Managing a power outage in a cellular radiocommunication network

Abstract

The invention relates to a method for managing a power outage in a site of a cellular radiocommunication network, said site being provided with an energy storage device, the method comprising: - receiving (300) a request for a power outage from the site, the request indicating a required outage duration; - predicting (302) an energy consumption of the site for a future prediction horizon, as a function of indicators representative of an operation of the site; - calculating (303) an autonomy of the storage device of the site as a function of the prediction of energy consumption of the site; - comparing (304) the calculated autonomy of the storage device and the required outage duration; - making a decision (305-308) on the request, as a function of a result of the comparison. Fig.3

Description

Title of the invention: Management of a power outage in a cellular radiocommunication network Field of invention

[0001] The field of the invention is that of cellular radiocommunication networks and more particularly the management of a power cut for voluntary load shedding in these networks. Prior art

[0002] Electricity suppliers face variations in energy consumption in an electricity network depending on the time of day, and generally distinguish between peak hours, during which energy consumption in the electricity network is high, and off-peak hours, during which energy consumption on the electricity network is lower.

[0003] 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 in fact imposes a high energy demand on the cellular radio communication network.

[0004] Cellular radio communication networks continue to expand to support increased data traffic as well as new technologies, such as the generation of 5G networks. This results in an increase in the energy requirements of cellular radio communication networks.

[0005] An existing solution for reducing the load imposed by a cellular radiocommunication network on an electrical network, in particular during peak hours, is a load shedding technique, according to which at least one site of the cellular radiocommunication network is required to be temporarily disconnected from the electrical network (therefore to cut off the power to the site) during a cut-off period, to be powered by an energy storage device of the site, such as a battery for example.

[0006] For this purpose, an operator in charge of the cellular radiocommunication network can negotiate with an aggregator the conditions of the power outage, in particular the duration of the outage. An aggregator acts as an intermediary between actors (individuals or companies) who have load shedding capabilities, as is the case for an operator of a cellular radiocommunication network, and an electricity network manager. An aggregator thus allows, for the network manager electrical, to group together the load shedding capacities of different consumers or electricity consumption sites.

[0007] However, it is essential to keep in mind that an energy storage device has limits in terms of autonomy. Thus, if the outage duration is greater than the autonomy of the storage device, this results in an inevitable interruption of the services permitted by the cellular radiocommunication network.

[0008] It is thus complex for the operator of the cellular radiocommunication network to know in advance the autonomy permitted by the storage device of a site, which depends on the activity of the site, which is variable and which is not known in advance.

[0009] This results in an inability for the network operator to be able to assess whether the duration of the outage in the context of voluntary load shedding is acceptable or not, i.e. whether or not it will impact the quality of the services provided by a site of a cellular radiocommunication network. Subject matter and summary of the invention

[0010] One of the aims of the invention is to remedy at least one of the drawbacks of the aforementioned state of the art by proposing a new technique for managing power outages at a cellular radiocommunication site, which makes it possible to take a decision whether or not to accept the outage, depending on the autonomy permitted by a storage device of the site. It is thus possible to accept a greater number of power outage requests, which facilitates balancing of the electrical network, in particular during peak hours, while guaranteeing stable performance for users of the communication services managed by the cellular radiocommunication site.

[0011] To this end, an object of the present invention relates to a method for managing a power outage in a site of a cellular radiocommunication network, said site being provided with an energy storage device, the method comprising: - receiving a request for a power outage from the site, the request indicating a required outage duration; - a prediction of the site's energy consumption for a future prediction horizon, based on indicators representative of the site's operation; - a calculation of the autonomy of the site's storage device based on the site's energy consumption prediction; - a comparison between the calculated autonomy of the storage device and the required cut-off duration; - a decision on the request, based on a result of the comparison.

[0012] Thus, the invention makes it possible to determine the actual autonomy of the storage device, which depends on the future energy consumption of the site. The decision-making process to accept or not a request for power cut from the site is thus improved: it is thus made it possible to accept a greater number of power outage requests, which facilitates the balancing of the electricity network, particularly during peak hours, while guaranteeing stable performance for users of the communication services managed by the cellular radiocommunication site. Advantageously, the prediction of the site's energy consumption is based on indicators representative of the operation of the site, such as, for example, key performance indicators or KPIs (in English, "Key Performance Indicator"), raw indicators (number of resources occupied per cell, quantity of data sent via the network, etc.), which allow an accurate prediction of energy consumption for the prediction horizon.

[0013] According to embodiments, the indicators representative of the operation of the site may comprise a history of past values ​​of an energy consumption indicator of the site.

[0014] Such a history allows the dynamic evolution of the energy consumption of the site to be taken into account, which improves the precision associated with the prediction of energy consumption for the prediction horizon.

[0015] In addition, the indicators representative of the operation of the site may further comprise at least one current value of at least one other indicator of operation of the site than the indicator of energy consumption of the site.

[0016] Such additional indicators make it possible to improve the accuracy associated with the prediction of energy consumption for the prediction horizon. Decision-making regarding the request is thus improved.

[0017] In addition or as a variant, the prediction of the energy consumption of the site can be based on a predictive model defined by parameters associated respectively with past values ​​of the historical energy consumption indicator and / or at least one current value of at least one other operating indicator of the site.

[0018] Thus, it is made possible to weight, by parameters, the contributions of each of the operating indicators in the prediction of the energy consumption of the site for the prediction horizon, which improves the accuracy of the prediction, and makes it possible to determine a predictive model specific to a given site.

[0019] Additionally, the parameters can be defined by machine learning, from a training data set and a test data set.

[0020] Thus, the accuracy enabled by the predictive model is improved. In addition, when training and testing data are collected on-site, the developed predictive model is site-specific.

[0021] According to embodiments, the site operation indicators may be global operation indicators of an entire site.

[0022] It is thus made possible to directly predict the energy consumption of the overall site, without differentiating between the elements making up the site.

[0023] Alternatively, the site may operate in at least one frequency band and / or at least one generation of cellular communication network, and the site operation indicators may be indicators associated with the at least one frequency band and / or the at least one generation of cellular communication network.

[0024] For example, the site operating indicators can be associated with respective “Remote Radio Units”, RRUs, each RRU corresponding to a group of cells which operate on the same frequency band with different network generations (2G and 3G on the 900 MHz band for example).

[0025] Thus, a predictive model is provided to determine the energy consumption of each frequency band and / or a network generation and / or an RRU. The energy consumption of the site is then obtained by summing the energy consumption predictions of all the frequency bands and / or network generations and / or RRUs of the site.

[0026] According to embodiments, if the autonomy of the storage device is less than the required cut-off duration, the decision may be a refusal of the power cut.

[0027] Thus, the request for power cut is refused in order not to degrade the quality of service for users of the cellular radiocommunication network.

[0028] Alternatively, if the autonomy of the storage device is less than the required cut-off duration, the decision may be a transmission of a power cut proposal indicating a new cut-off duration less than the autonomy of the storage device.

[0029] Thus, load shedding is made possible, but over a shorter period which makes it possible to maintain the quality of service provided to users of the cellular radiocommunication network.

[0030] As a further variant, the site may operate according to at least one frequency band and / or at least one generation of cellular communication network, and, if the autonomy of the storage device is less than the required outage duration, the decision comprises an acceptance of the power outage for the required outage duration. The method may further comprise, at the time of the power outage: - an activation of the storage device, - a calculation of a time of deactivation of said at least one frequency band or of said at least one network generation, said calculation being implemented according to a criterion for optimizing the energy stored in the storage device,

[0031] - a deactivation, at said calculated deactivation time, of said at least one frequency band or said at least one network generation.

