Method, apparatus and computer-readable storage medium for predicting base station traffic

Abstract

To provide a method, apparatus and computer-readable storage medium for predicting the traffic of a base station.SOLUTION: A method includes: acquiring past traffic data of a base station; and generating future traffic data of the base station based on the past traffic data using a predictor. The predictor is trained according to a first rule. The first rule is clustering N first training curves into M categories based on similarity, where each of the N first training curves graphically represents traffic values of N first training datasets changing over time, and M is less than or equal to N. The method further includes: selecting a representative curve for each category; identifying the M first training datasets corresponding to the respective representative curves as M final training datasets; and training the predictor using the M final training datasets.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present application relates to the field of artificial intelligence, and in particular to a method, an apparatus, and a non-transitory computer-readable storage medium for predicting the flow rate of a base station. [Background technology]

[0002] A base station is a radio transceiver station that transmits information between a mobile switching center and a mobile terminal in a specific radio coverage area. With the development of mobile communication network services, the deployment and management of mobile communication base stations becomes particularly important. For a deployed base station, the data flow rate and power consumption that need to be processed vary with time. For example, if the base station is always fully on during the time period when the data flow rate is very low, a large amount of power will actually be wasted.

[0003] To achieve the power saving goal, it is necessary to design a power saving strategy that can adapt to the changes in data traffic and flexibly control base stations to be turned on or off. Even if a unified power saving strategy is used to manage all base stations, problems will arise because there are a huge number of base stations across the country and the traffic patterns of base stations in different regions are very different, so a unified power saving strategy will definitely cause more problems.

[0004] Therefore, there is an urgent need for a method that can adaptively predict the traffic of different base stations in different regions, and pre-deploy and manage power-saving methods for base stations based on the traffic prediction results of each base station to save power consumption of base stations across the country. Summary of the Invention

[0005] The present application has been made in consideration of the above problems. In an exemplary aspect, the present disclosure provides a method for predicting the flow rate of a base station, comprising: obtaining past flow rate data of the base station; and generating future flow rate data of the base station based on the past flow rate data by a predictor; the predictor is trained according to a first rule, the first rule including: clustering N first training curves into M categories based on similarity, the N first training curves being graphs of flow rate values ​​in the N first training data that change over time, where M is less than or equal to N; selecting a representative curve for each category, and setting the M first training data corresponding to each representative curve as M final training data; and training the predictor by the M final training data.

[0006] In some embodiments, training the predictor with the M final training data includes the steps of adding M representative curves corresponding to the M final training data to a curve library and setting a label for each representative curve, and training the predictor with the M final training data and their labels.

[0007] In some embodiments, the step of generating future flow data for the base station by a predictor based on the past flow data includes a step of determining a curve from the curve library that best matches a curve of the past flow data, the curve of the past flow data being a graphical representation of flow values ​​in the past flow data that change over time, and a step of generating future flow data for the base station based on the past flow data and a label corresponding to the best matching curve.

[0008] In some embodiments, if the degree of match between the historical flow data curve and the best matching curve is less than a first threshold, the method further includes a step in which the generated future flow data is scaled.

[0009] In some embodiments, the step of scaling the generated future flow data includes the steps of calculating a ratio value corresponding to each time point based on the curve of the historical flow data and the best matching curve, and multiplying the generated future flow data by the ratio value of the corresponding time point on a point-by-point basis to generate scaled future flow data.

[0010] In some embodiments, if the degree of matching between the curve of the past flow data and the best matching curve is lower than a second threshold, the method further includes a step of dividing the curve of the past flow data into a plurality of partial curves, a curve that best matches the partial curve for each of the plurality of partial curves is determined from the curve library, and partial future flow data of the base station is generated based on the past flow data and a label corresponding to the best matching curve, and a step of combining a plurality of partial future flow data corresponding to the plurality of partial curves to generate the future flow data of the base station.

[0011] In some embodiments, the step of generating partial future flow data for the base station based on the past flow data and a label corresponding to the best matching curve includes a step of generating first future flow data for the base station based on the past flow data and a label corresponding to the best matching curve, and a step of extracting a portion corresponding to a time period of the partial curve from the first future flow data to obtain the partial future flow data.

[0012] In some embodiments, if the degree of matching between the curve of the historical flow data and the best matching curve is lower than a third threshold, the method further includes a step of identifying at least one segment in the curve of the historical flow data, where the degree of matching between the at least one segment and a corresponding segment of the best matching curve is lower than a fourth threshold; a step of individually replacing each of the at least one segment with the corresponding segment of the best matching curve to generate modified historical flow data; and a step of generating future flow data for the base station based on the modified historical flow data and a label corresponding to the best matching curve.

