Call bill monitoring method and system, computer equipment and readable storage medium

Abstract

The invention discloses a call ticket monitoring method and system, computer equipment and a readable storage medium, and relates to the technical field of data processing. The call ticket monitoring method comprises the following steps: acquiring input data according to historical call ticket data; and according to the input data and the prediction model, obtaining prediction data of the call ticket on that day. And according to the membership function, determining whether the predicted data of the call ticket on the day is abnormal, and giving an alarm when the predicted data of the call ticket on the day is abnormal. According to the call ticket monitoring method, the grey prediction algorithm and the membership function are combined, the call ticket monitoring efficiency and accuracy are improved, and early warning can be carried out in time.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a call bill monitoring method, a call bill monitoring system, a computer device and a readable storage medium. Background Art

[0002] The current call bill monitoring system manually screens data sets and discovers business anomalies by performing month-on-month analysis, year-on-year analysis, mean change analysis, mean and standard deviation analysis of data in the same time window, local data fluctuation analysis, and periodic analysis.

[0003] By analyzing a large amount of business indicator data in the database by daily, weekly, and monthly dimensions, such as averaging, weighted time series averaging, moving average, and special case statistics, we can predict future call bill trends and system resource utilization trends, providing a reasonable basis for system capacity management.

[0004] However, the existing call bill monitoring system can only manually screen the data set to observe data anomalies by calculating year-on-year and month-on-month changes, which is inefficient. Most of the existing call bill monitoring systems draw conclusions based on the average method or related simple numerical calculation methods. The data analysis results are not theoretical enough, and the data feature analysis is not clear enough. The prediction data obtained by processing the business indicator sequence using the average method, special case statistics and other methods has poor accuracy. When the data regularity is not strong, the average method is not effective and the accuracy is low. In addition, the current call bill monitoring system cannot give the next call bill change trend, and cannot warn of call bill anomalies in advance without human intervention, and the call bill warning delay is high. Summary of the invention

[0005] The technical problem to be solved by the present invention is that the call bill monitoring method in the prior art has the problems of low efficiency, low accuracy and high call bill warning delay.

[0006] In view of the above-mentioned deficiencies of the prior art, the following solutions are provided:

[0007] In a first aspect, the present invention provides a method for monitoring a call bill, comprising: obtaining input data according to historical call bill data; obtaining predicted data of the call bill of the day according to the input data and a prediction model; the prediction model is obtained according to a grey prediction algorithm; and determining whether the predicted data of the call bill of the day is abnormal according to a membership function, and issuing an alarm if the predicted data of the call bill of the day is abnormal.

[0008] Optionally, acquiring input data according to historical call bill data includes: determining multiple reference dates according to the date of the day, and acquiring actual call bill data or predicted call bill data of each reference date in the multiple reference dates. The historical call bill data includes actual call bill data or predicted call bill data of each reference date in the multiple reference dates. And, acquiring input data according to the actual call bill data or predicted call bill data of each reference date in the multiple reference dates.

[0009] Optionally, multiple reference dates are determined according to the date of the current day, including: in response to the current day being the first day of the current month, the reference date includes the first day of the four months before the current month. In response to the current day being the second day of the current month, the reference date includes the second day of the three months before the current month and the first day of the current month. In response to the current day being the third day of the current month, the reference date includes the third day of the two months before the current month and the first two days of the current month. In response to the current day being the fourth day of the current month, the reference date includes the fourth day of the month before the current month and the first three days of the current month. And, in response to the current day being a date other than the first four days of the current month, the reference date includes the first four days of the current day.

[0010] Optionally, the input data is obtained based on the actual bill data or predicted bill data of each reference date among the multiple reference dates, including: in response to the absence of abnormal data in the actual bill data of each reference date among the multiple reference dates, the actual bill data of each reference date among the multiple reference dates is used as input data. In response to the presence of normal data and abnormal data in the actual bill data of each reference date among the multiple reference dates, a reference date corresponding to the normal data and a reference date corresponding to the abnormal data are determined, and the actual bill data of the reference date corresponding to the normal data and the predicted bill data of the reference date corresponding to the abnormal data are used as input data. And, in response to the absence of normal data in the actual bill data of each reference date among the multiple reference dates, the predicted bill data of each reference date among the multiple reference dates is used as input data.

[0011] Optionally, obtaining the forecast data of the call list of the day according to the input data and the forecast model includes: determining the original sequence X according to the input data (0) , where X (0) =(x (0) (1),x (0) (2),...x (0) (n)), n is the number of data in the input data, x (0) (1), x (0) (2), …, x (0) (n) are input data, and n∈N, N is a set of natural numbers. According to the original sequence X (0) Get the r-order cumulative generation sequence X (r) , r-order adjacent mean generation sequence Z(r) and r-order cumulative reduction sequence X (-r) , r∈R + , R + is a set of positive real numbers. (r) =(x (r) (1), x (r) (2), …, x (r) (k), …, x (r) (n)), and Z (r) =(z (r) (1), z (r) (2), …, z (r) (k), …, z (r) (n)), and z (r) (k) = 0.5 × (x (r) (k)+x (r) (k-1)), k=2,3…,n. X (-r) =(x (-r) (1), x (-r) (2), …, x (-r) (k), …, x (-r) (n)), and Get the prediction model. The prediction model is x (r-1) (k)+az (r) (k) = b. The model parameters a and b of the prediction model are estimated using the least squares method. According to the whitened differential equation of the prediction model Get the time response function of the prediction model in, According to the time response function of the prediction model Get the time response series of the prediction model in, And, perform negative r-order cumulative reduction to 0-order to obtain the restored value of the prediction model in, The forecast data of the day's call list is the restored value of the forecast model

[0012] Optionally, obtaining a prediction model includes: determining multiple first values. Each of the multiple first values ​​is a positive real number. Using each of the multiple first values ​​as a quasi-fractional order r, for each quasi-fractional order r, determine a quasi-prediction model. Obtaining predicted call record data for the month of the current day one year before the year of the current day based on the actual call record data of the year before the year of the current day and each quasi-prediction model. Comparing the actual call record data for the month of the current day one year before the year of the current day with the predicted call record data for the month of the current day one year before the year of the current day corresponding to each quasi-prediction model. And, determining a prediction model from multiple quasi-prediction models based on the comparison result.

[0013] Optionally, determining a prediction model from a plurality of quasi-prediction models based on the comparison results includes: comparing the difference between the predicted call bill data of the month of the day in the year before the year in which the day corresponds to each quasi-prediction model and the actual call bill data of the month of the day in the year before the year in which the day corresponds to, with a first preset threshold. Obtaining the number of days on which the difference between the predicted call bill data of the month of the day in the year before the year in which the day corresponds to each quasi-prediction model and the actual call bill data of the month of the day in the year before the year in which the day corresponds to, is greater than the first preset threshold. And determining the minimum value of the number of days on which the difference between the daily call bill data is greater than the first preset threshold, and taking the quasi-prediction model corresponding to the minimum value of the number of days on which the difference between the daily call bill data is greater than the first preset threshold as the prediction model.

