High-load cell intelligent identification method, electronic equipment, device and storage medium

Abstract

The invention provides a high-load cell intelligent identification method, electronic equipment, a device and a storage medium. The method comprises the following steps: acquiring time sequence data of a plurality of load characteristic indexes of a to-be-identified cell in a historical time period; inputting the time sequence data into an Informer model to obtain a multi-index prediction result output by the Informer model, the multi-index prediction result comprising prediction results of the plurality of load characteristic indexes of the to-be-identified cell in a prediction time period; determining whether the to-be-identified cell is a high-load cell or not according to the prediction result and a preset high-load standard; wherein the Informer model is obtained by training based on multi-index historical time sequence data of a plurality of sample cells, and the multi-index historical time sequence data comprises historical time sequence data of the plurality of load characteristic indexes. Therefore, intelligent recognition of massive cells can be realized, and the efficiency and accuracy of high-load cell recognition are greatly improved.

Description

Technical Field

[0001] This application relates to the field of wireless communication technologies, and in particular, to a method for intelligently identifying high-load cells, an electronic device, a device, and a storage medium. Background Art

[0002] With the explosive growth of data service traffic, it brings huge pressure to the wireless network, directly affecting the user perception. How to reduce the investment cost and improve the expansion efficiency while ensuring the user perception has become an urgent problem in network optimization work, and this largely depends on the identification of high-load cells. If high-load cells can be intelligently identified, corresponding optimization and expansion can be carried out in a timely manner. Summary of the Invention

[0003] This application provides a method for intelligently identifying high-load cells, an electronic device, a device, and a storage medium, which is used to intelligently identify high-load cells and improve the expansion efficiency.

[0004] In a first aspect, this application provides a method for intelligently identifying high-load cells, including:

[0005] Obtain the time-series data of multiple load characteristic indicators of the cell to be identified in a historical time period;

[0006] Input the time-series data into the Informer model to obtain the multi-index prediction result output by the Informer model, where the multi-index prediction result includes the prediction results of multiple load characteristic indicators of the cell to be identified in a prediction time period;

[0007] Determine whether the cell to be identified is a high-load cell according to the prediction result and a preset high-load standard;

[0008] Wherein, the Informer model is trained based on the multi-index historical time-series data of multiple sample cells, and the multi-index historical time-series data includes the historical time-series data of multiple load characteristic indicators.

[0009] In some embodiments, obtaining the time-series data of multiple load characteristic indicators of the cell to be identified in a historical time period includes:

[0010] Perform data cleaning on the original data of multiple load characteristic indicators of the cell to be identified in a historical time period to obtain the first processed data of multiple load characteristic indicators;

[0011] Judge whether the cell to be identified is an abnormal cell according to the first processed data; abnormal cells include cells with a missing sampling point ratio greater than a first threshold, out-of-service cells, and test cells;

[0012] In the case where it is determined that the cell to be identified is not an abnormal cell, numerical processing is performed on the abnormal sampling points and missing sampling points in the first processed data to obtain the second processed data of multiple load characteristic indicators;

[0013] Perform standardization processing on the second processed data to obtain the time series data of multiple load characteristic indicators of the cell to be identified in the historical time period.

[0014] In some embodiments, performing numerical processing on the abnormal sampling points and missing sampling points in the first processed data includes:

[0015] Based on the Prophet model, obtain the prediction data of each load characteristic indicator of the cell to be identified in the historical time period;

[0016] According to the prediction data and the preset confidence interval, determine the abnormal sampling points in the first processed data;

[0017] Use the prediction data to perform numerical filling processing on the abnormal sampling points and missing sampling points in the first processed data.

[0018] In some embodiments, the determination method of the out-of-service cell includes:

[0019] If the monthly average traffic of the cell is 0 for at least one month, the cell is an out-of-service cell.

[0020] In some embodiments, the determination method of the test cell includes:

[0021] If the ratio between the maximum value and the minimum value of the monthly average traffic of the cell is greater than the second threshold, the cell is a test cell.

[0022] In some embodiments, the load characteristic indicators include the uplink traffic of the cell, the downlink traffic of the cell, the average utilization rate of the uplink physical resource block (PRB) of the cell, the average utilization rate of the downlink PRB of the cell, and the occupancy rate of the physical downlink control channel (PDCCH) control channel element (CCE) of the cell.

[0023] In some embodiments, the high load standard includes an uplink high load standard and a downlink high load standard;

[0024] The uplink high load standard includes: in the prediction time period, if the average value of the uplink traffic of the cell is greater than the uplink traffic threshold and the average utilization rate of the uplink PRB of the cell is greater than the uplink PRB average utilization rate threshold, the cell meets the uplink high load standard;

[0025] The downlink high-load standard includes: within a predicted time period, the average downlink traffic of the cell is greater than the downlink traffic threshold, and at least one of the following is satisfied: the average downlink PRB utilization rate of the cell is greater than the downlink PRB average utilization rate threshold, the PDCCH channel CCE occupancy rate of the cell is greater than the PDCCH channel CCE occupancy rate threshold, then the cell meets the downlink high-load standard.

[0026] In some embodiments, determining whether a cell to be identified is a high-load cell according to a prediction result and a preset high-load standard includes:

[0027] If it is determined according to the prediction result that the cell to be identified meets the uplink high-load standard and / or the downlink high-load standard, then it is determined that the cell to be identified is a high-load cell.

[0028] In some embodiments, the encoder of the Informer model includes multiple layers of encoding sub-modules, and each layer of encoding sub-module includes a multi-head probabilistic sparse self-attention sub-layer and a self-attention distillation sub-layer;

[0029] The decoder of the Informer model includes multiple layers of decoding sub-modules, and each layer of decoding sub-module includes a masked multi-head probabilistic sparse self-attention sub-layer and a multi-head attention sub-layer;

[0030] Wherein, the output features of the encoder are input into the masked multi-head probabilistic sparse self-attention sub-layer of the decoder.

[0031] In some embodiments, the training loss function of the Informer model includes the Log-Cosh loss function.

[0032] In some embodiments, the encoder of the Informer model includes 3 layers of encoding sub-modules, and the decoder of the Informer model includes 2 layers of decoding sub-modules.

[0033] In a second aspect, the present application further provides an electronic device, including a memory, a transceiver, and a processor;

[0034] The memory is used to store a computer program; the transceiver is used to transmit and receive data under the control of the processor; the processor is used to read the computer program in the memory and perform the following operations:

[0035] Obtain the time-series data of multiple load characteristic indicators of the cell to be identified in a historical time period;

[0036] Input the time-series data into the Informer model to obtain the multi-index prediction result output by the Informer model, and the multi-index prediction result includes the prediction results of multiple load characteristic indicators of the cell to be identified in a predicted time period;

[0037] Determine whether the cell to be identified is a high-load cell according to the prediction result and the preset high-load standard;

[0038] Among them, the Informer model is trained based on the multi-index historical time series data of multiple sample cells, and the multi-index historical time series data includes the historical time series data of multiple load characteristic indicators.

[0039] In some embodiments, obtaining the time series data of multiple load characteristic indicators of the cell to be identified in the historical time period includes:

[0040] Perform data cleaning on the original data of multiple load characteristic indicators of the cell to be identified in the historical time period to obtain the first processed data of multiple load characteristic indicators;

[0041] According to the first processed data, determine whether the cell to be identified is an abnormal cell; abnormal cells include cells with a proportion of missing sampling points greater than the first threshold, out-of-service cells, and test cells;

[0042] In the case of determining that the cell to be identified is not an abnormal cell, perform numerical processing on the abnormal sampling points and missing sampling points in the first processed data to obtain the second processed data of multiple load characteristic indicators;

[0043] Perform standardization processing on the second processed data to obtain the time series data of multiple load characteristic indicators of the cell to be identified in the historical time period.

[0044] In some embodiments, performing numerical processing on the abnormal sampling points and missing sampling points in the first processed data includes:

[0045] Based on the Prophet model, obtain the prediction data of each load characteristic indicator of the cell to be identified in the historical time period;

[0046] According to the prediction data and the preset confidence interval, determine the abnormal sampling points in the first processed data;

[0047] Use the prediction data to perform numerical filling processing on the abnormal sampling points and missing sampling points in the first processed data.