[0013] In another exemplary aspect, the present disclosure provides an apparatus for predicting the flow rate of a base station, comprising: a past flow data acquisition unit for acquiring past flow data of the base station; and a prediction unit for generating future flow data of the base station based on the past flow data, wherein the prediction unit is trained according to a first rule, the first rule including: clustering N first training curves into M categories based on similarity, the N first training curves being graphs representing the flow values ​​of the N first training data that change over time, where M is less than or equal to N; selecting a representative curve for each category, and setting the M first training data corresponding to each representative curve as M final training data; and training the predictor by the M final training data.

[0014] In yet another exemplary aspect, the present disclosure provides a non-transitory computer-readable storage medium having stored thereon computer instructions, the computer instructions, when executed by a processor, performing the method for predicting a flow rate of a base station as described above. [Brief description of the drawings]

[0015] [Figure 1] FIG. 1 illustrates a schematic diagram of an exemplary method for predicting a flow rate of a base station according to an embodiment of the present disclosure. [Diagram 2] FIG. 2 shows a schematic diagram of a first training method of a predictor for predicting a flow rate of a base station according to an embodiment of the present disclosure. [Diagram 3] FIG. 3 shows a schematic diagram of a second training method of a predictor for predicting a flow rate of a base station according to an embodiment of the present disclosure. [Figure 4] FIG. 4 illustrates a schematic diagram of an exemplary method for predicting a flow rate of a base station according to an embodiment of the present disclosure. [Diagram 5] FIG. 5 illustrates a schematic diagram of an exemplary method for predicting a flow rate of a base station according to an embodiment of the present disclosure. [Figure 6] FIG. 6 illustrates a schematic diagram of an exemplary method for adjusting a forecast result of a flow rate of a base station according to an embodiment of the present disclosure. [Figure 7] FIG. 7 illustrates a schematic diagram of an exemplary method for predicting the flow rate of a base station by time period, according to an embodiment of the present disclosure. [Figure 8] FIG. 8 illustrates a schematic diagram of an exemplary method for pre-processing historical flow data of a base station according to an embodiment of the present disclosure. [Figure 9] FIG. 9 illustrates a schematic diagram of performing power saving management on a base station according to predicted future flow data of the base station according to an embodiment of the present disclosure. [Figure 10] FIG. 10 illustrates a schematic diagram of an apparatus for predicting flow rate of a base station according to an embodiment of the present disclosure. [Figure 11] FIG. 11 illustrates a schematic diagram of a non-transitory computer-readable storage medium for predicting a flow rate of a base station according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] Preferred embodiments of the present disclosure are described in greater detail below.

[0017] It should be noted that the steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel, and method embodiments may include other steps and / or omit certain steps.

[0018] As used herein, the term "including" and variations thereof are not intended to be limiting and thus mean "including but not limited to." The term "based on" means "based at least in part on." The term "one embodiment" means "at least one embodiment," the term "another embodiment" means "at least one additional embodiment," and the term "some embodiments" means "at least some embodiments." Relevant definitions of other terms are provided in the description below.

[0019] It should be understood that concepts such as "first", "second", etc. referred to in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of functions performed by these devices, modules or units.

[0020] It should be noted that the modifications "a" and "multiple" referred to in this disclosure are exemplary and not limiting, and should be understood by one of ordinary skill in the art as "one or more" unless the context clearly dictates otherwise.

[0021] In order to flexibly control the on or off of a base station according to the change of data flow rate to achieve the purpose of power saving, the present disclosure provides a method that can predict the flow rate of a base station.

[0022] To predict the flow rate of a base station, a machine learning-based predictor needs to be used. In the training process of a conventional predictor, the training data is selected randomly. However, the method of randomly selecting training data often has the problem of low diversity and low representativeness. For example, there may be a large number of repetitions in the randomly selected training data, which means that even if trained for a long time by a large amount of training data, the obtained predictor still cannot learn comprehensive knowledge sufficiently, and the prediction accuracy of the predictor is greatly reduced. The usual way to solve this problem is to further expand the training dataset to increase the training amount and extend the training time.

[0023] In order to simplify the amount of training data so that the training dataset contains more diverse training data and at the same time achieve more effective training and obtain a better predictor, the present disclosure proposes that the predictor be trained first by selected representative training data.

[0024] Fig. 1 shows a schematic diagram of a method for predicting future flow data of a base station according to an embodiment of the present disclosure. Fig. 2 shows a schematic diagram of a first training method of a predictor for predicting flow rate of a base station according to an embodiment of the present disclosure.

[0025] As shown in FIG. 1, the method for predicting the flow rate of a base station proposed by the present disclosure includes: obtaining past flow rate data of the base station; and generating future flow rate data of the base station by a predictor based on the past flow rate data.

