CN110580544B - Traffic prediction method based on periodic dependence - Google Patents
Traffic prediction method based on periodic dependence Download PDFInfo
- Publication number
- CN110580544B CN110580544B CN201910740087.9A CN201910740087A CN110580544B CN 110580544 B CN110580544 B CN 110580544B CN 201910740087 A CN201910740087 A CN 201910740087A CN 110580544 B CN110580544 B CN 110580544B
- Authority
- CN
- China
- Prior art keywords
- traffic
- telephone traffic
- time period
- periodic
- date
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000000737 periodic effect Effects 0.000 title claims abstract description 52
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims abstract description 13
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 230000009021 linear effect Effects 0.000 claims description 17
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 5
- 230000005484 gravity Effects 0.000 claims description 3
- 238000010801 machine learning Methods 0.000 abstract description 7
- 230000007774 longterm Effects 0.000 abstract description 4
- 238000007405 data analysis Methods 0.000 abstract description 3
- 238000013499 data model Methods 0.000 abstract description 3
- 230000002354 daily effect Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 238000013459 approach Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 230000008092 positive effect Effects 0.000 description 3
- 239000000284 extract Substances 0.000 description 2
- 230000003442 weekly effect Effects 0.000 description 2
- 206010019799 Hepatitis viral Diseases 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000012067 mathematical method Methods 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 201000001862 viral hepatitis Diseases 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M3/00—Automatic or semi-automatic exchanges
- H04M3/22—Arrangements for supervision, monitoring or testing
- H04M3/2218—Call detail recording
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M3/00—Automatic or semi-automatic exchanges
- H04M3/22—Arrangements for supervision, monitoring or testing
- H04M3/36—Statistical metering, e.g. recording occasions when traffic exceeds capacity of trunks
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Signal Processing (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Development Economics (AREA)
- Data Mining & Analysis (AREA)
- Game Theory and Decision Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Telephonic Communication Services (AREA)
Abstract
The invention relates to a telephone traffic prediction method based on periodic dependence, which comprises the steps of collecting background data of a telephone traffic center, preprocessing the background data to obtain telephone traffic, extracting characteristics of the periodic dependence of the telephone traffic in different time units, increasing characteristic dimensionality of the extracted characteristics, inputting all the characteristics and corresponding target values into an algorithm model, obtaining a stable model after training, and inputting telephone traffic prediction request data into the model to obtain telephone traffic prediction. The method does not depend on a standard sequence model, fully considers the periodic dependence of the telephone traffic on different time units, and can well reflect the variation trend of the telephone traffic in different time periods; the method is based on big data analysis, and model training is carried out by adopting a machine learning algorithm, so that the telephone traffic of a future time period can be predicted more accurately, and the telephone traffic of each time period in a long term in the future can be predicted.
Description
Technical Field
The present invention relates to telephone communications; the technical field of automatic or semi-automatic switching offices, in particular to a telephone traffic prediction method based on periodic dependence.
Background
Traffic prediction (traffic forecast) refers to a mathematical method of obtaining the essential data in a telecommunications network, for long term development planning or for the near term adjustment of the organization of individual trunking circuits. The prediction content traffic prediction can be divided into total amount prediction and flow direction prediction, wherein the total amount prediction is the total amount and prediction of traffic of one office, one city, one province (district) or the whole country according to different requirements, and the flow direction prediction is the prediction of the traffic flowing from a certain calling office to a certain receiving office, and has directivity, so that the traffic distribution situation among the offices is predicted.
Nowadays, more and more companies are building their own call centers to help them handle customer requests over the phone, which is a great tendency for businesses. In fact, after the telephone traffic in different time periods in the future is known, the call center can be helped to allocate the staff in advance, the service response speed is further improved, the staff is reasonably allocated, and therefore the service quality is improved, and the staff cost is saved.
However, it is not easy to predict the amount of traffic for each time period over a long period in the future. In reality, the incoming call process is very complex and may be affected by different factors in different time periods, for example, many people call a taxi service center when they want to get into a company in the morning; while at the same time, some work at call centers often requires knowledge of the traffic volume for each time period over a long period of time in the future, for example, the call center may wish to schedule a shift for employees based on the traffic volume for each time period in the next week.
