TW202320503A - Number of repetitions prediction method and number of repetitions prediction device - Google Patents

Abstract

A number of repetitions prediction method and number of repetitions prediction device are provided. The number of repetitions prediction method includes the following steps. A plurality of data sets respectively corresponding to a plurality of user equipments are obtained, and the data sets are dimension-reduce into a plurality of custom feature sets. The custom feature sets are divided into a training set and a test set, and a classification algorithm is used to determine a plurality of first transmittable samples from the training set. A grouping algorithm is used to group the first transmittable samples into a plurality of first groups, the first groups are respectively corresponding to a plurality of different number of repetitions, and the first groups are used to generate the number of repetitions prediction model.

Description

Retransmission times prediction method and retransmission times prediction device

The present invention relates to a prediction method and a prediction device, in particular to a retransmission times prediction method and a retransmission times prediction device.

In Ultra-reliable and Low Latency Communications (uRLLC) of the fifth generation mobile communication system, user equipment with poor communication quality needs to use more retransmissions to compensate for additional signal attenuation. When the number of retransmissions is improperly configured, the transmission error rate will be high, or a lot of precious wireless resources will be wasted.

In view of the deficiencies in the prior art, the purpose of the present invention is to provide a retransmission number prediction method and a retransmission number prediction device, which can generate a model for predicting the retransmission times of the user equipment, and even enable the base station to use the model for different communications. Quality user equipment is configured with an appropriate number of retransmissions.

To achieve the above purpose, an embodiment of the present invention provides a method for predicting the number of retransmissions, including the following steps. A plurality of data sets respectively corresponding to a plurality of user equipments are obtained, and the data sets respectively include a plurality of communication quality parameters generated when the user equipments communicate with at least one base station. Using a dimensionality reduction algorithm to analyze the communication quality parameters to reduce the dimensionality of the data sets into a plurality of custom feature sets. The custom feature sets are divided into a training set and a test set, and a classification algorithm is used to binary classify the training set according to the transferability to determine a plurality of first transferable samples. Use a grouping algorithm to group the first transmittable samples into a plurality of first groups according to the plurality of retransmission times used when the user equipments communicate with the at least one base station, and the first groups correspond to different The number of retransmissions, and use the first groups to train the machine learning model to generate a prediction model of the number of retransmissions. The retransmission times prediction model is used to predict the retransmission times used when the user equipments communicate with the at least one base station.

In addition, an embodiment of the present invention provides an apparatus for predicting the number of retransmissions, including a memory and a processor. The storage is used for storing multiple data sets respectively corresponding to multiple user equipments. The processor is electrically connected to the storage, and is used for performing the following steps. The data sets respectively corresponding to the user equipments are acquired. The data sets respectively include a plurality of communication quality parameters generated when the user equipments communicate with at least one base station. Using a dimensionality reduction algorithm to analyze the communication quality parameters to reduce the dimensionality of the data sets into a plurality of custom feature sets. The custom feature sets are divided into a training set and a test set, and a classification algorithm is used to binary classify the training set according to the transferability to determine a plurality of first transferable samples. Use a grouping algorithm to group the first transmittable samples into a plurality of first groups according to the plurality of retransmission times used when the user equipments communicate with the at least one base station, and the first groups correspond to different The number of retransmissions, and use the first groups to train the machine learning model to generate a prediction model of the number of retransmissions. The retransmission times prediction model is used to predict the retransmission times used when the user equipments communicate with the at least one base station.

In order to further understand the features and technical content of the present invention, please refer to the following detailed description and drawings related to the present invention. However, the provided drawings are only for reference and description, and are not intended to limit the present invention.

The implementation of the present invention is illustrated through specific specific examples below, and those skilled in the art can understand the advantages and effects of the present invention from the content provided in this specification. The present invention can be implemented or applied through other different specific embodiments, and various modifications and changes can be made to the details in this specification based on different viewpoints and applications without departing from the concept of the present invention. In addition, the drawings of the present invention are only for simple illustration, and are not drawn according to the actual size, which is stated in advance. The following embodiments will further describe the relevant technical content of the present invention in detail, but the provided content is not intended to limit the protection scope of the present invention.

