Optimized management of the resources of a site of a cellular radiocommunication network

Abstract

The invention relates to a method for managing resources of a site of a cellular radiocommunication network, the site comprising radio resources for transporting site user data, the resources comprising K elements of the same nature, K being an integer greater than or equal to 2, the method comprising: - predicting (302) a volume of data traffic transported on the site, based on at least one indicator representative of operation of the site; - determining (303) a set of elements sufficient to transport the predicted volume of data traffic; - for at least one element of the site not belonging to the sufficient set of resources, disabling (304) the element of the site before a start time of the prediction period.

Description

Optimized resource management of a cellular radio communication network site Scope of the invention

[0001] The field of the invention is that of cellular radio communication networks and more particularly the management of resources of such networks. Prior art

[0002] A mobile cellular radio communication network consumes a significant amount of electricity due to the operation of sites comprising one or more base stations, antennas, and other network infrastructure. The continuous need for data transmission places a high energy demand on the cellular radio communication network.

[0003] Cellular radio communication networks continue to develop to support increased data traffic and new technologies, such as the generation of 5G networks. This results in an increase in the energy requirements of cellular radio communication networks.

[0004] The manager of a cellular radiocommunication network seeks to ensure better management of energy consumption across all mobile network equipment.

[0005] Some solutions involve reducing the number of resources at a radio site by putting one or more cells into sleep mode. For example, the document "Energy Efficiency of 5G Mobile Networks with Base Station Sleep Modes," by P. Lähdekorpi, M. Hronec, P. Jolma, and J. Moilanen, published by the Conference of Standards for Communications and Networking (CSCN) in 2017, pages 163 to 168, describes such a solution. Specifically, the operator can decide to put the cell into sleep mode when traffic on the deactivated cell is low or medium, thus shifting the traffic to the other resources at the radio site. However, putting the cell into sleep mode without prior knowledge of future traffic volume can lead to a loss of quality of service in the network, particularly if traffic increases after the cell is put into sleep mode and the remaining active resources at the radio site cannot carry all of it.

[0006] Other solutions involve switching off high-frequency bands, such as 2600 MHz (the so-called capacity bands), at night, while keeping low-frequency bands, such as 800 MHz, active. To achieve this, a predicted data traffic volume can be compared with a low traffic threshold. If the traffic is below this threshold, all capacity layers are deactivated. If the data traffic volume prediction then indicates a medium or high load on the radio site, all capacity layers are reactivated. Thus, in the case of a medium load, all capacity layers are activated, even though not all of them are necessary for carrying the traffic. This results in poor management of network resources based on data traffic volume.

[0007] Another solution for reducing the number of resources, particularly in the case of low traffic volume, is national roaming, which is described in the paper "National roaming as a fallback or default?", L. Weedage, SRC Magalhaes, S. Bayhan, IFIP Networking Conference, 2023. According to this solution, in the event of low traffic volume in a given geographic area Z, users of operator A accessing a first radio site can switch to a second radio site of operator B, thus putting the resources of the first radio site into standby mode. Conversely, in another geographic area Z', in the event of low traffic volume, the resources of a radio site of operator B are put into standby mode, and users of operator B in area Z' switch to the radio site of operator A. Similar solutions propose sharing the network infrastructure between several operators.The document “Network sharing and its energy benefits: A study of European mobile network operators”, IEEE Global Communications Conference (GLOBECOM), 2013, pages 2561-2567, describes such a solution.

[0008] However, these latter solutions have the disadvantage of a great complexity in the management of resources of the operators' cellular radiocommunication sites.

[0009] This results in a lack of optimization in the management of resources of a cellular radio communication network in relation to variations in traffic volume at each site of the network. Object and summary of the invention

[0010] One of the aims of the invention is to overcome at least one of the drawbacks of the aforementioned prior art by proposing a new resource management technique for a cellular radiocommunication site, adapted to changes in traffic volume at the site. This makes it possible to optimize the use of available resources at the site and consequently reduce its energy consumption without diminishing the quality of service delivered to site users.