[0014] Optionally, the expression of the membership function A(x) is: Among them, c, d, f and h are the parameters of the membership function.

[0015] In a second aspect, the present invention provides a call bill monitoring system, comprising an input data acquisition module, a prediction data acquisition module and an abnormal alarm module. The input data acquisition module is configured to acquire input data based on historical call bill data. The prediction data acquisition module is configured to acquire prediction data of the day's call bill based on input data and a prediction model. The prediction model is obtained based on a gray prediction algorithm. The abnormal alarm module is configured to determine whether the prediction data of the day's call bill is abnormal based on a membership function, and to issue an alarm if the prediction data of the day's call bill is abnormal.

[0016] In a third aspect, the present invention provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor runs the computer program stored in the memory, the processor executes the above-mentioned call record monitoring method.

[0017] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the processor executes the above-mentioned call record monitoring method.

[0018] The beneficial effects of a call bill monitoring method, system, computer device and readable storage medium provided by the embodiments of the present invention include: in the call bill monitoring method, by analyzing the call bill data set, the call bill data set is defined as a fuzzy set, and then the corresponding membership function is determined, the membership degree is obtained, and the final evaluation standard is obtained. According to the evaluation standard, the abnormality of the call bill data is timely judged, and the result of the membership function is used as a reference for the subsequent call bill prediction. The adaptive gray prediction algorithm is used to obtain a more accurate prediction result, capture real-time regularity, and timely discover the abnormality of the call bill data, and then synchronously warn. The gray prediction method only requires basic mathematical operations, usually only requires at least 4 call bill data, and there is no need to make too many strict assumptions on the sample data to establish a gray prediction model. By observing the call bill data set, the corresponding rules are extracted and the call bill data of the day are predicted to achieve the purpose of accurate call bill monitoring, which improves the efficiency and accuracy of call bill monitoring and can issue warnings in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A flowchart of a call bill monitoring method in an embodiment of the present invention;

[0020] Figure 2 A flowchart of another call bill monitoring method in an embodiment of the present invention;

[0021] Figure 3 A flowchart of another method for monitoring a call record in an embodiment of the present invention;

[0022] Figure 4 A flowchart of another method for monitoring a call record in an embodiment of the present invention;

[0023] Figure 5 A flowchart of another method for monitoring a call record in an embodiment of the present invention;

[0024] Figure 6 A function trend diagram of a membership function in an embodiment of the present invention;

[0025] Figure 7 A data schematic diagram of an original sequence and an accumulated sequence in an embodiment of the present invention;

[0026] Figure 8 A structural diagram of a call bill monitoring method in an embodiment of the present invention;

[0027] Fig. 9 A structural diagram of another call record monitoring method in an example of an embodiment of the present invention;

[0028] Fig.10 The figure is a structural diagram of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the technical solution of the present invention, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0030] It should be understood that the specific embodiments and drawings described herein are only used to explain the present invention rather than to limit the present invention.

[0031] It can be understood that, in the absence of conflict, the various embodiments of the present invention and the various features in the embodiments can be combined with each other.

[0032] It can be understood that, for the convenience of description, the drawings of the present invention only show the parts related to the present invention, while the parts irrelevant to the present invention are not shown in the drawings.

[0033] It can be understood that each unit and module involved in the embodiments of the present invention may correspond to only one physical structure, or may be composed of multiple physical structures, or multiple units and modules may be integrated into one physical structure.

[0034] It can be understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of the present invention may occur in an order different from that marked in the drawings.

[0035] It is understood that the flowcharts and block diagrams of the present invention illustrate the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to the various embodiments of the present invention. Each box in the flowchart or block diagram may represent a unit, module, program segment, or code, which contains executable instructions for implementing the specified functions. Moreover, each box or combination of boxes in the block diagram and flowchart may be implemented by a hardware-based system that implements the specified functions, or may be implemented by a combination of hardware and computer instructions.

[0036] It can be understood that the units and modules involved in the embodiments of the present invention can be implemented by software or hardware. For example, the units and modules can be located in a processor.

[0037] The embodiment of the present invention provides a method for monitoring a call record. Figure 1 As shown, the call record monitoring method includes steps 101 to 103.

[0038] Step 101: Obtain input data based on historical call record data.

[0039] like Figure 2As shown, step 101 includes steps 201 to 202.

[0040] In some embodiments, step 201, multiple reference dates are determined according to the date of the current day, and actual call record data or predicted call record data of each of the multiple reference dates is obtained respectively.

[0041] In step 201, the historical call record data includes actual call record data or predicted call record data of each reference date among multiple reference dates.

[0042] Understandably, in the grey prediction algorithm, only basic mathematical operations are required, usually at least 4 data are needed, and there is no need to make too many strict assumptions on the sample data to establish a grey prediction model. The sample data can be obtained at equal intervals or unequal intervals, and when the sample data is small, there will be no unpredictable situation.

[0043] It can be seen that the bill data of four reference dates can be selected as historical bill data to reduce the difficulty of calculation. In this case, in some embodiments, in step 201, in response to the current day being the first day of the current month (for example, the current day is June 1, 2024), the reference date includes the first day of the first four months of the current month (the reference date includes May 1, 2024, April 1, 2024, March 1, 2024, and February 1, 2024).

[0044] It can be understood that if the current day is the first day of the current month, and the current month has not yet begun, the first day of each month can be selected as the reference date, that is, the first day of the bill data of the four months before the current month is used as input data to predict the bill data of the current day, which can make the predicted value of the bill data of the current day more accurate. At this time, the intervals obtained by the sample data (the bill data of each reference date) are equal intervals, that is, the interval between every two adjacent reference dates is the same, which is one month.

[0045] In some embodiments, in step 201, in response to the current day being the second day of the month (for example, the current day is June 2, 2024), the reference dates include the second days of the three months before the current month and the first day of the current month (the reference dates include June 1, 2024, May 2, 2024, April 2, 2024, and March 2, 2024).

[0046] It can be understood that one day has already started in the current month, and the call data of the second day of the first three months of the current month and the first day of the current month are used as input data to predict the call data of the current day, which can make the predicted value of the call data of the current day more accurate. At this time, the intervals obtained by the sample data are non-equal intervals, that is, the intervals between each two adjacent reference dates are not exactly the same.

[0047] In some embodiments, in step 201, in response to the current day being the third day of the current month (for example, the current day is June 3, 2024), the reference date includes the third day of the two months before the current month and the first two days of the current month (the reference date includes June 1, 2024, June 2, 2024, May 3, 2024, and April 3, 2024). In response to the current day being the fourth day of the current month (for example, the current day is June 4, 2024), the reference date includes the fourth day of the month before the current month and the first three days of the current month (the reference date includes June 1, 2024, June 2, 2024, June 3, 2024, and May 4, 2024).