Please refer to FIG. 1 . FIG. 1 is a schematic diagram of a wireless communication system according to an embodiment of the present invention. As shown in FIG. 1 , the wireless communication system 1 may include M base stations 11 ₁ -11 _M and N user equipments 12 ₁ -12 _N . In this embodiment, M and N are integers greater than 1, and N may not be equal to M, but the present invention is not limited thereto. In other embodiments, M may also be equal to 1, and N is an integer greater than 1. In summary, the wireless communication system 1 includes at least one base station and multiple user equipments. In addition, each user equipment in this embodiment may be, for example, a smart phone, but the present invention is not limited thereto.

More specifically, the base stations 11 ₁ - 11 _M may respectively have signal coverage areas C ₁ -C _M , and in this embodiment, the signal coverage areas of adjacent base stations may partially overlap with each other, but the present invention does not for the limit. Therefore, in this case, each user equipment in the wireless communication system 1 is located at least in one of the signal coverage areas C ₁ -C _M and communicates with the corresponding base station. In this embodiment, each user equipment uses a repetitive transmission method (referred to as retransmission) to transmit data to its communication base station. The retransmission here refers to the continuous transmission of different retransmission versions in the time domain or the frequency domain, and under the protocol of the fifth generation mobile communication system, it is allowed to configure the retransmission times of the user equipment.

It should be understood that the UE closer to the edge of the signal coverage area will have poorer communication quality and need to use more retransmissions to compensate for the additional signal attenuation. Therefore, when the number of retransmission times configured for the user equipment is insufficient, data reception at the base station will fail, that is, the transmission error rate is high. Correspondingly, the closer the user equipment is to the center of the signal coverage area, the better the communication quality will be. Therefore, when the number of retransmissions configured for such user equipment is too large, many precious radio resources will be wasted.

In order to solve the above problems, the present invention generates a retransmission times prediction model for predicting the retransmission times of the user equipment. Please refer to FIG. 2A to FIG. 2B and FIG. 3. FIG. 2A to FIG. 2B are flow charts of steps of a method for predicting the number of retransmissions according to an embodiment of the present invention, and FIG. 3 is a functional block diagram of a device for predicting the number of retransmissions according to an embodiment of the present invention. The method for predicting the number of retransmissions in FIG. 2A to FIG. 2B is applicable to the wireless communication system 1 in FIG. 1 , and can be executed by the device 3 for predicting the number of retransmissions in FIG. 3 .

In this embodiment, the retransmission times prediction device 3 may be a self-organizing network (Self-organizing Networks, SON) server, a radio intelligent controller (Radio Intelligent Controller, RIC) or any base station of the wireless communication system 1, etc. A specific machine or device, but the present invention is not limited to the specific implementation of the specific machine or device. In a word, the retransmission times prediction device 3 at least includes a storage 31 and a processor 33 .

The storage 31 can be any storage device for storing data, such as random access memory, read only memory, flash memory or hard disk, etc., but the present invention is not limited thereto. In this embodiment, the storage 31 is configured to at least store data sets S _{1 -SN} corresponding to the user equipments 12 ₁ -12 _N respectively. In addition, the processor 33 is electrically connected to the storage 31 and is used to execute the steps in FIG. 2A to FIG. 2B . As shown in FIG. 2A to FIG. 2B , in step S201, the processor 33 obtains data sets S _{1 -SN} corresponding to user equipment 12 ₁ -12 _N respectively, and data sets S _{1 -SN} respectively include user equipment 12 ₁ A plurality of communication quality parameters generated when ~12 _N communicates with at least one base station, and in step S202, the processor 33 uses a dimensionality reduction algorithm to analyze these communication quality parameters to reduce the dimensionality of the data sets S ₁ ~S _N into self Define feature sets DS ₁ -DS _N .

In more detail, each data set can be a D-dimensional parameter set, that is, each data set can include D communication quality parameters, and D is an integer greater than 1. In addition, the dimensionality reduction algorithm of this embodiment can be, for example, High Correlation Filter, Random Forests, Forward Feature Construction, Backward Feature Elimination Elimination), Missing Values Ratio, Low Variance Filter and Principal Component Analysis, but the present invention is not limited thereto. Therefore, the processor 33 can use a dimensionality reduction algorithm to analyze the D communication quality parameters (for example, analyze the correlation and/or dependency between the D communication quality parameters, etc.) to reduce the dimensionality of the data sets S _{1 -SN} as The self-defined feature sets DS ₁ -DS _N , all of which only contain K communication quality parameters, are reduced from a D-dimensional parameter set to a K-dimensional parameter set, where K is an integer smaller than D.

