Modeling network performance and service quality in wireless networks

Patent metadata

Field	Value
Publication number	US-9955488-B2
Application number	US-201615087129-A
Country	US
Kind code	B2
Filing date	Mar 31, 2016
Priority date	Mar 31, 2016
Publication date	Apr 24, 2018
Grant date	Apr 24, 2018

Abstract

A recursive algorithm may be applied to group cells in a service network into a small number of clusters. For each of the clusters, different regression algorithms may be evaluated, and a regression algorithm generating a smallest error is selected. A total error for the clusters may be identified based on the errors from the selected regression algorithms and from degrees of separation associated with the cluster. If the total error is greater than a threshold value, the cells may be grouped into a larger number of clusters and the new clusters may be re-evaluated. A key performance indicator (KPI) may be estimated for a cell based on a regression algorithm selected for the cluster associated with the cell. A resources may be allocated to the cell based on the KPI value.

First claim

What is claimed is:

1. A method comprising:
collecting, by a processor, usage data related to a plurality of cells of a service network;
splitting, by the processor, the usage data into a training data portion and a test data portion, wherein the training data portion is smaller than the test data portion;
grouping, by the processor, cells of the plurality of cells into clusters;
selecting, by the processor, regression algorithms for the clusters, wherein the regression algorithms are selected based on the training data portion of the usage data, wherein the regression algorithms are selected from a group of regression algorithms, and wherein selecting the regression algorithms includes:
identifying prediction errors for the group of regression algorithms for each of the clusters, and
determining, as the regression algorithms, ones of the group of regression algorithms associated with the smallest prediction errors for each of the clusters;

identifying, by the processor, a key performance indicator (KPI) related to a communication resource for a cell of the plurality of cells;
identifying, by the processor, one of the clusters that includes the cell, wherein the one of the clusters is associated with one of the regression algorithms;
calculating, by the processor, a value for the KPI based on the test data portion of the usage data and the one of the regression algorithms; and
allocating, by the processor, the communication resource to the cell based on the calculated value for the KPI.

2. The method of claim 1 , wherein the group of regression algorithms includes at least two of:
a generalized additive model (GAM),
a gradient boost method (GBM),
a neural network method, and
multivariate adaptive regression splines (MARS) method.

3. The method of claim 1 , wherein the clusters are first clusters, wherein the first clusters include a first quantity of clusters, and wherein grouping the cells into the clusters includes:
calculating degrees of separation associated with pairs of the first clusters;
determining a total error for the first clusters based on the degrees of separation and the prediction errors; and
grouping the cells into second clusters when the total error is greater than a threshold value, wherein the second clusters include a second quantity of clusters that is greater than the first quantity.

4. The method of claim 3 , wherein each of the degrees of separation identifies a difference between a first KPI value calculated for one of the clusters and a second KPI value calculated for other ones of the clusters.

5. The method of claim 1 , wherein calculating the value for the KPI includes:
filtering the usage data to identify a portion of the usage data that is relevant to the KPI and the cell, wherein the KPI is calculated from the filtered usage data.

6. The method of claim 1 , wherein allocating the communication resource to the cell includes:
increasing an amount of the communication resource to the cell when the calculated KPI value is greater than a first threshold value; and
decreasing an amount of the communication resource to the cell when the calculated KPI value is less than a second threshold value that is less than the first threshold value.

7. The method of claim 1 , wherein the usage data includes a quantity of messages associated with the plurality of cells during a particular time period or an amount of data exchanged within the messages during the particular time period.

8. A device comprising:
a memory configured to store instructions; and
a processor configured to execute one or more of the instructions to:
collect usage data related to cells of a service network;
divide the usage data into a training data portion and a test data portion, wherein the training data portion is smaller than the test data portion;
group the cells into clusters;
select regression algorithms for the clusters, wherein the regression algorithms are selected based on the training data portion of the usage data, wherein the regression algorithms are selected from a group of regression algorithms, and wherein the processor, when selecting the regression algorithms, is further configured to:
identify prediction errors for the group of regression algorithms for each of the clusters, and
select, as the regression algorithms, regression algorithms of the group of regression algorithms associated with the smallest prediction errors for each of the clusters;

identify a key performance indicator (KPI) related to a communication resource for a cell of the cells;
identify a cluster of the clusters that includes the cell, wherein the cluster is associated with one of the regression algorithms;
calculate a value for the KPI based on the test data portion of the usage data and the one of the regression algorithms; and
allocate the communication resource to the cell based on the calculated value for the KPI.

9. The device of claim 8 , wherein the group of regression algorithms includes at least two of:
a generalized additive model (GAM),
a gradient boost device (GBM),
a neural network device, and
a multivariate adaptive regression splines (MARS) device.

10. The device of claim 8 , wherein the clusters are first clusters, wherein the first clusters include a first quantity of clusters, and wherein the processor, when grouping the cells into the clusters, is configured to:
calculate degrees of separation associated with pairs of the first clusters;
determine a total error for the first clusters based on the degrees of separation and the prediction errors; and
group the cells into second clusters when the total error is greater than a threshold value, wherein the second clusters include a second quantity of clusters that is greater than the first quantity.

11. The device of claim 10 , wherein each of the degrees of separation identifies a difference between a first KPI value calculated for one of the clusters and a second KPI value calculated for other ones of the clusters.

12. The device of claim 8 , wherein the processor, when calculating the value for the KPI, is further configured to:
filter the usage data to identify a portion of the usage data that is relevant to the KPI and the cell, wherein the KPI is calculated from the filtered usage data.

13. The device of claim 8 , wherein the processor, when allocating the communication resource to the cell, is configured to:
increase an amount of the communication resource to the cell when the calculated KPI value is greater than a first threshold value; and
decrease an amount of the communication resource to the cell when the calculated KPI value is less than a second threshold value that is less than the first threshold value.

14. The device of claim 8 , wherein the usage data includes a quantity of messages associated with the cells during a particular time period or an amount of data exchanged within the messages during the particular time period.

15. A non-transitory computer readable memory to store one or more of instructions that cause a processor to:
collect usage data related to cells of a service network;
divide the usage data into a training data portion and a test data portion, wherein the training data portion is smaller than the test data portion;
group the cells into clusters;
select regression algorithms for the clusters, wherein the regression algorithms are selected based on the training data portion of the usage data, wherein the regression algorithms are selected from a group of regression algorithms, and wherein the processor, when selecting the regression algorithms, is further configured

Assignees

• Verizon Patent & Licensing Inc

Inventors

• Ouyang Ye
• Becht Carol
• Iyer Krishna Pichumani

Classifications

• H04L41/16 Primary

  using machine learning or artificial intelligence · CPC title

• H04W72/542

  using measured or perceived quality · CPC title

• H04W16/18

  Network planning tools · CPC title

• H04W24/08

  Testing, {supervising or monitoring} using real traffic · CPC title

• H04L41/0896

  Bandwidth or capacity management, i.e. automatically increasing or decreasing capacities (flow or congestion control using dynamic resource allocation, e.g. in-call renegotiation, H04L47/76) · CPC title