[0048] It can be understood that the case where the current day is the third day of the month and the case where the current day is the fourth day of the month are similar to the case where the current day is the second day of the month.

[0049] In some embodiments, in step 201, in response to the current day being a date other than the first four days of the current month (for example, the current day is June 8, 2024), the reference date includes the first four days of the current day (the reference date includes June 7, 2024, June 6, 2024, June 5, 2024, and June 4, 2024).

[0050] It can be understood that if the current day is not the four days before today, then the four days before today are all data for this month. Using the call bill data of the four days before today as input data to predict the call bill data of the current day can make the predicted value of the call bill data of the current day more accurate.

[0051] Exemplarily, no matter what day of the month the current day is (for example, the current day is May 20, 2024), the four days before the current day can be used as reference dates (the reference dates include May 19, 2024, May 18, 2024, May 17, 2024, and May 16, 2024), or the dates corresponding to the current day in the four months before the current day can be used as reference dates (the reference dates include April 20, 2024, March 20, 2024, February 20, 2024, and January 20, 2024), or the dates corresponding to the current day in the four years before the current day can be used as reference dates (the reference dates include May 20, 2023, May 20, 2022, May 20, 2021, and May 20, 2020).

[0052] Step 202: Acquire input data according to actual call record data or predicted call record data of each reference date among multiple reference dates.

[0053] In some embodiments, the implementation method of step 202 includes: in response to the absence of abnormal data in the actual call record data of each reference date among the multiple reference dates, using the actual call record data of each reference date among the multiple reference dates as input data.

[0054] It can be understood that the judgment criteria for abnormal data can be set by the staff based on experience. When there is no abnormal data in the actual call bill data of each reference date among multiple reference dates, it means that the actual call bill data of each reference date is relatively in line with the norm and has reference value. Directly using the actual call bill data as input data is conducive to improving the degree of closeness between the predicted data of the day's call bill and the actual data of the day's call bill. In this way, when using the predicted data of the day's call bill to judge whether the actual data of the day's call bill is abnormal, the judgment result will be more accurate.

[0055] In some embodiments, the implementation method of step 202 includes: in response to the presence of normal data and abnormal data in the actual call record data of each reference date among multiple reference dates, determining a reference date corresponding to the normal data and a reference date corresponding to the abnormal data, and using the actual call record data of the reference date corresponding to the normal data and the predicted call record data of the reference date corresponding to the abnormal data as input data.

[0056] In some embodiments, the implementation method of step 202 includes: in response to the absence of normal data in the actual call record data of each reference date among the multiple reference dates, using the predicted call record data of each reference date among the multiple reference dates as input data.

[0057] It can be understood that abnormal data can be considered as erroneous data and therefore has no reference value. Therefore, using the predicted call record data of the reference date corresponding to the abnormal data instead of the actual call record data of the reference date corresponding to the abnormal data can make the input data more in line with the norm, which is conducive to improving the degree of closeness between the predicted data of the day's call record and the actual data of the day's call record. That is, when using the predicted data of the day's call record to determine whether the actual data of the day's call record is abnormal, the judgment result will be more accurate.

[0058] Step 102: Obtain the forecast data of the call record for the day based on the input data and the forecast model.

[0059] In step 102, the prediction model is obtained according to the grey prediction algorithm.

[0060] It can be understood that the framework of the grey prediction algorithm includes the following three parts:

[0061] 1) Gray generation: The uncertain data sequence is processed to change it from gray to white.

[0062] 2) Grey modeling: Use the regular sequence after grey generation to establish grey difference equations and grey differential equations.

[0063] 3) Grey prediction: Use mathematical models to make quantitative predictions about future changes in data.

[0064] Grey prediction model (GM model for short) is the theoretical basis of the prediction model in the present invention. The following is a brief introduction from three perspectives: definition, solution and application of GM model.

[0065] First, the GM(1,1) model definition is introduced, which includes the following three parts.

[0066] 1) Sequence Z (1) For sequence X (1) The sequence of the nearest mean of , where:

[0067] z (1) (k) = 0.5 × (x (1) (k)+x (1) (k+1)),k=2,3,...,n.

[0068] 2)x (0) (k)+ax (1) (k) = b is the original form of the GM (1,1) model, x (0) (k)+az (1) (k) = b is the basic form of the GM (1, 1) model, where a and b are model parameters and k = 2, 3, ..., n.

[0069] 3) For the GM(1,1) model x (0) (k)+az (1) (k) = the whitened differential equation of b.

[0070] Secondly, the solution method of GM(1,1) model is introduced, which includes the following three parts.

[0071] 1)x (0) (k)+az (1) The model parameters a and b in (k) = b can be estimated using the least squares method, is a parameter column, satisfying Among them, Y and B are:

[0072]

[0073] 2) Whitening differential equation The solution of is also called the time response function, which is:

[0074]

[0075] 3)GM(1,1) model x (0) (k)+az (1) The time response sequence of (k) = b is:

[0076]

[0077] Then, the steps to apply the GM(1,1) model to estimate the data series are as follows:

[0078] The first step is to list the observation sequence (i.e. the collected sample sequence), assuming that the collected sample sequence is X (0) =(x (0) (1),x (0) (2),...,x (0) (n)), n∈N, N is the set of natural numbers.

[0079] The second step is to perform a cumulative generation process on the listed observation sequences, and the new sequence X (1) =(x (1) (1),x (1) (2),...,x (1) (n)) is the sequence X (0) A cumulative generation sequence of , where

[0080] The third step is to solve the values ​​of model parameters a and b using the least squares method.

[0081] Step 4: Calculate the predicted value according to the time response formula, that is, the predicted value

[0082] The above-mentioned clock synchronization algorithm based on grey prediction is only based on the first-order accumulation of sample sequences and has a limited scope of application. In a call record monitoring method provided in an embodiment of the present invention, a method of expanding the modeling order from the first order to the fractional order is proposed. The accumulation number of the sequence can be accurately adjusted through the fractional order, thereby increasing the scope of application of the grey prediction model and enabling it to have higher prediction accuracy in different environments.

[0083] In this case, in some embodiments, such as Figure 3 As shown, the implementation method of step 102 includes steps 301 to 307.

[0084] Step 301: Determine the original sequence according to input data.

[0085] In step 301, the original sequence is recorded as X (0) , X (0) =(x (0) (1),x(0) (2),...x (0) (n)), n is the number of data in the input data, x (0) (1), x (0) (2), …, x (0) (n) are input data respectively, and n∈N, N is a set of natural numbers.