In the prior art, the incoming call amount in each time period is basically predicted, the average call duration is ignored, and the telephone traffic amount is usually related to the incoming call amount and the average call duration. For example:
1) modeling and predicting the time series, and modeling and predicting the future electric quantity according to the time series; the drawback of this approach is that modeling, which usually focuses on daily or even monthly total power, relies on a standard sequence model, and is not a prediction of traffic volume in each time period over a long period;
2) the method comprises the following steps of Poisson modeling prediction, wherein a call incoming process is modeled into a Poisson process, and a call incoming process of one day is modeled into a non-homogeneous Poisson process; the method has the defects that only the incoming call quantity of the next day is modeled, the periodic dependence factor is not considered, and although the quantity of each time period in the day is predicted, the time period is not long;
3) linear fixed and mixed modeling prediction is carried out, wherein the incoming call quantity of the same time period of the previous time period and the previous days is introduced for parameter estimation; the defects are that modeling is carried out according to the prediction day without considering the periodic dependence factor;
4) periodic average prediction, which is to take the average value according to different periods, such as the average value according to a day period; the defect is that the periodic average reflects the overall change trend, and the change of the future time period cannot be accurately predicted.
Disclosure of Invention
The invention solves the problems in the prior art and provides an optimized telephone traffic prediction method based on periodic dependence.
The invention adopts the technical scheme that a telephone traffic prediction method based on periodic dependence comprises the following steps:
step 1: background data of a telephone traffic center is collected; preprocessing the collected background data to obtain telephone traffic;
step 2: carrying out feature extraction on the periodic dependence of the telephone traffic in different time units;
and step 3: adding feature dimensions to the extracted features;
and 4, step 4: inputting all the characteristics and the corresponding target values into an algorithm model, and obtaining a stable model after training;
and 5: and inputting the telephone traffic prediction request data into the model to obtain the telephone traffic prediction.
Preferably, the step 1 comprises the steps of:
step 1.1: background data of a telephone traffic center is collected, wherein the background data is call data information in any time period;
step 1.2: extracting necessary information from the collected background data;
step 1.3: the traffic volume is calculated based on the necessary information.
Preferably, the necessary information includes a date and time callDatetime at the beginning of any time period, a total number of incoming calls callArrivals in any time period, and an average call duration callDuration in any time period; the traffic volume
Preferably, the date and time callDatetime of the beginning of any time period is a data identifier of an incoming call.
Preferably, in the step 2, feature extraction is performed on periodic dependence of traffic in different time units, and the features of the periodic dependence include year, month, day and time period.
Preferably, in step 3, the added dimension to the extracted features includes special date features and linear effect features.
Preferably, the special date characteristics include a specific gravity of the current date in the week, a position of the current date in the month, and whether it is a holiday.
Preferably, the linear effect features comprise an intra-day related feature and an inter-day related feature; the relevant characteristics in the day are telephone traffic of a plurality of time periods before the current time period; the daytime-related characteristic is traffic for the same time period several days prior to the current time period.
Preferably, in the step 5, the traffic prediction request data is obtained, the periodic dependency characteristic and the special date characteristic are obtained through date and time calculation, and the periodic dependency characteristic and the special date characteristic are input into the model to obtain the target traffic value.
Preferably, in the step 5, telephone traffic prediction request data is obtained, and the periodic dependency characteristic and the special date characteristic are obtained through date and time calculation; requesting the telephone traffic values of a plurality of previous time periods and the telephone traffic values of the same time periods of a plurality of previous days to obtain linear effect characteristics; and inputting the periodic dependence characteristic, the special date characteristic and the linear effect characteristic into the model to obtain the target telephone traffic value.
The invention provides an optimized telephone traffic prediction method based on periodic dependence, which comprises the steps of collecting background data of a telephone traffic center, preprocessing the background data to obtain telephone traffic, extracting features of the periodic dependence of the telephone traffic in different time units, adding feature dimensions to the extracted features, inputting all the features and corresponding target values into an algorithm model, training the algorithm model to obtain a stable model, and inputting telephone traffic prediction request data into the model to obtain telephone traffic prediction.
The method does not depend on a standard sequence model, fully considers the periodic dependence relationship of the telephone traffic on different time units, and can well reflect the change trend of the telephone traffic in different time periods; the method is based on big data analysis, and model training is carried out by adopting a machine learning algorithm, so that the telephone traffic of a future time period can be predicted more accurately, and the telephone traffic of each time period in a long term in the future can be predicted.