[0026] The past flow data of the base station shown in FIG. 1 may be, for example, the past flow data of the base station in the past, for example, the past week, the past month, or the past three months. The past flow data of the base station may also be expressed in the form of a sequence consisting of each time point and the corresponding flow value, or in the form of a curve. For clarity, an example of past flow data is shown in the form of a curve in FIG. 1.

[0027] In addition, in Fig. 1, the future flow rate data of the base station predicted by the predictor is shown by a dotted line immediately after the curve of the past flow rate data. The predicted future flow rate data of the base station can also be expressed in the form of a sequence consisting of each time point and its corresponding flow rate value.

[0028] The predictions shown in Figure 1 have been trained according to the first rule. Figure 2 shows a schematic diagram of how a predictor is trained according to the first rule.

[0029] As shown in Figure 2, the method in which a predictor is trained according to the first rule typically includes three steps: clustering (step 1), selecting representatives (step 2), and training (step 3).

[0030] In step 1, N first training curves are clustered into M categories based on similarity. The N first training curves are graphs of flow values ​​of N first training data that change over time, where M is less than or equal to N. For example, the N first training data can correspond to flow data obtained from N base stations in a certain geographical range. For example, in FIG. 2, for example, M=4, that is, N first training curves are clustered into 4 categories, and each training curve for each category is similar, and different categories usually have different curve appearances.

[0031] For example, the N first training data can be clustered using the Kmeans clustering method. Taking the number of categories as four as an example, first, in the first step, four first training curves can be randomly selected from the N first training curves as the initial center points of each of the four categories. Then, in the second step, for all the first training curves, the distance from the curve to each center point is calculated, and each first training curve is classified into the category of the center point with the closest distance. Then, in the third step, the center point for each of the four divided categories is recalculated (for example, the curve of the average value of the first training curves in the same category can be used as the new center point). Then, the above second and third steps can be repeated until the classification result does not change or a predetermined number of iterations is reached.

[0032] Optionally, the N first training curves may be clustered in any other manner known in the art.

[0033] In the second step, a representative curve is selected for each category, and M first training data corresponding to each representative curve are taken as M (for example, four as shown) final training data. As shown in the figure, the most representative training curve is selected from the four categories obtained by clustering (for example, the curve corresponding to base station 3, the curve corresponding to base station 8, the curve corresponding to base station 32, and the curve corresponding to base station 7), and the training data corresponding to the curves of these four base stations are taken as the final training data. This means that only one most representative training data is selected for each category, thereby avoiding excessive training time due to redundancy of training data of the same category.

[0034] For example, when using the above-mentioned K-means clustering method, for the final clustering results, the training curve that is closest to the final center point in each category can be determined as the representative curve for each category.

[0035] In the third step, a predictor is trained by the M (eg, 4) final training data selected in the second step.

[0036] In the example shown in FIG. 2, if there are initially 1000 pieces of first training data corresponding to 1000 base stations, the training time of each training data is Δ. In this case, if the predictor is directly trained by these 1000 pieces of first training data without sorting, a total training time of 1000xΔ is required. In contrast, if the training method based on clustering and sorting of representative curves shown in FIG. 2 is adopted, the predictor only needs to be trained by four representative training data in the end, and a total training time of only 4xΔ is required. In these two cases, the original 1000 pieces of first training data contain a large amount of similar redundant data, so the knowledge learned by the predictor is actually not very different, but compared with the case where clustering and sorting of training data is not performed, the training time has been significantly reduced after introducing clustering and sorting.

[0037] If we want to train the predictor with even more knowledge, we can further extend the first training data, cluster and filter it, and learn more knowledge in a shorter time.

[0038] The first rule for refining training data based on the clustering and sorting process proposed by the present disclosure, and an exemplary method for predicting base station flow rates by a trained predictor, have been described above in conjunction with Figures 1 and 2.

[0039] In addition, in order to avoid knowledge confusion caused by conflicting or similar waveforms during the training process, and to more effectively find knowledge related to the past data of the measurement waiting base station itself during the prediction process, the present disclosure also proposes another training rule, namely, during training, add a label to each training data to train a predictor, and during prediction, use the past flow data and corresponding labels as input data to predict future flow data.

[0040] Fig. 3 shows a schematic diagram of a second training method of a predictor for predicting the flow rate of a base station according to an embodiment of the present disclosure. Fig. 4 shows a schematic diagram of another exemplary method for predicting the flow rate of a base station according to an embodiment of the present disclosure.

[0041] As shown in Figure 3, in the present disclosure, a predictor for predicting the flow rate of a base station can also be trained according to a second rule. Training a predictor according to the second rule includes the following steps: Step 1: obtain K second training curves corresponding to K second training data. Step 2: K second training curves are added to the curve library and a label is assigned to each second training curve. Step 3: A predictor is trained by the K second training data and their labels.