[0086] Step 302: Obtain an r-order cumulative generation sequence, an r-order adjacent mean generation sequence, and an r-order cumulative reduction generation sequence according to the original sequence.

[0087] In step 302, the r-order cumulative generated sequence is recorded as X (r) , X (r) =(x (r) (1), x (r) (2), …, x (r) (k), …, x (r) (n)), and x (r) The expression of (k) is formula (1), and the r-order nearest mean generation sequence is denoted as Z (r) , Z (r) =(z (r) (1), z (r) (2), …, z (r) (k), …, z (r) (n)), and z (r) (k) is expressed as formula (2), and the r-order cumulative generation sequence is recorded as X (-r) , X (-r) =(x (-r) (1), x (-r) (2), …, x (-r) (k), …, x (-r) (n)), and x (-r) The expression of (k) is (3), r∈R + , R + is the set of positive real numbers.

[0088]

[0089] z (r) (k) = 0.5 × (x (r) (k)+x (r) (k-1)),k=2,3…,n (2)

[0090]

[0091] Step 303: Obtain a prediction model.

[0092] In step 303, the prediction model is x (r-1) (k)+az (r) (k) = b.

[0093] It can be understood that n∈R and is the Gamma function of real number n, defined as formula (4).

[0094]

[0095] By using the method of integration by parts, it can be deduced that the Gamma function has a recursive relationship as shown in formula (5).

[0096] Γ(n+1)=nΓ(n) (5)

[0097] When n∈R, equation (5) can be rewritten as equation (6).

[0098]

[0099] Available,

[0100] It can be seen that the Gamma function is a generalization of the factorial in the real number domain.

[0101] In formula (1), X (0) The coefficient of (i) is denoted by a k , then we can get formula (7).

[0102]

[0103] According to equation (7) and equation (1), by the properties of the gamma function, when n∈N, N is a set of natural numbers, equation (1) can be rewritten as equation (8).

[0104]

[0105] x (0) (k)+ax (r) (k) = b is the original form of the GM(1,1) model of the fractional-order operator. When r = 1, x (0) (k)+ax (r) (k) = b becomes x (0) (k)+ax (1) (k) = b, which is the original form of the GM(1,1) model.

[0106] According to formula (2), we can get X (r) The first-order differential equation of is the prediction model, also known as the GM(1,1) model of the fractional-order operator, as shown in formula (9).

[0107] x (r-1) (k)+az (r) (k) = b (9)

[0108] When r = 1, equation (9) is x (0) (k)+az (1) (k) = b, which is the basic form of the GM (1,1) model.

[0109] It can be understood that according to formula (9), the prediction model can be determined by determining the value of the fractional order r. Figure 4 As shown, the implementation method of step 303 includes steps 401 to 405.

[0110] Step 401: Determine a plurality of first values.

[0111] In step 401, each of the plurality of first values ​​is a positive real number.

[0112] Exemplarily, the value range of the first numerical value is the interval [1, 5].

[0113] Step 402: Use each of the plurality of first values ​​as a quasi-fractional order r, and determine a quasi-prediction model for each quasi-fractional order r.

[0114] Step 403: Obtain predicted call bill data for the month of the current day one year before the current day's year based on the actual call bill data of the year before the current day's year and each quasi-prediction model.

[0115] Step 404: compare the actual call record data of the month of the current day one year before the current day's year with the predicted call record data of the month of the current day one year before the current day's year corresponding to each quasi-prediction model.

[0116] Step 405: Determine a prediction model from among multiple quasi-prediction models according to the comparison result.

[0117] In some embodiments, Figure 5 As shown, the implementation method of step 405 includes steps 501 to arrangement 503.

[0118] Step 501: Compare the difference between the predicted call record data of the month of the day in the year before the year corresponding to each quasi-prediction model and the actual call record data of the month of the day in the year before the year corresponding to the day with a first preset threshold.

[0119] It is understandable that the data of the same month every year may be similar. Therefore, if a quasi-prediction model makes predictions based on the actual call record data of the same month of the year before the current year, and the predicted call record data is closest to the actual call record data, then it can be considered that the quasi-prediction model is applicable to the prediction of the current day.

[0120] It can be understood that the first preset threshold value can be in the form of a percentage (for example, 5%), or in the form of a specific numerical value (for example, 3), and its form and value can be set according to the actual situation.

[0121] Exemplarily, when the difference is manifested as the difference between the daily call record data of the predicted call record data of the month of the same day in the year before the year in which the current day corresponds to each quasi-prediction model and the actual call record data of the month of the same day in the year before the year in which the current day corresponds, the first preset threshold can be in the form of a specific numerical value.

[0122] Alternatively, exemplarily, when the difference is manifested as the ratio of the difference between the daily call record data of the predicted call record data of the month of the same day in the year before the year in which the current day corresponds to each quasi-prediction model and the actual call record data of the month of the same day in the year before the year in which the current day corresponds to, and the daily call record data, the first preset threshold can be in the form of a percentage.

[0123] For example, as shown in Table 1, the current day is June 6, 2024, and the difference is the ratio of the difference between the predicted call bill data of the month of the current day in the year before the current day corresponding to each quasi-prediction model and the actual call bill data of the month of the current day in the year before the current day. When the actual call bill data on June 1, 2023 is 25.86 and the first value is 1.1, the predicted call bill data on June 1, 2023 is 26.35. The difference is first calculated, that is, 26.35-25.86=0.49, and the ratio of the difference 0.49 to the actual call bill data 25.86 on June 1, 2023 is 1.89% as the difference between the actual call bill data on June 1, 2024 and the predicted call bill data. Then the first preset threshold can be set in the form of a percentage. For example, in Table 1, 5% is used as the first preset threshold.

[0124] Table 1 Examples of actual call record data and predicted call record data

[0125]

[0126] It can be understood that the difference in daily bill data is only used to indicate the closeness between the actual bill data and the predicted bill data. Therefore, regardless of the size of the actual bill data and the predicted bill data in the daily bill data, when calculating the difference in daily bill data, the absolute value of the difference between the actual bill data and the predicted bill data can be taken and compared with the actual bill data. For example, in Table 1, the actual bill data on June 2, 2023 is greater than the predicted bill data. When calculating the difference, the actual bill data can be subtracted from the predicted bill data to make the difference a positive number.

[0127] Step 502: Obtain the number of days in which the difference between the predicted call record data of the month of the day of the year before the year in which the day of the year corresponds to each quasi-prediction model and the actual call record data of the month of the day of the year before the year in which the day of the year corresponds to the day is greater than a first preset threshold.

[0128] For example, as shown in Table 1, when the first value is 1.1, the number of days when the difference exceeds 5% is 6 days. When the first value is 1.2, the number of days when the difference exceeds 5% is 8 days.

