Method of load forecasting via attentive knowledge transfer, and an apparatus for the same
A method of forecasting a future load may include: obtaining source data sets and a target data set that have been collected from a plurality of source base stations and a target base station, respectively; among a plurality of source machine learning models, selecting at least one machine learn source model that has a traffic load prediction performance higher than that of a target machine learning model through a negative transfer analysis; obtaining model weights to be applied to the target machine learning model and the selected at least one source machine learning model via an attention neural network that is jointly trained with the target machine learning model and the selected source machine learning models; obtaining a load forecasting model for the target base station by combining the target machine learning model and the selected at least one source machine learning model according to the model weights; and predicting a future communication traffic load of the target base station based on the load forecasting model.
1. A method of forecasting a future load by one or more processors, the method comprising:
obtaining source data sets and a target data set that have been collected from a plurality of source base stations and a target base station, respectively;
among a plurality of source machine learning models configured to predict a communication traffic load of the plurality of source base stations, selecting at least one source machine learning model that has a traffic load prediction performance higher than that of a target machine learning model configured to predict a communication traffic load of the target base station;
obtaining weights to be applied to the target machine learning model and the selected at least one source machine learning model via a neural network that determines the weights to the target machine learning model and the selected at least one source machine learning model, respectively for combining the target machine learning model and the selected at least one source machine learning model and that is jointly trained with the target machine learning model and the selected source machine learning models;
obtaining a load forecasting model for the target base station by combining the target machine learning model and the selected at least one source machine learning model according to the weights; and
predicting a future communication traffic load of the target base station based on the load forecasting model.
2. The method of claim 1 , further comprising:
transmitting information of the future communication traffic load to the target base station to enable the target base station to assign a plurality of user equipment (UEs) to a plurality of cells of the target base station according to the information of the future communication traffic load.
3. The method of claim 2 , wherein the information of the future communication traffic load comprises any one or any combination of a number of active UEs in each of the plurality of cells, a cell load ratio, an internet protocol (IP) throughput per cell, and a cell physical resource block (PRB) usage ratio.
4. The method of claim 1 , wherein at least one of the target machine learning model, the plurality of source machine learning model, and the load forecasting model is implemented as a Long Short-Term Memory (LSTM) model that comprises an input layer, a plurality of LSTM layers, a dense layer, and an output layer.
5. The method of claim 1 , wherein the neural network is implemented using a multi-layer perceptron (MLP).
6. The method of claim 1 , wherein the traffic load prediction performance is measured by calculating a mean average percentage error (MAPE) of each of the target machine learning model and the selected at least one source machine learning model.
7. The method of claim 1 , further comprising:
determining whether a target domain from which the target data set is collected, and source domains from which the source data sets are collected, have a same feature space; and
prior to obtaining the weights, performing iterative feature matching on the target data set and the source data sets in response to determining that the target domain and the source domains do not have the same feature space.
8. The method of claim 1 , further comprising:
inputting the predicted future communication traffic load and a communication traffic load that is observation from the target base station, to a loss function configured to calculate a mean average percentage error of the load forecasting model;
determining whether a performance degradation has occurred in the load forecasting model based on the mean average percentage error of the load forecasting model; and
retraining the load forecasting model in response to determining that the performance degradation has occurred in the load forecasting model.
9. An electronic device comprising:
at least one memory storing computer-readable instructions; and
at least one processor configured to execute the computer-readable instructions to:
obtain source data sets and a target data set that have been collected from a plurality of source base stations and a target base station, respectively;
among a plurality of source machine learning models configured to predict a communication traffic load of the plurality of source base stations, select at least one source machine learn model that has a traffic load prediction performance higher than that of a target machine learning model configured to predict a communication traffic load of the target base station;
obtain weights to be applied to the target machine learning model and the selected at least one source machine learning model via a neural network that determines the weights to the target machine learning model and the selected at least one source machine learning model, respectively for combining the target machine learning model and the selected at least one source machine learning model and that is jointly trained with the target machine learning model and the selected source machine learning models;
obtain a load forecasting model for the target base station by combining the target machine learning model and the selected at least one source machine learning model according to the weights; and
predict a future communication traffic load of the target base station based on the load forecasting model.