[0042] 3 shows four pieces of second training data corresponding to base station 1, base station 3, base station 4, and base station n. That is, K = 4. Label 1, label 2, label 3, and label 4 are set to the four pieces of second training data corresponding to base station 1, base station 3, base station 4, and base station n, respectively.

[0043] The label can be regarded as a unique ID for each training data, and is used to indicate the waveform shape corresponding to the training data, which makes it easier for the predictor to distinguish each training data during the training process. Therefore, when a label is added to each training data, the predictor can effectively distinguish different training data through the label during the training process, so that the predictor can learn more accurate knowledge while avoiding knowledge confusion caused by conflicting or similar waveforms.

[0044] Similarly, when the predictor is trained by the training method shown in FIG. 3, during the prediction process, the predictor can more effectively identify knowledge related to the past data of the measurement-waiting base station itself based on various labels, and predict future flow data with more accuracy.

[0045] For example, as shown in FIG. 4, the step of generating future flow rate data of a base station by a predictor based on past flow rate data includes the following steps: Step 1: Obtain a curve corresponding to the past flow rate data of the base station waiting for measurement. For example, the curve of the past flow rate data of the base station waiting for measurement is a graph representation of the flow rate values ​​in the past flow rate data that change over time. Step 2, from the curve library established during the training process, determine the curve that best matches the curve of the historical flow data of the base station waiting to be measured (e.g., the curve corresponding to label 4 shown in the figure). Step 3, based on the past flow data of the measurement-waiting base station and the label (e.g., label 4) corresponding to the best matching curve, future flow data of the measurement-waiting base station is generated.

[0046] It should be noted that the training and prediction method of creating a curve library for training data and adding labels has been described above in conjunction with FIGS. 3 and 4, but this does not mean that the training and prediction method described in conjunction with FIGS. 3 and 4 is necessarily performed independently of the embodiment of FIGS. 1 and 2. In fact, the training method described with reference to FIG. 3 can also be used in combination with the training method shown in FIG. 2. For example, N pieces of first training data can be first clustered and selected, thereby obtaining M pieces of final training data. Then, with the obtained M pieces of final training data, a curve library can be created and labeled to perform training, as shown in FIG. 3, that is, the M pieces of final training data in FIG. 2 are the K pieces of second training data in FIG. 3.

[0047] Specifically, for the M final training data, corresponding curves can be added to a curve library and corresponding labels can be added to each curve, and then a predictor can be trained with the M final training data and their corresponding labels by the base station.

[0048] In this case, the training of a predictor by M final training data may include the following steps: First, M representative curves corresponding to M final training data are added to a curve library, and a label is set for each representative curve. For example, referring back to FIG. 2, after selecting four representative curves (e.g., curves corresponding to base station 3, base station 8, base station 32, and base station 7) from the four categories obtained by clustering as shown in FIG. 2, these four representative curves can be added to a curve library, and labels 1, 2, 3, and 4 can be set for these four representative curves, respectively. Then, a predictor can be trained by M final training data (e.g., data corresponding to base station 3, base station 8, base station 32, and base station 7) and their labels.

[0049] In this combined training, it is equivalent to integrating the training data clustering step, the representative curve selection step, the curve library creation step, and the labeled training step, and finally the training process is performed only once.

[0050] Similarly, if the predictor creates a curve library as shown in FIG. 3 during the training process, the base station flow rate may be predicted using a prediction method involving a label matching process as shown in FIG. 4 during the prediction process.

[0051] The curve library and label-based training method proposed by the present disclosure and an exemplary method for predicting base station flow rate by a trained predictor have been described above in conjunction with FIG. 3 and FIG. 4 .

[0052] In some cases, even if the above embodiment of the training data screening method is used and labeled to train a predictor, it cannot ensure that all types of flow curves are included in the training set, and therefore cannot ensure that the predictor can learn comprehensive knowledge.

[0053] In addition, even if the training set contains a flow rate curve similar to the future flow rate data of the base station waiting for measurement, it is impossible to perfectly match the prediction results with the similar curve, and differences in details may cause the overall prediction performance to deteriorate.

[0054] To solve this problem, the present disclosure further proposes a method for further adjusting the prediction result when the degree of matching between the past flow rate data of the base station waiting for measurement and the curves in the curve library is low.

[0055] 5 shows a schematic diagram of another method for predicting the flow rate of a base station according to an embodiment of the present disclosure. FIG. 6 shows a schematic diagram of an exemplary method for adjusting the prediction result of the flow rate of a base station according to an embodiment of the present disclosure.

