtain a first tolerance coefficient.

[0103] Step S404: Adjust the prediction threshold sequence based on the first tolerance coefficient to obtain a target prediction threshold sequence for real-time anomaly warning of metric values. The target prediction threshold sequence includes multiple metric thresholds.

[0104] Step S405: When the sensitivity value is greater than the upper threshold of the preset threshold range, calculate the standard deviation of the multiple historical metric values, and adjust the prediction metric values in the prediction metric value sequence by a preset multiple of the standard deviation to obtain a target threshold sequence.

[0105] Among them, if the sensitivity value is greater than the upper threshold of the preset threshold range, it can be considered that the sensitivity value is large, and the tolerance of the prediction threshold sequence needs to be adjusted less. Therefore, an appropriate preset multiple of the standard deviation can be selected based on the numerical interval distribution of statistics, and the target threshold can be calculated with the prediction metric value as the benchmark. For example, since the probability that the numerical value is distributed in the interval (μ - 3σ, μ + 3σ) is about 0.9974, where μ is the mean value and σ is the standard deviation, that is, about 99.74% of the data values will fall within three times the standard deviation of the mean value, which means that under the assumption of normal distribution, the probability that the data outside three times the standard deviation from the mean value appears is very small and can be regarded as an outlier. Therefore, the preset multiple can be set to 3, and it is set that the target threshold sequence includes a target upper threshold sequence and a target lower threshold sequence. The corresponding calculation formulas are as follows:

[0106]

[0107] Among them, is the target upper threshold in the target upper threshold sequence, is the target lower threshold in the target lower threshold sequence, y predict is the prediction metric value in the prediction metric value sequence, and σ is the standard deviation of the historical metric values in the historical metric value sequence.

[0108] Step S406: Perform fitting calculations on the target threshold sequence and the prediction metric value sequence to obtain a second tolerance coefficient.

[0109] Among them, the second tolerance coefficient may include an upper-limit tolerance coefficient and a lower-limit tolerance coefficient. The upper-limit tolerance coefficient may include an upper-limit ratio coefficient and an upper-limit offset coefficient, and the lower-limit tolerance coefficient may include a lower-limit ratio coefficient and a lower-limit offset coefficient. The fitting calculation may adopt the least squares method, the gradient descent method, etc., which are not limited in this application. Taking the use of the least squares method as an example, the relevant calculation formula is as follows:

[0110]

[0111] Among them, a up is the upper-limit ratio coefficient, b up is the upper-limit offset coefficient, is the target upper-limit threshold in the target upper-limit threshold sequence, is the predicted index value in the predicted index value sequence, is the average value of the target upper-limit thresholds in the target upper-limit threshold sequence, is the average value of the predicted index values in the predicted index value sequence, and n is the number of elements in the predicted index value sequence. Similarly, the lower-limit ratio coefficient and the lower-limit offset coefficient can be calculated, which will not be elaborated in this application.

[0112] Step S407: Adjust the predicted threshold sequence based on the second tolerance coefficient to obtain the target predicted threshold sequence.

[0113] In the above, when the sensitivity value is greater than the upper-limit threshold of the preset threshold interval, the calculation method of the tolerance coefficient with a smaller tolerance degree can be selected following the numerical interval distribution of statistics, which is beneficial to ensuring a higher alarm sensitivity and is applicable to application scenarios with a smaller degree of data fluctuation.

[0114] Figure 5 This is a flowchart of a method for determining an index threshold provided by an embodiment of the present application, which includes a process of adjusting a predicted threshold sequence based on a third tolerance coefficient. As Figure 5 shown, it includes the following steps:

[0115] Step S501: Obtain a historical index value sequence, and calculate the predicted index value sequence and the predicted threshold sequence according to the set sequence prediction algorithm. Among them, the historical index value sequence includes multiple historical index values, and the predicted index value sequence includes multiple predicted index values.

[0116] Step S502: Perform eigenvalue weighted calculation with different statistical characteristics on multiple historical index values and multiple predicted index values to obtain a sensitivity value, where the sensitivity value is used to adjust the tolerance degree of the predicted threshold sequence.