[0129] Step 503: determine the minimum value of the number of days when the difference in daily call bill data is greater than the first preset threshold, and use the quasi-prediction model corresponding to the minimum value of the number of days when the difference in daily call bill data is greater than the first preset threshold as the prediction model.

[0130] Exemplarily, as shown in Table 1, it is obvious that the minimum value between 6 and 8 is 6, and the quasi-prediction model corresponding to 6 is a prediction model with a fractional order r value of 1.1.

[0131] Step 304: Estimate the model parameters of the prediction model using the least squares method.

[0132] In step 304, the model parameters include model parameter a and model parameter b.

[0133] Exemplarily, the estimation method of model parameter a and model parameter b in step 304 is consistent with the estimation method of model parameter a and model parameter b presented in the GM(1,1) model solving method. Reference may be made to the relevant description of the method for estimating model parameter a and model parameter b according to the least squares method in the above-mentioned GM(1,1) model solving method.

[0134] Step 305: Obtain the time response function of the prediction model according to the whitened differential equation of the prediction model.

[0135] In step 305, the whitened differential equation of the prediction model is shown in equation (10), and the time response function of the prediction model is recorded as The expression is formula (11).

[0136]

[0137] Step 306: Obtain the time response sequence of the prediction model according to the time response function of the prediction model.

[0138] In step 306, the time response sequence of the prediction model is recorded as The expression is formula (12).

[0139]

[0140] Step 307: Perform negative r-order cumulative reduction to 0-order to obtain the reduced value of the prediction model.

[0141] In step 307, the predicted data of the day's call list is the restored value of the prediction model, and the restored value of the prediction model is recorded as As shown in formula (13).

[0142]

[0143] Step 103: Determine whether the predicted data of the call bill for the day is abnormal based on the membership function, and issue an alarm if the predicted data of the call bill for the day is abnormal.

[0144] In some embodiments, the membership function is denoted as A(x) and is expressed as formula (14).

[0145]

[0146] In formula (14), c, d, f and h are the parameters of the membership function.

[0147] It can be understood that when a data is less than a certain data standard C (c≤C≤d) and greater than a certain data standard D (f≤D≤h), it can be determined as data anomaly. Through detailed observation and analysis of the call data, it can be concluded that the call data conforms to the trapezoidal fuzzy change trend. Therefore, the membership function shown in formula (14) is selected. The function trend diagram is shown in Figure 6 shown.

[0148] like Figure 6 As shown in the figure, if the predicted data of the day's call list is between d and f, it can be considered that the predicted data of the day's call list is infinitely close to the standard normal data. If the predicted data of the day's call list is between c and d or between f and h, it can be considered that the predicted data of the day's call list gradually deviates from the normal data, that is, it can be considered that the predicted data of the day's call list is abnormal.

[0149] In a call bill monitoring method provided by an embodiment of the present invention, a data set is defined as a fuzzy set. After the data is defined as a fuzzy set, the data can be judged to quickly and accurately locate the call bill anomaly. And the membership function obtained by the fuzzy set is used as a parameter for the subsequent gray prediction of the billing call bill data. Because the result obtained by the membership function changes according to the change of the data set, the parameters are automatically adjusted according to the actual situation. It is an adaptive gray prediction, and the prediction result will be more accurate. And the algorithm does not need to make too many strict assumptions on the sample data to establish a gray prediction model. The intervals obtained by the sample data can be equal intervals or non-equal intervals. Therefore, the gray prediction method is very suitable for the study of call bill monitoring data.

[0150] The following is an example to illustrate a call record monitoring method provided by an embodiment of the present invention.

[0151] This example takes the Jilin Province Unicom call bill monitoring system as an example. The Jilin Province Unicom call bill monitoring system integrates all aspects of the call bill from network elements to centralized procurement, billing sorting, duplicate sorting, and batch pricing, penetrates the entire process of switching-collection-billing-call bill revenue, realizes unified management, establishes a full-process call bill monitoring platform, and forms a closed-loop production control and monitoring system. The system realizes the timely discovery of abnormal fluctuations in billing revenue, call bills, and traffic, avoiding the risk of revenue loss caused by system vulnerabilities and other reasons.

[0152] The methods for monitoring conversational records include:

[0153] ① Determine the billing data caliber

[0154] ②Analyze call log data

[0155] The corresponding membership function and membership degree are derived based on the corresponding call record data, and then the evaluation criteria are determined.

[0156] ③ Predict call bill trends

[0157] According to the historical call record results, the grey prediction parameters are adaptively determined by the evaluation criteria of ② to achieve adaptive call record monitoring and prediction.

[0158] ④ Timely warning

[0159] Issue timely warnings after abnormal trends are predicted.

[0160] The following are respectively explained:

[0161] ①. Determine the billing side call data caliber (data model):

[0162] Determine the daily call bill data and sort out the data to form a template. Daily data is recorded through the Jilin Province United Call Bill Monitoring System. Due to the delay in the group's data transmission, the data is displayed on a T+1 basis.

[0163] ②. Analyze call data (determine membership function):

[0164] There are currently three methods for determining membership functions: 1. Fuzzy statistics; 2. Measurement using existing objective scales; 3. Assignment method.

[0165] Since the call bill monitoring data can be judged as abnormal when the data is less than a certain data standard C or greater than a certain data standard D, through detailed observation and analysis of the call bill data, it can be concluded that the data conforms to a trapezoidal fuzzy change trend, such as Figure 6 shown.

[0166] When the data is between d and f, we can regard it as infinitely close to the standard normal data; when the data is between c and d or f and h, we can regard it as gradually deviating from the normal data. Therefore, the relevant membership function can be obtained as shown in formula (14).

[0167] ③. Predict call bill trend (adaptive grey prediction model):

[0168] The prediction algorithm based on statistical processing needs to rely on a specific distribution model and cannot accurately predict the data results. The data determined by the membership function is more in line with the actual situation of the call data, and the optimal data sequence obtained after each algorithm calculation is used as the latest data sequence of the grey prediction model. The perfect combination of the two can obtain more accurate prediction data.

[0169] The following introduces the application of grey prediction method in this example.

[0170] 1. Grey prediction theory:

[0171] Grey prediction theory is a new modeling idea and method proposed from the perspective of control theory for the uncertain system of "partial information is known and part of the information is unknown". It has the advantages of small sample requirement (at least 4 data) and simple modeling process.

[0172] Normally, it is difficult for us to find the statistical laws of system development and evolution. In the grey prediction method, the randomness of the original data is weakened by accumulating the original sequence. The development trend of the grey accumulation process can be seen in the accumulative generation process, so that the laws contained in the discrete original data can be fully displayed, and then the grey differential or difference equation is established to generate new data.

[0173] like Figure 7 As shown, the original sequence is X (0) =(1,3,2,4,2), the sequence generated by accumulation is X (1) =(1,4,6,10,12), sequence X (0) X obtained after gray accumulation (1) It shows a monotonically increasing pattern.

[0174] In addition, in order to solve the problem that the model prediction results are inconsistent with the qualitative analysis conclusions, the algorithm applies the buffer operator theory to pre-process the modeling sequence, strengthen or weaken the development trend of the system, and integrate the qualitative analysis results into the future development trend of the system, so that the grey prediction model avoids the prediction error of "modeling for the sake of modeling".

[0175] Compared with the method based on statistical processing, the grey prediction method has four major advantages:

[0176] 1) Use "small data" for modeling (4 sample data are enough).

[0177] 2) There is no need to consider the probability distribution law and statistical characteristics of the original data.

[0178] 3) Less computational workload.

[0179] 4)Multi-factor analysis can be performed.

[0180] The grey prediction method only requires basic mathematical operations, usually at least 4 data points, and no strict assumptions are made on the sample data to establish a grey prediction model. The sample data can be obtained at equal intervals or unequal intervals, and when the sample data is small, there will be no unpredictable situation.

[0181] The grey prediction theory model architecture is as follows:

[0182] 1) Gray generation: Process the uncertain data sequence to turn it from gray to white.

[0183] 2) Grey modeling: Use the regular sequence after grey generation to establish grey difference equations and grey differential equations.

[0184] 3) Grey prediction: Use mathematical models to make quantitative predictions about future changes in data.

[0185] 2. GM(1,1) Grey Prediction Model

[0186] Grey prediction model (Grey Model), referred to as GM model.

[0187] GM(1,1) model definition:

[0188] 1) Sequence Z (1) For sequence X (1) The sequence of the nearest mean of , where:

[0189] z (1) (k) = 0.5 × (x (1) (k)+x (1) (k+1))),k=2,3,...,n.

[0190] 2)x (0) (k)+ax (1) (k) = b is the original form of the GM (1,1) model, x (0) (k)+az (1) (k) = b is the basic form of the GM (1, 1) model, where a and b are parameters, and k = 2, 3, ..., n.

[0191] 3) For the GM(1,1) model x(0) (k)+az (1) (k) = the whitened differential equation of b.

[0192] GM(1,1) model solution method:

[0193] 1)x (0) (k)+az (1) The parameters a and b in (k) = b can be estimated using the least squares method, is a parameter column, satisfying Among them, Y and B are:

[0194]

[0195] 2) Whitening differential equation The solution of is also called the time response function, which is:

[0196]

[0197] 3)GM(1,1) model x (0) (k)+az (1) The time response sequence of (k) = b is:

[0198]

[0199] The steps to estimate the data series using the GM(1,1) model are as follows:

[0200] Step 1. List the observation sequence. Let the collected sample sequence be X. (0) =(x (0) (1),x (0) (2),...,x (0) (n)),n∈N, N is the set of natural numbers.

[0201] Step 2: Perform a cumulative generation process on the listed sample sequences, and the new sequence X (1) =(x (1) (1),x (1) (2),...,x (1) (n)) is the sequence X (0) A cumulative generation sequence of , where

[0202]

[0203] Step 3. Solve the values ​​of a and b using the least squares method.

[0204] Step 4. Calculate the predicted value according to the time response formula:

[0205]

[0206] The current clock synchronization algorithm based on grey prediction is only based on the first-order accumulation of sample sequences and has a limited scope of application. This paper expands the modeling order from the first order to the fractional order. The fractional order can accurately adjust the accumulation number of the sequence, increase the scope of application of the grey prediction model, and enable it to have higher prediction accuracy in different environments.

[0207] 3. Fractional-order operator GM(1,1) prediction model

[0208] Fractional-order operator GM(1,1) model definition:

[0209] 1) Let X (0) =(x (0) (1),x (0) (2),...,x (0) (n)),(n∈N).X (r) =(x (r) (1),x (r) (2),...x (r) (n)),(r∈R + , R + is the set of positive real numbers), (n∈N, N is the set of natural numbers) is X (0) =(x (0) (1),x (0) (2),...,x (0) (n)), (n∈N), where:

[0210]

[0211] According to the properties of the gamma function, when n∈N, we have:

[0212] Γ(n)=Γ(n-1)!

[0213] The above formula can be rewritten as:

[0214]

[0215] x (0) (k)+ax ( r ) (k) = b is the original form of the GM(1,1) model of the fractional-order operator.

[0216] Especially when r = 1, x (0) (k)+ax (r) (k) = b becomes x (0) (k)+ax (1) (k) = b, which is the original form of the GM(1,1) model.

[0217] n∈R and is the Gamma function of real number n, defined as:

[0218] By using the method of partial integration, it can be deduced that the Gamma function has the following recursive relationship:

[0219] Γ(n+1)=nΓ(n)

[0220] In particular, when n∈R, we have

[0221] have, Γ(1)=Γ(2)=1, Γ(3)=2! ,Γ(4)=3! ,...,Γ(n)=(n-1)! .

[0222] It can be seen that the Gamma function is a generalization of the factorial in the real number domain.

[0223] X (0) The coefficient of (i) is:

[0224]

[0225] 2) Let X (0) With X (r) As defined above, X (r) Yes X (0) The fractional order cumulative generation sequence, Z (r) For X (r) The sequence of the nearest mean of , where:

[0226] z (r) (k) = 0.5 × (x (r) (k)+x (r) (k-1)),k=2,3…,n

[0227] You can get X ( r ) The first-order differential equation

[0228] x ( r- 1) (k)+az ( r ) (k) = b

[0229] Especially when r = 1, x ( r- 1) (k)+az ( r ) (k) = b becomes x (0) (k)+az (1) (k) = b, which is the mean GM(1,1) model.

[0230] 3) Solution method of fractional-order operator GM(1,1) model:

[0231] Let X (0) =(x (0) (1),x (0) (2),...x (0) (n)) is the original sequence, n∈N,r∈R + .

[0232] X ( r ) =(x ( r ) (1),x ( r ) (2),...,x ( r ) (n)) is X (0) =(x (0) (1),x (0) (2),...,x (0) (n)) is an r-order cumulative generation sequence, where:

[0233]

[0234] Z ( r ) =(z ( r ) (2),z ( r ) (3),...,z ( r ) (n)), n∈N, where:

[0235] z (r) (k) = 0.5 × (x (r) (k)+x (r) (k-1)),k=2,3…,n

[0236] X (-r) =(x (-r) (1),x (-r) (1),...,x (-r) (n)) is X (0) The r-order cumulative reduction generating operator of

[0237]

[0238] Fractional stage sub-GM(1,1) model x (r-1) (k)+az (r) The parameters a and b in (k) = b can be estimated using the least squares method, is a parameter column, satisfying Where Y and B are:

[0239]

[0240] Fractional-order operator GM(1,1) model x (r-1) (k)+az (r) The whitened differential equation for (k) = b is:

[0241]

[0242] The solution of the whitened differential equation of the fractional-order operator GM(1,1) model, i.e., the time response function, is:

[0243]

[0244] Fractional-order operator GM(1,1) model x (r-1) (k)+az (r) The time response sequence of (k) = b is:

[0245]

[0246] The restored value obtained by the fractional-order operator GM(1,1) model is:

[0247]

[0248] The final prediction result of the call bill monitoring data is obtained by restoring the value.