[0048] In some embodiments, the determination method of out-of-service cells includes:

[0049] If the monthly average traffic of a cell is 0 for at least one month, the cell is an out-of-service cell.

[0050] In some embodiments, the determination method of test cells includes:

[0051] If the ratio between the maximum value and the minimum value of the monthly average traffic of a cell is greater than the second threshold, the cell is a test cell.

[0052] In some embodiments, the load characteristic indicators include the uplink traffic of the cell, the downlink traffic of the cell, the average utilization rate of the uplink physical resource blocks (PRBs) of the cell, the average utilization rate of the downlink PRBs of the cell, and the occupancy rate of the control channel elements (CCEs) of the physical downlink control channel (PDCCH) of the cell.

[0053] In some embodiments, the high-load criteria include an uplink high-load criterion and a downlink high-load criterion;

[0054] The uplink high-load criterion includes: within a predicted time period, if the average value of the uplink traffic of the cell is greater than the uplink traffic threshold and the average utilization rate of the uplink PRBs of the cell is greater than the uplink PRB average utilization rate threshold, then the cell meets the uplink high-load criterion;

[0055] The downlink high-load criterion includes: within a predicted time period, if the average value of the downlink traffic of the cell is greater than the downlink traffic threshold and at least one of the following is satisfied: the average utilization rate of the downlink PRBs of the cell is greater than the downlink PRB average utilization rate threshold, the occupancy rate of the CCEs of the PDCCH channel of the cell is greater than the PDCCH channel CCE occupancy rate threshold, then the cell meets the downlink high-load criterion.

[0056] In some embodiments, determining whether the cell to be identified is a high-load cell according to the prediction result and the preset high-load criteria includes:

[0057] If it is determined according to the prediction result that the cell to be identified meets the uplink high-load criterion and / or the downlink high-load criterion, then it is determined that the cell to be identified is a high-load cell.

[0058] In some embodiments, the encoder of the Informer model includes multiple layers of encoding sub-modules, and each layer of encoding sub-module includes a multi-head probabilistic sparse self-attention sub-layer and a self-attention distillation sub-layer;

[0059] The decoder of the Informer model includes multiple layers of decoding sub-modules, and each layer of decoding sub-module includes a masked multi-head probabilistic sparse self-attention sub-layer and a multi-head attention sub-layer;

[0060] Wherein, the output features of the encoder are input into the masked multi-head probabilistic sparse self-attention sub-layer of the decoder.

[0061] In some embodiments, the training loss function of the Informer model includes the Log-Cosh loss function.

[0062] In some embodiments, the encoder of the Informer model includes 3 layers of encoding sub-modules, and the decoder of the Informer model includes 2 layers of decoding sub-modules.

[0063] In a third aspect, the present application further provides a high-load cell intelligent identification device, including:

[0064] An acquisition unit for acquiring time series data of multiple load characteristic indicators of a cell to be identified within a historical time period;

[0065] A prediction unit for inputting the time series data into an Informer model to obtain a multi-index prediction result output by the Informer model, where the multi-index prediction result includes prediction results of multiple load characteristic indicators of the cell to be identified within a prediction time period;

[0066] A determination unit for determining whether the cell to be identified is a high-load cell according to the prediction result and a preset high-load standard;

[0067] Wherein, the Informer model is trained based on multi-index historical time series data of multiple sample cells, and the multi-index historical time series data includes historical time series data of multiple load characteristic indicators.

[0068] In some embodiments, acquiring the time series data of multiple load characteristic indicators of the cell to be identified within a historical time period includes:

[0069] Performing data cleaning on the original data of multiple load characteristic indicators of the cell to be identified within a historical time period to obtain first-processed data of multiple load characteristic indicators;

[0070] Judging whether the cell to be identified is an abnormal cell according to the first-processed data; abnormal cells include cells with a proportion of missing sampling points greater than a first threshold, out-of-service cells, and test broadcast cells;

[0071] In the case of determining that the cell to be identified is not an abnormal cell, performing numerical processing on abnormal sampling points and missing sampling points in the first-processed data to obtain second-processed data of multiple load characteristic indicators;

[0072] Performing standardization processing on the second-processed data to obtain the time series data of multiple load characteristic indicators of the cell to be identified within a historical time period.

[0073] In some embodiments, performing numerical processing on abnormal sampling points and missing sampling points in the first-processed data includes:

[0074] Respectively obtaining prediction data of each load characteristic indicator of the cell to be identified within a historical time period based on the Prophet model;

[0075] Determining abnormal sampling points in the first-processed data according to the prediction data and a preset confidence interval;

[0076] Performing numerical filling on abnormal sampling points and missing sampling points in the first-processed data using the prediction data.

[0077] In some embodiments, the determination method of out-of-service cells includes:

[0078] If the monthly average traffic of a cell is 0 for at least one month, the cell is an out-of-service cell.

[0079] In some embodiments, the method for determining a broadcast measurement cell includes:

[0080] If the ratio between the maximum value and the minimum value of the monthly average traffic of a cell is greater than a second threshold, the cell is a broadcast measurement cell.

[0081] In some embodiments, the load characteristic indicators include the uplink traffic of a cell, the downlink traffic of a cell, the average utilization rate of the uplink physical resource blocks (PRBs) of a cell, the average utilization rate of the downlink PRBs of a cell, and the occupancy rate of control channel elements (CCEs) of the physical downlink control channel (PDCCH) of a cell.

[0082] In some embodiments, the high-load criteria include an uplink high-load criterion and a downlink high-load criterion;

[0083] The uplink high-load criterion includes: within a prediction time period, if the average value of the uplink traffic of a cell is greater than an uplink traffic threshold and the average utilization rate of the uplink PRBs of the cell is greater than an uplink PRB average utilization rate threshold, then the cell meets the uplink high-load criterion;

[0084] The downlink high-load criterion includes: within a prediction time period, if the average value of the downlink traffic of a cell is greater than a downlink traffic threshold and at least one of the following is satisfied: the average utilization rate of the downlink PRBs of the cell is greater than a downlink PRB average utilization rate threshold, the occupancy rate of CCEs of the PDCCH channel of the cell is greater than a PDCCH channel CCE occupancy rate threshold, then the cell meets the downlink high-load criterion.

[0085] In some embodiments, determining whether a cell to be identified is a high-load cell according to a prediction result and a preset high-load criterion includes:

[0086] If it is determined according to the prediction result that the cell to be identified meets the uplink high-load criterion and / or the downlink high-load criterion, then it is determined that the cell to be identified is a high-load cell.

[0087] In some embodiments, the encoder of the Informer model includes multiple layers of encoding sub-modules, and each layer of encoding sub-module includes a multi-head probabilistic sparse self-attention sub-layer and a self-attention distillation sub-layer;

[0088] The decoder of the Informer model includes multiple layers of decoding sub-modules, and each layer of decoding sub-module includes a masked multi-head probabilistic sparse self-attention sub-layer and a multi-head attention sub-layer;

[0089] Wherein, the output features of the encoder are input into the masked multi-head probabilistic sparse self-attention sub-layer of the decoder.

[0090] In some embodiments, the training loss function of the Informer model includes the Log-Cosh loss function.

[0091] In some embodiments, the encoder of the Informer model includes three encoding sub-modules, and the decoder of the Informer model includes two decoding sub-modules.

[0092] Fourthly, the present application further provides a non-transitory readable storage medium storing a computer program for causing a processor to execute the high-load cell intelligent identification method described in the first aspect above.

[0093] Fifthly, the present application further provides a communication device storing a computer program for causing the communication device to execute the high-load cell intelligent identification method described in the first aspect above.

[0094] Sixthly, the present application further provides a processor-readable storage medium storing a computer program for causing a processor to execute the high-load cell intelligent identification method described in the first aspect above.

[0095] Seventhly, the present application further provides a chip product storing a computer program for causing the chip product to execute the high-load cell intelligent identification method described in the first aspect above.