[0056] As shown in Figure 5, when the degree of matching between the curve of the past flow data of the base station waiting for measurement and the curve that best matches it (e.g., the curve corresponding to label 2) is lower than the first threshold, if the labeled prediction method shown in Figure 4 is directly used, a large deviation will occur between the prediction result (i.e., future flow data) and the true value ref, leading to the prediction result not matching the true value.

[0057] As shown in the figure, the first half of the results output by the predictor is more consistent with the true value ref. However, for the second half, there is an obvious deviation between the two. This deviation is usually caused by the fact that the curves in the curve library do not cover all types of curves. Due to the incomplete knowledge coverage, the curve corresponding to the best label 2 matched during the prediction process cannot actually reflect the actual historical flow data of the measurement waiting base station, which will cause further deviation in the prediction results.

[0058] In order to address the deviation problem of the prediction results as described above, the present disclosure further proposes scaling the generated future flow data, i.e., adjusting the waveform of the future flow data, if the degree of matching between the curve of the past flow data of the base station waiting for measurement and the best matching curve (e.g., the curve corresponding to label 2) is lower than a first threshold.

[0059] For example, the matching degree between the curve of the past flow data of the base station waiting for measurement and the best matching curve can be calculated by the difference in flow values ​​at each time point between the two curves. For example, the curve of the past flow data of the base station waiting for measurement and the best matching curve both have 100 corresponding flow values ​​at 100 time points. If the difference between the 100 corresponding flow values ​​between the two curves is all within the allowable error, it can be determined that the matching degree between the two is 100%. If the difference between only 50 corresponding flow values ​​between the two curves is within the allowable error, it can be determined that the matching degree is 50%, and so on.

[0060] For example, in this case, the first threshold can be set to 80%, and future flow data generated will be scaled if the difference between corresponding flow values ​​less than 80 is less than the tolerance.

[0061] In some examples, the waveform of the generated future flow data can be adjusted based on downscaling or upscaling. The downscaling or upscaling ratio may be calculated, for example, by a formula summarizing experiences, or may be obtained by finding a certain regularity using other statistical methods. For example, it is considered that the base station is usually in a state where the flow rate is low in the early morning hours and is in a state where the flow rate is high in the morning peak hours. In this case, the flow rate value corresponding to the early morning hours of the generated future flow data can be downscaled by a predetermined ratio, and the flow rate value corresponding to the morning peak hours of the generated future flow data can be upscaled by a predetermined ratio.

[0062] In another example, the generated future flow data can be scaled based on the ratio value between the curve of the base station's historical flow data and the curve that best matches. For example, Figure 6 shows a schematic diagram of an exemplary method for adjusting the prediction result of the flow rate of a base station according to an embodiment of the present disclosure.

[0063] As shown in FIG. 6, the corresponding ratio value A / A' at each time t between the curve of the past flow data of the base station waiting for measurement and the curve that best matches (e.g., the curve corresponding to label 2) is calculated for each point, and a scaling ratio dictionary including the corresponding ratio value at each time is generated. A represents the flow value at time t of the curve of the past flow data of the base station waiting for measurement, and A' represents the flow value at the same time t of the curve that best matches (e.g., the curve corresponding to label 2). For example, as shown in FIG. 6, the curve of the past flow data of the base station waiting for measurement and the curve that best matches (e.g., the curve corresponding to label 2) both include 24 flow values ​​from 0 to 23 o'clock, and the generated scaling ratio dictionary can include various ratio values ​​corresponding to 0 o'clock, 1 o'clock, ..., 23 o'clock. Then, referring to the generated scaling ratio dictionary, the future flow data of the base station waiting for measurement generated by the predictor is multiplied by the ratio value at the corresponding time for each point to generate scaled future flow data.

[0064] An example was described above with reference to FIG. 6 in which future flow data generated is scaled if the curve of the historical flow data of the base station waiting to be measured does not closely match the best matching curve, but this is actually a post-processing technique that corrects the prediction results if the match is not close.

[0065] In some cases, if the degree of matching between the curve of the past flow data of the base station not waiting for measurement and the best matching curve is not high, a time-period prediction method in which the past flow data of the base station not waiting for measurement is matched by time period can also be considered.

[0066] FIG. 7 illustrates a schematic diagram of an exemplary method for predicting the flow rate of a base station by time period, according to an embodiment of the present disclosure.

[0067] For example, as shown in FIG. 7, if the degree of matching between the curve of the past flow data of the base station not waiting for measurement and the best matching curve is lower than the second threshold, first divide the curve of the past flow data into multiple partial curves, that is, into multiple segments. For convenience of explanation, FIG. 7 shows an example in which the past flow data of the base station not waiting for measurement is divided into two segments. That is, if the past flow data of the base station not waiting for measurement includes flow values ​​for the past 24 hours, the curve of the previous 12 hours (i.e., corresponding to 0 to 11 hours) is used as the first partial curve, and the curve of the next 12 hours (i.e., corresponding to 12 to 23 hours) is used as the second partial curve.