[0117] Step S503: When the sensitivity value is within the preset threshold range, extract multiple first metric values that meet the preset alarm rules from the historical metric value sequence, and extract multiple first predicted thresholds with the same timestamps as the multiple first metric values from the predicted threshold sequence. Perform fitting calculations on the multiple first metric values and the multiple first predicted thresholds to obtain the first tolerance coefficient.

[0118] Step S504: Adjust the predicted threshold sequence based on the first tolerance coefficient to obtain the target predicted threshold sequence for anomaly warning of real-time metric values.

[0119] Step S505: When the sensitivity value is less than the lower threshold of the preset threshold range, extract multiple second metric values that meet the preset anomaly rules from the historical metric value sequence, and extract multiple second predicted thresholds with the same timestamps as the multiple second metric values from the predicted threshold sequence. Perform fitting calculations on the multiple second metric values and the multiple second predicted thresholds to obtain the third tolerance coefficient.

[0120] Among them, if the sensitivity value is less than the lower threshold of the preset threshold range, it can be regarded as a small sensitivity value, and the tolerance degree of the predicted threshold sequence needs to be adjusted more. Multiple second metric values that meet the preset anomaly rules can be extracted from the historical metric value sequence. The preset anomaly rule can be to regard the historical metric values that exceed the preset threshold limit in the historical metric values as the second metric values. The specific fitting calculation process can refer to the relevant description of the first tolerance coefficient in the foregoing embodiments, and will not be elaborated herein.

[0121] Step S506: Adjust the predicted threshold sequence based on the third tolerance coefficient to obtain the target predicted threshold sequence.

[0122] Above, adjusting the predicted threshold sequence based on the third tolerance coefficient can make the original second metric values that meet the preset anomaly rules no longer regarded as anomaly objects, ensuring a small alarm sensitivity and being applicable to application scenarios with large data fluctuation degrees.

[0123] Figure 6 This is a flowchart of a method for determining metric thresholds including a process of calculating sensitivity values provided by an embodiment of the present application. As Figure 6 shown, it includes the following steps:

[0124] Step S601: Obtain the historical metric value sequence, and perform calculations on the historical metric value sequence according to the set sequence prediction algorithm to obtain the predicted metric value sequence and the predicted threshold sequence, where the historical metric value sequence includes multiple historical metric values, and the predicted metric value sequence includes multiple predicted metric values.

[0125] Step S602: Calculate multiple index feature values corresponding to different statistical features based on multiple historical index values and multiple predicted index values, where the index feature values are used to represent the degree of fluctuation of the historical index values or the degree of similarity between the historical index value sequence and the predicted index value sequence.

[0126] Among them, the selected statistical features can be used to represent the degree of fluctuation of the historical index values or the degree of similarity between the historical index value sequence and the predicted index value sequence. This application does not make any limitations here. Each statistical feature can be correspondingly calculated to obtain an index feature value. For example, for the fluctuation feature, the standard deviation or coefficient of variation can be correspondingly calculated as the fluctuation feature value. Another example is that for the similarity feature, the coefficient of determination or mean absolute percentage error can be correspondingly calculated as the trend similarity feature value. Of course, there are other characteristic calculation methods for measuring the trend change and fluctuation degree of index data. This application does not make any limitations here.

[0127] Step S603: Perform weighted calculation on multiple index feature values to obtain a sensitivity value.

[0128] Step S604: When the sensitivity value is within the preset threshold range, extract multiple first index values that meet the preset alarm rules from the historical index value sequence, and extract multiple first predicted thresholds with the same timestamps as the multiple first index values from the predicted threshold sequence. Perform fitting calculation on the multiple first index values and the multiple first predicted thresholds to obtain a first tolerance coefficient.

[0129] Step S605: Adjust the predicted threshold sequence based on the first tolerance coefficient to obtain a target predicted threshold sequence for real-time index value anomaly warning.

[0130] As described above, by calculating multiple index feature values corresponding to different statistical features, the trend change and fluctuation degree of index data can be effectively measured. And by performing weighted calculation on multiple index feature values to obtain a sensitivity value, the characteristic information of different dimensions can be reasonably integrated to determine an appropriate sensitivity value, which adapts to the actual trend change and fluctuation degree of the index and determines the tolerance degree of the predicted threshold sequence.