Drawings
FIG. 1 is a schematic diagram of the distribution of traffic in different years, months, days and time periods in the embodiment of the present invention;
FIG. 2 is a schematic diagram of the distribution of traffic in each time period of two consecutive weeks according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of the average traffic distribution per time period per day of the week in an embodiment of the present invention;
FIG. 4 is a diagram illustrating the distribution of traffic in different years, months and days according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of the average distribution of traffic volume per day per month in an embodiment of the present invention;
FIG. 6 is a schematic diagram illustrating the average daily traffic distribution in each time segment according to an embodiment of the present invention;
FIG. 7 is a flow chart of the present invention for predicting traffic using a circular direct prediction approach;
FIG. 8 is a flow chart of the present invention for predicting traffic volume using incremental prediction;
FIG. 9 is a flow chart of the present invention.
Detailed Description
The present invention is described in further detail with reference to the following examples, but the scope of the present invention is not limited thereto.
The invention relates to a telephone traffic prediction method based on periodic dependence, which is characterized in that a large amount of call information of a telephone traffic center in the past time is analyzed, the call volume and the average call duration are considered simultaneously in the same time period, so that corresponding telephone traffic is obtained, corresponding time dependence characteristics are obtained according to the distribution situation of the telephone traffic in different time units, a prediction model is trained by a machine learning method by combining the past telephone traffic characteristics, and the telephone traffic value of each time period in the future long time period is predicted by adopting two modes of direct prediction and incremental prediction.
The method comprises the following steps.
Step 1: background data of a telephone traffic center is collected; and preprocessing the collected background data to obtain the telephone traffic.
The step 1 comprises the following steps:
step 1.1: background data of a telephone traffic center is collected, wherein the background data is call data information in any time period;
step 1.2: extracting necessary information from the collected background data;
the necessary information includes the date and time callDatetime at the beginning of any time period, the total number of calls callArrivals in any time period, and the average call duration in any time period callDuration; the traffic volume
The date and time callDatetime of the beginning of any time period is the data identification of the call.
Step 1.3: the traffic volume is calculated based on the necessary information.
In the invention, step 1 is a data preparation stage, which preprocesses the call information of the service center in the past time, so as to analyze the telephone traffic more simply and conveniently in the following.
In order to solve the problem with the lowest cost, the invention extracts the following data from a background database of a traffic center:
1) a start time of each time segment;
2) a total number of incoming calls in a time period;
3) average call duration over a period of time;
and calculating the telephone traffic in a time period according to the total number of the incoming calls in the time period, the average call time and the time of the time period.
Step 2: feature extraction is performed on the periodic dependence of traffic on different time units.
In the step 2, feature extraction is carried out on the periodic dependence of the traffic in different time units, and the features of the periodic dependence comprise year, month, day and time period.
And step 3: adding feature dimensions to the extracted features.
In the step 3, the added dimension of the extracted features includes special date features and linear effect features.
The special date characteristics include the specific gravity of the current date in the week, the position of the current date in a month, and whether the current date is a holiday.
The linear effect features comprise an intra-day related feature and an inter-day related feature; the relevant characteristics in the day are telephone traffic of a plurality of time periods before the current time period; the daytime-related characteristic is traffic for the same time period several days prior to the current time period.
In the invention, the steps 2 and 3 are the processes of extracting the characteristics of the periodic dependence of the telephone traffic in different time units and increasing the characteristic dimension.
In the invention, aiming at the periodic dependence characteristics of the telephone traffic, the telephone traffic is divided into years, months, days, time periods of the telephone traffic, and the like, through statistical analysis, the telephone traffic has different values in different years, months, dates and time periods, and the periodic dependence of daily, weekly and yearly in the day is displayed under the influence of the normal work time and the scheduling of a call center of people.
In the invention, aiming at the special date characteristic of the telephone traffic, the telephone traffic on the special date is distributed differently by the distribution of the telephone traffic on the time units of week, month and year:
1) a weekly profile; by distributing the traffic volume according to the day of the week, the traffic volume distribution on different days is different, particularly on weekends, so that the day of the week and whether the day of the week is weekend or not are taken as characteristics;
2) a monthly characteristic; by distributing the telephone traffic according to the daily telephone traffic of one month, the telephone traffic at the beginning of the month, in the middle of the month and at the end of the month is obviously more than that of other days, so that whether the telephone traffic is at the beginning of the month, in the middle of the month and at the end of the month is taken as a characteristic;
3) the annual characteristics; by distributing the traffic on a daily basis for one year, the traffic is significantly less in some holidays than on other days, so whether a holiday is characterized.