[0096] The high-load cell intelligent identification method, electronic device, device and storage medium provided by the present application use the Informer model to predict multiple load characteristic indicators of the cell to be identified, and then determine the high-load cell according to the prediction result in combination with the preset high-load standard, realizing the intelligent identification of a large number of cells without the investment of a large amount of human resources, and greatly improving the efficiency and accuracy of high-load cell identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0098] Figure 1 It is a schematic flowchart of the high-load cell intelligent identification method provided by the embodiment of the present application;

[0099] Figure 2It is an example diagram of the daily total traffic index data of the out-of-service cells and the test cells provided by the embodiments of the present application;

[0100] Figure 3 It is an example diagram of the result of the Prophet model for predicting the downlink traffic provided by the embodiments of the present application;

[0101] Figure 4 It is a comparison diagram before and after processing the abnormal sampling points and missing sampling points of the downlink traffic provided by the embodiments of the present application;

[0102] Figure 5 It is a flow chart of the training and prediction of the Informer model provided by the embodiments of the present application;

[0103] Figure 6 It is an example diagram of the structure and training of the Informer model provided by the embodiments of the present application;

[0104] Figure 7 It is a schematic structural diagram of the electronic device provided by the embodiments of the present application;

[0105] Figure 8 It is a schematic structural diagram of the high-load cell intelligent identification device provided by the embodiments of the present application. Detailed implementation manners

[0106] In the embodiments of the present application, the term "and / or" describes the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after.

[0107] In the embodiments of the present application, the term "a plurality of" means two or more, and other quantifiers are similar.

[0108] In the embodiments of the present application, terms such as "first" and "second" are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that such terms can be interchanged under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first" and "second" are usually of the same type, and the number of objects is not limited. For example, the first object can be one or more.

[0109] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0110] For the identification of high-load cells, manual methods require a large amount of human resources input, are time-consuming and cumbersome in process, so they can only be applied to a small range and cannot be applied to the optimization and expansion work of a large range and the entire network. By collecting cell traffic information and using an Artificial Intelligence (AI) model to predict future traffic volume is a feasible method. Currently, commonly used AI models are time series prediction models such as Auto-Regressive Moving Average (ARMA), Prophet, and Long Short-Term Memory (LSTM). There are still certain limitations in using these models to predict high-load cells. For example, ARMA model and Prophet model can only consider a single element for single-index prediction, while communication metrics complement and are interrelated with each other, and single-index time prediction cannot meet the application requirements; although the LSTM model can consider multiple features, its effect for predicting long cycles is average. For example, when predicting relevant metrics for the next month, the accuracy of the LSTM model is relatively low.

[0111] In view of the above problems, each embodiment of the present application provides a solution. Based on the Informer model (or an improved Informer model), long-term and short-term predictions of multiple load characteristic indicators of cells are performed, and high-load cells are intelligently identified according to the prediction results combined with a preset high-load standard, which can effectively improve the expansion efficiency and ensure user perception.

[0112] Figure 1 FIG. is a schematic flowchart of the high-load cell intelligent identification method provided by an embodiment of the present application. As Figure 1 shown, the method includes the following steps:

[0113] Step 100: Obtain the time series data of multiple load characteristic indicators of the cell to be identified in the historical time period.

[0114] Specifically, the execution subject of each step in the method can be a high-load cell intelligent identification device, which can be implemented by software and / or hardware, and can be integrated in an electronic device. The electronic device can be a terminal device (such as a smart phone, a personal computer, etc.), a server (such as a local server or a cloud server, or a server cluster, etc.), a network device (for example: a base station), or other devices.

[0115] When identifying high-load cells, for the cell to be identified, first obtain the time-series data of multiple load characteristic indicators of the cell within a historical time period. Among them, the historical time period can be selected according to the requirements of the Informer model for the input data, which is not limited here. For example, it can be the data of the past 1 year. The load characteristic indicators refer to the characteristic indicators that can be used to characterize the cell load, and the load characteristic indicators used for the training and inference of the Informer model are kept consistent. The time-series data is the time-sequence data, which is the data arranged in the order of time. For example, the data of a certain load characteristic indicator of a cell every day in the past 1 year arranged in the order of time is the time-series data of the load characteristic indicator of the cell in the past 1 year.

[0116] In some embodiments, the load characteristic indicators include the uplink traffic of the cell, the downlink traffic of the cell, the average utilization rate of the uplink physical resource block (PRB) of the cell, the average utilization rate of the downlink PRB of the cell, and the occupancy rate of the control channel element (CCE) of the physical downlink control channel (PDCCH) of the cell. The data of these load characteristic indicators can be obtained by collecting the key performance indicator (KPI) data of the cell.

[0117] In some embodiments, obtaining the time-series data of multiple load characteristic indicators of the cell to be identified within a historical time period includes:

[0118] Perform data cleaning on the original data of multiple load characteristic indicators of the cell to be identified within a historical time period to obtain the first processed data of multiple load characteristic indicators;

[0119] According to the first processed data, determine whether the cell to be identified is an abnormal cell; the abnormal cells include the cells with the proportion of missing sampling points greater than the first threshold, the out-of-service cells, and the test cells;

[0120] In the case of determining that the cell to be identified is not an abnormal cell, perform numerical processing on the abnormal sampling points and missing sampling points in the first processed data to obtain the second processed data of multiple load characteristic indicators;

[0121] Perform standardization processing on the second processed data to obtain the time-series data of multiple load characteristic indicators of the cell to be identified within a historical time period.

[0122] Specifically, since the collected original data may have problems such as out-of-order time series, missing timestamps, abnormal sampling points, and missing sampling points, it is necessary to preprocess the original data of each load characteristic index. For example, data cleaning, abnormal cell elimination, processing of abnormal sampling points and missing sampling points, etc. The time series data of each load characteristic index obtained after preprocessing is input into the Informer model for multi-index prediction. The training data set used for training the Informer model is also obtained through the same preprocessing process.

[0123] Data cleaning refers to the preliminary arrangement of the original data of each load characteristic index, including sorting the data in chronological order, filling in the missing timestamps (the index values corresponding to the missing timestamps are empty), etc.

[0124] After data cleaning, next, based on the first processed data of multiple load characteristic indexes obtained after data cleaning, it is judged whether the cell to be identified is an abnormal cell. If it is an abnormal cell, prediction is not performed on this cell. During training, if a certain sample cell is judged to be an abnormal cell, the data of this sample cell is not used to participate in model training. In the embodiments of the present application, abnormal cells include cells with a missing sampling point ratio greater than the first threshold (hereinafter referred to as cells with more missing values), out-of-service cells, and broadcast measurement cells. That is, if a certain cell belongs to at least one of these three types of cells: cells with more missing values, out-of-service cells, and broadcast measurement cells, then this cell can be considered an abnormal cell. Among them, the first threshold can be set flexibly, and the present application does not make any limitations. For example, the first threshold is 10%, 15%, etc.

[0125] In some embodiments, the sampling point data of the single-day total traffic of the cell (uplink traffic + downlink traffic) can be used to judge abnormal cells. For example, for the judgment of "cells with more missing values", obtain the single-day total traffic index data of a certain cell for 1 year. If the proportion of the number of missing sampling points in the total number of sampling points in 1 year (such as 365) of this index data is greater than the first threshold (such as 10%), then judge this cell as a "cell with more missing values", which belongs to an abnormal cell. For out-of-service cells and broadcast measurement cells, the sampling point data of the single-day total traffic of the cell can also be used for judgment.

[0126] For example, if the monthly average traffic of a certain cell is 0 for at least one month, then judge this cell as an out-of-service cell. Figure 2 This is an example diagram of the single-day total traffic index data of out-of-service cells and broadcast measurement cells provided by the embodiments of the present application. As Figure 2 shown, in the left example in the figure, the monthly average traffic of the cell is 0 for more than 2 months, then this cell can be determined as an out-of-service cell.