[0011] To this end, an object of the present invention relates to a method for managing resources of a site of a cellular radiocommunication network according to claim 1. Claims 2 to 10 describe preferred embodiments of said method.

[0012] The site includes radio resources for transporting user data from the site, said resources comprising K elements of the same type, K being an integer greater than or equal to 2. The method may include: - a prediction of a volume of data traffic transported on said site for a future prediction horizon, based on indicators representative of site operation; - a determination of a set of elements sufficient to transport the predicted volume of data traffic in the future prediction horizon; - for at least one element of the site not belonging to the sufficient set of resources, a deactivation of the site element before a start time of the future prediction horizon.

[0013] Thus, the site's resources are adapted based on a prediction of the volume of data traffic in a future prediction horizon, which makes it possible to optimize the use of the site's resources, and in particular to reduce the overall energy consumption of the site.

[0014] According to some embodiments, all site elements not belonging to the determined sufficient set of elements can be deactivated before the start time of the future prediction horizon.

[0015] Thus, it is possible to minimize the overall energy consumption of the site by optimizing resource allocation based on a prediction of the volume of data traffic transported on the site.

[0016] According to some embodiments, the determination of the sufficient element set may include obtaining the maximum traffic volumes of the site elements, and the determined sufficient element set may be the smallest set, in number of elements, for which a sum of the maximum traffic volumes of the elements in the set is greater than the data traffic volume predicted for the future prediction horizon.

[0017] Thus, it becomes possible to optimize resource management based on a prediction of data traffic volume without reducing the quality of service for site users.

[0018] According to some embodiments, indicators representing the operation of the site may include a history of past values ​​of a site data traffic volume indicator.

[0019] Such a history allows for the consideration of the dynamic evolution of the volume of data traffic transported on the site, or for each element of the site, which improves the accuracy associated with the prediction of the volume of data traffic for the prediction horizon.

[0020] In addition, the indicators representing the operation of the site may also include at least one current value of at least one other indicator of the operation of the site in addition to the indicator of the site's data traffic volume.

[0021] Such complementary indicators improve the accuracy associated with predicting the site's data traffic volume for the prediction horizon, thereby improving resource management based on such a prediction.

[0022] In addition or as an alternative, the prediction of the site's data traffic volume can be based on at least one predictive model defined by parameters associated respectively with past values ​​of the historical data traffic volume indicator and / or at least one current value of at least one other site performance indicator.

[0023] Thus, it is made possible to weight, by parameters, the contributions of each of the operating indicators in the prediction of the site's data traffic volume for the prediction horizon, which improves the accuracy of the prediction, and makes it possible to determine at least one predictive model specific to a given site.

[0024] In addition, the parameters can be defined by machine learning, from a training dataset and a test dataset.

[0025] Thus, the accuracy afforded by at least one predictive model is improved. Furthermore, when training and test data are collected on-site, the at least one predictive model developed is site-specific.

[0026] According to embodiments, the site can operate on at least one frequency band and / or at least one generation of cellular radio communication network, and the site's operating indicators can be indicators associated with at least one frequency band and / or at least one generation of cellular radio communication network.

[0027] For example, site operating indicators can be associated with respective "Remote Radio Units", RRUs, each RRU corresponding to a group of cells that operate on the same frequency band with different network generations (2G and 3G on the 900 MHz band for example).

[0028] Thus, a predictive model is designed to determine the volume of data traffic carried by each frequency band and / or each network generation and / or each RRU. The site's data traffic volume is then obtained by summing the predictions for all frequency bands and / or network generations and / or RRUs at the site.

[0029] In addition, a remote radio unit, RRU, can be a group of radio cells operating on the same frequency band for at least two generations of cellular radio communication network, and the K elements of the site can be K RRUs.

[0030] Disabling at least one RRU has a significant impact on the site's energy consumption, due to the fixed portion of an RRU's energy consumption (called the intercept), which is independent of the volume of data traffic carried by that RRU. Resource management according to the invention thus enables a significant reduction in energy consumption at the site.