[0131] Figure 7 This is a flowchart of an index threshold determination method provided by an embodiment of this application, which includes the process of calculating index feature values. As Figure 7 shown, it includes the following steps:

[0132] Step S701: Obtain a historical index value sequence, and calculate the historical index value sequence according to the set sequence prediction algorithm to obtain a predicted index value sequence and a predicted threshold sequence, where the historical index value sequence includes multiple historical index values, and the predicted index value sequence includes multiple predicted index values.

[0133] Step S702: Calculate the standard deviation and mean of multiple historical indicator values, divide the standard deviation by the mean to obtain a volatility eigenvalue, perform a linear correlation calculation on the multiple historical indicator values and multiple predicted indicator values to obtain a trend similarity eigenvalue, and perform a relative error calculation on the multiple historical indicator values and multiple predicted indicator values to obtain an absolute similarity eigenvalue.

[0134] Among them, the linear correlation calculation can be to calculate the coefficient of determination, Pearson correlation coefficient, etc. as the trend similarity eigenvalue, and the relative error calculation can be to calculate the mean absolute percentage error, mean square error, etc. as the absolute similarity eigenvalue. Here, the present application does not make any limitations.

[0135] Step S703: Perform a weighted calculation on the volatility eigenvalue, trend similarity eigenvalue, and absolute similarity eigenvalue to obtain a sensitivity value.

[0136] Among them, the specific formula for the weighted calculation is as follows:

[0137] Sensitivity = k′1×(1 - Volatility) + k2′×Similarity + k3′×(1 - MAPE),

[0138] Among them, Sensitivity is the sensitivity value, Volatility is the volatility eigenvalue, Similarity is the trend similarity eigenvalue, MAPE is the absolute similarity eigenvalue, and k′1, k′2, and k′3 are weights. For example, k′1 = 1 / 3, k′2 = 1 / 3, k′3 = 1 / 3. Here, the present application does not make any limitations.

[0139] Step S704: When the sensitivity value is within a preset threshold range, extract multiple first indicator values that meet the preset warning rules from the historical indicator value sequence, and extract multiple first predicted thresholds with the same timestamps as the multiple first indicator values from the predicted threshold sequence, and perform a fitting calculation on the multiple first indicator values and multiple first predicted thresholds to obtain a first tolerance coefficient.

[0140] Step S705: Adjust the predicted threshold sequence based on the first tolerance coefficient to obtain a target predicted threshold sequence for abnormal warning of real-time indicator values.

[0141] As described above, by performing a weighted calculation on the volatility eigenvalue, trend similarity eigenvalue, and absolute similarity eigenvalue to obtain a sensitivity value, it is possible to effectively select representative statistical features for calculating the sensitivity value, ensure the reliability of the sensitivity in measuring the tolerance degree for adjusting the predicted threshold sequence, and contribute to improving the warning accuracy.

[0142] Figure 8The structural block diagram of an indicator threshold determination device provided by an embodiment of the present application. The device is configured to execute the indicator threshold determination method provided by the above embodiment, and has corresponding functional modules and beneficial effects for executing the method. As Figure 8 shown, the device includes:

[0143] A prediction sequence determination module 101, configured to obtain a historical indicator value sequence, calculate the historical indicator value sequence according to a set sequence prediction algorithm, and obtain a predicted indicator value sequence and a predicted threshold sequence. The historical indicator value sequence includes multiple historical indicator values, and the predicted indicator value sequence includes multiple predicted indicator values;

[0144] A sensitivity value determination module 102, configured to perform eigenvalue weighted calculation of different statistical characteristics on multiple historical indicator values and multiple predicted indicator values to obtain a sensitivity value, and the sensitivity value is used to adjust the tolerance degree of the predicted threshold sequence;

[0145] A tolerance coefficient determination module 103, configured to, when the sensitivity value is within a preset threshold interval, extract multiple first indicator values that meet the preset warning rule from the historical indicator value sequence, and extract multiple first predicted thresholds with the same timestamp as the multiple first indicator values from the predicted threshold sequence, and perform fitting calculation on the multiple first indicator values and the multiple first predicted thresholds to obtain a first tolerance coefficient;

[0146] A first indicator threshold determination module 104, configured to adjust the predicted threshold sequence based on the first tolerance coefficient to obtain a target predicted threshold sequence for performing abnormal warning of real-time indicator values. The target predicted threshold sequence includes multiple indicator thresholds.