[0127] For example, if the ratio between the maximum monthly average traffic volume and the minimum monthly average traffic volume of a certain cell is greater than the second threshold, then the cell is a test cell. Herein, the second threshold can be flexibly set and is not limited in this application. For example, the second threshold is 10, 20, etc. As Figure 2 shown, the maximum monthly average traffic volume of the cell in the right example in the figure is much greater than the minimum monthly average traffic volume of the cell, and this cell can be determined as a test cell.

[0128] If it is determined that the cell to be recognized is not an abnormal cell, then the subsequent processing of abnormal sampling points (the index values of the sampling points are abnormal) and missing sampling points (the sampling points only have timestamps and the index values are empty) is performed. In the case where there are abnormal values and missing values in the data, not processing will affect the accuracy of model prediction. Therefore, it is necessary to perform numerical processing on the abnormal sampling points and missing sampling points in the first processed data. Performing numerical processing means adjusting the index values of the abnormal sampling points and filling the index values for the missing sampling points, etc.

[0129] In some embodiments, performing numerical processing on the abnormal sampling points and missing sampling points in the first processed data includes:

[0130] respectively obtaining the prediction data of each load characteristic index of the cell to be recognized within the historical time period based on the Prophet model;

[0131] determining the abnormal sampling points in the first processed data according to the prediction data and the preset confidence interval;

[0132] performing numerical filling processing on the abnormal sampling points and missing sampling points in the first processed data using the prediction data.

[0133] Specifically, in this embodiment, the Prophet model is used to perform single-index prediction on each load characteristic index respectively to obtain the prediction data of each load characteristic index of the cell to be recognized within the historical time period, and then these prediction data and the preset confidence interval are used to screen out the abnormal sampling points.

[0134] The principle of the Prophet model is to analyze various time series characteristics: periodicity, trend, holiday effect, and some outliers. In terms of trend, it supports adding change points to achieve piecewise linear fitting. In terms of period, it uses Fourier series to establish a period model (sin + cos). In terms of holidays and emergencies, users can specify holidays and the relevant N days before and after them in the form of a table. The Prophet model can be regarded as an integrated solution for time series. However, it can only perform single-index prediction and the long-period prediction is not accurate enough. In the embodiments of this application, it is used for numerical processing of abnormal sampling points and missing sampling points.

[0135] Taking the load characteristic index of downlink traffic as an example (the abnormal sampling points and missing sampling points of other load characteristic indexes can be processed similarly), Figure 3 Figure 3 is an example diagram of the prediction result of the downlink traffic by the Prophet model provided by the embodiment of the present application. The curve in the figure is the data curve of the downlink traffic of a certain cell fitted by the Prophet model within a period of time. The solid dots in the figure are the scatter points of the real data of the downlink traffic within the same period of time. The gray area is the confidence interval (for example, a confidence interval of 0.95 can be set). The points outside the confidence interval are abnormal points, marked as hollow dots. Figure 3 The Prophet model parameters used in the example prediction result are set as follows: holidays = legal holiday time, yearly_seasonality = 8, weekly_seasonality = 8, daily_seasonality = 20, changepoint_prior_scale = 0.04, growth = 'linear', interval_width = 0.95.

[0136] Then, for abnormal sampling points and missing sampling points, the numerical values of the prediction indexes with the same time stamp are used to fill the numerical values of the sampling points. For example, in the data of a certain load characteristic index of a certain cell, the time stamp of an abnormal sampling point 1 is recorded as time stamp 1, and the time stamp of a missing sampling point 1 is recorded as time stamp 2. Then, the prediction index value of time stamp 1 in the prediction data of this load characteristic index of this cell can be used to replace the original index value of the abnormal sampling point 1, and the prediction index value of time stamp 2 in the prediction data of this load characteristic index of this cell can be used as the index value of the missing sampling point 1.

[0137] Figure 4 Figure 4 is a comparison diagram of the downlink traffic abnormal sampling points and missing sampling points before and after processing provided by the embodiment of the present application. As Figure 4 shown, the processed data in the figure (represented by the solid line) is smoother and more complete than the data before processing (represented by the hollow line), which is beneficial to improving the prediction accuracy of the subsequent Informer model.

[0138] Since there are significant differences in the dimensions, orders of magnitude, etc. of each load characteristic index, after numerically processing the abnormal sampling points and missing sampling points in the first processed data to obtain the second processed data of multiple load characteristic indexes, for the convenience of prediction by the Informer model, the second processed data can be standardized to constrain the data to a unified dimension.

[0139] Normalization refers to converting the original data according to a certain ratio through a certain mathematical transformation method, so that it falls into a small specific interval, such as the interval [0, 1] or [-1, 1], eliminating the differences in characteristic attributes such as nature, dimension, and order of magnitude between different variables, and converting it into a dimensionless relative value, that is, the normalized value, so that the values of each index are at the same order of magnitude, thus facilitating the comprehensive analysis and comparison of indexes with different units or orders of magnitude.

[0140] In some embodiments, the normalization calculation formula is: Normalized value = (Original value - Mean) / Standard deviation. Among them, the normalized value, original value, mean, and standard deviation are all for a single load characteristic index of a single cell, that is, when performing normalization, the respective means and standard deviations are used for each load characteristic index. For example, for the normalization of uplink traffic data, the mean and standard deviation used are the mean and standard deviation of the uplink traffic of the cell; for the normalization of the average utilization rate of uplink PRB data, the mean and standard deviation used are the mean and standard deviation of the average utilization rate of uplink PRB.

[0141] Step 101: Input the time series data into the Informer model to obtain the multi-index prediction results output by the Informer model. The multi-index prediction results include the prediction results of multiple load characteristic indexes of the cell to be identified during the prediction time period.

[0142] Among them, the Informer model is trained based on the multi-index historical time series data of multiple sample cells, and the multi-index historical time series data includes the historical time series data of multiple load characteristic indexes.

[0143] Specifically, the Informer model is a time series prediction model improved based on the Transformer model. By adopting the ProbSpare self-attention mechanism, self-attention distilling mechanism, and generative style decoder mechanism, it reduces the model complexity and memory usage, improves the calculation efficiency and the inference speed of long sequence prediction. Compared with the defects that the ARMA model and Prophet model can only perform single-index prediction, and the LSTM has a general effect for predicting long periods, the Informer model has good effects for both multi-index prediction and long and short period prediction.

[0144] Before identifying high-load cells, the Informer model is pre-trained using the historical time-series data of multiple indicators of a large number of sample cells. The historical time-series data of multiple indicators includes the historical time-series data of multiple load characteristic indicators. The load characteristic indicators used for training and inference of the Informer model are consistent. The historical time-series data of a certain load characteristic indicator refers to the time-series data of this load characteristic indicator in the past period (such as the past year).

[0145] Input the time-series data of multiple load characteristic indicators of the cell to be identified in the historical time period into the pre-trained Informer model, and the prediction results of multiple load characteristic indicators of the cell to be identified in the prediction time period can be obtained. The prediction time period is a future time period that the model needs to predict. The Informer model can perform long-term prediction (such as predicting the results for the next month) or short-term prediction (such as predicting the results for the next day).

[0146] Step 102: Determine whether the cell to be identified is a high-load cell according to the prediction results and the preset high-load standard.

[0147] Specifically, after obtaining the prediction results, it is possible to determine whether the cell to be identified is a high-load cell according to the prediction results and the preset high-load standard (the setting of this high-load standard corresponds to the load characteristic indicators). For example, if the prediction results of the cell to be identified meet the high-load standard, it can be determined that the cell to be identified is a high-load cell.

[0148] In some embodiments, the high-load standard includes an uplink high-load standard and a downlink high-load standard;

[0149] The uplink high-load standard includes: within the prediction time period, the average uplink traffic of the cell is greater than the uplink traffic threshold, and the average uplink PRB utilization rate of the cell is greater than the average uplink PRB utilization rate threshold, then the cell meets the uplink high-load standard;

[0150] The downlink high-load standard includes: within the prediction time period, the average downlink traffic of the cell is greater than the downlink traffic threshold, and at least one of the following is satisfied: the average downlink PRB utilization rate of the cell is greater than the average downlink PRB utilization rate threshold, the occupancy rate of the PDCCH channel CCE of the cell is greater than the PDCCH channel CCE occupancy rate threshold, then the cell meets the downlink high-load standard.