In the invention, aiming at the linear effect characteristic of the telephone traffic, the telephone traffic has similar daily change trend by distributing the telephone traffic of each day, the change trend of the increase or decrease of the telephone traffic of continuous time periods is almost the same from the dimension of the time period in one day, and the telephone traffic of the time periods with approximately the same change trend is subjected to correlation calculation to obtain:
1) the telephone traffic correlation between adjacent time periods in one day is strong and has positive effects;
2) the longer the time interval is, the smaller the correlation is;
traffic has an intra-day dependency and is characterized by the traffic of the previous time periods.
And 4, step 4: and inputting all the characteristics and the corresponding target values into an algorithm model, and obtaining a stable model after training.
In the invention, a machine learning algorithm for supervised learning is adopted to train the model, the characteristic value is the characteristic extracted for each time period, the target value is the telephone traffic of the corresponding time period, the characteristics and the target values of a large amount of telephone traffic data are input into the machine learning algorithm, the model can be obtained after the training and stabilization, and the types of the characteristics can be correspondingly combined according to specific conditions.
And 5: and inputting the telephone traffic prediction request data into the model to obtain the telephone traffic prediction.
In the step 5, the telephone traffic prediction request data is obtained, the periodic dependency characteristics and the special date characteristics are obtained through the calculation of the date and time, and the periodic dependency characteristics and the special date characteristics are input into the model to obtain the target telephone traffic value.
In the step 5, telephone traffic prediction request data are obtained, and the periodic dependence characteristic and the special date characteristic are obtained through the calculation of date and time; requesting the telephone traffic values of a plurality of previous time periods and the telephone traffic values of the same time periods of a plurality of previous days to obtain linear effect characteristics; and inputting the periodic dependence characteristic, the special date characteristic and the linear effect characteristic into the model to obtain the target telephone traffic value.
In the invention, two modes are adopted to predict the telephone traffic:
(1) the method does not contain linear effect characteristics, can directly extract periodic characteristics and special date characteristics from the time period of the period to be predicted, and directly input the periodic characteristics and the special date characteristics into a model trained by corresponding characteristics to obtain corresponding telephone traffic values;
(2) the method comprises the steps of predicting the telephone traffic of a next time period by adopting incremental prediction, directly extracting all features of the next time period, inputting the features into a feature training model to obtain the telephone traffic of the next time period, taking the predicted telephone traffic as the features of the telephone traffic of the next time period, obtaining the telephone traffic of a corresponding time period according to the previous prediction mode, and repeating the steps to predict the telephone traffic of all the time periods in the period.
In the present invention, an example is given.
Take 2016 and 2017 call data information of a certain telephone center for each time slot as an example.
Firstly, the data information needed by us is obtained from the call data information:
1) callDatetime, the date and time when each time period starts, and using the date and time as the data identification of one-time incoming call;
2) callArrivals, total number of incoming calls in a time period;
3) callDuration, average call duration over a period of time;
will be a timeThe time length of the segment is recorded as callPeriod, 15 minutes is taken as an example, the traffic callTraffic in the corresponding time segment is calculated,
secondly, the data is analyzed and relevant feature information is extracted.
As shown in fig. 1, the traffic distribution is from "2016-01-1500: 00" to "2016-02-1423: 45" and from "2017-01-1500: 00" to "2017-02-1423: 45" for two consecutive time periods; the traffic is obviously different in different years, months, days and time periods, so the total minutes (totalMinutes) of the year (year), month (month), day (day) and time corresponding to callDatetime is taken as a feature, for example: the above-mentioned characteristic values corresponding to "2016-01-1507: 15" are "2016", "1", "15", and "450", and the total number of minutes of time can be calculated from the number of hours and the time period of the number of minutes, totalMinutes ═ hour × 60+ minute.
As shown in fig. 2, in order to more accurately analyze the traffic distribution of two days on weekend and other days, as shown in fig. 3, the traffic of each day of one week is different and the traffic of two days on weekend is significantly smaller than that of other days, although the traffic of each day has similar distribution, the traffic of two days on weekend is significantly smaller than that of other days, so that the days on week (dayofweek) and whether it is weekend (isweekend) are used as features, and the specific values are shown in table 1 below.