[0147] As described above, by obtaining a historical indicator value sequence, calculating the historical indicator value sequence according to a set sequence prediction algorithm, a predicted indicator value sequence and a predicted threshold sequence are obtained, where the historical indicator value sequence includes multiple historical indicator values, and the predicted indicator value sequence includes multiple predicted indicator values; eigenvalue weighted calculations of different statistical characteristics are performed on the multiple historical indicator values and the multiple predicted indicator values to obtain a sensitivity value; in the case where the sensitivity value is within a preset threshold interval, multiple first indicator values that meet the preset alarm rule are extracted from the historical indicator value sequence, and multiple first predicted thresholds with the same timestamps as the multiple first indicator values are extracted from the predicted threshold sequence, and a first tolerance coefficient is obtained by performing fitting calculations on the multiple first indicator values and the multiple first predicted thresholds; the predicted threshold sequence is adjusted based on the first tolerance coefficient to obtain a target predicted threshold sequence for abnormal warning of real-time indicator values. In the above solution, by calculating the historical indicator value sequence according to the set sequence prediction algorithm, a predicted indicator value sequence and a predicted threshold sequence that conform to the changes in historical indicator data can be initially determined. By obtaining the sensitivity value through eigenvalue weighted calculations of different statistical characteristics based on the historical indicator values in the historical indicator value sequence and the predicted indicator values in the predicted indicator value sequence, the correlation characteristics between the historical indicator value sequence and the predicted indicator value sequence can be effectively combined, and the sensitivity value for reasonably determining the tolerance degree of specifically adjusting the predicted threshold sequence can be determined. By judging the threshold interval where the sensitivity value is located, a tolerance coefficient matching it can be calculated accordingly. Based on this first tolerance coefficient, the predicted threshold sequence can be adjusted to a target predicted threshold sequence, which can adapt to the trend changes of different indicator data for threshold adjustment, provide a reliable indicator threshold for indicator warning, and improve the warning accuracy.

[0148] In a possible embodiment, the multiple first predicted thresholds include multiple upper limit predicted thresholds, the first tolerance coefficient includes an upper limit tolerance coefficient, the upper limit tolerance coefficient includes an upper limit ratio coefficient and an upper limit offset coefficient, and the tolerance coefficient determination module 103 is further configured to:

[0149] Calculate the standard deviation of the multiple historical indicator values;

[0150] Perform a first ratio calculation on each first indicator value and the upper limit predicted threshold with the same timestamp to obtain multiple first result values, and determine the maximum value among the multiple first result values as the upper limit ratio coefficient;

[0151] Based on the standard deviation and the upper limit ratio coefficient, perform a first integration calculation on each first indicator value and the upper limit predicted threshold with the same timestamp to obtain multiple second result values, and determine the maximum value among the multiple second result values as the upper limit offset coefficient.

[0152] In a possible embodiment, the multiple first prediction thresholds include multiple lower prediction thresholds, the first tolerance coefficient includes a lower tolerance coefficient, the lower tolerance coefficient includes a lower ratio coefficient and a lower offset coefficient, and the tolerance coefficient determination module 103 is further configured to:

[0153] Calculate the standard deviation of the multiple historical metric values;

[0154] Based on the standard deviation, perform a second ratio calculation on each first metric value and the lower prediction threshold with the same timestamp to obtain multiple third result values, and determine the minimum value among the multiple third result values as the lower ratio coefficient;

[0155] Based on the lower ratio coefficient, perform a second integration calculation on each first metric value and the lower prediction threshold with the same timestamp to obtain multiple fourth result values, and determine the minimum value among the multiple fourth result values as the lower offset coefficient.

[0156] In a possible embodiment, it further includes a second metric threshold determination module, configured to:

[0157] When the sensitivity value is greater than the upper threshold of the preset threshold interval, calculate the standard deviation of the multiple historical metric values, and adjust the prediction metric values in the prediction metric value sequence according to the standard deviation of the preset multiple to obtain a target threshold sequence;

[0158] Perform a fitting calculation on the target threshold sequence and the prediction metric value sequence to obtain a second tolerance coefficient;

[0159] Based on the second tolerance coefficient, adjust the prediction threshold sequence to obtain a target prediction threshold sequence.