[0032] Such a variant makes it possible to accept more power outage requests, while optimally preserving the storage resources of the storage device, for the duration of the power outage, by scheduling a time for deactivating said at least one frequency band or said at least one network generation of the cellular radiocommunication site. Thus, thanks to this variant, the storage device avoids discharging too quickly, for the benefit, during the power outage, of maintaining the quality of service or QoS perceived by the users, whose communications received or transmitted from their communication terminals transit via the cellular radiocommunication site. The site can then be quickly restored to full capacity once the power outage is over.

[0033] In addition, when said site operates according to at least two frequency bands and / or at least two generations of cellular radiocommunication network, the method may comprise, at the time of the power cut, once the storage device is activated:

[0034] - a calculation, according to said optimization criterion, of at least two successive instants deactivation of respectively said at least two frequency bands or respectively said at least two network generations,

[0035] - a deactivation, at said at least two calculated deactivation times, of respectively said at least two frequency bands or respectively said at least two network generations.

[0036] Such an embodiment is based on an effective strategy implementing a progressive reduction of the frequency bands or generations of cellular radiocommunication network used on the site, thus making it possible to reduce energy consumption and increase the duration during which the storage device can take over during the power outage.

[0037] Thus, for example, in the case where a radio site uses four cells, such as for example LTE-2600, LTE-800, 3G-2100 and 2G-900, when a power outage is required with an outage duration greater than the autonomy of the storage device, rather than keeping these four cells active throughout the outage period, the site implements a gradual reduction strategy. At a first instant, it is for example the 2100 MHz frequency band of the 3G-2100 cell which is deactivated, which reduces the consumption of the site and prolongs the autonomy of the storage device. Other frequency bands can then be progressively deactivated. By following this progressive approach, the radio site prolongs the duration during which the storage device can support the services, while optimizing the use of energy stored at the storage device level.

[0038] The invention also relates to a power outage management device in a cellular radiocommunication site, the power outage management device being configured to:

[0039] - receive a request for power outage from the site, the request indicating a duration required cut-off; - predict the site's energy consumption for a future prediction horizon, based on indicators representative of the site's operation; - calculate the autonomy of the site's storage device based on the site's energy consumption prediction; - compare the calculated autonomy of the storage device and the required cut-off time; - make a decision on the request, based on a result of the comparison.

[0040] Such a device is in particular configured to implement the aforementioned control method, according to one or other of its embodiments.

[0041] The invention also relates to a computer program comprising instructions for implementing the power outage management method according to the invention, according to any one of the particular embodiments described above, when said program is executed by a processor.

[0042] Such instructions can be stored permanently in a non-transitory memory medium of the power outage management device implementing the power outage management method according to the invention.

[0043] This program may use any programming language, and be in the form of source code, object code, or intermediate code between source code and object code, such as in a partially compiled form, or in any other desirable form.

[0044] The invention also relates to a recording medium or information medium readable by a computer, and comprising instructions of a computer program as mentioned above.

[0045] The recording medium may be any entity or device capable of storing the program. For example, the medium may comprise 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 medium, a hard disk or an SSD.

[0046] On the other hand, the recording medium may be a transmissible medium such as an electrical or optical signal, which may be conveyed via an electrical or optical cable, by radio or by other means, so that the computer program that it contains is executable remotely. The program according to the invention can in particular be downloaded over a network, for example an Internet-type network.

[0047] 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 power failure management method.

[0048] According to an exemplary embodiment, the present technique is implemented by means of software and / or hardware components. In this regard, the term “device” or “module” may correspond in this document to a software component, a hardware component or a set of hardware and software components. Brief description of the drawings

[0049] Other characteristics and advantages will appear on reading particular embodiments of the invention, given as illustrative and non-limiting examples, and the appended drawings, among which:

[0050] [Fig.l] represents an architecture in which the method for managing power outages in a cellular radiocommunication network is implemented, according to a particular embodiment of the invention,

[0051] [Fig.2] represents a device for managing power outages in a cellular radiocommunication network, according to one embodiment of the invention; [Fig.3] represents the main steps implemented in the power outage management method, according to one embodiment of the invention, as implemented in the architecture of [Fig.l]; [Fig.4] represents a device for controlling the consumption of a cellular radiocommunication network during a power cut, according to a particular embodiment of the invention;

[0052] [Fig.5] represents additional steps for controlling the consumption of the cellular radiocommunication network, implemented in the method for managing a power outage of a radiocommunication network, according to a particular embodiment of the invention;

[0053] [Fig.6] represents a curve of the evolution, in a real situation, of the energy consumption of an element of the cellular radiocommunication network, as a function of time;

[0054] [Fig.7A] represents the main actions implemented during one of the complementary steps of the method for managing a power outage of a cellular radiocommunication site, as illustrated in Figures 3 and 5, according to one embodiment;

[0055] [Fig.7B] represents the main actions implemented during one of the complementary stages of the power outage management process of a site cellular radiocommunication, as illustrated in Figures 3 and 5, according to another embodiment;

[0056] [Fig.8] represents a schematic curve illustrating the strategy for deactivating RRUs of the cellular radiocommunication network, in accordance with the other embodiment of [Fig.7B].

[0057] Detailed description of an embodiment of the invention

[0058] [Fig.l] represents an architecture in which a method for managing a PO power outage is implemented in the context of voluntary load shedding, according to one embodiment of the invention. Such an RC network is for example of the 3G, 4G, 5G, etc. type.

[0059] Such an architecture includes: - a power outage management device DIS, configured to receive a power outage request REQ from a site S installed at a given geographical location and comprising the equipment necessary for the transmission and reception of communication data via the RC network. The request REQ is received in the context of voluntary load shedding and can identify a power outage start time t0 as well as a required power outage duration Te. The request REQ is received, prior to the power outage PO, from an entity EN_PR which may be a management entity of an electricity network in charge of supplying electricity to the site S, or which may belong to an aggregator, which is an intermediary between actors who have load shedding capabilities and the manager of the electricity network;

[0060] - an STO energy storage device capable of storing energy in a form given, and to return electrical energy, on command, to the site S. The storage device STO can for example be an electric battery. Alternatively, the storage device STO can be a generator, an inertial flywheel, or any other storage device capable of storing energy in a given form and returning electrical energy on command. The maximum value of the energy stored in the storage device STO is noted Emax. Before the power cut PO (or during reception of the request REQ), the energy stored in the storage device STO has an initial value Eo, which is less than or equal to Emax.

[0061] According to certain embodiments of the invention, the architecture may optionally comprise:

[0062] - a DME calculation device configured to estimate the total energy consumed by the S site;

[0063] - a DCC device configured to control the energy consumption of the site S during a PO power outage.

[0064] The DME and DCC devices are described in the following with reference to certain embodiments of the invention.

[0065] According to the invention, the power cut management device DIS is configured to predict the energy consumption PR of the site S over a future period of duration œ for a prediction horizon H (i.e. expiring with a duration H in the future after a current instant), and to deduce therefrom a maximum duration AT during which the storage device STO can serve the site S before the device STO is completely discharged. The maximum duration AT thus corresponds to the autonomy of the storage device STO. From the deduced maximum duration AT and from the required power cut duration Te indicated in the request REQ coming from the entity EN_PR, the power cut management device DIS is configured to take a decision DEC to accept or not the power cut PO, the decision DEC being transmitted in return to the entity EN_PR, and being described in the following.