For example, when using the principal component analysis method to analyze these communication quality parameters, the processor 33 will establish a covariance matrix (Covariance Matrix) according to the data sets S _{1 -SN} , and decompose the covariance matrix into eigenvectors (Eigenvectors ) and eigenvalues (Eigenvalues). Next, the processor 33 selects K eigenvectors corresponding to the K largest eigenvalues, and sorts the selected K eigenvectors. Then, the processor 33 uses the sorted K feature vectors to establish a projection matrix (Project Matrix), and uses the projection matrix to transform the data sets S ₁ -S _N to obtain custom feature sets DS ₁ -DS _N .

It can be seen that the purpose of the processor 33 to analyze these communication quality parameters using a dimensionality reduction algorithm is to find out more critical parameters in the data sets S _{1 -SN} for subsequent training models, thereby avoiding using too many parameters to The overfitting phenomenon produced by the training model can improve the accuracy of machine learning. Please refer to Table 1 and Table 2 together. Table 1 is the data set of the embodiment of the present invention, and Table 2 is the custom feature set of the embodiment of the present invention.

As shown in Table 1, the communication quality parameters of each data set may at least include a reference signal received power (Reference Signal Received Power, RSRP), a received signal strength indicator (Received Signal Strength Indication, RSSI), a bit An error rate (Bit Error Rate, BER), a packet error rate (Packet Error Rate, PER) and a data rate (Data Rate), but the present invention is not limited thereto. [Table 1] RSRP RSSI BER PER Data Rate … Data set S ₁ -83dBm 5dB 0.000014 0.216 … … Data set S ₂ -92dBm 1dB 0.000010 0.114 … … Data set S ₃ -81dBm -8dB 0.000231 0.243 … … Data set S ₄ -97dBm 3dB 0.000039 0.016 … … ... ... [Table 2] RSRP RSSI BER Custom Feature Collection DS ₁ -83dBm 5dB 0.000014 Custom Feature Collection DS ₂ -92dBm 1dB 0.000010 Custom Feature Collection DS ₃ -81dBm -8dB 0.000231 Custom Feature Collection DS ₄ -97dBm 3dB 0.000039 ... ...

In addition, as shown in Table 2, after the dimension reduction processing in step S202, the processor 33 can obtain the self-defined feature sets DS ₁ -DS _N that only include the reference signal received power, received signal strength index and bit error rate. , but the present invention is not limited thereto. Next, in step S203, the processor 33 divides the self-defined feature set DS ₁ ˜DS _N into a training set Tr and a test set Te, and in step S204, uses a classification algorithm to classify the training set Tr according to the transferability. Binary classification to determine a plurality of first transmittable samples.

Please also refer to FIG. 4 . FIG. 4 is a schematic diagram of a custom feature set in an embodiment of the present invention divided into a training set and a test set. As shown in Figure 4, each custom feature set can be represented as a point in space, and the processor 33 can use random sampling (Random Sampling) to select a part from the custom feature sets DS ₁ ~DS _N as a training set Tr and the other part as the test set Te, but the present invention is not limited thereto. In a word, the present invention does not limit the specific implementation manner in which the processor 33 divides the self-defined feature set DS ₁ -DS _N into a training set Tr and a testing set Te. In addition, the classification algorithm of this embodiment can be, for example, one of Support Vector Machine, Linear Classification and K-Nearest Neighbor, but the present invention does not rely on this for the limit.

In more detail, after determining the training set Tr, the processor 33 can use a classification algorithm to classify each custom feature set in the training set Tr as the first transferable sample or the first non-transferable sample according to transferability. sample. For example, when the i-th custom feature set in the training set Tr can successfully transmit data when the number of retransmissions is more than 128, the processor 33 can classify the i-th custom feature set in the training set Tr as the first Samples can be transferred. In contrast, when the i-th custom feature set in the training set Tr cannot successfully transmit data when the number of retransmissions is greater than 128, the processor 33 can classify the i-th custom feature set in the training set Tr as the i-th custom feature set in the training set Tr. 1. Samples cannot be transmitted. Please also refer to FIG. 5 . FIG. 5 is a schematic diagram of binary classification of the training set to determine a plurality of first transmittable samples according to an embodiment of the present invention.