[0151] Taking 5G cells as an example, for cells with different coverage types, different working frequency bands, different transceiver modes, and different main service packet types, the thresholds of each load characteristic indicator in the high-load standard are different. Table 1 below exemplifies the high-load standard thresholds for cells corresponding to different coverage types, different working frequency bands, different transceiver modes, and different main service packet types.

[0152] Table 1 Example of high-load standard thresholds for different cells

[0153]

[0154]

[0155] Among them, the cells of different packet types are defined in Table 2 below.

[0156] Table 2 Definition of cells of different packet types

[0157] Packet type Traffic volume (KB) of each Quality of Service (QoS) flow in the cell Large packet cell >3000 Medium packet cell (1500,3000] Small packet cell <＝1500

[0158] Among them,

[0160] In some embodiments, according to the prediction result and the preset high-load standard, determining whether the cell to be identified is a high-load cell includes:

[0161] If it is determined according to the prediction result that the cell to be identified meets the uplink high-load standard and / or the downlink high-load standard, then determine that the cell to be identified is a high-load cell.

[0162] Specifically, in this embodiment, when the high-load standard includes the uplink high-load standard and the downlink high-load standard, as long as at least one of the uplink high-load standard and the downlink high-load standard is met, it is determined that the cell is a high-load cell.

[0163] For example, assume that a certain cell is a macro station, 700 MHz, 4TR, large packet cell. According to the multi-index prediction result, the average uplink traffic in the next month is greater than the uplink traffic threshold of 2 GB, and the average uplink PRB utilization rate is greater than the uplink PRB average utilization rate threshold of 50%. Then this cell meets the uplink high-load standard, and at this time, it can be determined that this cell is a high-load cell.

[0164] For example, assume that a certain cell is a macro station, 700 MHz, 4TR, large packet cell. According to the multi-index prediction result, the average downlink traffic in the next month is greater than the downlink traffic threshold of 8 GB, and it meets: the average downlink PRB utilization rate is greater than the downlink PRB average utilization rate threshold of 70%, and / or the PDCCH channel CCE occupancy rate is greater than the PDCCH channel CCE occupancy rate threshold of 50%. Then this cell meets the downlink high-load standard, and at this time, it can be determined that this cell is a high-load cell.

[0165] For another example, if a cell meets both the uplink high-load standard and the downlink high-load standard at the same time, it can also be determined that this cell is a high-load cell.

[0166] The intelligent high-load cell identification method provided by the embodiments of this application uses the Informer model to predict multiple load characteristic indicators for the cell to be identified, and then determines the high-load cells according to the prediction results in combination with the preset high-load standard. It can realize the intelligent identification of a large number of cells without the investment of a large amount of human resources, greatly improving the efficiency and accuracy of high-load cell identification.

[0167] In some embodiments, the encoder of the Informer model includes multiple encoding sub-modules. Each encoding sub-module includes a multi-head probabilistic sparse self-attention sub-layer and a self-attention distillation sub-layer;

[0168] The decoder of the Informer model includes multiple decoding sub-modules. Each decoding sub-module includes a masked multi-head probabilistic sparse self-attention sub-layer and a multi-head attention sub-layer;

[0169] Among them, the output features of the encoder are input into the masked multi-head probabilistic sparse self-attention sub-layer of the decoder.

[0170] Specifically, the Informer model used in this embodiment is improved compared with the existing Informer model. The Informer model in this embodiment includes four parts: a feature processing module, an encoder, a decoder, and a fully connected layer. The core of the improvement lies in the encoder and decoder structures. The encoder of the Informer model in this embodiment includes multiple encoding sub-modules. Each encoding sub-module includes a multi-head probabilistic sparse self-attention sub-layer and a self-attention distillation sub-layer; the decoder includes multiple decoding sub-modules. Each decoding sub-module includes a masked multi-head probabilistic sparse self-attention sub-layer and a multi-head attention sub-layer. Moreover, different from the existing Informer model, the output features of the encoder of the Informer model in this embodiment are not input into the multi-head self-attention sub-layer of the decoder, but into the masked multi-head probabilistic sparse self-attention sub-layer of the decoder. By using the improved Informer model, it can adapt to the multi-index prediction of a large number of cells and further improve the identification efficiency of high-load cells.

[0171] In some embodiments, the encoder of the Informer model includes 3 encoding sub-modules, and the decoder of the Informer model includes 2 decoding sub-modules. By reasonably setting the number of layers of the encoding sub-modules and the decoding sub-modules, the model complexity can be further reduced on the basis of ensuring the prediction accuracy, and the model inference speed can be improved.

[0172] In some embodiments, the training loss function of the Informer model includes the Log-Cosh loss function.

[0173] Specifically, the Log-Cosh loss function is a commonly used function in regression tasks. It is smoother than the L2 loss function. The Log-Cosh loss function is the logarithm of the hyperbolic cosine of the prediction error. When the variable x is small, the loss function value log(cosh(x)) is approximately equal to (x 2 ) / 2; when x is large, the loss function value log(cosh(x)) is approximately equal to abs(x) - log(2). This means that "Log-Cosh" works very similarly to MSE (mean squared error), but is not affected by occasional outliers. It has all the advantages of the Huber loss and is differentiable in any case. By using the Log-Cosh loss function to train the Informer model, the trained Informer model can have better performance in multi-metric prediction.

[0174] Figure 5 This is the flowchart of the training and prediction of the Informer model provided by the embodiment of the present application. As Figure 5 shown, in this example, first, the KPI data of a certain city for 1 year is collected for model training. The collected data is first cleaned, then abnormal cells are removed, then feature selection is performed, and then the Prophet algorithm is used to detect outliers and fill in missing values for each load feature index. The data of each cell is standardized, and finally, it is input into the improved Informer model. During prediction, the original data of multiple load feature indexes of the cell to be identified is first cleaned, and then goes through processes such as abnormal cell judgment, outlier detection, missing value filling, and standardization processing. Finally, the obtained data is input into the improved Informer model to obtain the prediction result.

[0175] Figure 6 This is the structure and training example diagram of the Informer model provided by the embodiment of the present application. As Figure 6 shown, the figure exemplifies the structure of the improved Informer model. Each encoding sub-module in the encoder of the improved Informer model is composed of a combination of a one-layer multi-head probabilistic sparse self-attention mechanism and a one-layer self-attention distillation. Each decoding sub-module in the decoder is composed of a combination of a one-layer masked multi-head probabilistic sparse self-attention mechanism and a one-layer multi-head attention mechanism. When training the improved Informer model, the Log-Cosh loss function is used for backpropagation.

[0176] When making predictions, the multi-index data of the cell to be recognized after preprocessing is input into the improved Informer model to capture the long-term correlation between the input and output of the index sequence. Finally, the prediction result is obtained through the fully connected layer. The prediction results show that the error value of the improved Informer model provided by the embodiments of the present application for predicting the index in the next month is within 0.3, and the average accuracy of each index is 84%. While the error value of predicting the index in the next day is within 0.3, and the average accuracy of each index is 89%. The model performance is good.

[0177] The methods and devices provided in the embodiments of the present application are based on the same inventive concept. Since the principles of solving problems by the methods and devices are similar, the implementation of the devices and methods can be referred to each other, and the repeated parts will not be described again.

[0178] Figure 7 It is a schematic structural diagram of the electronic device provided by the embodiments of the present application, as Figure 7 shown, the electronic device includes a memory 720, a transceiver 710, and a processor 700; wherein, the processor 700 and the memory 720 can also be physically separated.

[0179] The memory 720 is used to store computer programs; the transceiver 710 is used to receive and send data under the control of the processor 700.

[0180] Specifically, the transceiver 710 is used to receive and send data under the control of the processor 700.

[0181] Among them, in Figure 7 the bus architecture may include any number of interconnected buses and bridges, specifically, various circuits represented by one or more processors represented by the processor 700 and the memory represented by the memory 720 are linked together. The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art. Therefore, the present application will not further describe them. The bus interface provides an interface. The transceiver 710 can be multiple components, that is, including a transmitter and a receiver, and provides a unit for communicating with various other devices on the transmission medium, and these transmission media include wireless channels, wired channels, optical cables and other transmission media.