[0068] After the division, the first and second partial curves as shown in the figure are obtained, and then, for each partial curve, a curve that best matches each partial curve is determined from the curve library. For example, for the first partial curve of the previous 12 hours (i.e., corresponding to 0 to 11 hours), the determined best matching curve is the curve corresponding to label 1, and for the second partial curve of the later 12 hours (i.e., corresponding to 12 to 23 hours), the determined best matching curve is the curve corresponding to label 2.

[0069] Then, for each partial curve, partial future flow data of the base station waiting to be measured can be generated based on the past flow data of the base station waiting to be measured and the label corresponding to the best matching curve. That is, as shown in the figure, the past flow data of the base station waiting to be measured and label 1 can be input to a predictor to generate first future flow data, and the past flow data of the base station waiting to be measured and label 2 can be input to a predictor to generate second future flow data.

[0070] Then, a portion of the first partial curve corresponding to the time period (0:00-11:00) may be extracted from the first future flow data to be used as the first partial future flow data generated for the first partial curve. A portion of the second partial curve corresponding to the time period (12:00-23:00) may be extracted from the second future flow data to be used as the second partial future flow data generated for the second partial curve.

[0071] Finally, the multiple parts of the future flow data corresponding to the multiple partial curves are combined to generate the future flow data of the base station waiting for measurement. That is, in the example shown in Figure 7, the first partial future flow data (0 to 11 o'clock) and the second partial future flow data (12 to 23 o'clock) are combined to form the final prediction result of the base station waiting for measurement.

[0072] For example, as shown in Fig. 7, when predictions are made by time period and then combined to obtain the final prediction result, the degree of agreement between the final prediction result and the true value ref is very high.

[0073] For convenience of explanation, in the example of Figure 7, the curve of the past flow data of the base station waiting for measurement is divided equally into two segments, but this is merely an example and is not limited to this. For example, according to the actual situation, the curve of the past flow data of the base station waiting for measurement can be divided equally or unevenly into more segments, such as three segments, four segments, five segments, etc., thereby realizing more accurate prediction.

[0074] In addition to the method of forecasting by time period and combining the results proposed above, in order to improve the forecast accuracy, the input data (i.e., the past flow data of the measurement waiting base station) is preprocessed to improve the matching degree with the representative curve in the curve library. This is because even if the best matching curve is found in the curve library, the problem of the matching degree being too low may still exist. For example, only a part of the past flow data may be similar to the best matching curve, while the other part may not be similar, which may cause errors in the final forecast result.

[0075] In order to solve the above problem, in some embodiments, for example, when the matching degree between the curve of the past flow data of the measurement waiting base station and the best matching curve is lower than a third threshold, the matching degree between the corresponding segment of the best matching curve is lower than a fourth threshold and at least one segment in the curve of the past flow data can be identified.For example, the fourth threshold can be the same as the third threshold, or can be different.

[0076] After identifying at least one segment having a lower degree of matching, each of the at least one segment of the curve of the historical flow data can be individually replaced with a corresponding segment of the best matching curve to generate modified historical flow data.

[0077] Then, referring to FIG. 4, the corrected historical flow data and the label of the best matching curve can be input to generate future flow data for the base station waiting for measurements.

[0078] FIG. 8 illustrates a schematic diagram of an exemplary method for pre-processing historical flow data of a base station according to an embodiment of the present disclosure.

[0079] For example, as shown in the figure, only the first half of the curve of the historical flow data of the base station waiting for measurement is similar to the best matching curve labeled 1, and the second half is significantly different. In this case, the second half of the curve of the historical flow data of the base station waiting for measurement can be replaced with the second half of the best matching curve labeled 1, that is, the first half of the curve of the historical flow data of the base station waiting for measurement and the second half of the best matching curve labeled 1 are combined to obtain the modified historical flow data.

[0080] By combining, the modified historical flow data as shown in Fig. 8 is obtained, and then compared with the original historical flow data of the base station waiting for measurement, the modified historical flow data obviously has a higher matching degree with the curve with label 1, which is the best match. Then, the modified historical flow data and label 1 can be input into the predictor to obtain the prediction result of the future flow data of the base station waiting for measurement.

[0081] In the example of FIG. 8, since the input data has been modified, that is, only the first half of the final input data is the actual past flow data, only the corresponding first half of the prediction result output by the predictor is close to the true value. As for the second half of the prediction result, since the corresponding part of the input data has been modified, there is a possibility that there will be a large difference between the true value. Therefore, during actual use, it is usually necessary to extract and use the first half of the prediction result output by the predictor, and discard the second half.