Table 1: traffic tables featuring day of the week (dayofweek) and whether it is weekend (isweekend)
Monday | Zhou Di | Wednesday | Week four | ZhouWu for treating viral hepatitis | Saturday medicine | (Sunday) | |
|
1 | 2 | 3 | 4 | 5 | 6 | 7 |
|
0 | 0 | 0 | 0 | 0 | 1 | 1 |
As shown in fig. 4, the daily traffic distribution in 2016 and 2017 is shown, the traffic at the beginning and end of each month is very large, and the traffic in the month of each month is also larger than that of other dates, as shown in fig. 5, the average traffic distribution per day per month more intuitively reflects that the traffic at the beginning, middle and end of the month is larger than that of other dates, so the date of each month is divided into several parts and characterized according to the distribution of the specific traffic in each month, for example, the following part (sectionofmonth) of a month is characterized in table 2.
Table 2: table of feature values of a part (section) to one of dates in a month
Date of each month | sectionofmonth |
No. 1 to No. 5 | 1 |
16 # to 22 # | 2 |
Last two |
3 |
|
0 |
As shown in fig. 1 and 4, the traffic volume during the spring festival of each year is significantly less than the traffic volume on other dates, it is likely that most people are on vacation and only a few are on duty, and we also feature a special date like the spring festival as it is not a fixed date, as in table 3.
Table 3: feature value table in which the date of the spring festival (isfestrival) is associated with a feature
Whether it is spring festival | isfestival |
Is that | 1 |
Whether or not | 0 |
As shown in fig. 1, 2, and 3, the daily distribution of the traffic is similar to each other, and in order to better show the traffic distribution in one day, fig. 6 shows the average traffic distribution in each time period in two years. In fig. 6, the traffic volume changes in a continuous period of time almost in the same trend, in order to prove that the traffic volume in the current time period has a dependency relationship with the traffic volume in the previous time periods, Pearson correlation coefficients are calculated between the time periods of two years to measure the degree of correlation, and table 4 shows the Pearson correlation coefficient values between the intercepted time periods, and it is concluded that:
1) the telephone traffic correlation between adjacent time periods in one day is strong and has positive effects;
2) the farther the two time periods are different, the weaker the correlation of the telephone traffic is;
therefore, an intra-day correlation feature, i.e., the traffic volume of the previous several time periods of the current time period, will be introduced.
Table 4: table of Pearson correlation coefficient values between truncated time segments
As shown in fig. 2 and 3, the traffic shows periodicity of day and week, the value and the variation trend of the traffic have similarity in each day of the week, and in order to prove the correlation of the traffic between days, Pearson correlation coefficient between days is calculated every day of the week for two years to measure the correlation degree between days, as shown in table 5, the conclusion is that:
1) the correlation of telephone traffic between two adjacent days is strong and has positive effect;
2) the farther two days are apart, the weaker the correlation of the telephone traffic is;
a time-of-day correlation feature, i.e., the amount of traffic for the same time period over the previous days of the current time period, is introduced.
Table 5: pearson correlation coefficient-based day-to-day correlation degree table
Thirdly, training the model by machine learning, such as random forest algorithm, inputting the characteristics and target values of the traffic data into the algorithm, and obtaining the model after training, as shown in table 6, showing the characteristics and target values of a time period, wherein tbd,iRepresents the i-th time period on day d, dtd,iRepresenting a corresponding time period tbd,iPeriodic feature of (1), sdd,iRepresenting a corresponding time period tbd,iSpecial date feature of (1), td,iRepresenting a corresponding time period tbd,iTraffic of (2).
Table 6: characteristic and target value correspondence table for one time period
Finally, traffic prediction is performed by using two strategies, as shown in fig. 7 and 8, the flows of the two strategies are respectively shown. FIG. 7 is a strategy approach without linear effect features, in which all features can be obtained directly by calculating the date and time, so that the target traffic value can be obtained by inputting the features calculated for the time period to be predicted into the model trained from the corresponding features; fig. 8 is a policy paradigm including linear effect characteristics, because in this way, the traffic characteristic value of the time period to be predicted must know the traffic values of the previous time periods, so that only incremental prediction can be adopted, the traffic of the first time period is predicted, and then the traffic value is taken as the characteristic of the next time period, so as to predict the traffic of the next time period, and the traffic values of all the time periods to be predicted are sequentially predicted according to this way.