[0182] The processor 700 is responsible for managing the bus architecture and general processing, and the memory 720 can store the data used by the processor 700 when performing operations.

[0183] The processor 700 may be a Central Processing Unit (CPU), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or a Complex Programmable Logic Device (CPLD). The processor may also adopt a multi-core architecture.

[0184] The processor 700 is used to execute any of the methods provided in the embodiments of the present application by calling a computer program stored in the memory 720 according to the obtained executable instructions. For example: obtaining time series data of multiple load characteristic indicators of a cell to be identified in a historical time period; inputting the time series data into an Informer model to obtain a multi-index prediction result output by the Informer model, where the multi-index prediction result includes prediction results of multiple load characteristic indicators of the cell to be identified in a prediction time period; determining whether the cell to be identified is a high-load cell according to the prediction result and a preset high-load standard; wherein, the Informer model is trained based on multi-index historical time series data of multiple sample cells, and the multi-index historical time series data includes historical time series data of multiple load characteristic indicators.

[0185] In some embodiments, obtaining time series data of multiple load characteristic indicators of a cell to be identified in a historical time period includes:

[0186] Performing data cleaning processing on the original data of multiple load characteristic indicators of the cell to be identified in a historical time period to obtain first-processed data of multiple load characteristic indicators;

[0187] Judging whether the cell to be identified is an abnormal cell according to the first-processed data; the abnormal cells include cells with a proportion of missing sampling points greater than a first threshold, out-of-service cells, and broadcast measurement cells;

[0188] In the case of determining that the cell to be identified is not an abnormal cell, performing numerical processing on abnormal sampling points and missing sampling points in the first-processed data to obtain second-processed data of multiple load characteristic indicators;

[0189] Performing standardization processing on the second-processed data to obtain time series data of multiple load characteristic indicators of the cell to be identified in a historical time period.

[0190] In some embodiments, performing numerical processing on abnormal sampling points and missing sampling points in the first-processed data includes:

[0191] Based on the Prophet model, prediction data of each load characteristic index of the cell to be identified in the historical time period is obtained respectively;

[0192] According to the prediction data and the preset confidence interval, abnormal sampling points in the first processed data are determined;

[0193] The prediction data is used to perform numerical filling processing on the abnormal sampling points and missing sampling points in the first processed data.

[0194] In some embodiments, the determination method of the out-of-service cell includes:

[0195] If the monthly average traffic of the cell is 0 for at least one month, the cell is an out-of-service cell.

[0196] In some embodiments, the determination method of the broadcast measurement cell includes:

[0197] If the ratio between the maximum value and the minimum value of the monthly average traffic of the cell is greater than the second threshold, the cell is a broadcast measurement cell.

[0198] In some embodiments, the load characteristic indexes include the uplink traffic of the cell, the downlink traffic of the cell, the average utilization rate of the uplink physical resource block (PRB) of the cell, the average utilization rate of the downlink PRB of the cell, and the occupancy rate of the control channel element (CCE) of the physical downlink control channel (PDCCH) of the cell.

[0199] In some embodiments, the high-load standard includes an uplink high-load standard and a downlink high-load standard;

[0200] The uplink high-load standard includes: in the prediction time period, the average value of the uplink traffic of the cell is greater than the uplink traffic threshold, and the average utilization rate of the uplink PRB of the cell is greater than the uplink PRB average utilization rate threshold, then the cell meets the uplink high-load standard;

[0201] The downlink high-load standard includes: in the prediction time period, the average value of the downlink traffic of the cell is greater than the downlink traffic threshold, and at least one of the following is satisfied: the average utilization rate of the downlink PRB of the cell is greater than the downlink PRB average utilization rate threshold, the occupancy rate of the CCE of the PDCCH channel of the cell is greater than the PDCCH channel CCE occupancy rate threshold, then the cell meets the downlink high-load standard.

[0202] In some embodiments, determining whether the cell to be identified is a high-load cell according to the prediction result and the preset high-load standard includes:

[0203] If it is determined according to the prediction result that the cell to be identified meets the uplink high-load standard and / or the downlink high-load standard, it is determined that the cell to be identified is a high-load cell.

[0204] In some embodiments, the encoder of the Informer model includes multiple layers of encoding sub-modules, and each layer of encoding sub-module includes a multi-head probabilistic sparse self-attention sub-layer and a self-attention distillation sub-layer;

[0205] The decoder of the Informer model includes multiple layers of decoding sub-modules, and each layer of decoding sub-module includes a masked multi-head probabilistic sparse self-attention sub-layer and a multi-head attention sub-layer;

[0206] Among them, the output features of the encoder are input into the masked multi-head probabilistic sparse self-attention sub-layer of the decoder.

[0207] In some embodiments, the training loss function of the Informer model includes the Log-Cosh loss function.

[0208] In some embodiments, the encoder of the Informer model includes 3 layers of encoding sub-modules, and the decoder of the Informer model includes 2 layers of decoding sub-modules.

[0209] It should be noted here that the above-mentioned electronic device provided by the embodiments of the present application can implement all the method steps implemented by the above method embodiments, and can achieve the same technical effects. The same parts and beneficial effects as those in the method embodiments will not be specifically described in this embodiment.

[0210] Figure 8 It is a schematic structural diagram of a high-load cell intelligent identification device provided by an embodiment of the present application, as Figure 8 shown, the device includes:

[0211] An acquisition unit 800, configured to acquire the time-series data of multiple load characteristic indicators of the cell to be identified in a historical time period;

[0212] A prediction unit 810, configured to input the time-series data into the Informer model to obtain a multi-index prediction result output by the Informer model, where the multi-index prediction result includes the prediction results of multiple load characteristic indicators of the cell to be identified in a prediction time period;

[0213] A determination unit 820, configured to determine whether the cell to be identified is a high-load cell according to the prediction result and a preset high-load standard;

[0214] Among them, the Informer model is trained based on the multi-index historical time-series data of multiple sample cells, and the multi-index historical time-series data includes the historical time-series data of multiple load characteristic indicators.

[0215] In some embodiments, acquiring the time-series data of multiple load characteristic indicators of the cell to be identified in a historical time period includes:

[0216] Perform data cleaning on the original data of multiple load characteristic indicators of the cell to be identified within a historical time period to obtain the first processed data of the multiple load characteristic indicators;

[0217] Based on the first processed data, determine whether the cell to be identified is an abnormal cell; abnormal cells include cells with a proportion of missing sampling points greater than a first threshold, out-of-service cells, and test cells;

[0218] In the case where it is determined that the cell to be identified is not an abnormal cell, perform numerical processing on the abnormal sampling points and missing sampling points in the first processed data to obtain the second processed data of the multiple load characteristic indicators;

[0219] Perform standardization processing on the second processed data to obtain the time series data of multiple load characteristic indicators of the cell to be identified within a historical time period.

[0220] In some embodiments, performing numerical processing on the abnormal sampling points and missing sampling points in the first processed data includes:

[0221] Based on the Prophet model, obtain the prediction data of each load characteristic indicator of the cell to be identified within a historical time period;

[0222] According to the prediction data and a preset confidence interval, determine the abnormal sampling points in the first processed data;

[0223] Use the prediction data to perform numerical filling processing on the abnormal sampling points and missing sampling points in the first processed data.

[0224] In some embodiments, the determination method of out-of-service cells includes:

[0225] If the monthly average traffic of a cell is 0 for at least one month, the cell is an out-of-service cell.

[0226] In some embodiments, the determination method of test cells includes:

[0227] If the ratio between the maximum value and the minimum value of the monthly average traffic of a cell is greater than a second threshold, the cell is a test cell.

[0228] In some embodiments, the load characteristic indicators include the uplink traffic of the cell, the downlink traffic of the cell, the average utilization rate of the uplink physical resource blocks (PRBs) of the cell, the average utilization rate of the downlink PRBs of the cell, and the occupancy rate of the physical downlink control channel (PDCCH) control channel elements (CCEs) of the cell.