[0082] For the latter part, a more accurate prediction result can be obtained in a similar manner. For example, a curve that best matches the latter part of the curve of the past flow data of the base station waiting for measurement can be found in the curve library (e.g., the curve labeled 2, not shown). Assuming that only the latter part of the past flow data of the base station waiting for measurement is similar to the corresponding part of the curve labeled 2 and the first half is not similar, the same method can be used to replace the first half of the curve of the past flow data of the base station waiting for measurement with the corresponding first half of the curve labeled 2 to obtain the corrected past flow data. Then, the corrected past flow data and label 2 can be input into the predictor to extract the latter half of the prediction result as the final prediction result for the latter half.

[0083] Optionally, the first and second parts extracted as above can be combined to obtain a complete prediction result.

[0084] In addition, in the curve of the past flow rate data shown in FIG. 8, only the latter half does not match the curve labeled 1, but this is just an example. In reality, there may be multiple discrete segments in the curve of the past flow rate data that do not match the curve labeled 1. In this case, referring to the above method, the multiple non-matching discrete segments can be replaced with the corresponding segments of the curve labeled 1 to obtain the corrected past flow rate data.

[0085] If multiple non-matching segments are involved, the prediction results can be extracted and combined according to a similar method as above, so as to obtain a complete prediction result of future flow data of the measurement-waiting base station.

[0086] Various embodiments of the present disclosure for predicting the flow rate of a base station have been described above in conjunction with Figures 1 to 8. It should be noted that, unless clearly inappropriate, multiple training or prediction methods described above can be used in combination.

[0087] After predicting the future flow data of base station according to various embodiments described above, can perform power saving management for base station according to the future flow data. Figure 9 shows a schematic diagram of performing power saving management for base station according to the predicted future flow data of base station according to the embodiment of the present disclosure.

[0088] As shown in Figure 9, the power saving level of the base station, i.e., the ES level, can be determined based on the predicted future flow data of the base station. Specifically, for example, the current ES level of the base station can be determined from the provided ES level table according to the lookup table. For example, if the predicted future flow rate of the base station at time t is 20,000, it can be determined that the power saving level at time t is the ES2 level, and the base station can be switched on and off based on ES2 and the set related strategy, thereby achieving the purpose of power saving.

[0089] In another aspect, the present disclosure further provides an apparatus for predicting a flow rate of a base station. Figure 10 shows a schematic diagram of an apparatus 1000 for predicting a flow rate of a base station according to an embodiment of the present disclosure.

[0090] As shown in Fig. 10, the device 1000 includes a past flow data acquiring unit 1010 and a prediction unit 1020. The past flow data acquiring unit 1010 is used to acquire the past flow data of the base station, and the prediction unit 1020 is used to generate the future flow data of the base station according to the past flow data.

[0091] The prediction unit 1010 is trained according to a first rule. Specifically, the first rule includes: clustering N first training curves into M categories based on similarity, the N first training curves being graphs of flow values ​​of N first training data that change over time, where M is less than or equal to N; selecting a representative curve for each category, and taking the M first training data corresponding to each representative curve as M final training data; and training the predictor by the M final training data.

[0092] The first rule above is similar to the first rule in the method for training a predictor with reference to FIG.

[0093] The prediction unit 1010 may also be trained according to the second rule described with respect to FIG.

[0094] In addition, the features of the various embodiments of the method for predicting the flow rate of a base station described with reference to Figures 1 to 8 are also applicable to the device 1000 for predicting the flow rate of a base station, unless expressly inapplicable.

[0095] In another aspect, the present disclosure further provides a non-transitory computer-readable storage medium for predicting the flow rate of a base station. Figure 11 shows a schematic diagram of a non-transitory computer-readable storage medium 1100 for predicting the flow rate of a base station according to an embodiment of the present disclosure.

[0096] The non-transitory computer-readable storage medium 1100 stores computer program instructions 1110, which, when executed by a processor, perform various methods for predicting the flow rate of a base station and various methods for managing a base station, as described with reference to Figures 1 to 9, provided by an embodiment of the present disclosure, and will not be described repeatedly here.

[0097] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or a combination of hardware and software. Such hardware or software may be referred to as a "data block," "module," "engine," "unit," "component," or "system." Additionally, each aspect of the present application may be realized as a computer product including computer-readable program code and disposed on one or more computer-readable mediums.