The method comprises the steps of collecting background data of a telephone traffic center, preprocessing the background data to obtain telephone traffic, extracting features of the periodic dependence of the telephone traffic in different time units, increasing feature dimensions of the extracted features, inputting all the features and corresponding target values into an algorithm model, training the algorithm model to obtain a stable model, and inputting telephone traffic prediction request data into the model to obtain telephone traffic prediction.
The method does not depend on a standard sequence model, fully considers the periodic dependence of the telephone traffic on different time units, and can well reflect the variation trend of the telephone traffic in different time periods; the method is based on big data analysis, and model training is carried out by adopting a machine learning algorithm, so that the telephone traffic of a future time period can be predicted more accurately, and the telephone traffic of each time period in a long term in the future can be predicted.
Claims (8)
1. A traffic prediction method based on periodic dependence is characterized in that: the method comprises the following steps:
step 1: background data of a telephone traffic center is collected; preprocessing the collected background data to obtain the telephone traffic, comprising the following steps:
step 1.1: background data of a telephone traffic center is collected, wherein the background data is call data information in any time period;
step 1.2: extracting necessary information from the collected background data; the necessary information includes the date and time callDatetime at the beginning of any time period, the total number of calls callArrivals in any time period, and the average call duration in any time period callDuration;
step 1.3: calculating a traffic volume based on the necessary information; the traffic volume Wherein callPeriod is the duration of a time period;
step 2: extracting the characteristics of the periodic dependence of the telephone traffic in different time units;
and step 3: adding feature dimensions to the extracted features;
and 4, step 4: inputting all the characteristics and the corresponding target values into an algorithm model, and obtaining a stable model after training;
and 5: and inputting the telephone traffic prediction request data into the model to obtain the telephone traffic prediction.
2. The traffic prediction method based on periodic dependency as claimed in claim 1, wherein: the date and time callDatetime of the beginning of any time period is the data identification of the call.
3. A traffic prediction method based on periodic dependency according to claim 1, characterized in that: in the step 2, feature extraction is carried out on the periodic dependence of the traffic in different time units, and the features of the periodic dependence comprise year, month, day and time period.
4. The traffic prediction method based on periodic dependency as claimed in claim 1, wherein: in the step 3, the added dimension to the extracted features includes special date features and linear effect features.
5. The traffic prediction method based on periodic dependency as claimed in claim 4, wherein: the special date characteristics include the specific gravity of the current date in the week, the position of the current date in a month, and whether the current date is a holiday.
6. The traffic prediction method based on periodic dependency as claimed in claim 4, wherein: the linear effect features comprise an intra-day related feature and an inter-day related feature; the relevant characteristics in the day are telephone traffic of a plurality of time periods before the current time period; the daytime-related characteristic is traffic volume for the same time period several days prior to the current time period.
7. A traffic prediction method based on periodic dependency according to claim 5, characterized in that: in the step 5, the telephone traffic prediction request data is obtained, the periodic dependency characteristics and the special date characteristics are obtained through the calculation of the date and time, and the periodic dependency characteristics and the special date characteristics are input into the model to obtain the target telephone traffic value.