[0229] In some embodiments, the high-load standards include an uplink high-load standard and a downlink high-load standard;

[0230] The uplink high-load standard includes: within the predicted time period, the average uplink traffic of the cell is greater than the uplink traffic threshold, and the average uplink PRB utilization rate of the cell is greater than the uplink PRB average utilization rate threshold, then the cell meets the uplink high-load standard;

[0231] The downlink high-load standard includes: within the predicted time period, the average downlink traffic of the cell is greater than the downlink traffic threshold, and at least one of the following is satisfied: the average downlink PRB utilization rate of the cell is greater than the downlink PRB average utilization rate threshold, the occupancy rate of PDCCH channel CCE of the cell is greater than the PDCCH channel CCE occupancy rate threshold, then the cell meets the downlink high-load standard.

[0232] In some embodiments, determining whether the cell to be identified is a high-load cell according to the prediction result and the preset high-load standard includes:

[0233] If it is determined according to the prediction result that the cell to be identified meets the uplink high-load standard and / or the downlink high-load standard, then it is determined that the cell to be identified is a high-load cell.

[0234] In some embodiments, the encoder of the Informer model includes multiple layers of encoding sub-modules, and each layer of encoding sub-module includes a multi-head probabilistic sparse self-attention sub-layer and a self-attention distillation sub-layer;

[0235] The decoder of the Informer model includes multiple layers of decoding sub-modules, and each layer of decoding sub-module includes a masked multi-head probabilistic sparse self-attention sub-layer and a multi-head attention sub-layer;

[0236] Among them, the output features of the encoder are input into the masked multi-head probabilistic sparse self-attention sub-layer of the decoder.

[0237] In some embodiments, the training loss function of the Informer model includes the Log-Cosh loss function.

[0238] In some embodiments, the encoder of the Informer model includes 3 layers of encoding sub-modules, and the decoder of the Informer model includes 2 layers of decoding sub-modules.

[0239] It should be noted that the division of units in the embodiments of the present application is illustrative, only a logical function division, and there may be other division methods in actual implementation. In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0240] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs.

[0241] It should be noted here that the above-mentioned device provided in the embodiments of this application can implement all the method steps implemented in the above-mentioned method embodiments and can achieve the same technical effects. Therefore, the parts and beneficial effects that are the same as those in the method embodiments will not be specifically described in this embodiment.

[0242] On the other hand, the embodiments of this application also provide a non-transitory readable storage medium. The non-transitory readable storage medium stores a computer program, and the computer program is used to cause a processor to execute the high-load cell intelligent identification method provided in the above-mentioned various embodiments.

[0243] It should be noted here that the non-transitory readable storage medium provided in the embodiments of this application can implement all the method steps implemented in the above-mentioned method embodiments and can achieve the same technical effects. Therefore, the parts and beneficial effects that are the same as those in the method embodiments will not be specifically described in this embodiment.

[0244] The non-transitory readable storage medium can be any available medium or data storage device that a computer can access, including but not limited to magnetic memories (such as floppy disks, hard disks, magnetic tapes, magneto-optical discs (MO), etc.), optical memories (such as CDs, DVDs, BDs, HVDs, etc.), and semiconductor memories (such as ROM, EPROM, EEPROM, non-volatile memories (NANDFLASH), solid-state drives (SSD)).

[0245] The technical solutions provided by the embodiments of the present application can be applied to a variety of systems, especially 5G systems. For example, the applicable systems can be Global System of Mobile Communication (GSM) systems, Code Division Multiple Access (CDMA) systems, Wideband Code Division Multiple Access (WCDMA) General Packet Radio Service (GPRS) systems, Long Term Evolution (LTE) systems, LTE Frequency Division Duplex (FDD) systems, LTE Time Division Duplex (TDD) systems, Long Term Evolution Advanced (LTE-A) systems, Universal Mobile Telecommunication System (UMTS), Worldwide Interoperability for Microwave Access (WiMAX) systems, 5G New Radio (NR) systems, etc. Both terminal devices and network devices are included in these various systems. The system may also include a core network part, such as an Evolved Packet System (EPS), a 5G System (5GS), etc.

[0246] The network device involved in the embodiments of the present application may be a base station, which may include multiple cells that provide services to terminals. Depending on the specific application scenarios, the base station may also be referred to as an access point, or may be a device in the access network that communicates with wireless terminal devices through one or more sectors over the air interface, or other names. The network device can be used to mutually replace the received air frames and Internet Protocol (IP) packets, and act as a router between the wireless terminal device and the rest of the access network, where the rest of the access network may include an Internet Protocol (IP) communication network. The network device can also coordinate the management of the attributes of the air interface. For example, the network device involved in the embodiments of the present application may be a network device (Base Transceiver Station, BTS) in a Global System for Mobile communications (GSM) or Code Division Multiple Access (CDMA), or may be a network device (NodeB) in a Wide-band Code Division Multiple Access (WCDMA), or may also be an evolved network device (evolutional Node B, eNB or e-NodeB) in a Long Term Evolution (LTE) system, a 5G base station (gNB) in a 5G network architecture (next generation system), or may be a Home evolved Node B (HeNB), a relay node, a femto, a pico, etc. The embodiments of the present application do not limit this. In some network architectures, the network device may include a centralized unit (centralized unit, CU) node and a distributed unit (distributed unit, DU) node, and the centralized unit and the distributed unit may also be arranged separately geographically.

[0247] Those skilled in the art should understand that the embodiments of the present application may be provided as a method, a system, or a computer program product. Therefore, the present application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) that contain computer-usable program code.

[0248] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer-executable instructions. These computer-executable instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0249] These processor-executable instructions can also be stored in a processor-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the processor-readable memory produce a manufactured article including instruction means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0250] These processor-executable instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are performed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0251] Obviously, those skilled in the art can make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalent technologies, this application is also intended to include these changes and modifications.

Claims

1. An intelligent identification method for high-load communities, characterized in that, it includes: Obtaining the time-series data of multiple load characteristic indicators of the community to be identified within a historical time period; Inputting the time-series data into the Informer model to obtain the multi-index prediction results output by the Informer model, where the multi-index prediction results include the prediction results of the multiple load characteristic indicators of the community to be identified within the prediction time period; Determining whether the community to be identified is a high-load community according to the prediction results and a preset high-load standard; wherein, the Informer model is trained based on the multi-index historical time-series data of multiple sample communities, and the multi-index historical time-series data includes the historical time-series data of the multiple load characteristic indicators.

2. The intelligent identification method for high-load communities according to claim 1, characterized in that, The obtaining of the time-series data of multiple load characteristic indicators of the community to be identified within a historical time period includes: Performing data cleaning on the original data of multiple load characteristic indicators of the community to be identified within a historical time period to obtain the first processed data of the multiple load characteristic indicators; Judging whether the community to be identified is an abnormal community according to the first processed data; the abnormal communities include communities with a proportion of missing sampling points greater than a first threshold, out-of-service communities, and test communities; In the case of determining that the community to be identified is not an abnormal community, performing numerical processing on the abnormal sampling points and missing sampling points in the first processed data to obtain the second processed data of the multiple load characteristic indicators; Performing standardization processing on the second processed data to obtain the time-series data of the multiple load characteristic indicators of the community to be identified within the historical time period.

3. The intelligent identification method for high-load communities according to claim 2, characterized in that, The performing of numerical processing on the abnormal sampling points and missing sampling points in the first processed data includes: Respectively obtaining the prediction data of each load characteristic indicator of the community to be identified within the historical time period based on the Prophet model; Determining the abnormal sampling points in the first processed data according to the prediction data and a preset confidence interval; Using the prediction data to perform numerical filling processing on the abnormal sampling points and missing sampling points in the first processed data.

4. The intelligent identification method for high-load communities according to claim 2, characterized in that, The determination method of the out-of-service community includes: If the monthly average traffic of a community is 0 for at least one month, then the community is an out-of-service community.

5. The intelligent identification method for high-load communities according to claim 2, characterized in that, The determination method of the test community includes: If the ratio between the maximum value of the monthly average traffic of a community and the minimum value of the monthly average traffic of the community is greater than a second threshold, then the community is a test community.