As shown in FIG. 5 , the processor 33 can use a distinguishing curve L to separate the first transmissible samples and the first non-transmissible samples in the training set Tr, and for the convenience of understanding, the first transmissible samples The set of and the set of the first non-transmittable samples can be denoted by symbols T11 and T12 respectively. Then, in step S205, the processor 33 uses a grouping algorithm to group the first transmittable samples into a plurality of first transmittable samples according to the plurality of retransmission times used when the user equipment 12 ₁ -12 _N communicates with the at least one base station. Groups, the first groups respectively correspond to different retransmission times, and in step S206, the processor 33 uses the first groups to train the machine learning model to generate a retransmission number prediction model. The retransmission times prediction model is used to predict the retransmission times used when the user equipments 12 ₁ -12 _N communicate with the at least one base station.

The clustering algorithm in this embodiment can be, for example, one of K-means, Agglomerative Clustering and Divisive Clustering, but the present invention is not limited thereto. Please also refer to FIG. 6 . FIG. 6 is a schematic diagram of grouping the first transmittable samples into a plurality of first groups according to an embodiment of the present invention. As shown in FIG. 6 , in this embodiment, it can be assumed that the processor 33 will use a grouping algorithm to group the first transmittable samples into three first groups G11-G13, and the first groups G11-G13 Corresponding to 128, 256 and 1024 retransmission times respectively. Therefore, after using the first groups G11-G13 to train the machine learning model, the processor 33 can obtain at least one function that can predict the number of retransmissions to be 128, 256 or 1024 according to the communication quality of the user equipment, The at least one function is a retransmission number prediction model generated by training the machine learning model. Since the training principle of the machine learning model is well known to those skilled in the art, the details thereof will not be repeated here.

It should be understood that, since the retransmission times prediction model at this time is an unoptimized model, the method for retransmission times prediction in FIG. 2A to FIG. 2B may further include steps S207-S212. In step S207, the processor 33 may use the test set Te to test the retransmission times prediction model to obtain an accuracy rate, and in step S208, determine whether the accuracy rate meets an accuracy rate standard. When the accuracy rate does not reach the accuracy rate standard, the processor 33 will first execute step S209 to select a subset of the training set Tr as the verification set Va, and in step S210, use a classification algorithm to classify the verification set Va according to the transferability Binary classification is performed to determine a plurality of second transmittable samples.

Next, in step S211, the processor 33 uses a grouping algorithm to group the second transmittable samples into a plurality of second transmittable samples according to the retransmission times used when the user equipment 12 ₁ -12 _N communicates with the at least one base station. Groups, these second groups correspond to different retransmission times respectively, and in step S212, the processor 33 uses these second groups to train the retransmission times prediction model again to update the retransmission times predictive model. After step S212, the processor 33 returns to execute steps S207-S208, and when the accuracy rate has not reached the accuracy rate standard, the processor 33 repeats steps S209-S212 and S207-S208 until the accuracy rate reaches the accuracy rate standard.

More specifically, the present invention provides three different implementation manners for how the processor 33 selects a subset of the training set Tr as the verification set Va. In the first embodiment, the processor 33 can calculate the distance between each self-defined feature set in the training set Tr and the distinguishing curve L, and select a distance smaller than a threshold from these self-defined feature sets in the training set Tr value as the validation set Va. Please also refer to FIG. 7 . FIG. 7 is a schematic diagram of a subset of the training set selected as a verification set according to the first embodiment of the present invention.

As shown in Figure 7, the custom feature set whose distance is smaller than the threshold value will be closer to the distinguishing curve L. Therefore, in the first implementation manner, the processor 33 can also be regarded as selecting a plurality of custom feature sets that are closer to the distinguishing curve L from the training set Tr as the verification set Va. It should be noted that, for the processor 33 at this time, the closer the user-defined feature set is to the distinguishing curve L, the more difficult it is to accurately classify it as the first transmissible sample or the first non-transmissible sample. Therefore, the purpose of the processor 33 selecting a plurality of custom feature sets closer to the distinguishing curve L as the verification set Va is to use these custom feature sets to determine a new distinguishing curve L.

Please also refer to FIG. 8 . FIG. 8 is a schematic diagram of determining a plurality of second transferable samples from the validation set in FIG. 7 through binary classification. As shown in FIG. 8, after the verification set Va is determined, the processor 33 can use the classification algorithm to classify each custom feature set in the training verification set Va as the second transferable sample or the second non-transferable sample. . Therefore, the processor 33 can use the self-defined feature sets in the verification set Va to determine a new distinguishing curve L, and use the new distinguishing curve L to separate the second transmittable samples in the verification set Va from the Some second non-transmissible samples.