[0066] The DIS cut-off management device is for example a server, a platform, for example of the CSON type (“Centralized - self-organizing networks” in English), RIC intelligent controller (“Radio access network Intelligent Controller” in English) or OSS (“Operations Support System” in English), etc. It can be installed on the site S or remotely.

[0067] The site S is composed of K elements ELi, EL2, ..., Ek, ..., ELK of the same nature.

[0068] The RC network being a cellular radiocommunication network, the K elements ELi, EL2, ..., ELk can be respectively K different frequency bands, for example 1800 MHz and 2100 MHz or at least two different cellular network generations, for example 3G and 4G, in the case for example where K=2. The elements ELi to ELK can also be K RRUs (in English "Remote radio Unit") each operating on a different frequency band. Conventionally, an RRU corresponds to a group of cells which operate on the same frequency band with different network generations (2G and 3G on the 900 MHz band for example).

[0069] We will now describe, with reference to [Fig.2], the simplified structure of the cut-off management device DIS.

[0070] The DIS cut-off management device comprises: - a first communication interface C0M1 configured to communicate, via the RC network of [Fig.l] or another network, with in particular the DME calculation device and the DCC device; - a second COM2 communication interface configured to communicate, via the RC network or another network, with the EN_PR entity described previously; - a COL1 module for collecting DR data relating to the operation of said RC network, including in particular indicators representative of the operation of site S; - a COB1 module for controlling activation / deactivation of the STO storage device; - a PRED1 module for predicting the energy consumption PR of site S for a prediction horizon H, based on the indicators representative of the operation of site S included in the DR data; - a CALC1 module for calculating the maximum duration AT described previously, during which the storage device STO can serve the site S before the STO device is completely discharged, depending on the energy consumption PR for the prediction horizon H; - a COMP1 module for comparing the maximum duration AT and the required cut-off duration Te; - a DEC1 decision-making module DEC based on the result of the comparison from the COMP1 module.

[0071] At initialization, computer program code instructions PG1 are for example loaded into a RAM memory (not shown) before being executed by a processor PROC1 of the power outage management device DIS. The processor PROC1 of the processing unit UTR1 notably implements the following actions, within the framework of the power outage management method of the site S, which will be described below, according to the instructions of the computer program PG: - receive, from the entity EN_PR, the request REQ requesting a power outage and indicating a required outage duration Te as well as an outage time t0; - collect the DR data relating to the operation of the RC network, in particular the site S; - receive from the storage device STO, the value of the energy Eo stored before the power cut PO; - predict the energy consumption PR of site S for the prediction horizon H; - calculate the maximum AT duration, or autonomy, described previously; - compare the maximum duration AT and the required cut-off duration Te; - take the DEC decision, based on the comparison between the maximum duration AT and the required cut-off duration Te; - transmit the DEC decision, following receipt of the REQ request, to the EN_PR entity; - command the activation of the STO storage device based on the DEC decision. Optionally, according to certain embodiments, and according to the decision DEC, the cut-off management device DIS is configured to transmit the energy consumption prediction PR for the horizon H and the required cut-off duration Te, to the control device DCC.

[0072] We now describe, in relation to [Fig. 3], together with Figures 1 and 2, the progress of a method for managing a power outage of the cellular radiocommunication network, according to a particular embodiment of the invention. Such a power outage may for example be requested by the entity EN_PR in the PR request.

[0073] At a step 300, the power cut management device DIS receives the request REQ from the entity EN_PR, the request REQ requesting a power cut PO of the site S, within the framework of voluntary load shedding, the request REQ indicating the required power cut duration Te, and the power cut instant t0.

[0074] In a step 301, the cut-off management device DIS receives, or collects, the operating data DR of the network RC, comprising indicators representative of the operation of the site S, at a current time, which is prior to the time t0 indicated in the request REQ.

[0075] Such DR data may include raw indicators relating to the operation of the RC network, in particular of the site S, such as for example the number of resources occupied per cell covered by the site S, the quantity of data sent via the RC network, etc.

[0076] Alternatively, such DR data may comprise performance indicators or KPIs relating to the RC network, in particular to the site S, such as, for example, depending on the cellular network generation considered 2G, 3G, 4G, 5G, etc.: the traffic volume, the throughput for each cell, the average number of users for each cell, the cell load, etc. The energy consumption KPI is also collected by the outage management device DIS. Conventionally, such an energy consumption KPI is available for the radio site S, for each of the frequency bands or network generations of the radio site S. The KPIs are sent back to the DIS device with a granularity œ which is variable depending on the context of the modeling, which may for example be a time interval of 15 min, 30 min, 1 hour, etc.

[0077] In a step 302, the cut-off management device DIS predicts the energy consumption PR of the site S for a prediction horizon H, from the indicators representative of the operation of the site S collected in the previous step 301.

[0078] Steps 301 and 302 can be implemented in parallel with step 300 of receiving the request REQ. Thus, the request REQ can be received after the prediction step 302, as long as the required cut-off duration Te is less than a duration remaining before the end of the prediction horizon H, when the prediction horizon starts at the current time (the implementation of step 301). Alternatively, steps 301 and 302 are implemented following the reception of the request REQ in step 300.

[0079] In the following, an example of prediction of the energy consumption PR of the site S for the prediction horizon H is described, for illustrative purposes.

[0080] In this example, the prediction PR is a prediction of the total energy consumption of the site S. The prediction PR is based on Pt KPIs collected in step 301, which are noted in which » corresponds to a time index, in which 1 < _ / < pr denotes an index of the KPI and in which the superscript ( T ) refers to the entire site S.

[0081] As indicated previously, the site S is made up of K RRUs, each RRU corresponding to a group of cells which operate on the same frequency band with different network generations (2G and 3G on the 900 MHz band for example).

[0082] Step 302 may also include predicting the energy consumption of each RRU.

[0083] For an RRU 1 k < K, pk KPIs are obtained in step 301, which are noted in which' corresponds to the time index, in which 1 jPk designates the index of the KPI and in which the exponent (k) refers to the k'th RRU of the site S.

[0084] For example, the KPIs considered can be any combination of the following indicators:

[0085] - For 2G network generation: traffic volume and throughput for each cell;

[0086] - For 3G network generation: traffic volume, throughput and average number of users for each cell;

[0087] - For 4G network generation: traffic volume, cell load and number average number of users for each cell.

[0088] In this example, no indicator is considered for the 5G network generation. However, according to the invention, indicators relating to the 5G network generation may be taken into account for the prediction of the energy consumption PR of step 302.

[0089] In addition, an energy consumption KPI is generally available per radio site and per RRU, which makes it possible to access a current energy consumption value of the site and / or of each RRU, but also to store a history of the energy consumption of the site S and / or of each RRU. Such an energy consumption KPI is noted jp / Ù for the entire site S and for the k'th RRU, in which! refers to the time index of the measurement.

[0090] As previously indicated, all KPIs can be collected by the DIS cut-off management device with a granularity of (0 (w =15, 30, 60 min etc.). Thus, respectively corresponds to the energy consumption of site S, respectively of the RRU, during the reference period œ-

[0091] To take into account the granularity of implementation of step 301, the index * in the different KPIs is incremented every period 01. Thus, if the value of the index i = 10, then the measurements and j^W were taken at time / =10 w, expressed in minutes or hours.