[0249] ④ Timely warning

[0250] When the predicted result is greater than or less than the given threshold C or D (C and D take values ​​between the domain of the membership function c to d, and f to h), the system triggers an alarm in time and updates the predicted initial value sequence after eliminating abnormal data.

[0251] In the prior art, the data set for call record monitoring is not defined, and there is no systematic theoretical basis. In this example, after defining the data as a fuzzy set, the data can be judged to quickly and accurately locate the abnormality of the call record.

[0252] In addition, in the prior art, no prediction alarm is given for the call sheet results; the traditional grey prediction algorithm only predicts based on the historical values ​​of the samples, and the prediction effect is not very accurate. In this example, the membership function obtained by the fuzzy set is used as the parameter for the subsequent grey prediction of the billing call sheet data. Because the result obtained by the membership function changes according to the change of the data set, the parameters are automatically adjusted according to the actual situation. It is an adaptive grey prediction, and the prediction result will be more accurate. In addition, the algorithm does not need to make too many strict assumptions on the sample data to establish a grey prediction model. The intervals obtained by the sample data can be equal intervals or non-equal intervals. Therefore, the grey prediction method is very suitable for the study of call sheet monitoring data.

[0253] An embodiment of the present invention provides a call bill monitoring system, such as Figure 8 As shown, the call record monitoring system 800 includes an input data acquisition module 801 , a prediction data acquisition module 802 and an abnormal alarm module 803 .

[0254] The input data acquisition module 801 is configured to acquire input data according to historical call record data.

[0255] In some embodiments, the input data acquisition module 801 is configured to: determine multiple reference dates according to the date of the day, and respectively acquire the actual bill data or predicted bill data of each reference date in the multiple reference dates. And, acquire input data according to the actual bill data or predicted bill data of each reference date in the multiple reference dates. Among them, the historical bill data includes the actual bill data or predicted bill data of each reference date in the multiple reference dates.

[0256] In some embodiments, in the case where multiple reference dates are determined according to the date of the day, the input data acquisition module 801 is configured as follows: in response to the day being the first day of the month, the reference date includes the first day of the four months before the month. In response to the day being the second day of the month, the reference date includes the second day of the three months before the month and the first day of the month. In response to the day being the third day of the month, the reference date includes the third day of the two months before the month and the first two days of the month. In response to the day being the fourth day of the month, the reference date includes the fourth day of the month before the month and the first three days of the month. And, in response to the day being a date other than the first four days of the month, the reference date includes the first four days of the day.

[0257] In some embodiments, when input data is obtained based on actual bill data or predicted bill data of each reference date among multiple reference dates, the input data acquisition module 801 is configured as follows: in response to the absence of abnormal data in the actual bill data of each reference date among multiple reference dates, the actual bill data of each reference date among multiple reference dates is used as input data. In response to the presence of normal data and abnormal data in the actual bill data of each reference date among multiple reference dates, a reference date corresponding to the normal data and a reference date corresponding to the abnormal data are determined, and the actual bill data of the reference date corresponding to the normal data and the predicted bill data of the reference date corresponding to the abnormal data are used as input data. And, in response to the absence of normal data in the actual bill data of each reference date among multiple reference dates, the predicted bill data of each reference date among multiple reference dates is used as input data.

[0258] The prediction data acquisition module 802 is configured to acquire the prediction data of the call bill of the day according to the input data and the prediction model, wherein the prediction model is obtained according to the grey prediction algorithm.

[0259] In some embodiments, the prediction data acquisition module 802 is configured to: determine the original sequence according to the input data. Obtain the r-order cumulative generation sequence, the r-order adjacent mean generation sequence and the r-order cumulative subtraction generation sequence according to the original sequence. Obtain the prediction model. Estimate the model parameters of the prediction model using the least squares method. Obtain the time response function of the prediction model according to the whitened differential equation of the prediction model. Obtain the time response sequence of the prediction model according to the time response function of the prediction model. And, perform negative r-order cumulative reduction to 0th order to obtain the restored value of the prediction model.

[0260] It can be understood that the detailed scheme and technical effects of the prediction data acquisition module 802 can refer to the relevant description of steps 301 to 307 in a call record monitoring method provided in an embodiment of the present invention, and will not be repeated here.

[0261] In some embodiments, Fig. 9 As shown, the prediction data acquisition module 802 includes a prediction model acquisition unit 8021. The prediction model acquisition unit 8021 is configured to: determine multiple first numerical values. Each of the multiple first numerical values ​​is a positive real number. Use each of the multiple first numerical values ​​as a quasi-fractional order r, and determine a quasi-prediction model for each quasi-fractional order r. Obtain the predicted call bill data of the month of the current day one year before the year of the current day based on the actual call bill data of the year before the year of the current day and each quasi-prediction model. Compare the actual call bill data of the month of the current day one year before the year of the current day with the predicted call bill data of the month of the current day one year before the year of the current day corresponding to each quasi-prediction model. And, determine the prediction model among the multiple quasi-prediction models based on the comparison result.

[0262] In some embodiments, the prediction model acquisition unit 8021 is configured to: respectively compare the difference between the predicted call bill data of the month of the day of the year before the year in which the current day corresponds to each quasi-prediction model and the actual call bill data of the month of the day of the year before the year in which the current day corresponds to, with a first preset threshold. Respectively obtain the number of days in which the difference between the predicted call bill data of the month of the day of the year before the year in which the current day corresponds to each quasi-prediction model and the actual call bill data of the month of the day of the year before the year in which the current day corresponds to, is greater than the first preset threshold. And, determine the minimum value of the number of days in which the difference in daily call bill data is greater than the first preset threshold, and use the quasi-prediction model corresponding to the minimum value of the number of days in which the difference in daily call bill data is greater than the first preset threshold as the prediction model.

[0263] It can be understood that the detailed scheme and technical effects of the prediction model acquisition unit 8021 can refer to steps 401 to 405 in a call record prediction method provided in an embodiment of the present invention.

[0264] The abnormal alarm module 803 is configured to determine whether the predicted data of the call bill of the day is abnormal according to the membership function, and to issue an alarm if the predicted data of the call bill of the day is abnormal.

[0265] In some embodiments, the expression of the membership function A(x) is as shown in formula (14).

[0266] The specific scheme and beneficial effects of a call record monitoring system provided by an embodiment of the present invention can be referred to the relevant description of a call record monitoring method provided by an embodiment of the present invention, and will not be repeated here.