[0160] In a possible embodiment, it further includes a third metric threshold determination module, configured to:

[0161] When the sensitivity value is less than the lower threshold of the preset threshold interval, extract multiple second metric values that meet the preset anomaly rule from the historical metric value sequence, and extract multiple second prediction thresholds with the same timestamp as the multiple second metric values from the prediction threshold sequence, and perform a fitting calculation on the multiple second metric values and the multiple second prediction thresholds to obtain a third tolerance coefficient;

[0162] Based on the third tolerance coefficient, adjust the prediction threshold sequence to obtain a target prediction threshold sequence.

[0163] In a possible embodiment, the sensitivity value determination module 102 is further configured to:

[0164] Multiple index feature values corresponding to different statistical features are calculated based on multiple historical index values and multiple predicted index values. The index feature values are used to represent the degree of fluctuation of the historical index values or the degree of similarity between the historical index value sequence and the predicted index value sequence;

[0165] The multiple index feature values are weighted and calculated to obtain a sensitivity value.

[0166] In a possible embodiment, the sensitivity value determination module 102 is further configured to:

[0167] Calculate the standard deviation and average value of the multiple historical index values, and divide the standard deviation by the average value to obtain a fluctuation feature value;

[0168] Perform a linear correlation calculation on the multiple historical index values and the multiple predicted index values to obtain a trend similarity feature value;

[0169] Perform a relative error calculation on the multiple historical index values and the multiple predicted index values to obtain an absolute similarity feature value;

[0170] The fluctuation feature value, the trend similarity feature value, and the absolute similarity feature value are weighted and calculated to obtain a sensitivity value.

[0171] In a possible embodiment, the first tolerance coefficient includes a proportional coefficient and an offset coefficient. The first index threshold determination module 104 is further configured to:

[0172] Integrate each prediction threshold in the prediction threshold sequence based on the proportional coefficient, the offset coefficient, and the set linear function relationship to obtain a target prediction threshold sequence.

[0173] Figure 9 The structural schematic diagram of an index threshold determination device provided by the embodiments of the present application is shown in Figure 9 As shown, the device includes a processor 201, a memory 202, an input device 203, and an output device 204; the number of processors 201 in the device can be one or more, Figure 9 Taking one processor 201 as an example; the processor 201, the memory 202, the input device 203, and the output device 204 in the device can be connected through a bus or other means, Figure 9Take the bus connection as an example. The memory 202, as a computer-readable storage medium, can be configured to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the method for determining the index threshold in the embodiments of the present application. The processor 201 executes various functional applications and data processing of the device by running the software programs, instructions, and modules stored in the memory 202, that is, implements the above-mentioned method for determining the index threshold. The input device 203 can be configured to receive input digital or character information, and generate key signal inputs related to the user settings and function controls of the device. The output device 204 can include display devices such as a display screen.

[0174] The embodiments of the present application also provide a non-volatile storage medium containing computer-executable instructions. When the computer-executable instructions are executed by a computer processor, they are configured to execute a method for determining an index threshold described in the above embodiments, including: obtaining a historical index value sequence, calculating the historical index value sequence according to a set sequence prediction algorithm to obtain a predicted index value sequence and a predicted threshold sequence, where the historical index value sequence includes multiple historical index values, and the predicted index value sequence includes multiple predicted index values; performing eigenvalue weighted calculations with different statistical characteristics on the multiple historical index values and the multiple predicted index values to obtain a sensitivity value, where the sensitivity value is used to adjust the tolerance level of the predicted threshold sequence; in the case where the sensitivity value is within a preset threshold interval, extracting multiple first index values that meet the preset warning rule from the historical index value sequence, and extracting multiple first predicted thresholds with the same timestamps as the multiple first index values from the predicted threshold sequence, and performing fitting calculations on the multiple first index values and the multiple first predicted thresholds to obtain a first tolerance coefficient; adjusting the predicted threshold sequence based on the first tolerance coefficient to obtain a target predicted threshold sequence for real-time abnormal warning of index values, and the target predicted threshold sequence includes multiple index thresholds.

[0175] It should be noted that in the embodiments of the above-mentioned index threshold determination device, the included units and modules are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual distinction and are not configured to limit the protection scope of the embodiments of the present application.