[0092] The prediction at the current time index * (the current time of step 301) of the future energy consumption PR for a horizon H = hxœ , that is to say for a horizon ending at the time index i + h, is carried out by considering a history of the energy consumption (of the site S and of each RRU) as well as the current values ​​at time i of the KPIs according to the following prediction models:

[0093] For the entire radio site [°°941 (1)

[0095] For the RRU [°°961 1^=^+^^+^^ (2)

[0097] In which with (a) = (T) or (k) is the history of the values previous energy consumption measurements of site S or of the RRU of index k, the history being made up of L previous measurements. To take into account the granularity of collection of the measurements of step 301, the prediction horizon H is preferably a multiple of w. If h = 1 and w = 60 min, then the prediction horizon H = 60 min. If h = 4 and uj = 30 min, then H = 120 min, etc.

[0098] The prediction horizon H (and therefore the value h) can be determined from the cut-off time L indicated in the request, in order to estimate the consumption energy in a future period of duration w including the cut-off time t0.

[0099] In (1) and (2), the parameters are unknown, but can be obtained during a preliminary learning phase, not shown in [Fig.3]. The preliminary phase may include machine learning of the parameters, based on an artificial intelligence-based technique such as linear regression, LASSO regression (for "Least Absolute Shrinkage and Selection Operation"), or any other technique.

[0100] Machine learning can be performed on a training database, storing values ​​and ^'7, as well as the corresponding target values ​​jdd (the ground truth).

[0101] Thus, during the learning phase, the future value (for the horizon of ^i+h prediction hxw) is considered the variable of interest. The history of the energy consumption and current values ​​of other KPIs are predictor variables as shown in Table 1 where N is the number of observations used during the training phase, i.e. the number of datasets in the training database. Predictor variables Variable of interest v(a) Al,l ^i+h xX^) ^3-L ^Z+h A;V,1 ^N+iL a N+h

[0102] However, no restrictions are attached to the way in which the parameters ^a) a(a) ) j are determined, which can be obtained by a technique other than a technique based on machine learning.

[0103] Following the learning, the cut-off management device DIS stores the trained parameters da- ) and ) and is thus able to implement, by the PRED1 prediction module, a prediction of the energy consumption of site S, or RRU by RRU.

[0104] For the entire radio site, we obtain: 101051 (1)

[0106] The value obtained is a prediction of the energy consumption of the site S Ei+h (therefore corresponding to PR) over a period w for the prediction horizon H, that is to say for a period of duration w expiring at the end of the prediction horizon H.

[0107] For the k'th RRU, we obtain: 101081 (2)

[0109] The obtained value is a prediction of the energy consumption of the k-th ^i+h RRU over a period œ for the prediction horizon H.

[0110] According to a first embodiment of the invention, the cut-off management device DIS stores the predictive model corresponding to equation (1) in the prediction module PRED1 described previously. The application of the model to the history of the [YES]

[0112]

[0113]

[0114]

[0115]

[0116]

[0117]

[0118]

[0119] energy consumption of site S and the current values ​​of the other KPIs thus makes it possible to predict the energy consumption PR of site S over the period a' for the prediction horizon H. According to a second embodiment, the cut-off management device DIS stores the predictive model corresponding to equation (2) for each RRU of the site S, and the energy consumption prediction PR of the site S over the period w for the prediction horizon H can be obtained by summing all the individual predictions ^ki of the K RRUs of the site S. According to another embodiment not described, the outage management device DIS stores a prediction model of the energy consumption for each network generation (2G, 3G, 4G, 5G for example) on the site S, and the prediction of energy consumption PR of the site S for a duration œ for the prediction horizon H can be obtained by summing all the individual predictions of the network generations of the site S. The predictive models described above were tested, after learning the parameters, on a set of real test data from a cellular radiocommunication network, for a prediction horizon H, with a granularity value œ of 60 minutes, and a horizon H of also 60 minutes (therefore with h=l), in the case where the outage would occur at the outage instant t0 which is less than one hour after the current instant of index i. The test period includes data between 10:00 and 17:00 of day d. The parameter learning period includes the two days preceding day d. A learning model is created per site for the predictive model (1) and per RRU for the predictive model (2). The results show that the predicted energy consumption values, per site or per RRU, are very close or even identical to the actual energy consumption values. A second experiment was conducted to evaluate the performance of the predictive model (2) by RRU as a function of the absolute value of the relative error RE, between the actual energy consumption values ​​from the test data and the predicted energy consumption values ​​obtained from the predictive model (2): Actual value-predicted value actual value Table 2 below gives the results obtained expressed as a rate (or probability or percentage) that RE is less than 15% for several values ​​of the prediction horizon. Table 2 shows that the error on the prediction of energy consumption per RRU is low.

[0120] Table 2: Prediction horizon H - 12Q min H = 180 min H = 240 min Percentage that RE<15% 91% 86% 81%

[0121] Thus, the predictive models (1) and (2) are particularly suitable for predicting the energy consumption of a site or an RRU, as a function of radio KPIs such as traffic, throughput, load and the number of active users.

[0122] Referring again to Figure 3, the cut-off management method further comprises, following steps 300 to 302, a step 303 of calculation by the calculation module CALC1 of the maximum duration AT during which the storage device STO can serve the site S before the device STO is completely discharged, from the prediction of the energy consumption PR of the site S for the horizon H.

[0123] In particular, the maximum duration AT, or autonomy, is calculated from the value Eo of the energy stored in the storage device STO and the prediction of the electrical consumption PR of the site S during the period for the prediction horizon H.

[0124] For example, the calculation module CALC1 applies the following formula to determine the maximum duration AT;

[0125]

[0126] We recall that AO is the prediction of energy consumption from the model E'i+h predictive (1) or (2) over a period for the prediction horizon H. Alternatively, AT can be reduced by X% compared to the result obtained from the formula above, for example 10%, in order to take into account possible errors in the prediction.

[0127] In a step 304, the comparison module COMP1 of the cut-off management device DIS is able to compare the maximum duration AT with the required cut-off duration Te in the request REQ.

[0128] Depending on the result of the comparison, the decision module DEC1 is able to take a decision DEC, which is returned to the entity EN_PR in response to the request REQ.

[0129] If the maximum duration AT is greater than the required outage duration Te, then the decision DEC can be an acceptance of the request REQ. The outage management device therefore accepts the power outage PO at a step 305, and can activate the storage device STO following the power outage during this step 305.

[0130] If the maximum duration AT is less than the required cut-off duration Te, the cut-off management device DIS can take one of the following decisions DEC: - either refuse the request REQ during a step 306; - either conditionally accept the request REQ, replacing the required cut-off duration Te with a new duration less than the maximum duration AT, during a step 307. In this case, the cut-off management device DIS can transmit a cut-off duration proposal indicating the new duration less than the maximum duration AT; - either accept the request REQ during a step 308, then transmit, during a step 309, the required cut-off duration Te, which has been accepted, to the control device DCC for implementation of the additional steps of the cut-off management method, described with reference to the following figures, in particular to [Fig.5].

[0131] The DCC control device and the DME calculation device are now described. As a reminder, such devices are optional, and are advantageous in the embodiments where the decision according to steps 308 and 309 is taken by the cut-off management device DIS.