It can be seen that if these self-defined feature sets in the verification set Va are used to determine the new distinguishing curve L, then these self-defined feature sets in the verification set Va can be more accurately classified as the second transferable sample or the first 2. Samples cannot be transmitted. On the other hand, in the second implementation manner, the processor 33 may determine layers with different application types from the training set Tr, and select at least one custom feature set from each layer as the verification set Va. Specifically, each custom feature set may have application type information for indicating the application type of the corresponding user equipment. Therefore, the processor 33 can group the custom feature sets in the training set Tr into multiple third groups according to the application type information of each custom feature set, and the third groups correspond to different There are multiple application types, and at least one is selected from the custom feature sets of each third group as the verification set Va.

Please also refer to FIG. 9 . FIG. 9 is a schematic diagram of a subset of the training set selected as the verification set according to the second embodiment of the present invention. As shown in FIG. 9 , in this embodiment, it can be assumed that the processor 33 will group the custom feature sets in the training set Tr into three third groups G31-G33, and the third groups G31-G33 are respectively Corresponding to the first application type, the second application type and the third application type. For the convenience of understanding, the self-defined feature sets of the third group G31 are represented by square symbols, the self-defined feature sets of the third group G32 are represented by triangle symbols, and the third group G33 These custom feature sets are represented by circular symbols. Therefore, in the second implementation manner, the third groups G31-G33 can also be regarded as stratifications with different application types, and the processor 33 can use random sampling to select from the self-groups of each third group. Select at least one of the defined feature sets as the verification set Va.

It can be seen that the purpose of the processor 33 using the second implementation is to break the dependency of application types, so that the processor 33 can take into account the influence of the application type on the retransmission times when updating the retransmission times prediction model. Similarly, in the third implementation manner, the processor 33 may determine clusters with different regions from the training set Tr, and select at least one custom feature set from each cluster as the verification set Va. Specifically, each custom feature set may have area information, which is used to indicate the area where the corresponding user equipment is located, such as the signal coverage of which base station it is located in, but the present invention is not limited thereto. Therefore, the processor 33 can group these self-defined feature sets in the training set Tr into multiple third groups according to the area information of each self-defined feature set, and these third groups correspond to different multi-groups respectively. regions, and at least one of these custom feature sets of each third group is selected as the verification set Va.

Please also refer to FIG. 10 . FIG. 10 is a schematic diagram of a subset of the training set selected as the verification set according to the third embodiment of the present invention. As shown in FIG. 10 , in this embodiment, it can be assumed that the processor 33 will group the custom feature sets in the training set Tr into three third groups G34-G36, and the third groups G34-G36 are respectively Corresponds to the first area, the second area and the third area. For the convenience of understanding, the self-defined feature sets of the third group G34 are represented by square symbols, the self-defined feature sets of the third group G35 are represented by triangle symbols, and the third group G36 These custom feature sets are represented by circular symbols. Therefore, in the third implementation manner, the third groups G34-G36 can also be regarded as clusters with different regions, and the processor 33 can use random sampling to obtain the self-defined features from each third group Select at least one of them as the verification set Va.

It can be seen that the purpose of the processor 33 using the third embodiment is to break the interdependence of the regions, so that the processor 33 can take into account the influence of the region on the retransmission times when updating the retransmission times prediction model. In addition, after determining the verification set Va no matter whether the second or third implementation is used, the processor 33 will also use the classification algorithm to classify each custom feature set in the training verification set Va as the second A transmissible sample or a second non-transmissible sample.

Next, in this embodiment, it may be assumed that the processor 33 will use a grouping algorithm to group the second transmittable samples into three second groups G21 - G23 . The second groups G21 - G23 may also correspond to 128, 256 and 1024 retransmission times respectively, but the present invention is not limited thereto. Therefore, after using the second groups G21 - G23 to train the machine learning model again, the processor 33 can optimize the retransmission times prediction model, thereby improving the accuracy of the retransmission times prediction model. Finally, when the accuracy rate reaches the accuracy rate standard, the processor 33 may execute step S213 to output the at least one base station with a prediction model of the number of retransmissions, so that the at least one base station can configure user equipments with different communication qualities according to the prediction model of the number of retransmissions Appropriate number of retransmissions.