[0031] In addition, the prediction of the traffic volume for the site can be based on K prediction models defined respectively for the K RRUs of the site.

[0032] Thus, the accuracy associated with predicting the site's traffic volume is improved, allowing for better optimization of site resources.

[0033] The invention also relates to a resource management device for a site of a cellular radio communication network according to claim 11.

[0034] The site includes radio resources for transporting user data from the site, said resources comprising K elements of the same type, K being an integer greater than or equal to 2. The resource management device can be configured to: - predict a volume of data traffic transported on said site for a future prediction horizon, based on indicators representative of site operation; - determine a set of elements sufficient to transport the predicted volume of data traffic in the future prediction horizon; - for at least one element of the site not belonging to the sufficient set of resources, deactivate the element of the site before a time at the start of the future prediction horizon.

[0035] Such a device is specifically configured to implement the aforementioned resource management process, according to one or the other of its embodiments.

[0036] The invention also relates to a computer program comprising instructions for implementing the resource management method according to the invention, according to any one of the particular embodiments described above, when said program is executed by a processor.

[0037] Such instructions can be stored permanently in a non-transient memory medium of the resource management device implementing the resource management process according to the invention.

[0038] This program can use any programming language, and be in the form of source code, object code, or code somewhere between source code and object code, such as in a partially compiled form, or in any other desirable form.

[0039] The invention also relates to a recording medium or information medium readable by a computer, and comprising instructions for a computer program as mentioned above.

[0040] The recording medium can be any entity or device capable of storing the program. For example, the medium can include a storage means, such as a ROM, for example a CD-ROM or a microelectronic circuit ROM, or a magnetic recording means, for example a mobile device, a hard drive or an SSD.

[0041] On the other hand, the recording medium can be a transmissible medium such as an electrical or optical signal, which can be transmitted via an electrical or optical cable, by radio, or by other means, so that the computer program it contains can be executed remotely. The program according to the invention can, in particular, be uploaded to a network, for example, an Internet-type network.

[0042] Alternatively, the recording medium may be an integrated circuit in which the program is incorporated, the circuit being adapted to execute or to be used in the execution of the aforementioned resource management process.

[0043] In an example implementation, the present technique is implemented using software and / or hardware components. In this context, the term "device" or "module" may refer in this document to a software component, a hardware component, or a set of hardware and software components.

[0044] Other features and advantages will become apparent from particular embodiments of the invention, given by way of illustrative and non-limiting examples, and the accompanying drawings, among which: Figure 1 represents an architecture in which the resource management process of a site of a cellular radiocommunication network is implemented, according to a particular embodiment of the invention; Figure 2 represents a resource management device of a site of a cellular radiocommunication network, according to an embodiment of the invention; Figure 3 represents the main steps implemented in the resource management process of a site of a cellular radiocommunication network, according to an embodiment of the invention, as implemented in the architecture of Figure 4; Figure 5 represents a curve of the evolution, in real-world conditions, of the energy consumption of an element of the cellular radiocommunication network, as a function of time.

[0045] Detailed description of an embodiment of the invention

[0046] Lare represents an architecture in which a resource management method is implemented for a site S of a cellular RC radiocommunication network, according to an embodiment of the invention. Such an RC network is, for example, of the 3G, 4G, 5G, etc. type.

[0047] Such an architecture includes: - a resource management device (DGR), configured to manage radio resources at site S, based on DR data relating to the operation of the RC network, described below. The DGR at site S is, for example, a server, a platform (e.g., a CSON, "Centralized Self-Organizing Network"), a RIC (Radio Access Network Intelligent Controller), or an OSS (Operations Support System), etc. It can be installed on site S or remotely; - the radio resources at site S, comprising K elements EL1, EL2, ..., E k , …, EL K of the same nature.

[0048] Since the RC network is a cellular radio communication network, the K elements EL1, EL2, …, EL Kcan be K different frequency bands respectively, for example 1800 MHz and 2100 MHz, or at least two different generations of cellular network, for example 3G and 4G, in the case where K=2. The elements EL1 to EL K They can also be K RRUs (Remote Radio Units), each operating on a different frequency band. Typically, an RRU corresponds to a group of cells operating on the same frequency band with different network generations (2G and 3G on the 900 MHz band, for example).

[0049] In what follows, it is considered, for illustrative purposes, that the K elements are K RRUs.

[0050] We will now describe, with reference to the, the simplified structure of the DGR resource management system.

[0051] The DGR resource management system includes: - a COL module for collecting DR data relating to the operation of said RC network, including indicators representative of the operation of site S; - a PRED module for predicting a volume V i+h of data traffic carried on site S for a future prediction horizon H, based on DR data representative of the operation of site S; - a DET module for determining a set E of K* elements sufficient to carry the predicted data traffic volume in the future prediction horizon H, K* being strictly less than K when the predicted data traffic volume is medium or low (and K* being more generally less than or equal to K); - a DES module for disabling at least one EL element k of the site not belonging to set E, before a starting moment of the future prediction horizon H.

[0052] At initialization, computer program code instructions PG are for example loaded into RAM memory (not shown) before being executed by a PROC processor of the resource management device DGR.The PROC processor of the UTR processing unit implements the following actions, within the framework of the resource management process, which will be described below, according to the instructions of the computer program PG: - collect the DR data relating to the operation of the RC network, in particular of site S; - predict the volume of data traffic carried on site S for the future prediction horizon H, according to the DR data representative of the operation of site S; - determine the set E of K* elements sufficient to carry the predicted volume of data traffic in the future prediction horizon H, K* being strictly less than K when the predicted volume of data traffic is medium or low; - deactivate at least one EL element. k of the site not belonging to set E, before the start time of the future prediction horizon H.

[0053] We now describe, in relation to the, together figures 1 and 2, the process of managing resources of a site S of the cellular radiocommunication network RC, according to a particular embodiment of the invention.

[0054] At a step 301, the resource management system receives, or collects, the DR operating data of the RC network, including indicators representative of the operation of site S, at a current time.

[0055] Such DR data may include raw indicators relating to the operation of the RC network, particularly of site S, such as, for example, the number of resources occupied per cell covered by site S, the amount of data sent via the RC network, etc.

[0056] Alternatively, such DR data may include performance indicators or KPIs relating to the RC network, particularly to site S, such as, for example, for each RRU: throughput, average number of users, RRU load, RRU energy consumption, etc.

[0057] The KPI of data traffic volume transported is also collected by the resource management system.

[0058] The KPIs are sent back to the DGR system with a granularity which is variable depending on the context of the modeling, which can for example be a time interval of 15 minutes (mn), 30 min, 1 hour, etc.

[0059] At a step 302, the DGR resource management device predicts the data traffic volume of site S for a prediction horizon H, based on indicators representative of the operation of site S collected in the previous step 301.

[0060] In what follows, an example of PR prediction of the data traffic volume of site S for the prediction horizon H is described, for illustrative purposes.

[0061] In this example, the PR prediction is a prediction of the total data traffic volume of site S, which is obtained from K PR predictions. k obtained each for one of the K RRUs of site S.

[0062] PR prediction k is based on KPIs collected in step 301, which are rated in which corresponds to a temporal index, in which denotes a KPI index and in which the exponent refers to RRU of site S.

[0063] As previously mentioned, site S consists of RRUs, each RRU corresponding to a group of cells that operate on the same frequency band with different network generations (2G and 3G on the 900 MHz band for example).

[0064] For an RRU KPIs are obtained in step 301, which are noted in which corresponds to the temporal index, in which denotes the KPI index and in which the exponent refers to RRU of site S.

[0065] Alternatively, the PR prediction is obtained from a single prediction of the total data traffic volume of site S. The PR prediction is then based on KPIs collected in step 301, which are rated in which corresponds to a temporal index, in which denotes a KPI index and in which the exponent refers to the entirety of the S site.

[0066] For example, the KPIs considered can be any combination of the following indicators:

[0067] - For 2G network generation: throughput for each cell;

[0068] - For 3G network generation: throughput and average number of users for each cell;

[0069] - For 4G network generation: cell load and average number of users for each cell; - for the entire RRU, an RRU energy consumption KPI.

[0070] In this example, no indicators are considered for 5G network generation. However, according to the invention, indicators relating to 5G network generation can be taken into account for PR predictions. k of data traffic volume of each RRU with index k of stage 302.

[0071] Furthermore, a data traffic volume KPI is obtained, such a KPI being generally available per technology (per network generation). Thus, for a given RRU of index k, the data traffic volume at a current time i, denoted , is given by the following formula (1):

[0072] (1) in which i thus refers to the temporal index of the measurement, i.e. the current instant included in the period current; in which the index n in refers to the technology (network generation), representing the data traffic volume KPI for index generation n, over the period including the current time index i. Data traffic volume KPIs can be stored by the DGR resource management device for the different times i, so as to constitute a history of the volume of data traffic for each RRU of index k and for each time i.

[0073] As previously mentioned, all KPIs can be collected by the DGR resource management system with a granularity of ( min etc.). Thus, corresponds to the volume of data traffic carried by the RRU with index k during the reference period including instant i.

[0074] To account for the granularity of implementation of step 301, the index The various KPIs are incremented every period Thus, if the value of the index , then the measures were taken just now expressed in minutes or hours.

[0075] The prediction at the current time index (the current time of step 301) of the future PR data traffic volume k for the RRU with index k and for a horizon that is, for a horizon including the time index (for example, ending at time index i+h), is performed by considering a history of the data traffic volume of each RRU as well as the current values ​​at time i of the KPIs according to the following prediction model, PR k being noted for the RRU: (2)

[0076] in which is the history of previous data traffic volume values ​​for the RRU with index k, the history being comprised of previous measures and in which are parameters. To account for the granularity of measurement collection in step 301, the prediction horizon H is preferably a multiple of . If , then the prediction horizon If , SO , etc.

[0077] The prediction horizon H (and therefore the value h) can be determined from the current instant of index i in order to estimate the volume of data traffic in a future period of duration including the future instant i+h (for example ending at instant i+h).

[0078] In (2), the parameters are unknown, but can be obtained during a preliminary learning phase, not shown on the. The preliminary phase may include machine learning of the parameters, based on an artificial intelligence technique such as linear regression, LASSO regression, for "Least Absolute Shrinkage and Selection Operation", or any other technique.

[0079] Machine learning can be performed on a training database, storing values And , as well as the corresponding target values (the truth on the ground).

[0080] Thus, during the learning phase, the future value (for the prediction horizon) ) is considered the variable of interest. Historical data traffic volumes and the current values ​​of the other KPIs are predictive variables as shown in Table 1 where is the number of observations used during the learning phase, i.e. the number of data sets in the training database.

[0081] Table 1: Predictive Variables Variable of Interest … … … … ............... … …

[0082] However, there are no restrictions on how the parameters are determined. , which can be obtained by a technique other than a machine learning-based technique.

[0083] Following the learning process, the DGR resource management system stores the parameters trained, and is thus able to implement, via the PRED prediction module, a prediction of the data traffic volume RRU by RRU. Note that a set of parameters can be determined for each RRU of index k, and the parameters can thus be denoted with (k) as a superscript, as in formula (3) below.

[0084] For the RRU, we obtain:

[0085]

[0086] The value obtained is a prediction of the data traffic volume of the k-th RRU over a period for the prediction horizon H.

[0087] Thus, for the entire site S, the prediction of the data traffic volume, noted , is given by the following formula (4), by summing the different predictions obtained for the K RRUs of site S:

[0088] (4)

[0089] Thus, the PRED prediction module obtains the data traffic volume prediction. from site S during step 302.

[0090] Alternatively, the prediction The predictions for the entire site are not obtained by determining them. RRU by RRU, but is obtained from a single model of formula (5): (5)

[0091] in which the exponent T indicates that the variables concern the entire site, and no longer a single RRU.

[0092] The parameters can be determined for the entire site, in the same way as the parameters described previously with the exponent (k) for each k-th RRU.

[0093] Thus, in a first embodiment, the DGR resource management system stores the predictive models corresponding to equations (2) in the PRED prediction module described above, and the prediction of the data traffic volume of site S over the period The prediction horizon H can be obtained by summing all the individual predictions. K RRUs from site S.

[0094] In a second embodiment, the DGR resource management device stores the unique predictive model (5) in the PRED prediction module described above.

[0095] According to another undescribed embodiment, the DGR resource management device stores a data traffic volume prediction model for each network generation (e.g., 2G, 3G, 4G, 5G) at site S, and the data traffic volume prediction for site S over a period of time for the prediction horizon H can be obtained by summing all the individual predictions of the network generations of site S.

[0096] The predictive models described above were tested, after parameter training, on a real test dataset of a cellular radio communication network, for a prediction horizon H, with a granularity value of of 60 minutes, and a horizon H also of 60 minutes (so with h=1).

[0097] The test period includes data between 10am and 5pm on day d. The parameter training period includes the two days preceding day d.

[0098] A learning model is created by RRU for the predictive model (2).

[0099] The results show that the predicted values ​​of data traffic volume, per site or per RRU, are very close or even identical to the actual values ​​of data traffic volume.

[0100] In Table 2 below, consider the metric of the absolute value of the relative error RE, between the actual values ​​of data traffic volume from the test data and the predicted values ​​of data traffic volume obtained from the predictive models (2): (6)

[0101] Table 2 thus gives the rate (or probability or percentage) that the relative error RE is less than 25% for H=60min and The rate is initially calculated on a set S1 of test samples from all radio sites in a cellular radio communication network, and then on a set S2 containing all traffic samples from all radio sites, removing samples for which the data traffic volume is very low (e.g., less than 5% of the maximum data traffic volume). For samples with very low data traffic volumes, a larger relative error is acceptable because the predicted data traffic volume remains low.

[0102] Table 2 shows that the error in predicting the data traffic volume for the entire site S is small.

[0103] Table 2: Set S1 Set S2 Percentage that RE<25% 77% 91%

[0104] Thus, predictive models (2) are particularly suited to predicting the volume of data traffic from RRUs of a site S, based on radio KPIs such as throughput, load and number of active users.

[0105] Referring again to Figure 3, the resource management process further includes, following steps 301 and 302, a step 303 of determination, by the DET determination module, of a set E of K* elements sufficient to carry the volume of data traffic predicted within the prediction horizon H. Indeed, in many situations, particularly when the volume of data traffic If the predicted value is low or medium, it can be transmitted by a number K* of RRUs strictly less than K. This is particularly the case when the individual predictions of the RRUs are low or medium.

[0106] The set E can be determined from the maximum data traffic volumes that the K RRUs can respectively support. The maximum data traffic volumes are denoted in what follows. The values can be known in advance (predetermined) or can be obtained from collected data (for example, they can correspond to peak values ​​of data traffic volume over a given period, for example, several days).

[0107] According to some embodiments, K* is the exact and minimum number of RRUs required to satisfy the predicted data traffic volume for site S over the prediction horizon H. Thus, K* can be obtained as follows:

[0108] which means that K* is the smallest value satisfying , when the RRU indices are reordered (to simplify understanding of the invention) so that the first K* RRUs are part of set E. In this case, according to the invention, at a step 304, at least one of the RRUs indexed from K*+1 to K is deactivated (or switched off) by the DES deactivation module, before time i+h (for example, in the period preceding time i+h), which allows the energy consumption of site S to be reduced for the period of the prediction horizon H, while maintaining quality of service for users. Preferably, the DES deactivation module disables all RRUs indexed from K*+1 to K, before time i+h (for example, from the period preceding time i+h or at the beginning of the period including time i+h), which minimizes the energy consumption of site S for the period of the prediction horizon H, while maintaining quality of service for users.