[0176] In some possible implementation manners, various aspects of the method provided by the present application can also be implemented in the form of a program product, which includes program code. When the program product runs on a computer device, the program code is configured to cause the computer device to execute the steps in the method according to various exemplary embodiments of the present application described in this specification. For example, the computer device can execute the method for determining the index threshold recorded in the embodiments of the present application. The program product can be implemented by any combination of one or more readable media.

Claims

1. A method for determining an index threshold, characterized in that Including: Obtain a historical index value sequence, calculate the historical index value sequence according to a set sequence prediction algorithm to obtain a predicted index value sequence and a predicted threshold sequence, the historical index value sequence includes a plurality of historical index values, and the predicted index value sequence includes a plurality of predicted index values; Perform eigenvalue weighted calculation of different statistical characteristics on the plurality of historical index values and the plurality of predicted index values to obtain a sensitivity value, and the sensitivity value is used to adjust the tolerance degree of the predicted threshold sequence; When the sensitivity value is within a preset threshold interval, extract a plurality of first index values that meet the preset alarm rule from the historical index value sequence, and extract a plurality of first predicted thresholds with the same timestamp as the plurality of first index values from the predicted threshold sequence, and perform fitting calculation on the plurality of first index values and the plurality of first predicted thresholds to obtain a first tolerance coefficient; Adjust the predicted threshold sequence based on the first tolerance coefficient to obtain a target predicted threshold sequence for abnormal warning of real-time index values, and the target predicted threshold sequence includes a plurality of index thresholds.

2. The method for determining the index threshold according to claim 1, wherein The plurality of first predicted thresholds include a plurality of upper limit predicted thresholds, the first tolerance coefficient includes an upper limit tolerance coefficient, the upper limit tolerance coefficient includes an upper limit ratio coefficient and an upper limit offset coefficient, and the performing fitting calculation on the plurality of first index values and the plurality of first predicted thresholds to obtain a first tolerance coefficient includes: Calculate the standard deviation of the plurality of historical index values; Perform a first ratio calculation on each of the first index values and the upper limit predicted threshold with the same timestamp to obtain a plurality of first result values, and determine the maximum value among the plurality of first result values as the upper limit ratio coefficient; Based on the standard deviation and the upper limit ratio coefficient, perform a first integration calculation on each of the first index values and the upper limit predicted threshold with the same timestamp to obtain a plurality of second result values, and determine the maximum value among the plurality of second result values as the upper limit offset coefficient.

3. The method for determining the index threshold according to claim 1, wherein, The plurality of first predicted thresholds include a plurality of lower limit predicted thresholds, the first tolerance coefficient includes a lower limit tolerance coefficient, the lower limit tolerance coefficient includes a lower limit ratio coefficient and a lower limit offset coefficient, and the performing fitting calculation on the plurality of first index values and the plurality of first predicted thresholds to obtain a first tolerance coefficient includes: Calculate the standard deviation of the plurality of historical index values; Perform a second ratio calculation on each of the first index values and the lower limit predicted threshold with the same timestamp based on the standard deviation to obtain a plurality of third result values, and determine the minimum value among the plurality of third result values as the lower limit ratio coefficient; Based on the lower limit ratio coefficient, perform a second integration calculation on each of the first index values and the lower limit predicted threshold with the same timestamp to obtain a plurality of fourth result values, and determine the minimum value among the plurality of fourth result values as the lower limit offset coefficient.

4. The method for determining the index threshold according to claim 1, characterized in that, After performing eigenvalue weighted calculation of different statistical characteristics on the plurality of historical index values and the plurality of predicted index values to obtain a sensitivity value, it further includes: When the sensitivity value is greater than the upper threshold of the preset threshold range, calculate the standard deviation of the multiple historical index values, and adjust the prediction index values in the prediction index value sequence according to the standard deviation of a preset multiple to obtain a target threshold sequence; Perform a fitting calculation on the target threshold sequence and the prediction index value sequence to obtain a second tolerance coefficient; Adjust the prediction threshold sequence based on the second tolerance coefficient to obtain a target prediction threshold sequence.