8. A traffic prediction method based on periodic dependency according to claim 6, characterized in that: in the step 5, telephone traffic prediction request data are obtained, and the periodic dependence characteristic and the special date characteristic are obtained through the calculation of date and time; requesting the telephone traffic values of a plurality of previous time periods and the telephone traffic values of the same time periods of a plurality of previous days to obtain linear effect characteristics; and inputting the periodic dependence characteristic, the special date characteristic and the linear effect characteristic into the model to obtain the target telephone traffic value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910740087.9A CN110580544B (en) | 2019-08-12 | 2019-08-12 | Traffic prediction method based on periodic dependence |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910740087.9A CN110580544B (en) | 2019-08-12 | 2019-08-12 | Traffic prediction method based on periodic dependence |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110580544A CN110580544A (en) | 2019-12-17 |
CN110580544B true CN110580544B (en) | 2022-06-17 |
Family
ID=68810575
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910740087.9A Active CN110580544B (en) | 2019-08-12 | 2019-08-12 | Traffic prediction method based on periodic dependence |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110580544B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113038302B (en) * | 2019-12-25 | 2022-09-30 | 中国电信股份有限公司 | Flow prediction method and device and computer storage medium |
CN113949358B (en) * | 2021-12-16 | 2022-03-22 | 广东省新一代通信与网络创新研究院 | Power amplifier voltage adjusting method and system based on machine learning |
CN113973156B (en) * | 2021-12-22 | 2022-04-08 | 杭州远传新业科技有限公司 | Telephone traffic prediction method and system and telephone traffic prediction device |
CN114781678B (en) * | 2021-12-30 | 2022-12-16 | 友邦人寿保险有限公司 | Incoming call quantity prediction method, system, terminal and medium based on prediction model |
CN116436002B (en) * | 2023-06-13 | 2023-09-05 | 成都航空职业技术学院 | Building electricity utilization prediction method |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101132447A (en) * | 2007-08-13 | 2008-02-27 | ***通信集团安徽有限公司 | Hot line call prediction method for large call centre |
JP2015158861A (en) * | 2014-02-25 | 2015-09-03 | Kddi株式会社 | Communication traffic prediction device and program |
CN106713677A (en) * | 2016-05-24 | 2017-05-24 | 国家电网公司客户服务中心 | Prediction method for incoming call traffic of power client service center |
-
2019
- 2019-08-12 CN CN201910740087.9A patent/CN110580544B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101132447A (en) * | 2007-08-13 | 2008-02-27 | ***通信集团安徽有限公司 | Hot line call prediction method for large call centre |
JP2015158861A (en) * | 2014-02-25 | 2015-09-03 | Kddi株式会社 | Communication traffic prediction device and program |
CN106713677A (en) * | 2016-05-24 | 2017-05-24 | 国家电网公司客户服务中心 | Prediction method for incoming call traffic of power client service center |
Non-Patent Citations (3)
Title |
---|
Forecasting Long-Term Call Traffic Based on Seasonal Dependencies;Longchun Cao等;《LNICST》;20190818;第292卷;第231-246页 * |
用周期模型和近邻算法预测话务量时间序列;刘童 等;《吉林大学学报(信息科学版)》;20070531;第25卷(第3期);第239-245页 * |
面向长期多步话务量的预测算法研究;曹龙春;《中国优秀博硕士学位论文全文数据库(硕士) 信息科技辑》;20210215(第2期);第I136-457页 * |
Also Published As
Publication number | Publication date |
---|---|
CN110580544A (en) | 2019-12-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110580544B (en) | Traffic prediction method based on periodic dependence | |
Ibrahim et al. | Modeling and forecasting call center arrivals: A literature survey and a case study | |
CN111507650B (en) | Computing power distribution scheduling method and system for edge computing platform | |
CN101132447A (en) | Hot line call prediction method for large call centre | |
JP4928962B2 (en) | How to predict call volume | |
EP1998277A1 (en) | System and method for multi-week scheduling | |
EP1998278A1 (en) | System and method for long term forecasting | |
US20070265904A1 (en) | Method and apparatus for improved forecasting using multiple sources | |
CA2488079A1 (en) | System and method for event-based forecasting | |
US20050138153A1 (en) | Method and system for predicting network usage in a network having re-occurring usage variations | |
EP3770841A1 (en) | Demand prediction system and method | |
Balasubramanian et al. | Churn prediction in mobile telecom system using data mining techniques | |
CN109685234B (en) | Processing method and system for hotel room reservation | |
US7406171B2 (en) | Agent scheduler incorporating agent profiles | |
CN107977855B (en) | Method and device for managing user information | |
US20060072733A1 (en) | System and method for merchant contact and lead management | |
JP2002149931A (en) | Personnel disposing processing method and system thereof, and profit and loss evaluation processing method and system thereof | |
CN111178948B (en) | Method for realizing dynamic borrowing of shared automobile | |
CN109242665B (en) | Business rule multi-channel sharing method, device, equipment and storage medium | |
CN105809276A (en) | Number taking prediction method and system | |
CN116402362A (en) | Electricity fee charging method and device based on user portrait | |
CN110880987A (en) | Packet domain network capacity prediction method and system based on time characteristics | |
Cao et al. | Forecasting long-term call traffic based on seasonal dependencies | |
Um-In et al. | Agent scheduling of call center using decomposition technique | |
Arntz et al. | Weather‐related Employment Subsidies as a Remedy for Seasonal Unemployment? Evidence from Germany |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |