CN113780398A - Wireless network link quality prediction method and system - Google Patents
Wireless network link quality prediction method and system Download PDFInfo
- Publication number
- CN113780398A CN113780398A CN202111027557.0A CN202111027557A CN113780398A CN 113780398 A CN113780398 A CN 113780398A CN 202111027557 A CN202111027557 A CN 202111027557A CN 113780398 A CN113780398 A CN 113780398A
- Authority
- CN
- China
- Prior art keywords
- data
- wireless network
- link quality
- model
- characteristic
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses a wireless network link quality prediction method and a system, belonging to the technical field of wireless network link quality prediction and comprising the following steps: s1: data acquisition and pretreatment; s2: feature engineering and construction sample labels; s3: training a model; s4: wireless network link quality is predicted. According to the invention, a prediction model of the wireless network link quality is established, the quality of the wireless link can be predicted more efficiently, the service application can be found to sense the wireless link state more quickly, the user service code rate or the cache setting is adjusted in advance, video blocking and cache delay are avoided, and the user experience of the video service is effectively improved; various characteristics in the wireless link time sequence data are extracted, a fusion regression algorithm is adopted, the quality of the wireless link is analyzed and predicted from influence factors in various aspects, the quality of the wireless network link is more accurately reflected, a prediction result is decomposed and opened to a service, the optimization of a service side is facilitated, the cross-layer intelligent optimization of user experience is realized, and the service perception of a user is improved.
Description
Technical Field
The invention relates to the technical field of wireless network link quality prediction, in particular to a wireless network link quality prediction method and a wireless network link quality prediction system.
Background
5G is being gradually deployed and commercialized as a focus of development of the current information communication industry. The 5G service application scenes are more and more diverse, and different scenes have differentiated service requirements. In order to improve the future wireless network service capability and provide deterministic service guarantee for users, the network needs to have the multidimensional intelligent environment perception capability from a resource layer to a service layer, so that the user experience cross-layer intelligent optimization of network and service cooperation is achieved. The downlink rate of the user service is a main factor influencing the experience of the user video service. The real-time prediction of the user downlink rate can help business application to sense the state of a wireless link, the business code rate or the cache setting of the user is adjusted in advance, video blocking and cache delay are avoided, and the user experience of the video business is effectively improved.
The user service downlink rate is a measurement index of the user wireless link quality, and the user service downlink rate is predicted, so that the early warning of network quality faults or abnormalities is realized. During operation, the wireless network generates network-side parameters such as: the uplink signal to interference plus noise ratio, the MAC downlink rate information, the total number of downlink PDCPSDU packets, the downlink modulation and coding strategy, the downlink PRB occupation number of the base station and the like. The traditional wireless network quality prediction is based on historical sampling values of user service downlink rate data, and adopts a time sequence algorithm to predict data of the next time point, and has the defects of low prediction efficiency, poor accuracy of prediction results, single considered influence factor and the like. According to the conventional method, the order of a model is determined according to different time sequences by utilizing a traditional time sequence algorithm ARIMA, and then data fitting is carried out, so that the number of users is large, the order and fitting need to be determined one by one, and the prediction efficiency is low. In addition, although prediction is performed based on network-side parameter values using a regression tree in machine learning, characteristics in terms of time continuity are easily ignored, and the effect is not good. Therefore, the application range thereof has a certain limitation. Therefore, a wireless network link quality prediction method and system based on fusion of a time sequence convolutional neural network and a lightGBM algorithm are provided.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: the method solves the problems that the existing prediction method is low in prediction efficiency, can only capture the linear relation of data or neglects the time sequence characteristics and the like, provides a wireless network link quality prediction method, builds a prediction model of user service downlink rate data, can predict the operation quality of a wireless link more efficiently and can solve the problems in the operation process of the wireless link more quickly; various characteristics in the time series data of the wireless link are extracted, the quality of the wireless link is analyzed and predicted from influence factors in various aspects by adopting a method of fusing a time sequence convolutional neural network and a lightGBM algorithm, the operation quality of the wireless link is more accurately reflected, an operator is helped to find and solve problems in advance, the optimization of a service side is helped to realize cross-layer intelligent optimization of user experience, and the service perception of the user is improved.
The invention solves the technical problems through the following technical scheme, and the invention comprises the following steps:
s1: data acquisition and preprocessing
Acquiring network side parameters of a wireless network link and historical data of user service downlink rate, preprocessing the data, and generating a historical data table of wireless network link quality;
s2: feature engineering and construction sample labels
Constructing a sample label by using the user service downlink rate, performing feature engineering according to network side parameter historical data of the wireless network link, performing feature extraction, construction and extraction, and generating a historical data table containing wireless network link quality data features and labels of each user;
s3: model training
Outputting a training sample set with a label according to the characteristic engineering, and adopting an algorithm training model integrating a time sequence convolutional neural network and a lightGBM algorithm to obtain a prediction model;
s4: predicting wireless network link quality
And outputting a test sample set without a label according to the characteristic engineering, performing service downlink rate prediction on the test sample by using the prediction model, and evaluating the model.
Further, in the step S1, the network side parameters of the wireless network link include data collection time, cell or user identifier, network side parameter name and network side parameter value data.
Further, in step S1, the collected historical data is the historical data of the network side parameters of the wireless network link and the user service downlink rate collected a period of time before the time to be predicted.
Further, the specific process in step S1 is as follows:
s11: firstly, acquiring millisecond-level network side data and user downlink rate data in different time periods;
s12: and processing the abnormal data value and the null value of the data to generate a historical data table of the wireless network link quality.
Further, in step S2, feature engineering is to convert the raw data into a data set for model training; the characteristic engineering sequentially comprises three steps of characteristic selection, characteristic construction and characteristic extraction; the characteristics are selected as characteristics for deleting redundancy or irrelevance; the feature construction is to construct a new feature manually on the basis of original data; the feature extraction is used for converting original data into data features which can be identified by a model in a mode based on principal component analysis and time window mapping.
Further, in the step S3, the process of training the model is to use the sample feature and tag data table generated in the step S2 as a training set, construct a wireless network link quality data prediction model based on a fusion method of the machine learning time series convolutional neural network and the lightGBM algorithm, and predict quality data of the wireless network link for a period of time in the future.
Further, in step S4, the predicting the wireless network link quality data specifically includes the following steps:
s41: constructing a characteristic data table of wireless network link quality data to be predicted;
s42: and inputting the characteristic data of the wireless network link quality data to be predicted into the prediction model, and outputting the prediction result of the user service downlink rate data.
Further, in the step S41, the manner of constructing the characteristic data table of the wireless network link quality data to be predicted is the same as that in the step S2, but no sample label is set.
Further, the steps S1 to S4 are a cyclic process, and the network side parameter data of the wireless network link and the user service downlink rate data of a period are adopted to construct data and perform feature engineering, construct a data set, obtain a labeled training sample and train a model, predict the user service downlink rate of the unlabeled test sample by using the trained model, and evaluate the model.
The invention also provides a system for predicting the quality of the wireless link, which adopts the prediction method to predict the quality of the wireless link and comprises the following steps:
the acquisition and preprocessing module is used for acquiring network side parameters of a wireless network link and historical data of user service downlink rate, preprocessing the data and generating a historical data table of the quality of the wireless network link;
the system comprises a label and characteristic construction module, a label and characteristic construction module and a characteristic construction module, wherein the label and characteristic construction module is used for constructing a sample label by utilizing the user service downlink rate, performing characteristic engineering according to network side parameter historical data of a wireless network link, performing characteristic extraction, construction and extraction, and generating a historical data table containing wireless network link quality data characteristics and labels of each user;
the model training module is used for outputting a training sample set with a label according to the characteristic engineering, and adopting an algorithm training model integrating a time sequence convolutional neural network and a lightGBM algorithm to obtain a prediction model;
the quality prediction module is used for outputting a test sample set without a label according to the characteristic engineering, performing service downlink rate prediction on the test sample by using a prediction model, and evaluating the model;
the central processing module is used for sending instructions to each module to complete related actions;
the acquisition and preprocessing module, the label and feature construction module, the model training module and the quality prediction module are all electrically connected with the central processing module.
Compared with the prior art, the invention has the following advantages: the wireless network link quality prediction method can predict the quality of a wireless network link from multidimensional index data based on the network side parameters of the wireless network link and the user service downlink rate; compared with the traditional time sequence prediction algorithm, the method of fusing the time sequence convolutional neural network and the lightGBM algorithm is adopted, the quality of the wireless link is analyzed and predicted from multiple influencing factors, the training efficiency is high, and model training is only needed to be performed once in each prediction, so that the prediction efficiency is improved to a great extent; the regression algorithm is adopted to predict the time sequence, besides the influence of time factors, factors such as correlation among users in different cells are also considered, more correlation characteristics can be expanded according to actual service scenes, the requirement on time sequence data is low, the nonlinear relation among the data can be found, the accuracy and the reliability of a prediction result are improved, the operation quality of a wireless link is more accurately reflected, an operator is helped to find and solve problems in advance, the optimization of a service side is facilitated to realize the cross-layer intelligent optimization of user experience, the service perception of a user is improved, and the method is worthy of popularization and use.
Drawings
Fig. 1 is a flowchart illustrating a wireless network link quality prediction method according to a second embodiment of the present invention.
Detailed Description
The following examples are given for the detailed implementation and specific operation of the present invention, but the scope of the present invention is not limited to the following examples.
Example one
The embodiment provides a technical scheme: a wireless network link quality prediction method comprises the following steps:
s1: acquiring and preprocessing the wireless network link quality and network side parameter historical data;
s2: constructing a sample label and sample characteristics according to the user service downlink rate and the historical data of the network side parameters;
s3: training a model by adopting a lightGBM algorithm and TNC algorithm fusion method according to the extracted features;
s4: and predicting user service downlink rate data in a future period of time by using the model.
In step S1, the collected historical data is data of the downlink rate of the wireless link network side and the user service at a time before the time to be predicted.
In step S1, the data preprocessing is to perform outlier and null processing on the network-side data of the wireless network link. In the step S2, the feature engineering includes three parts, feature selection, feature construction, and feature extraction. The feature selection comprises deleting the features with missing values exceeding 50%, deleting constant values exceeding 90%, and deleting redundant or irrelevant features; the feature construction is mainly to construct new features by manpower, such as elementary transformation between features, feature statistics in a unit period: data characteristics of minimum value, quarter fraction, median, mean, three quarters to three, maximum value, and the like; the feature extraction is mainly based on a PCA method to extract features, and original data which cannot be identified and processed by a machine learning algorithm is converted into unidentifiable data features.
In step S3, the sample label is the user traffic downlink rate data at a certain time.
In step S3, the sample feature generation labeled data table generated in step S2 is used as a training set, and a wireless link quality data prediction model is constructed based on a method of fusing the time-series convolutional neural network and the machine learning lightGBM algorithm, so as to predict quality data of the wireless link for a period of time in the future.
In step S4, the sample features generated in step S2 are generated into a test sample set without labels, a prediction model is called to predict the test sample set, the model is evaluated, and whether feature optimization or model optimization is performed is determined according to the evaluation result of the model.
The step S4 of predicting the downlink rate data of the user service mainly includes the following steps:
s41: constructing a characteristic data table of user service downlink rate data to be predicted;
s42: and inputting the characteristic data table of the user service downlink rate data to be predicted into the prediction model, and outputting the prediction result of the user service downlink rate data.
In the step S41, the manner of constructing the downlink rate data characteristic of the user traffic to be predicted is consistent with the manner described in the step S2.
The steps S1 to S4 are a cyclic process, wireless network side parameter data and user service downlink rate data of a period are adopted, data are built, characteristic engineering is carried out, a data set is constructed, a labeled training sample is obtained, a model is trained, the user service downlink rate of a label-free test sample is predicted by using the trained model, and the model is evaluated; if the evaluation effect is not good, the model can be evaluated by training again in modes of optimizing feature engineering or optimizing the model and the like to form a closed loop.
In step S4, the sample features generated in step S2 are generated into a test sample set without labels, a prediction model is called to predict the test sample set, the model is evaluated, and whether feature optimization or model optimization is performed is determined according to the evaluation result of the model.
Example two
As shown in fig. 1, a method for predicting the link quality of a wireless network based on the fusion of a time-series convolutional neural network and a lightGBM algorithm includes the following steps:
s1: acquiring and preprocessing the wireless network link quality and network side parameter historical data;
the specific process is as follows:
firstly, network side data and user downlink rate data of millisecond levels in different time periods are collected. Table 1 is an example of network side data, table 2 is a table of network parameters, and table 3 is an example of a sample tag table containing user downlink rate data.
Table 1 network side data example
| TimeStamp | UeCRNTI | ParamName | ParamValue | ID |
| 1618377939 | 2787 | DLMACRate | 5511167 | 2787 |
| 1618377939 | 2787 | MCS | 19 | 2787 |
| 1618377939 | 2787 | PDCPOccupBuffer | 44 | 2787 |
| 1618377939 | 2787 | ULSINR | 3072 | 2787 |
| 1618377939 | 2787 | DLPDCPSDUNum | 599 | 2787 |
| 1618377939 | 0 | DLOccupyPRBNum | 72 | 2787 |
| 1618377939 | 0 | CellDLMACRate | 34562607 | 2787 |
Wherein TimeStamp represents data acquisition time; UeCRTI is user identification, 0 represents cell identification, corresponding network parameters are cell level, non-0 represents user identification, and corresponding network parameters are UE level.
Table 2 network parameter table
Wherein, ParamValue corresponds to the value of each parameter at the current TimeStamp time. The unit of the TimeStamp is millisecond, all network parameters are collected once per second, and the time error in millisecond level can be ignored.
Table 3 example of sample tag table for user downlink rate data
| TimeStamp | dlBw | UeCRNTI |
| 1618378059 | 2529 | 2787 |
| 1618378060 | 5439 | 2787 |
| 1618378062 | 5532 | 2787 |
| 1618378064 | 2266 | 2787 |
| 1618378066 | 2333 | 2787 |
| 1618378068 | 1667 | 2787 |
| 1618378070 | 3463 | 2787 |
| 1618378072 | 2046 | 2787 |
| 1618378074 | 1818 | 2787 |
And the network side data and the user downlink rate are associated with the UeCRTI through a time stamp.
The historical data preprocessing mainly comprises data abnormal value and null value processing. The data outlier and null processing comprises outlier replacement and null mean replacement. Finally, a historical data table of the wireless link quality is generated.
S2: constructing a sample label and sample characteristics according to the user service downlink rate and the historical data of the network side parameters;
the specific process is as follows:
s21: setting a sample label
The user service downlink rate is a main factor influencing the user video service experience; predicting the user service downlink rate in real time can help service application to sense the state of a wireless link, adjust the user service code rate or the cache setting in advance, avoid video blocking and cache delay, and effectively improve the user experience of the video service. Therefore, the user service downlink rate is taken as a label of the sample and is taken as an index for measuring the quality of the user wireless link.
S22: feature engineering-feature selection
Firstly, a network side parameter table is made according to a time stamp, a UeCRTI and a network side parameter value in the network side parameter table.
Table 4 network side parameter table
Deleting the characteristics of missing values exceeding 50 percent, deleting constant values exceeding 90 percent, and deleting redundancy or irrelevance from the characteristics of network side parameter values CellDLMACRate, DLOccupyPRBNum, ULSINR, DLMACRate, MCS and the like;
s22: feature engineering-feature construction
The feature construction is mainly to construct new features manually, and the main constructed features are as follows:
1. the elementary transformation between the parameter characteristics of the network side mainly comprises the addition, the subtraction, the multiplication and the division between the characteristics; form feature table 0, denoted df0_ feat;
2. the ring ratio rate of change of the network-side parameter characteristics,
form feature table 1, denoted df1_ feat;
3. taking the time stamp of the label sample as a time point, extracting a data set of parameter values of the network side parameters of the previous N seconds, and recording the data set as df 2; firstly, converting rows into columns for each network side parameter for data in df2 to form N characteristic columns, and forming a characteristic sub table 1, which is recorded as df2_ 1; then taking the minimum value, the maximum value, the mean value, the median and the variance sum of parameter value statistics in N seconds as 6 statistical characteristic columns to form a characteristic sub-table 2 which is recorded as df2_ 2; then, the maximum time difference between the time stamp of the tag sample and df2 is used as a time feature list to form a feature sub-table 3, which is designated as df2_ 3; then, the hour, minute, second and millisecond of the tag TimeStamp are used as a time feature column to form a feature sub-table 4, which is recorded as df2_4, and finally, the sub-tables df2_1, df2_2, df2_3 and df2_4 are associated with the sample tag through the TimeStamp (TimeStamp) and the UeRNTI to form a feature table 2, which is recorded as df2_ feat.
4. Taking the time stamp of the label sample as a time point, decomposing the time stamp into hours, minutes, seconds and milliseconds, counting the data characteristics of the minimum value, the quartile, the median, the mean value, the third quarter and the maximum value of the parameter values of the network side parameters under different dimensions, wherein the data characteristics comprise situation characteristic sub-tables under 5 different dimensions of hour-minute-second, hour-minute, hour, minute and second, and then associating the sub-tables with the label sample set to form a characteristic table 3 which is recorded as df3_ feat;
5. and taking the time stamp of the label sample as a time point, extracting a data set of parameter values of the network side parameters 2 seconds before and after the time point, and filling the data set with missing values when the missing condition (namely the condition that the adjacent time point is more than or equal to 2 seconds) occurs, wherein the missing values are recorded as df 4. Firstly, carrying out numerical difference on each network side parameter for data in df4, and then converting rows into columns to form a characteristic sub-table 1, which is recorded as df4_ 1; then, taking the minimum value, the maximum value, the mean value, the median and the variance sum of the df4 parameter value statistics of each column as 6 statistical characteristic columns to form a characteristic sub-table 2, and recording the characteristic sub-table as df4_ 2; (ii) a Then, the sub-tables df4_1 and df4_2 are associated with the label sample set to form a feature table 4, which is marked as df4_ feat;
6. the method comprises the steps of carrying out downsampling on network side parameter tables and sample label data, collecting data with frequencies of 5S, 10S, 20S, 30S, 1Min, 2Min, 3Min, 4Min and 5Min, carrying out six statistical characteristics (minimum value, quartile, median, mean, three quarters and maximum value) on each network side parameter, associating subtables with different frequencies with sample label samples to form a characteristic table 5, and recording the characteristic table as df5_ feat;
7. and taking the time stamp of the label sample as a time point, performing characteristic construction on the user service downlink rate 24 hours before the time point, and extracting a data set of the user service downlink rate N seconds before, and recording the data set as df 6. Firstly, converting rows into columns for the downlink rate of user service of data in df6 to form N characteristic columns, and forming a characteristic sub-table 1, which is recorded as df6_ 1; then taking the minimum value, the maximum value, the mean value, the median and the variance sum of parameter value statistics in N seconds as 6 statistical characteristic columns to form a characteristic sub-table 2 which is recorded as df6_ 2; then, counting the data characteristics of the minimum value, the quartile, the median, the mean value, the quarter-third and the maximum value of the parameter values of the network side parameters under 5 different dimensions of hour-minute-second, hour-minute, hour, minute and second to form a characteristic sublist 3 which is recorded as df6_ 3; then, sub-tables df6_1, df6_2 and df6_3 are associated with the label sample set, and finally a feature table 6 is formed and is recorded as df6_ feat;
s23: feature engineering-feature extraction
Performing feature extraction on each statistical feature with network side parameter frequencies of 5S, 10S, 20S, 30S, 1Min, 2Min, 3Min, 4Min and 5Min in df5_ feat through Principal Component Analysis (PCA), reducing the dimension into m dimension, and forming a feature table df5_ PCA _ feat;
s24: formation of the broad table:
the data sets of df0_ feat, df1_ feat, df2_ feat, df3_ feat, df4_ feat, df5_ pca _ feat, df6_ feat and the sample tags are associated by a TimeStamp (TimeStamp) and a UeRNTI to form a wide table 1, denoted final _ df _ feat;
s3: training a model by adopting a lightGBM algorithm and TNC algorithm fusion method according to the extracted features;
and (4) training the final _ df _ feat characteristic data constructed in the step (S2) by using a machine learning algorithm lightGBM to obtain a lightGBM model. The lightGBM model mainly uses the negative gradient of the penalty function as the residual approximation of the current decision tree to fit the new decision tree. And simultaneously, training final _ df _ feat characteristic data constructed in the step S2 by adopting a time sequence convolutional neural network (TCN) to obtain a TCN model. The TCN model is mainly composed of Causal Convolution (cause Convolution), expanded Convolution (scaled Convolution) and Residual Connection (Residual Connection), wherein the Causal Convolution is a value at the time t of the previous layer and depends only on the values at the time t and before the time t of the next layer. Unlike conventional convolutional neural networks, causal convolution does not see future data, is a unidirectional structure, and is not bidirectional. That is, only the former cause has the latter effect, and the model is a strict time constraint model. The dilation convolution further solves the problem that the modeling length of the causal convolution to the time is limited by the size of a convolution kernel, and allows the input during convolution to have interval sampling, so that a large receptive field can be obtained. Residual chaining, in turn, is an efficient method of training deep networks, allowing the networks to pass information in a cross-layer manner.
S4: predicting user service downlink rate data in a future period of time by using a model;
for constructing the feature data of a period of time to be predicted using the same method as step S2, no sample label is set. Inputting the constructed feature data into the lightGBM model and the TCN model established in step S3, outputting predicted values for predicting a period of time, and averaging the predicted values to obtain a final predicted value.
To sum up, the method for predicting quality of a wireless network link based on fusion of a time-series convolutional neural network and a lightGBM algorithm according to the above embodiment utilizes fusion of the time-series convolutional neural network and the machine-learned lightGBM regression algorithm, can predict quality of the wireless link from multidimensional index data based on wireless network side parameters and historical user service downlink rates, realizes prediction of a time series, considers factors such as spatial dimensions, correlation among users, cells and network side parameters in addition to influence of time factors, can expand more relevant features according to an actual service scenario, has low requirement on time-series data, can find nonlinear relationship among data, improves accuracy and reliability of prediction results, thereby more accurately reflecting operation quality of the wireless link, and helps operators to find problems in advance, The method and the device solve the problems, are beneficial to the optimization of the service side so as to realize cross-layer intelligent optimization of user experience and improve the service perception of the user.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.
Claims (10)
1. A wireless network link quality prediction method is characterized by comprising the following steps:
s1: data acquisition and preprocessing
Acquiring network side parameters of a wireless network link and historical data of user service downlink rate, preprocessing the data, and generating a historical data table of wireless network link quality;
s2: feature engineering and construction sample labels
Constructing a sample label by using the user service downlink rate, performing feature engineering according to network side parameter historical data of the wireless network link, performing feature extraction, construction and extraction, and generating a historical data table containing wireless network link quality data features and labels of each user;
s3: model training
Outputting a training sample set with a label according to the characteristic engineering, and adopting an algorithm training model integrating a time sequence convolutional neural network and a lightGBM algorithm to obtain a prediction model;
s4: predicting wireless network link quality
And outputting a test sample set without a label according to the characteristic engineering, performing service downlink rate prediction on the test sample by using the prediction model, and evaluating the model.
2. The method of claim 1, wherein the step of predicting the link quality of the wireless network comprises: in step S1, the network side parameters of the wireless network link include data collection time, cell or user identifier, network side parameter name, and network side parameter value data.
3. The method of claim 1, wherein the step of predicting the link quality of the wireless network comprises: in step S1, the collected historical data is the historical data of the network side parameters of the wireless network link and the user service downlink rate at a time before the time to be predicted.
4. The method of claim 1, wherein the step of predicting the link quality of the wireless network comprises: the specific process in step S1 is as follows:
s11: firstly, acquiring millisecond-level network side data and user downlink rate data in different time periods;
s12: and processing the abnormal data value and the null value of the data to generate a historical data table of the wireless network link quality.