10. The electronic device of claim 9 , further comprising a transceiver,
wherein the at least one processor is further configured to execute the computer-readable instructions to:
control the transceiver to transmit information of the future communication traffic load to the target base station to enable the target base station to assign a plurality of user equipment (UEs) to a plurality of cells of the target base station according to the information of the future communication traffic load.
11. The electronic device of claim 10 , wherein the information of the future communication traffic load comprises any one or any combination of a number of active UEs in each of the plurality of cells, a cell load ratio, an internet protocol (IP) throughput per cell, and a cell physical resource block (PRB) usage ratio.
12. The electronic device of claim 9 , wherein at least one of the target machine learning model, the plurality of source machine learning model, and the load forecasting model is implemented as a Long Short-Term Memory (LSTM) model that comprises an input layer, a plurality of LSTM layers, a dense layer, and an output layer.
13. The electronic device of claim 9 , wherein the neural network is implemented using a multi-layer perceptron (MLP).
14. The electronic device of claim 9 , wherein the at least one processor is further configured to execute the computer-readable instructions to:
measure the traffic load prediction performance by calculating a mean average percentage error (MAPE) of each of the target machine learning model and the selected at least one source machine learning model.
15. The electronic device of claim 9 , wherein the at least one processor is further configured to execute the computer-readable instructions to:
determine whether a target domain from which the target data set is collected, and source domains from which the source data sets are collected, have a same feature space; and
prior to obtaining the weights, perform iterative feature matching on the target data set and the source data sets in response to determining that the target domain and the source domains do not have the same feature space.
16. The electronic device of claim 9 , wherein the at least one processor is further configured to execute the computer-readable instructions to:
input the predicted future communication traffic load and a communication traffic load that is observation from the target base station, to a loss function configured to calculate a mean average percentage error of the load forecasting model;
determine whether a performance degradation has occurred in the load forecasting model based on the mean average percentage error of the load forecasting model; and
retrain the load forecasting model in response to determining that the performance degradation has occurred in the load forecasting model.
17. A non-transitory computer-readable storage medium having recorded thereon instructions to be executed by at least one processor to perform a method of forecasting a future load, the method comprising:
obtaining source data sets and a target data set that have been collected from a plurality of source base stations and a target base station, respectively;
among a plurality of source machine learning models configured to predict a communication traffic load of the plurality of source base stations, selecting at least one source machine learn model that has a traffic load prediction performance higher than that of a target machine learning model configured to predict a communication traffic load of the target base station;
obtaining weights to be applied to the target machine learning model and the selected at least one source machine learning model via a neural network that determines the weights to the target machine learning model and the selected at least one source machine learning model, respectively for combining the target machine learning model and the selected at least one source machine learning model and that is jointly trained with the target machine learning model and the selected source machine learning models;
obtaining a load forecasting model for the target base station by combining the target machine learning model and the selected at least one source machine learning model according to the weights; and
predicting a future communication traffic load of the target base station based on the load forecasting model.
18. The non-transitory computer-readable storage medium of claim 17 , wherein the method further comprises:
transmitting information of the future communication traffic load to the target base station to enable the target base station to assign a plurality of user equipment (UEs) to a plurality of cells of the target base station according to the information of the future communication traffic load,
wherein the information of the future communication traffic load comprises any one or any combination of a number of active UEs in each of the plurality of cells, a cell load ratio, an internet protocol (IP) throughput per cell, and a cell physical resource block (PRB) usage ratio.
19. The non-transitory computer-readable storage medium of claim 17 , wherein the method further comprises:
determining whether a target domain from which the target data set is collected, and source domains from which the source data sets are collected, have a same feature space; and
prior to obtaining the weights, performing iterative feature matching on the target data set and the source data sets in response to determining that the target domain and the source domains do not have the same feature space.
20. The non-transitory computer-readable storage medium of claim 17 , wherein the method further comprises:
inputting the predicted future communication traffic load and a communication traffic load that is observed from the target base station, to a loss function configured to calculate a mean average percentage error of the load forecasting model;
determining whether a performance degradation has occurred in the load forecasting model based on the mean average percentage error of the load forecasting model; and
retraining the load forecasting model in response to determining that the performance degradation has occurred in the load forecasting model.