6. The intelligent identification method for high-load communities according to any one of claims 1 to 3, characterized in that, The load characteristic indicators include the uplink traffic of the cell, the downlink traffic of the cell, the average utilization rate of the uplink physical resource blocks (PRBs) of the cell, the average utilization rate of the downlink PRBs of the cell, and the occupancy rate of the control channel elements (CCEs) of the physical downlink control channel (PDCCH) of the cell.

7. The intelligent identification method for high-load cells according to claim 6, wherein, the high-load criteria include an uplink high-load criterion and a downlink high-load criterion; the uplink high-load criterion includes: within the prediction time period, the average value of the uplink traffic of the cell is greater than the uplink traffic threshold, and the average utilization rate of the uplink PRBs of the cell is greater than the uplink PRB average utilization rate threshold, then the cell meets the uplink high-load criterion; the downlink high-load criterion includes: within the prediction time period, the average value of the downlink traffic of the cell is greater than the downlink traffic threshold, and at least one of the following is satisfied: the average utilization rate of the downlink PRBs of the cell is greater than the downlink PRB average utilization rate threshold, the occupancy rate of the CCEs of the PDCCH channel of the cell is greater than the PDCCH channel CCE occupancy rate threshold, then the cell meets the downlink high-load criterion.

8. The intelligent identification method for high-load cells according to claim 7, wherein, determining whether the cell to be identified is a high-load cell according to the prediction result and the preset high-load criteria includes: if it is determined according to the prediction result that the cell to be identified meets the uplink high-load criterion and / or the downlink high-load criterion, then it is determined that the cell to be identified is a high-load cell.

9. The intelligent identification method for high-load cells according to claim 1, wherein, the encoder of the Informer model includes multiple layers of encoding sub-modules, and each layer of the encoding sub-module includes a multi-head probabilistic sparse self-attention sub-layer and a self-attention distillation sub-layer; the decoder of the Informer model includes multiple layers of decoding sub-modules, and each layer of the decoding sub-module includes a masked multi-head probabilistic sparse self-attention sub-layer and a multi-head attention sub-layer; wherein, the output features of the encoder are input into the masked multi-head probabilistic sparse self-attention sub-layer of the decoder.

10. The intelligent identification method for high-load cells according to claim 9, wherein, the training loss function of the Informer model includes the Log-Cosh loss function.

11. The intelligent identification method for high-load cells according to claim 9, wherein, the encoder of the Informer model includes 3 layers of the encoding sub-modules, and the decoder of the Informer model includes 2 layers of the decoding sub-modules.

12. An electronic device, wherein, it includes a memory, a transceiver, and a processor; the memory is used to store computer programs; the transceiver is used to transmit and receive data under the control of the processor; the processor is used to read the computer programs in the memory and perform the following operations: obtain the time-series data of multiple load characteristic indicators of the cell to be identified in the historical time period; Input the time-series data into the Informer model to obtain the multi-metric prediction results output by the Informer model. The multi-metric prediction results include the prediction results of the multiple load characteristic metrics of the cell to be identified within the prediction time period. Determine whether the cell to be identified is a high-load cell according to the prediction results and a preset high-load standard. Among them, the Informer model is trained based on the multi-metric historical time-series data of multiple sample cells. The multi-metric historical time-series data includes the historical time-series data of the multiple load characteristic metrics.

13. The electronic device according to claim 12, characterized in that the obtaining of the time-series data of multiple load characteristic metrics of the cell to be identified within the historical time period includes: performing data cleaning processing on the original data of multiple load characteristic metrics of the cell to be identified within the historical time period to obtain the first processed data of the multiple load characteristic metrics; judging whether the cell to be identified is an abnormal cell according to the first processed data; the abnormal cells include cells with a proportion of missing sampling points greater than a first threshold, out-of-service cells, and test cells; in the case of determining that the cell to be identified is not an abnormal cell, performing numerical processing on the abnormal sampling points and missing sampling points in the first processed data to obtain the second processed data of the multiple load characteristic metrics; performing standardization processing on the second processed data to obtain the time-series data of the multiple load characteristic metrics of the cell to be identified within the historical time period.

14. The electronic device according to claim 13, characterized in that the performing of numerical processing on the abnormal sampling points and missing sampling points in the first processed data includes: respectively obtaining the prediction data of each load characteristic metric of the cell to be identified within the historical time period based on the Prophet model; determining the abnormal sampling points in the first processed data according to the prediction data and a preset confidence interval; using the prediction data to perform numerical filling processing on the abnormal sampling points and missing sampling points in the first processed data.

15. The electronic device according to claim 13, characterized in that the determination method of the out-of-service cell includes: if the monthly average traffic of a cell is 0 for at least one month, then the cell is an out-of-service cell.

16. The electronic device according to claim 13, characterized in that the determination method of the test cell includes: if the ratio between the maximum value and the minimum value of the monthly average traffic of a cell is greater than a second threshold, then the cell is a test cell.

17. The electronic device according to any one of claims 12 to 14, characterized in that the load characteristic metrics include the uplink traffic of the cell, the downlink traffic of the cell, the average utilization rate of the uplink physical resource blocks (PRBs) of the cell, the average utilization rate of the downlink PRBs of the cell, and the occupancy rate of the physical downlink control channel (PDCCH) control channel elements (CCEs) of the cell.

18. The electronic device according to claim 17, characterized in that the high-load standard includes an uplink high-load standard and a downlink high-load standard; The uplink high-load standard includes: within the predicted time period, the average uplink traffic of the cell is greater than the uplink traffic threshold, and the average uplink PRB utilization rate of the cell is greater than the uplink PRB average utilization rate threshold, then the cell meets the uplink high-load standard; The downlink high-load standard includes: within the predicted time period, the average downlink traffic of the cell is greater than the downlink traffic threshold, and at least one of the following is satisfied: the average downlink PRB utilization rate of the cell is greater than the downlink PRB average utilization rate threshold, the PDCCH channel CCE occupancy rate of the cell is greater than the PDCCH channel CCE occupancy rate threshold, then the cell meets the downlink high-load standard.

19. The electronic device according to claim 18, characterized in that determining whether the cell to be identified is a high-load cell according to the prediction result and a preset high-load standard includes: if it is determined according to the prediction result that the cell to be identified meets the uplink high-load standard and / or the downlink high-load standard, then it is determined that the cell to be identified is a high-load cell.

20. The electronic device according to claim 12, characterized in that the encoder of the Informer model includes multiple layers of encoding sub-modules, and each layer of the encoding sub-module includes a multi-head probabilistic sparse self-attention sub-layer and a self-attention distillation sub-layer; the decoder of the Informer model includes multiple layers of decoding sub-modules, and each layer of the decoding sub-module includes a masked multi-head probabilistic sparse self-attention sub-layer and a multi-head attention sub-layer; wherein, the output features of the encoder are input into the masked multi-head probabilistic sparse self-attention sub-layer of the decoder.

21. The electronic device according to claim 20, characterized in that the training loss function of the Informer model includes a Log-Cosh loss function.

22. The electronic device according to claim 20, characterized in that the encoder of the Informer model includes 3 layers of the encoding sub-modules, and the decoder of the Informer model includes 2 layers of the decoding sub-modules.

23. A high-load cell intelligent identification device, characterized in that including: an acquisition unit, configured to acquire the time-series data of multiple load characteristic indicators of the cell to be identified in a historical time period; a prediction unit, configured to input the time-series data into an Informer model to obtain a multi-index prediction result output by the Informer model, where the multi-index prediction result includes the prediction results of the multiple load characteristic indicators of the cell to be identified in a predicted time period; a determination unit, configured to determine whether the cell to be identified is a high-load cell according to the prediction result and a preset high-load standard; wherein, the Informer model is trained based on the multi-index historical time-series data of multiple sample cells, and the multi-index historical time-series data includes the historical time-series data of the multiple load characteristic indicators.

24. A non-transitory readable storage medium, characterized in that The non-transitory readable storage medium stores a computer program for causing a processor to execute the method according to any one of claims 1 to 11.

Images