[0132] According to these embodiments, the control device DCC is in particular configured to activate the storage device STO at the time of the power outage PO and to calculate, optimally from a quality of service point of view and in a single pass, K instants of deactivation of respectively K elements ELi, EL2, ..., Ek, ..., ELk of the same nature which make up the site S, such that l <k<K.

[0133] The DCC device is for example a server, a platform, for example of the CSON type (“Centralized - self-organizing networks” in English), RIC (“Radio access network Intelligent Controller” in English) intelligent controller or OSS (“Operations Support System” in English), etc. It can be installed on the site S or remotely.

[0134] According to the invention, the DME calculation device is an artificial intelligence module which generates a mathematical model of the energy ET consumed on the site S, from the collection of DR data relating to the operation of the RC network.

[0135] Such generation is implemented at time t0 at the start of the power outage, or just before such a time (for example a few minutes before t0), so as to have the most recent DR data. The DME calculation device is configured to transmit to the DCC calculation device the energy ET estimated according to the aforementioned modeling. Note that the energy ET corresponds to the energy consumed on the site at the time of the estimation, and is different from the energy consumption predictions implemented by the prediction module PRED1 of the power outage management device DIS.

[0136] According to the invention, the DCC calculation device is configured to:

[0137] - calculate K deactivation instants Afj, AG, ..., AZæ or ^j, h of respectively the K elements ELi, EL2, ..ELK, as a function of the energy Er of the value of the energy Eo stored in the storage device at the time of the PO outage, of the duration Te of the power outage, as well as of DR data relating to the operation of the RC network, collected in a time interval preceding the PO outage,

[0138] - command the deactivation of the K elements ELi, EL2, ..., ELK, respectively at said K calculated deactivation times.

[0139] We will now describe, with reference to [Fig.4], the simplified structure of the DCC control device.

[0140] The DCC device comprises:

[0141] - a COM communication interface configured to communicate, via the network RC of [Fig.l] or another network, including the DME calculation device, the DIS cut-off management device, and the S site,

[0142] - a COL module for collecting DR data relating to the operation of said RC network,

[0143] - a COB module for controlling activation / deactivation of the storage device STO,

[0144] - a CAL module for calculating K deactivation instants Atj, A / 2, or t], G, • • •, tg of respectively K elements ELi, EL2, ..., ELK of the same nature which compose the site S,

[0145] - a COE module for controlling the deactivation of the K elements ELi, EL2, ..., ELk, at said K calculated deactivation times.

[0146] At initialization, the code instructions of the computer program PG are for example loaded into a RAM memory (not shown) before being executed by the processor PROC. The processor PROC of the processing unit UTR notably implements the following actions, within the framework of the complementary steps of the cut-off management method, which will be described below with reference to [Fig. 5], according to the instructions of the computer program PG:

[0147] - receive, from the DIS cut-off management device, information Ipo indicating a power outage PO of site S, as well as the duration Te of the power outage, accepted from the entity EN_PR during step 308;

[0148] - command the activation of the STO storage device, such capacity activation is optional, since the cut-off management device can itself activate the storage device STO when accepting the request REQ in step 308;

[0149] - receive, from the DME calculation device, an estimated value of the energy ET consumed by the site S before the power outage PO, and / or receive, from the power outage management device DIS, the predicted energy consumption PR of the site S;

[0150] - receive, from a communication terminal associated with the site S or from the STO storage device if it is equipped with a dedicated communication module, the value of the energy Eo stored in the STO storage device at the time of the power cut PO;

[0151] - collect DR data relating to the operation of said RC network,

[0152] - calculate K deactivation instants A / ;, At2, ..., A / Aoufi,•••' of respectively K elements ELi, EL2, ..., ELK of the same nature which make up the site S.

[0153] Optionally, the DCC control device is configured to:

[0154] - command the deactivation of the STO storage device after restoration of the fluent,

[0155] - command the activation of the K elements ELi, EL2, ..., ELK, after reestablishment of the fluent.

[0156] We now describe, in relation to [Fig.5], together with figures 1 and 4, the complementary steps of a cut-off management method, according to a particular embodiment of the invention following the decision of steps 308 and 309 described previously.

[0157] The additional steps may include a step S0 of estimating the energy ET consumed on the site S, from the collection of DR data relating to the operation of the RC network. Such a phase is implemented by the DME calculation device at the time of the outage, at the outage instant t0. These are therefore not the same DR data as those used in formulas (1) and (2) to predict the energy consumption of the site S during the steps of [Fig.3]. The DR data correspond to the data collected for the instant t0 of the outage or shortly before the instant t0.

[0158] As explained previously, the DR data may include raw indicators relating to the operation of the network, such as for example the number of resources occupied per cell covered by the site S, the quantity of data sent via the RC network, etc.

[0159] Alternatively, such DR data may include performance indicators or KPIs relating to the RC network, such as, for example, depending on the cellular network generation considered 2G, 3G, 4G, 5G, etc.: the traffic volume, the throughput for each cell, the average number of users for each cell, the cell load, etc. The energy consumption KPI is also collected by the DME calculation device. Typically, such an energy consumption KPI is available for radio site S, for each of the frequency bands or network generations of radio site S. The KPIs are collected with a granularity of œ which is variable depending on the modeling context, for example a time interval of 15 min, 30 min, 1 hour, etc.

[0160] The aim of phase S0 is to accurately and reliably model the energy consumption ET using an artificial intelligence technique. The accuracy of the modeling is crucial to avoid unwanted service interruptions. By having an accurate estimation of the power consumption, the strategies for deactivating the K elements ELb EL2, ..., ELK which will be detailed later in the description will be particularly suitable for selectively deactivating certain network generations or certain frequency bands, starting with those which are the least critical for QoS.

[0161] According to the invention, the DME calculation device uses a linear regression method for modeling ET energy consumption, for example the LASSO method (in English, “Least Absolute Shrinkage and Selection Operator”), Ridge, etc.

[0162] In the embodiment described below, it is the LASSO method which is considered, this type of method being particularly well suited to the processing of indicators representative of the operation of the site, whether these are KPIs, or raw indicators (number of resources occupied per cell, quantity of data sent via the network, etc.).

[0163] The LASSO regression which links the variable of interest, which is here the energy consumed E^ with the explanatory variables, which are here the KPIs or the raw indicators mentioned above, is expressed as follows:

[0164] py yMwyX») (3)

[0165] where:

[0166] - Et is the total energy consumed by site S,

[0167] - corresponds to the energy consumed when there are no users on the site radio and therefore no traffic,

[0168] - n corresponds to a generation of cellular network used on site S,

[0169] - N(n) corresponds to a set of different frequency bands used on the site S and corresponding to network generation n,

[0170] - p(n) is the number of KPIs considered for the generation of network n,

[0171] - aiJ is a linear regression coefficient which is trained in relation to a ith frequency band considered for a network generation, such that <i<^V(w ), et pour un j-ième KPI considéré, tel que \<y<p(n ).