[0098] The present application uses certain words to describe embodiments of the present application. For example, "one embodiment," "an embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic associated with at least one embodiment of the present application. Thus, it should be noted that "one embodiment," "one embodiment," or "one alternative embodiment" referenced two or more times in different places in this specification do not necessarily refer to the same embodiment. Also, certain features, structures, or characteristics in one or more embodiments of the present application may be combined as appropriate.

[0099] Unless otherwise specified, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art to which this disclosure belongs. It should also be understood that terms as defined in ordinary dictionaries should be interpreted to have a meaning consistent with the meaning in the context of the relevant technology, and should not be interpreted in an idealized or highly formalized sense unless expressly stated herein.

[0100] The above is a description of the present disclosure and should not be construed as limiting the present disclosure. Although some exemplary embodiments of the present disclosure have been described, those skilled in the art can easily understand that many modifications of the exemplary embodiments are possible without departing from the teachings and advantages of the present disclosure. Therefore, all such modifications are included within the scope of the present disclosure. It should be understood that the above is a description of the present disclosure and should not be construed as being limited to the specific embodiments disclosed.

Claims

1. Historical flow data of a base station is obtained; A predictor generates future flow data of the base station based on the past flow data; The predictor is trained according to a first rule; The first rule is: Clustering the N first training curves into M categories based on similarity, the N first training curves being graphical representations of flow rate values ​​of the N first training data over time, where M is less than or equal to N; A representative curve is selected for each category, and M pieces of first training data corresponding to each representative curve are set as M pieces of final training data; training the predictor with the M final training data.

2. The predictor is trained by the M final training data, Adding M representative curves corresponding to the M final training data to a curve library and setting a label for each representative curve; The method for predicting the flow rate of a base station according to claim 1, comprising: training the predictor by the M final training data and their labels.

3. The step of generating future flow rate data of the base station based on the past flow rate data by a predictor includes: determining a curve from the curve library that best matches the historical flow data curve, the historical flow data curve being a graphical representation of flow values ​​in the historical flow data as they change over time; The method for predicting the flow rate of a base station according to claim 2, further comprising: generating future flow rate data of the base station based on the past flow rate data and a label corresponding to the best matching curve.

4. If the degree of match between the historical flow data curve and the best matching curve is less than a first threshold, the method further comprises: The method for predicting base station flow as recited in claim 3, further comprising the step of scaling the generated future flow data.

5. The step in which the generated future flow data is scaled is calculating a ratio value corresponding to each time point based on the historical flow data curve and the best matching curve; The method for predicting base station flow rate as claimed in claim 4, further comprising: multiplying the generated future flow rate data by a ratio value at a corresponding time point for each point to generate scaled future flow rate data.

6. If the degree of match between the historical flow data curve and the best matching curve is less than a second threshold, the method further comprises: dividing the curve of the historical flow data into a plurality of sub-curves; For each of the plurality of partial curves, a curve that best matches the partial curve is determined from the curve library, and partial future flow data for the base station is generated based on the past flow data and a label corresponding to the best matching curve; The method for predicting a flow rate of a base station as described in claim 3, further comprising a step of combining a plurality of partial future flow rate data corresponding to the plurality of partial curves to generate the future flow rate data of the base station.

7. The step of generating partial future flow data of the base station based on the past flow data and the label corresponding to the best matching curve includes: generating a first future flow data for the base station based on the historical flow data and a label corresponding to the best matching curve; A method for predicting a flow rate of a base station as described in claim 6, comprising a step of cutting out a portion corresponding to a time period of the partial curve from the first future flow rate data to obtain the partial future flow rate data.

8. If the degree of match between the historical flow data curve and the best matching curve is less than a third threshold, the method further comprises: identifying at least one segment of the historical flow data curve, the at least one segment matching a corresponding segment of the best matching curve being less than a fourth threshold; replacing each of the at least one segment individually with the corresponding segment of the best matching curve to generate modified historical flow data; The method for predicting the flow rate of a base station according to claim 3, further comprising: generating future flow rate data of the base station based on the corrected past flow rate data and a label corresponding to the best matching curve.

9. An apparatus for predicting a flow rate of a base station, comprising: A past flow data acquisition unit for acquiring past flow data of the base station; a prediction unit for generating future flow data of the base station based on the past flow data; The prediction unit is trained according to a first rule; The first rule is: Clustering the N first training curves into M categories based on similarity, the N first training curves being graphical representations of flow rate values ​​of the N first training data over time, where M is less than or equal to N; A representative curve is selected for each category, and M pieces of first training data corresponding to each representative curve are set as M pieces of final training data; The predictor is trained by the M final training data.

10. A non-transitory computer-readable storage medium having computer instructions stored thereon, comprising: A non-transitory computer-readable storage medium, in which the computer instructions, when executed by a processor, perform the method for predicting a base station flow rate according to any one of claims 1 to 8.

Images