[0267] An embodiment of the present invention provides a computer device, such as Fig.10 As shown, the computer device 1000 includes a memory 1001 and a processor 1002. The memory 1001 stores a computer program. When the processor 1002 runs the computer program stored in the memory 1001, the processor 1002 executes the above-mentioned call record monitoring method.

[0268] The specific scheme and beneficial effects of a computer device 1000 provided by an embodiment of the present invention can refer to the relevant description of a call record monitoring method provided by an embodiment of the present invention, which will not be repeated here.

[0269] An embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the processor executes the above-mentioned call record monitoring method.

[0270] The specific scheme and beneficial effects of a computer-readable storage medium provided by an embodiment of the present invention can be referred to the relevant description of a call record monitoring method provided by an embodiment of the present invention, and will not be repeated here.

[0271] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A call bill monitoring method, characterized in that: include: Obtain input data based on historical call log data; Obtaining forecast data of the day's call records based on the input data and the forecast model; The prediction model is obtained according to the grey prediction algorithm; as well as Determine whether the predicted data of the day's call bill is abnormal based on the membership function, and issue an alarm if the predicted data of the day's call bill is abnormal.

2. The method for monitoring call records according to claim 1, characterized in that: The step of obtaining input data according to historical call record data includes: Determine multiple reference dates according to the date of the day, and obtain actual call record data or predicted call record data of each reference date in the multiple reference dates respectively; the historical call record data includes the actual call record data or predicted call record data of each reference date in the multiple reference dates; and The input data is acquired according to the actual call record data or the predicted call record data of each reference date among the multiple reference dates.

3. The method for monitoring call records according to claim 2, characterized in that: Determining multiple reference dates according to the date of the day includes: In response to the current date being the first day of the current month, the reference date includes the first days of the four months preceding the current month; In response to the current day being the second day of the current month, the reference date includes the second days of the first three months of the current month and the first day of the current month; In response to the current day being the third day of the current month, the reference date includes the third days of the first two months of the current month and the first two days of the current month; In response to the current day being the fourth day of the current month, the reference date includes the fourth day of the month before the current month and the first three days of the current month; and In response to the current day being a date other than the first four days of the current month, the reference date includes the four days before the current day.

4. The method for monitoring call records according to claim 2, characterized in that: The acquiring input data according to actual call record data or predicted call record data of each reference date of the multiple reference dates includes: In response to the absence of abnormal data in the actual call record data of each reference date among the multiple reference dates, using the actual call record data of each reference date among the multiple reference dates as the input data; In response to the presence of normal data and abnormal data in the actual bill data of each reference date among the multiple reference dates, determining a reference date corresponding to the normal data and a reference date corresponding to the abnormal data, and using the actual bill data of the reference date corresponding to the normal data and the predicted bill data of the reference date corresponding to the abnormal data as the input data; and In response to the absence of normal data in the actual call record data of each of the multiple reference dates, the predicted call record data of each of the multiple reference dates is used as the input data.

5. The method for monitoring call records according to claim 1, characterized in that: The obtaining predicted data of the call record of the day according to the input data and the prediction model includes: Determine the original sequence X according to the input data (00 , where X (0) =(x (0) (1),x (0) (2),...x (0) (n)), n is the number of data in the input data, x (0) (1), x (0) (2), …, x (0) (n) are input data, 且 n∈N; N is the set of natural numbers; According to the original sequence X (0) Get the r-order cumulative generation sequence X ( r ) , r-order adjacent mean generation sequence Z (r) and r-order cumulative reduction sequence X (-r) ; where r∈R + , R + is the set of positive real numbers; X (r) =(x (r) (1), x (r) (2),…,x (r) (k), …, x (r) (n)), and Z (r) = (z (r) (1), z (r) (2), …, z (r) (k), …, z (r) (n)), and z (r) (k)=0.5×(x (r) (k)+x (r) (k-1)),k=2,3…,n X (-r) =(x (-r) (1), x (-r) (2)),…,x (-r) (k), …, x (-r) (n)), and Get the prediction model; the prediction model is x (r-1) (k)+az (r) (k) = b; Use the least squares method to estimate the model parameters a and b of the prediction model; According to the whitened differential equation of the prediction model Get the time response function of the prediction model in, According to the time response function of the prediction model Get the time response sequence of the prediction model in, as well as Perform negative r-order cumulative reduction to 0-order to obtain the restored value of the prediction model in, The predicted data of the day's call record is the restored value of the prediction model 6. The method for monitoring call bills according to claim 5, characterized in that: The obtaining of the prediction model comprises: Determine a plurality of first values; each value in the plurality of first values ​​is a positive real number; Using each of the plurality of first values ​​as a quasi-fractional order r, and determining a quasi-prediction model for each of the quasi-fractional order r; Obtain the predicted call bill data of the month of the current day one year before the current day's year based on the actual call bill data of the year before the current day's year and each quasi-prediction model; Respectively comparing the difference between the predicted call record data of the month of the day of the year before the year of the day corresponding to each of the quasi-prediction models and the actual call record data of the month of the day of the year before the year of the day; and A prediction model is determined from among the plurality of quasi-prediction models according to the comparison result.

7. The method for monitoring call bills according to claim 6, characterized in that: Determining a prediction model from among the plurality of quasi-prediction models according to the comparison result comprises: Compare the difference between the predicted call record data of the month of the day of the year before the year of the day corresponding to each of the quasi-prediction models and the actual call record data of the month of the day of the year before the year of the day with the first preset threshold; Respectively obtaining the number of days in which the difference between the predicted call record data of the month of the day of the year before the year of the day corresponding to each of the quasi-prediction models and the actual call record data of the month of the day of the year before the year of the day is greater than a first preset threshold; and Determine the minimum value of the number of days when the difference in the daily call bill data is greater than the first preset threshold, and use the quasi-prediction model corresponding to the minimum value of the number of days when the difference in the daily call bill data is greater than the first preset threshold as the prediction model.

8. The method for monitoring call bills according to claim 1, characterized in that: The expression of the membership function A(x) is: Among them, c, d, f and h are parameters of the membership function.

9. A call bill monitoring system, characterized in that: include: An input data acquisition module is configured to: acquire input data according to historical call record data; A prediction data acquisition module is configured to: acquire the prediction data of the call bill of the day according to the input data and the prediction model; the prediction model is obtained according to the grey prediction algorithm; and The abnormal alarm module is configured to: determine whether the predicted data of the call list of the day is abnormal according to the membership function, and issue an alarm when the predicted data of the call list of the day is abnormal.

10. A computer device, characterized in that: It comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor runs the computer program stored in the memory, the processor executes the call record monitoring method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor executes the call record monitoring method according to any one of claims 1 to 8.

Images