To sum up, one of the beneficial effects of the present invention is to use the dimensionality reduction algorithm to find the more critical parameters in the data set of the user equipment, thereby avoiding the phenomenon of over-fitting caused by training the model with too many parameters. In addition, if the accuracy rate of the retransmission number prediction model does not meet the accuracy rate standard, the retransmission number prediction method and the retransmission number prediction device of the present invention will further consider other factors (such as distance from the distinguishing curve, application type or area) To select a subset of the training set as the verification set, and use the verification set to optimize the model to improve the accuracy of the retransmission prediction model. In this way, the base station can also configure appropriate retransmission times for user equipments with different communication qualities according to the retransmission times prediction model of the present invention, so as to reduce transmission error rate and improve resource utilization and transmission rate.

The content provided above is only a preferred feasible embodiment of the present invention, and does not therefore limit the scope of the patent application of the present invention, so all equivalent technical changes made by using the description and drawings of the present invention are included in the application of the present invention within the scope of the patent.

1: wireless communication system 11 ₁ ~11 _M : base station 12 ₁ ~12 _N : user equipment C ₁ ~C _M : signal coverage S ₁ ~S _N : data set DS ₁ ~DS _N : custom feature set 3: heavy Transmission number prediction device 31: storage 33: processor Tr: training set Te: test set L: distinguishing curve T11: first set of transmittable samples T12: first set of non-transmittable samples G11~G13: first group Va: verification set G21~G23: second group G31~G33, G34~G36: third group S201~S213: process steps

FIG. 1 is a schematic diagram of a wireless communication system according to an embodiment of the present invention.

FIG. 2A to FIG. 2B are flowcharts of the steps of the method for predicting the number of retransmissions according to the embodiment of the present invention.

FIG. 3 is a schematic functional block diagram of an apparatus for predicting the number of retransmissions according to an embodiment of the present invention.

Fig. 4 is a schematic diagram of a custom feature set in an embodiment of the present invention divided into a training set and a test set.

FIG. 5 is a schematic diagram of binary classification of a training set to determine a plurality of first transmittable samples according to an embodiment of the present invention.

FIG. 6 is a schematic diagram of grouping first transmittable samples into a plurality of first groups according to an embodiment of the present invention.

FIG. 7 is a schematic diagram of a subset of the training set selected as a verification set according to the first embodiment of the present invention.

FIG. 8 is a schematic diagram of binary classification of the verification set in FIG. 7 to determine a plurality of second transferable samples.

FIG. 9 is a schematic diagram of a subset of the training set selected as a verification set according to the second embodiment of the present invention.

FIG. 10 is a schematic diagram of a subset of the training set selected as the verification set according to the third embodiment of the present invention.

S201~S208, S213: process steps

Claims (14)

A method for predicting the number of retransmissions, comprising the following steps: Obtaining multiple data sets respectively corresponding to multiple user equipments, wherein the data sets respectively include multiple communication quality parameters generated when the user equipments communicate with at least one base station; Analyzing the communication quality parameters using a dimensionality reduction algorithm to reduce the dimensionality of the data sets into a plurality of custom feature sets; dividing the custom feature sets into a training set and a test set, and using a classification algorithm to binary classify the training set according to the transferability to determine a plurality of first transferable samples; and using a grouping algorithm to group the first transmittable samples into a plurality of first groups according to the plurality of retransmission times used when the user equipments communicate with the at least one base station, and the first groups respectively correspond to different retransmission times, and use the first groups to train a machine learning model to generate a retransmission number prediction model, wherein the retransmission number prediction model is used to predict the relationship between the user equipments and the at least one The number of retransmissions used by the base station for communication. The method for predicting the number of retransmissions described in claim 1 further includes the following steps: Using the test set to test the retransmission times prediction model to obtain an accuracy rate, and judging whether the accuracy rate meets an accuracy rate standard. The method for predicting the number of retransmissions as described in claim 2, wherein when the accuracy rate does not meet the accuracy standard, the method for predicting the number of retransmissions further includes the following steps: selecting a subset of the training set as a validation set, and using the classification algorithm to binary classify the validation set according to the transferability to determine a plurality of second transferable samples; and using the grouping algorithm to group the second transmittable samples into a plurality of second groups according to the retransmission times used when the user equipments communicate with the at least one base station, and the second groups respectively correspond to different retransmission times, and use the second groups to train the machine learning model again to update the retransmission number prediction model. The method for predicting the number of retransmissions as described in Claim 1, wherein the communication quality parameters include a reference signal received power, a received signal strength index, a bit error rate, a packet error rate, and a data rate. The method for predicting the number of retransmissions as described in claim item 3, wherein the step of selecting the subset of the training set as the verification set includes: Calculate a distance between each of the self-defined feature sets in the training set and a distinguishing curve, and select those with the distance smaller than a threshold value from the self-defined feature sets in the training set as the verification set. The method for predicting the number of retransmissions as described in claim 3, wherein each of the custom feature sets has an application type information, the application type information is used to indicate an application type of the corresponding user equipment, and the training set is selected The steps for using this subset as the validation set include: According to the application type information of each of the custom feature sets, the custom feature sets in the training set are grouped into a plurality of third groups, and from the self-defined feature sets of each of the third groups At least one of the defined feature sets is selected as the verification set. The method for predicting the number of retransmissions as described in claim 3, wherein each of the self-defined feature sets has an area information, the area information is used to indicate an area where the corresponding user equipment is located, and the training set is selected The steps for subsetting this validation set include: According to the area information of each of the custom feature sets, the custom feature sets in the training set are grouped into a plurality of third groups, and the custom feature sets of each of the third groups are selected At least one of the feature sets is selected as the verification set. A device for predicting the number of retransmissions, comprising: a storage for storing a plurality of data sets respectively corresponding to a plurality of user equipments; and A processor is electrically connected to the storage and configured to perform the following steps: Obtain the data sets respectively corresponding to the user equipments, wherein the data sets respectively include a plurality of communication quality parameters generated when the user equipments communicate with at least one base station; Analyzing the communication quality parameters using a dimensionality reduction algorithm to reduce the dimensionality of the data sets into a plurality of custom feature sets; dividing the custom feature sets into a training set and a test set, and using a classification algorithm to binary classify the training set according to the transferability to determine a plurality of first transferable samples; and using a grouping algorithm to group the first transmittable samples into a plurality of first groups according to the plurality of retransmission times used when the user equipments communicate with the at least one base station, and the first groups respectively correspond to different retransmission times, and use the first groups to train a machine learning model to generate a retransmission number prediction model, wherein the retransmission number prediction model is used to predict the relationship between the user equipments and the at least one The number of retransmissions used by the base station for communication. The device for predicting the number of retransmissions as described in Claim 8, wherein the processor further performs the following steps: Using the test set to test the retransmission times prediction model to obtain an accuracy rate, and judging whether the accuracy rate meets an accuracy rate standard. The device for predicting the number of retransmissions as described in Claim 9, wherein when the accuracy rate does not meet the accuracy rate standard, the processor further performs the following steps: selecting a subset of the training set as a validation set, and using the classification algorithm to binary classify the validation set according to the transferability to determine a plurality of second transferable samples; and using the grouping algorithm to group the second transmittable samples into a plurality of second groups according to the retransmission times used when the user equipments communicate with the at least one base station, and the second groups respectively correspond to different retransmission times, and use the second groups to train the machine learning model again to update the retransmission number prediction model. The device for predicting the number of retransmissions as described in Claim 8, wherein the communication quality parameters include a reference signal received power, a received signal strength index, a bit error rate, a packet error rate and a data rate. The device for predicting the number of retransmissions as described in claim 10, wherein the step of selecting the subset of the training set as the verification set includes: Calculate a distance between each of the self-defined feature sets in the training set and a distinguishing curve, and select those with the distance smaller than a threshold value from the self-defined feature sets in the training set as the verification set. The device for predicting the number of retransmissions as described in claim 10, wherein each of the custom feature sets has an application type information, the application type information is used to indicate an application type of the corresponding user equipment, and the training set is selected The steps for using this subset as the validation set include: According to the application type information of each of the custom feature sets, the custom feature sets in the training set are grouped into a plurality of third groups, and from the self-defined feature sets of each of the third groups At least one of the defined feature sets is selected as the verification set. The device for predicting the number of retransmissions as described in claim item 10, wherein each of the self-defined feature sets has a region information, and the region information is used to indicate a region where the corresponding user equipment is located, and select the training set The steps for subsetting this validation set include: According to the area information of each of the custom feature sets, the custom feature sets in the training set are grouped into a plurality of third groups, and the custom feature sets of each of the third groups are selected At least one of the feature sets is selected as the verification set.

Images