[0172]

[0173]

[0174]

[0175]

[0176]

[0177]

[0178]

[0179]

[0180]

[0181]

[0182]

[0183]

[0184] Note that the linear regression coefficients aiJ are different from the coefficients ai determined for the predictive models (1) and (2) described previously. In the example shown, three generations of 2G, 3G, 4G networks are considered, such that n E {2,3, 4}. It goes without saying that this number can be less than 3, for example 3G or 3G and 4G, or greater than 3, for example 2G, 3G, 4G, 5G. The expression of the energy consumed per network generation n is represented in the following form: ^nG ^nG^ where 0„g is the energy consumed when data traffic is zero or very low, as is the case for example at night. The expression of the energy consumed by a considered RRU k or a considered frequency band k is represented in the following form: ^k ~ Û}.k +5-^(2,3,4)53^1 where represents the intercept which is a fixed value of the energy consumed for the RRU or the frequency band k. As is known per se, such a linear regression is subjected to Lasso regularization which is a regularization technique to penalize the coefficients anJJ (expression (3)), ^ / (expression (4)), a»J (expression (5)) of the less important features, by bringing these coefficients to zero. Thus, Lasso regularization adds a penalty proportional to the absolute value of the coefficients to the loss function. Such regularization is applied during model processing and when the model processes new samples from i to i+1. The modeling of the consumed energy ET, EnG? Ek, according to the Lasso regression above is implemented based on a set of real data (KPIs or raw indicators mentioned above) collected according to a time interval where, as a non-exhaustive example, w = 60 min, according to this modeling. The performance indicator RE of the modeling of the consumed energy is expressed according to the relation (6) below, in the form of a ratio between the difference in absolute value of the real value of En EnG or Ek, ​​and the estimated value of ET, EnG or Ek, ​​and the real value of ET, EnG or E^ i.e. the relative error RE between the real and estimated values ​​of En EnG or Ek; P p Actual value-estimated value! The training of the model for estimating the consumed energy ET, EnG, Ek was carried out for different metrics: mean RE(%) which represents the average of the RE values ​​over all the test samples, RE<15% which represents the probability (or rate) that the RE value is less than 15% over all the samples test, RE<10% which represents the probability (or rate) that the value of RE is less than 10% over all the test samples, considering the estimate of Et consumed by the site S, the estimate of EnG consumed for each of the network generations 2G, 3G, 4G and the estimate of Ek for an RRU operating at a low frequency, for example 800 MHz or 900 MHz, or high, for example 2100 MHz or 2600 MHz. The results of this training, as indicated in Table 3 below, show that the error on the energy estimation is low, whether this energy is estimated per site, per network generation or per frequency band.

[0185] Table 3: Relative error in estimating energy consumed SiteS 2G 3G 4G RRU Mean RE (%) 5.98 2.61 7.98 6.07 6.53 RE<15% (%) 97.83 100 90.83 99.57 96.92 RE<10% (%) 92.3 97.45 79.19 97.09 84.17

[0186] The results shown in Table 3 show that the error on the energy estimation is low. Thus, the energy estimation modeling based on the aforementioned Lasso regression is robust to accurately estimate the energy consumption based on the aforementioned KPIs or raw indicators.

[0187] Lasso regression, which incorporates a regularization term as defined above, is particularly effective in dealing with multicollinearity of predictive values ​​of KPIs or raw indicators, i.e., the high correlation between these predictive values. In the context of the aforementioned modeling, based on linear regression with several KPIs or raw indicators, including traffic, with some KPIs or raw indicators being predominant compared to other KPIs or raw indicators considered, Lasso regression introduces a penalty term according to which some regression coefficients, trained for non-dominant KPIs or raw indicators, are pushed to reach exactly zero. Such a penalization is notably described in the paper PK Matthew, FA Chama, and NS Agog, “Penalization Techniques for Remedying Multicollinearity in Multiple Regression ModeV KASU Journal of Mathematical Science, vol. 3.1, pp. 41-49, 2022. 。

[0188] Thus, when Lasso regression is applied to a set of raw KPIs or indicators, it is likely to identify and retain one or more of the most relevant raw KPIs or indicators, for example the “traffic” KPI or the “amount of data sent” raw indicator, while forcing the regression coefficients associated with the least relevant or highly correlated raw KPIs or indicators to be harmed. This not only improves the interpretability of the model, but also to improve its performance by focusing on the most informative explanatory variables.

[0189] We now describe, with reference to Figure 6, a curve representing the evolution, in a real situation, of the energy consumption, for example per RRU, as a function of time and over several days “Day 1”, “Day 2”, “Day 3”, “Day 4”. It is noted that the fixed value of the energy which does not depend on the traffic, is quite high compared to the consumption linked to the traffic. It is also noted the accuracy of the value of the intercept in equation (5), since late at night, users do not transmit data via their mobile terminals and the energy consumed corresponds well to On the other hand, we note that exceeds half of the energy consumption.

[0190] Of course, correspondingly, the evolution, in real situation, of the energy consumption, by site S or generation of cellular radiocommunication network, as a function of time and over several days, can also be obtained to verify the accuracy of the values ​​Vit and

[0191] The representative curve of [Fig.6] gives an indication of the deactivation strategy to be executed, although the curves representing the evolution, in a real situation, of the energy consumption, by site S or generation of cellular radiocommunication network, as a function of time and over several days, have not been represented. Indeed, this curve shows that it is not recommended to proceed with a cell-by-cell extinction of the RC network. Indeed, if one deactivates a cell, for example 3G-2100 MHz, then another cell, for example 3G-1800 MHz, then yet another cell, for example 4G-2100 MHz, the energy management of the STO storage device is not optimal. The RRU operating at the 2100 MHz frequency remains functional even when the 3G-2100 MHz cell is deactivated, because the other 4G-2100 MHz cell is still switched on.This means that the fixed part of the RRU (intercept) still consumes power even after the 3G-2100 MHz cell has been deactivated. And the intercept consumes a lot of power. The optimal strategy to use is to deactivate RRU by RRU or frequency band by frequency band, to save the power of the fixed part, once the RRU is deactivated.

[0192] Referring again to Figure 5, once the preliminary phase S0 of estimating the energy ET consumed on the site S has been implemented, the storage device STO is activated in SI when the power outage PO begins following acceptance of step 308. The activation SI is implemented by the aforementioned COB module of the control device DCC. Alternatively, step S2 can be implemented by the outage management device DIS during step 308, in which case steps 308 and S2 are combined.

[0193] The detection of the power outage PO may for example comprise a reception by the control device DCC, via its communication interface COM, of information IP0 indicating the power outage PO, coming from the power outage management device DIS.

[0194] The method continues by implementing, in S2, a calculation of K deactivation instants AZj, Az2, .... AZ^ou^i?^ •••' of respectively said K elements ELi, EL2, ..., ELk of the same nature which make up the site S. The calculation S2 is implemented by the aforementioned CAL module.

[0195] In S3, said K elements ELi, EL2, ..., ELK are then deactivated respectively at said K deactivation times AZh AZ2, ..., AtK or t.... tg calculated, using the aforementioned control module COE.

[0196] Step S3 can be implemented for a single element (K=1), in the case where the site S comprises only one frequency band or one network generation, as well as for several elements, i.e. at least two elements of the same nature. Thus, the complementary steps S0 to S3 which have just been described above can of course be applied to two, three, four, five etc. different frequency bands, different RRUs or different cellular radiocommunication network generations, which implies the calculation in S2 of respectively two, three, four, five, etc. deactivation instants Az2, AZj, AZ4, A / s. etc. or ^2' h' etc.

[0197] An embodiment of the calculation step S2 of said K deactivation instants Azb Az2, ..., AtK is now described with reference to FIG. 7A. In accordance with this embodiment, the calculation step S2 is implemented so as to maximize one of the aforementioned KPIs or raw indicators, for example according to the volume of data, the quantity of data sent via the RC network, etc. during the power outage PO.

[0198] The first embodiment will be described by considering the maximization of the data volume passed by RRU of the radio site S. It is considered that the radio site S operates with K RRUs, where K>1 and (Az1? AZ2, .... Az^) are the switching-off times of the K RRUs. According to the embodiment shown, the DCC control device is configured to determine the optimal switching-off times A / *, A , A making it possible to maximize the traffic volume passed during the duration Te of the power outage PO.

[0199] A set of data relating to the operation of the radio site S, as used in the calculation S2, comprises the following data:

[0200] - Eq the energy stored in the STO storage device at the time of the cut-off PO current, - The duration of the power outage PO, indicated by the power outage management device DIS;

[0201]

[0202]

[0203]

[0204]

[0205]

[0206]

[0207]

[0208]

[0209]

[0210]

[0211]

[0212]

[0213]

[0214]

[0215] -Vk=VklG+Vk,k4G The volume of traffic flowed at the time of the power outage PO, for a kth RRU among K, during a time interval where ^k^iG is the KPI corresponding to the volume of traffic of the network generation nG collected at the time of the outage, where n is for example such that 2 < n < 4. - Er the total energy consumed by site S, as estimated in S0. For this purpose, the method comprises, at the time of the power cut PO, a reception step S20, during which the DCC control device receives, via its communication interface COM: - an estimated value of the energy ET consumed by the site S before or during the power cut PO at time t0, from the DME calculation device, - a value of the energy Eo stored in the STO storage device at the time of the power outage PO, coming from a communication terminal associated with the site S or from the STO storage device if the latter is equipped with a dedicated communication module, - a value of the duration Te of the power outage PO, coming from the power outage management device DIS. Of course, as an alternative, the estimated value of the energy ET, the value of the energy Eo and the value of the duration Te of the power outage PO can be received simultaneously or each at different times. The method further comprises, at the time of the power outage PO, a step S21 of collecting data DR relating to the operation of said RC network, such as the value of the traffic volume Vk = V^k^kAG' in the example shown, energy consumption indicators associated with the k-th RRU, etc. During a step S22, the calculation module CAL solves the following optimization problem (7): Af .. K=i submitted to (7) At tk > 0, for l <k<K, Vk <V kjn- pour l<k<K, where vk ~ vk2G + vk3G + vk,4G to be optimized here corresponds to the amount of traffic transmitted by the k-th RRU during the reference time interval œ and vkjiG corresponds to the KPI of the amount of traffic of the k-th RRU of the network generation «G, in equation (5). Thus, is the total number of bits emitted by the k-th RRU during the cutoff w of PO current. It should be noted that if one or more RRUs is / are deactivated, its users will switch to the other RRUs still in operation. This results in a redistribution of traffic and users across the remaining bands after deactivation of one or more RRUs. Thus, vk may vary due to traffic transfer from a powered-down RRU to another working RRU.

[0216] Furthermore, in the optimization problem (7) above:

[0217] - is the amount of traffic of the k-th RRU that is collected at the time of the PO power outage,

[0218] - V is the maximum amount of traffic that the k-th RRU can support, the maximum traffic quantity being obtained from the data collected in S21,

[0219] - Ek is the corresponding energy consumed during the time interval of reference w,

[0220] - is the energy consumed by the k-th RRU during Af / f, starting from the beginning of the power outage PO, when this k-th RRU transmits vk bits instead of Vwhere Ek.v is the energy consumed when the k-th RRU transmits vk bits during the reference time interval is calculated by considering the above expression (5) and using vkjiG instead of V ^q. Recall that vk / iG and Vare used in (5) as the traffic KPIs of nG technology. For the other KPI values ​​to be used in expression (5), it is sufficient to assume that they change proportionally in the same way as the traffic KPI or to create a learning model that links each KPI to the traffic.

[0221] In order to solve the optimization problem (7), an energy efficiency metric is defined for each RRU, measured in bits / Joule. For a k-th RRU, the energy efficiency metric, denoted Pk, is expressed as follows:

[0222] _ (8) Pk “ Ek

[0223] Without loss of generality, we assume for this approach that p^ p^... > pK. Thus the calculation of p^ p^... ~ PK defines a priority order in which to successively deactivate the K elements by first deactivating the element ELK with the lowest energy efficiency and last deactivating the element ELi with the highest energy efficiency. Note that Pk is not constant and varies from one measurement to another in reality. However, from several experiments based on a real data set, it is found that when P[ (the average value of Pp, such that l <i<K, est significativement supérieure à Pf tel que l<i<K, alors P; > Pj almost all the time. The case where the means of Pi and Pj are close is not a problem since the optimal solution will not differ much.

[0224] Given that the energy stored in the STO storage device is insufficient to electrically supply the site for the entire duration Te of the outage, it is no longer possible to keep all the RRUs functional for the entire duration Te of the PO power outage. It is clear that the higher P^, the higher the volume of traffic transferred, because more bits are transmitted per unit of energy. Thus, the solution to the optimization problem (7) consists of keeping the RRUs that have a high energy efficiency for the duration Te of the PO power outage. These RRUs that have a high energy efficiency will also transmit with their maximum capacity. Conversely, the RRUs with a low energy efficiency are deactivated because they consume more and transmit fewer bits.

[0225] Thus, the solution (9) to the optimization problem (7) is as follows:

[0226] Vta,if l <k<N-l [°2271 A4 = (?)

[0228] a A = ?k = °'if k>N' K

[0229] where^ • \ and is the energy ïv — £31 min i / „ .iv üni iw } <L^K^ consumed by the k-th RRU during w when it transfers The energy is calculated using expression (5), when the traffic volume is equal to V^.

[0230] We now describe, with reference to FIG. 7B, another embodiment of the calculation step S2 of the K deactivation instants. According to this second embodiment, the calculation step S2 is implemented so as to optimize the autonomy of the storage device STO during the duration Te of the power outage PO, the order in which the K RRUs are deactivated being predefined by the manager of the network RC. For this purpose, a preliminary step SP2 is implemented, during which is defined from which threshold Th of autonomy of the storage device STP, it is appropriate to switch off a first RRU, then a second and so on. The threshold Th is defined in minutes in the embodiment shown. Of course, in other embodiments, the threshold Th can be defined in hours. In the example shown, for reasons of simplification, it is defined that:

[0231] - the first RRU, RRUi, is deactivated at ti,

[0232] - the second RRU, RRU2, is deactivated at t2

[0233] -...,

[0234] - the K-th RRU, RRUK, is deactivated at tK.

[0235] It goes without saying that this example is not exhaustive. It could be defined for example as:

[0236] - the third RRU, RRU3, is deactivated at tb

[0237] - the K-th RRU, RRUK, is deactivated at tK,

[0238] -...,

[0239] - the first RRU, RRUb is deactivated at tK-i,

[0240] - etc.

[0241] In this other embodiment, it is considered that the radio site S operates with K RRUs, where K>1 and are the instants of deactivation of the K RRUs. According to the embodiment shown, the DCC control device is configured to determine in a single pass the deactivation times at time of the PO power outage, allowing the autonomy of the storage device to be maximized during the duration Te of the PO power outage.