[0109] It should be noted that the small error in the prediction of the total site data traffic volume, shown in Table 2, does not impact the exact number K* of RRUs required to carry the predicted volume across the entire site, since the experiments that led to the results in Table 2 show that the calculation of K* is accurate in almost 99% of cases.

[0110] It should also be noted that, as illustrated in the figure, the fixed part of an RRU's energy consumption, which does not depend on the volume of data traffic, consumes a lot of energy: about 50% or even more of an RRU's total consumption.

[0111] This graph presents a curve representing the evolution, in a real-world scenario, of energy consumption, for example per RRU, as a function of time over several days: Day 1, Day 2, Day 3, and Day 4. It can be observed that the fixed energy value, which does not depend on the data traffic volume, is quite high compared to the consumption linked to the data traffic volume. It is noted that the intercept (the fixed energy consumption value of an RRU, independent of the data traffic volume) exceeds half of the total energy consumption.

[0112] Referring again to the, the steps of the resource management process according to the invention can be iterated, so as to continuously adapt the activated and deactivated resources of site S, to the predicted evolution of the volume of data traffic.

Claims

Method for managing resources of a site (S) of a cellular radiocommunication (CR) network, the site comprising radio resources for the transport of site user data, said resources comprising K elements (EL1-EL1 K ) of the same nature, K being an integer greater than or equal to 2, the process comprising: - a prediction (302) of a volume of data traffic transported on said site, as a function of at least one indicator representative of the operation of the site; - a determination (303) of a set of elements sufficient to transport the predicted volume of data traffic; - for at least one element of the site not belonging to the sufficient set of resources, a deactivation (304) of the element of the site before a start time of the prediction period. A method according to claim 1, wherein all elements (EL1-EL K) of the site (S) not belonging to the set of sufficient elements determined are deactivated before the start time of the prediction period. A method according to claim 1 or claim 2, wherein the determination (303) of the sufficient element set includes obtaining the maximum traffic volumes of the elements (EL1-EL K ) of the site (S), and in which the determined sufficient set of elements is the smallest set, in number of elements, for which a sum of the maximum traffic volumes of the elements of the set is greater than the predicted data traffic volume. A method according to any one of claims 1 to 3, wherein said at least one indicator representative of the operation of the site (S) includes a history of at least one past value of a data traffic volume indicator of the site. Method according to claim 4, wherein said at least one indicator representative of the operation of the site (S) further comprises at least one current value of at least one other indicator of the operation of the site other than the indicator of the data traffic volume of the site. A method according to claim 4 or claim 5, wherein the prediction of the site's data traffic volume is based on at least one predictive model defined by parameters associated respectively with at least one past value of the historical data traffic volume indicator and / or at least one current value of at least one other site operating indicator. A method according to claim 6, wherein the parameters are defined by machine learning, from a training data set and a test data set. A method according to any one of the preceding claims, wherein the site (S) operates according to at least one frequency band and / or at least one generation of cellular radio communication network, and wherein said at least one site operating indicator is an indicator associated with at least one frequency band and / or at least one generation of cellular radio communication network. A method according to claim 8, wherein a remote radio unit, RRU, is a group of radio cells operating on the same frequency band for at least two generations of cellular radio communication network, and wherein the K elements (EL1-EL K ) of the site (S) are K RRUs. Method according to claim 6 and claim 9, wherein the prediction (302) of the traffic volume for the site (S) is based on K prediction models defined respectively for the K RRUs of the site. Resource management device (RMD) of a site (S) of a cellular radiocommunication (CR) network, the site comprising radio resources for the transport of site user data, said resources comprising K elements (EL1-EL1 K ) of the same nature, K being an integer greater than or equal to 2, the resource management device being configured to: - predict a volume of data traffic transported on said site, based on at least one indicator representative of site operation; - determine a set of elements sufficient to transport the predicted volume of data traffic; - for at least one element of the site not belonging to the sufficient set of resources, deactivate the element of the site before a time at the start of the prediction period. Computer program comprising program code instructions for implementing the resource management method according to any one of claims 1 to 10, when executed on a computer. Computer-readable information carrier, and comprising instructions for a computer program according to claim 12.

Images