5. The method of claim 1, wherein the step of predicting the link quality of the wireless network comprises: in step S2, feature engineering is to convert the raw data into a data set for model training; the characteristic engineering sequentially comprises three steps of characteristic selection, characteristic construction and characteristic extraction; the characteristics are selected as characteristics for deleting redundancy or irrelevance; the feature construction is to construct a new feature manually on the basis of original data; the feature extraction is used for converting original data into data features which can be identified by a model in a mode based on principal component analysis and time window mapping.
6. The method of claim 3, wherein the step of predicting the link quality of the wireless network comprises: in the step S3, the process of training the model is to use the sample features and the tag data table generated in the step S2 as a training set, construct a wireless network link quality data prediction model based on a fusion method of the machine learning time series convolutional neural network and the lightGBM algorithm, and predict quality data of the wireless network link for a period of time in the future.
7. The method of claim 6, wherein the step of predicting the link quality of the wireless network comprises: in step S4, the predicting the wireless network link quality data specifically includes the following steps:
s41: constructing a characteristic data table of wireless network link quality data to be predicted;
s42: and inputting the characteristic data table of the user service downlink rate data to be predicted into the prediction model, and outputting the prediction result of the user service downlink rate data.
8. The method of claim 7, wherein the step of predicting the link quality of the wireless network comprises: in the step S41, the manner of constructing the characteristic data table of the wireless network link quality data to be predicted is the same as that in the step S2, but no sample label is set.
9. The method of claim 1, wherein the step of predicting the link quality of the wireless network comprises: and the steps S1 to S4 are a cyclic process, network side parameter data and user service downlink rate data of a wireless network link in a period are adopted, data and characteristic engineering are built, a data set is constructed, a labeled training sample is obtained and a model is trained, the trained model is used for predicting the user service downlink rate of an unlabeled test sample, and the model is evaluated.
10. A wireless link quality prediction system, wherein the prediction method according to any one of claims 1 to 9 is used to predict the wireless link quality, and the method comprises:
the acquisition and preprocessing module is used for acquiring network side parameters of a wireless network link and historical data of user service downlink rate, preprocessing the data and generating a historical data table of the quality of the wireless network link;
the system comprises a label and characteristic construction module, a label and characteristic construction module and a characteristic construction module, wherein the label and characteristic construction module is used for constructing a sample label by utilizing the user service downlink rate, performing characteristic engineering according to network side parameter historical data of a wireless network link, performing characteristic extraction, construction and extraction, and generating a historical data table containing wireless network link quality data characteristics and labels of each user;
the model training module is used for outputting a training sample set with a label according to the characteristic engineering, and adopting an algorithm training model integrating a time sequence convolutional neural network and a lightGBM algorithm to obtain a prediction model;
the quality prediction module is used for outputting a test sample set without a label according to the characteristic engineering, performing service downlink rate prediction on the test sample by using a prediction model, and evaluating the model;
the central processing module is used for sending instructions to each module to complete related actions;
the acquisition and preprocessing module, the label and feature construction module, the model training module and the quality prediction module are all electrically connected with the central processing module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111027557.0A CN113780398A (en) | 2021-09-02 | 2021-09-02 | Wireless network link quality prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111027557.0A CN113780398A (en) | 2021-09-02 | 2021-09-02 | Wireless network link quality prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113780398A true CN113780398A (en) | 2021-12-10 |
Family
ID=78840737
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111027557.0A Pending CN113780398A (en) | 2021-09-02 | 2021-09-02 | Wireless network link quality prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113780398A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114880637A (en) * | 2022-07-07 | 2022-08-09 | 深圳竹云科技股份有限公司 | Account risk verification method and device, computer equipment and storage medium |
| CN117241071A (en) * | 2023-11-15 | 2023-12-15 | 北京浩瀚深度信息技术股份有限公司 | Method for sensing video katon quality difference based on machine learning algorithm |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2011013768A1 (en) * | 2009-07-29 | 2011-02-03 | 京セラ株式会社 | Wireless terminal and transmission speed prediction method |
| CN112448827A (en) * | 2019-08-27 | 2021-03-05 | 中国移动通信有限公司研究院 | Information processing method, device, equipment and computer readable storage medium |
| US20210173993A1 (en) * | 2019-12-10 | 2021-06-10 | Microchip Technology Inc. | Machine Learning Based Methods and Apparatus for Integrated Circuit Design Delay Calculation and Verification |
-
2021
- 2021-09-02 CN CN202111027557.0A patent/CN113780398A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2011013768A1 (en) * | 2009-07-29 | 2011-02-03 | 京セラ株式会社 | Wireless terminal and transmission speed prediction method |
| CN112448827A (en) * | 2019-08-27 | 2021-03-05 | 中国移动通信有限公司研究院 | Information processing method, device, equipment and computer readable storage medium |
| US20210173993A1 (en) * | 2019-12-10 | 2021-06-10 | Microchip Technology Inc. | Machine Learning Based Methods and Apparatus for Integrated Circuit Design Delay Calculation and Verification |
Non-Patent Citations (2)
| Title |
|---|
| YUANYUAN WANG，ET AL: "Short-Term Load Forecasting for Industrial Customers Based on TCN-LightGBM", IEEE TRANSACTIONS ON POWER SYSTEMS》 * |
| 李飞，等: "基于LightGBM 算法的基站小区智慧节能研究", 《现代计算机》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114880637A (en) * | 2022-07-07 | 2022-08-09 | 深圳竹云科技股份有限公司 | Account risk verification method and device, computer equipment and storage medium |
| CN114880637B (en) * | 2022-07-07 | 2022-09-20 | 深圳竹云科技股份有限公司 | Account risk verification method and device, computer equipment and storage medium |
| CN117241071A (en) * | 2023-11-15 | 2023-12-15 | 北京浩瀚深度信息技术股份有限公司 | Method for sensing video katon quality difference based on machine learning algorithm |
| CN117241071B (en) * | 2023-11-15 | 2024-02-06 | 北京浩瀚深度信息技术股份有限公司 | Method for sensing video katon quality difference based on machine learning algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110943857A (en) | Power communication network fault analysis and positioning method based on convolutional neural network | |
| CN109872003B (en) | Object state prediction method, object state prediction system, computer device, and storage medium | |
| CN113780398A (en) | Wireless network link quality prediction method and system | |
| CN112818035B (en) | Network fault prediction method, terminal equipment and storage medium | |
| EP4050528A1 (en) | Model update system, model update method, and related device | |
| CN113709779B (en) | Cellular network fault diagnosis method | |
| CN114221790A (en) | BGP (Border gateway protocol) anomaly detection method and system based on graph attention network | |
| CN112217674A (en) | Alarm root cause identification method based on causal network mining and graph attention network | |
| CN109783553A (en) | A kind of power distribution network mass data increased quality system | |
| CN115877068A (en) | Voltage sag propagation track identification method of regional power grid based on deep learning | |
| CN116596033A (en) | Transformer ozone concentration prediction method based on window attention and generator | |
| CN103258255B (en) | A kind of Methods of Knowledge Discovering Based being applicable to grid management systems | |
| CN112232985B (en) | Power distribution and utilization data monitoring method and device for ubiquitous power Internet of things | |
| CN107666403B (en) | Index data acquisition method and device | |
| CN115883424B (en) | Method and system for predicting flow data between high-speed backbone networks | |
| CN115733780A (en) | Dynamic self-adaptive method, system, equipment and medium based on flexible Ethernet | |
| CN115713044A (en) | Method and device for analyzing residual service life of electromechanical equipment under multi-working-condition switching | |
| CN114915563A (en) | Network flow prediction method and system | |
| CN117692026B (en) | Link sensing method and device for power line communication | |
| Shan et al. | Root Cause Analysis of Failures for Power Communication Network Based on CNN | |
| CN117978551B (en) | Interaction abnormal behavior analysis method for transformer substation monitoring network | |
| CN113269368B (en) | Civil aircraft safety trend prediction method based on data driving | |
| CN113762485B (en) | Attention mechanism-based multi-dataset joint prediction method | |
| Qian et al. | Research on Optimization of 4G-LTE Wireless Network Cells Anomaly Diagnosis Algorithm based on Multidimensional Time Series Data. | |
| WO2024066720A1 (en) | Indicator threshold determination method and apparatus, storage medium, and electronic apparatus |
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 |