5. The method for determining the index threshold according to claim 1, characterized in that, After calculating the sensitivity value by performing eigenvalue weighted calculations with different statistical characteristics on the multiple historical index values and the multiple prediction index values, it further includes: When the sensitivity value is less than the lower threshold of the preset threshold range, extract multiple second index values that meet the preset anomaly rules from the historical index value sequence, and extract multiple second prediction thresholds with the same timestamps as the multiple second index values from the prediction threshold sequence, and perform a fitting calculation on the multiple second index values and the multiple second prediction thresholds to obtain a third tolerance coefficient; Adjust the prediction threshold sequence based on the third tolerance coefficient to obtain a target prediction threshold sequence.

6. The method for determining the index threshold according to claim 1, wherein The calculation of the sensitivity value by performing eigenvalue weighted calculations with different statistical characteristics on the multiple historical index values and the multiple prediction index values includes: Calculate multiple index eigenvalues corresponding to different statistical characteristics based on the multiple historical index values and the multiple prediction index values, where the index eigenvalues are used to represent the degree of fluctuation of the historical index values or the degree of similarity between the historical index value sequence and the prediction index value sequence; Perform a weighted calculation on the multiple index eigenvalues to obtain the sensitivity value.

7. The method for determining an index threshold according to claim 6, wherein The multiple index eigenvalues include a fluctuation eigenvalue, a trend similarity eigenvalue, and an absolute similarity eigenvalue. The calculation of multiple index eigenvalues corresponding to different statistical characteristics based on the multiple historical index values and the multiple prediction index values includes: Calculate the standard deviation and average value of the multiple historical index values, and divide the standard deviation by the average value to obtain the fluctuation eigenvalue; Perform a linear correlation calculation on the multiple historical index values and the multiple prediction index values to obtain the trend similarity eigenvalue; Perform a relative error calculation on the multiple historical index values and the multiple prediction index values to obtain the absolute similarity eigenvalue; Correspondingly, the performance of a weighted calculation on the multiple index eigenvalues to obtain the sensitivity value includes: Perform a weighted calculation on the fluctuation eigenvalue, the trend similarity eigenvalue, and the absolute similarity eigenvalue to obtain the sensitivity value.

8. The method for determining an index threshold according to claim 1, wherein The first tolerance coefficient includes a proportional coefficient and an offset coefficient. The adjustment of the prediction threshold sequence based on the first tolerance coefficient to obtain a target prediction threshold sequence includes: Integrate each prediction threshold in the prediction threshold sequence based on the proportional coefficient, the offset coefficient, and the set linear function relationship to obtain a target prediction threshold sequence.

9. An index threshold determination device, characterized in that, It includes: A prediction sequence determination module, configured to obtain a historical index value sequence, calculate the historical index value sequence according to a set sequence prediction algorithm, and obtain a predicted index value sequence and a prediction threshold sequence, where the historical index value sequence includes multiple historical index values, and the predicted index value sequence includes multiple predicted index values; A sensitivity value determination module, configured to perform eigenvalue weighted calculation of different statistical characteristics on the multiple historical index values and the multiple predicted index values to obtain a sensitivity value, where the sensitivity value is used to adjust the tolerance degree of the prediction threshold sequence; A tolerance coefficient determination module, configured to, when the sensitivity value is within a preset threshold interval, extract multiple first index values that meet a preset alarm rule from the historical index value sequence, and extract multiple first prediction thresholds with the same timestamps as the multiple first index values from the prediction threshold sequence, and perform fitting calculation on the multiple first index values and the multiple first prediction thresholds to obtain a first tolerance coefficient; A first index threshold determination module, configured to adjust the prediction threshold sequence based on the first tolerance coefficient to obtain a target prediction threshold sequence for abnormal warning of real-time index values, where the target prediction threshold sequence includes multiple index thresholds.

10. An index threshold determination device, the device comprising: One or more processors; A storage device, configured to store one or more programs, and when the one or more programs are executed by the one or more processors, enable the one or more processors to implement the index threshold determination method according to any one of claims 1-8.

11. A non-volatile storage medium storing computer-executable instructions, where the computer-executable instructions are configured to execute the index threshold determination method according to any one of claims 1-8 when executed by a computer processor.

12. A computer program product, comprising a computer program, characterized in that, The computer program is stored in a computer-readable storage medium, and at least one processor of the device reads and executes the computer program from the computer-readable storage medium, enabling the device to execute the index threshold determination method according to any one of claims 1-8.

Images