[0242] As in the embodiment of [Fig.7A], a set of data relating to the operation of the radio site S, as used in the calculation S2 comprises the following data:

[0243] - Eü the energy stored in the STO storage device at the time of the cut-off PO current,

[0244] - The duration of the PO power outage, indicated by the power management device DIS cut,

[0245] + the volume of traffic flowed at the time of the cut of current PO, for a k-th RRU among K, during a time interval % where V is the KPI corresponding to the volume of traffic of the network generation nG collected at the time of the outage, where n is for example such that 2 < « < 4.

[0246] - And the total energy consumed by site S, as estimated in S0.

[0247] For this purpose, the method comprises, at the time of the power cut PO, a step S200 reception, during which the DCC control device receives, via its COM communication interface:

[0248] - an estimated value of the energy ET consumed by the site S before the cut-off of PO current, from the DME calculation device,

[0249] - a value of the energy Eo stored in the storage device STO at the time from the PO cut, coming from a communication terminal associated with the S site or from the STP storage device if the latter is equipped with a dedicated communication module,

[0250] - a value of the duration Te of the power cut PO, coming from the device DIS cut-off management.

[0251] Of course, as an alternative, the estimated value of the energy ET, the value of the energy Eo and the value of the duration Te of the power cut PO can be received simultaneously or each at different times.

[0252] Also as in the embodiment of Figure 7A, the steps complementary to steps 308 and 309, according to the other embodiment further comprise, at the time of the power cut PO, a step S201 of collecting the data DR relating to the operation of said RC network, such as the value of the traffic volume Vk = V^2G + ^k^G^ Vk4G> in the example shown, energy consumption indicators associated with the k-th RRU, etc.

[0253] During a step S202, the calculation module CAL then calculates the instant ti+i of deactivation of the (i+l)-th RRU, RRUi+i, as follows; / ,+i = A;-Th + ti (10)

[0254] - where h is the previous instant of deactivation of the ith RRU, RRU;,

[0255] - where A,- is the autonomy of the STO storage device, calculated at

[0256] During a step S203, the calculation module CAL calculates the remaining energy at the storage device STO as follows:

[0257] y* (11)

[0258] - where is the remaining energy at the STO storage device, which has been calculated at

[0259] At ^ / +1, the (i+l)-th RRU, RRUi+i being deactivated, the traffic or data volume associated with it is distributed between the RRUs not yet deactivated, i.e. RRUi+2, ...,rruk

[0260] During a step S204, the calculation module CAL calculates the energy consumed by each of the RRUs, RRUi+2 to RRUK, from the aforementioned expression (5):

[0261] c _ û , y V DT

[0262] During a step S205, the calculation module CAL calculates the autonomy of the storage device STO at ti+\ according to the following relationship:

[0263] . (12)

[0264] Steps S202 to S205 are iterated for remaining RRUs, i.e., RRUi+2 to RRUk.

[0265] The last instant tK of deactivation of RRUK is then calculated as follows in S202:

[0266] + (13)

[0267] When initializing the calculation according to steps S202 to S205, it is considered that the residual energy of the storage device STO is such that pt°) — jr at t0 = 0, where A) corresponds to the instant of the electrical cut PO. In addition, at ^o, a parameter 2 for re-parameterizing the threshold Th is initialized, such that 0 < 2 < 1 and such that if Ao < Th, then Th =

[0268] We will now describe, in relation to Figure 8, a schematic curve illustrating the strategy for deactivating K RRUs in accordance with the calculation S2 implemented according to the embodiment, which has just been described above with reference to Figure 7B. In the example shown, K=4. Figure 6 shows from which autonomy threshold Th of the storage device STO, it is necessary to deactivate a first RRU, RRUb then a second RRU, RRU2, then a third RRU, RRU3, then possibly

[0269]

[0270]

[0271]

[0272]

[0273]

[0274]

[0275]

[0276]

[0277] a fourth RRU, RRU4. Such a deactivation strategy implies that during a PO power failure, the autonomy of the STO storage device should be continuously monitored, as shown in Figure 8. Thus, when the autonomy of the STO storage device reaches the threshold Th. RRUi is deactivated. This increases the autonomy of the STO storage device. Then, the autonomy of the STO storage device is monitored again and RRU2 is deactivated when the autonomy of the STO storage device reaches the threshold Th. and so on as shown in Figure 8. Note that the fourth RRU, RRU4, may not be deactivated when the autonomy of the STO storage device reaches the threshold Th, as symbolized by the dot-dash circle in [Fig.8]. The fourth RRU4 may indeed remain activated until the STO storage device is empty. As shown in [Fig.8], at the time of the PO power outage, the RRUs RRUi to RRU4 are all active. In S205, the autonomy of the STO storage device is calculated as follows from the aforementioned relation (12): y* „ In S202, the deactivation time of RRUi is then calculated as follows, from the above-mentioned relation (9): t^-A^-Th At 6, the autonomy of the STO storage device reaches the value of Th. The remaining energy (in joules) at the level of the STO storage device is then calculated in S203 as follows, from the aforementioned relation (11): n(DD e,- =e0-la=” As already explained above in relation to [Fig.7B], when an RRU is deactivated, the assumption is made that its traffic or data volume is distributed equally among the remaining RRUs, provided that the maximum capacity, in terms of traffic, of each of these remaining RRUs is not exceeded. Such an assumption is well-founded since a positive error on the traffic estimation added on one RRU will be corrected by a negative error on the traffic estimation added on another RRU. In addition, according to [Fig.6], the fixed part of the RRU is the one that consumes the most. This means that a slight error on the estimation of the new traffic distribution following a deactivation of an RRU does not have a significant impact on the estimation of the energy consumed by the RRU. The indicators (KPIs or raw indicators) related to traffic on an active RRU increase when another RRU is deactivated. For other KPIs, the assumption is made that they increase proportionally in the same way. than traffic-related indicators. It is also possible to use a learning model, based for example on linear regression, which expresses each indicator (KPI (apart from traffic) or raw indicator) as a function of the traffic transmitted on the active RRU. Using the new values ​​of the different indicators following deactivation of an RRU, a recalculation of the energy consumed by the RRUs still in activity is implemented in accordance with equation (5).

[0278] At ^2, the autonomy of the STO storage device reaches the value of Th. The remaining energy (in joules) at the level of the STO storage device is then calculated in S203 as follows, from the aforementioned relation (11):

[0279] Wrt) J-'r — J-'r L-1^2 w

[0280] Once RRU2 is deactivated at the new values ​​of energy E^ 3 <k<K sont calculées en S204 à partir de l’équation (3) précitée et de la répartition du trafic du RRU2 vers les autres RRU3 à RRU4 encore en activité.

[0281] In S205, the autonomy of the storage device STO is calculated again as follows from the aforementioned relation (12):

[0282] . rUr

[0283] In S202, the instant h of deactivation of RRU3 is then calculated as follows, from the aforementioned relation (10):

[0284] t3 = A2-Th+t2

[0285] At h, the autonomy of the STO storage device reaches the value of Th. The remaining energy at the level of the STO storage device is then calculated in S203 as follows, from the aforementioned relation (11):

[0286] yK e^2) — Z- <j{_3 <!'

[0287] Once RRU3 is deactivated at a new value of the energy E4 is calculated in S204 from the aforementioned equation (3) and the transfer of traffic from RRU3 to the still active RRU4.

[0288] In S205, the autonomy of the storage device STO is calculated again as follows from the aforementioned relation (12):

[0289] Â a^~ëT-

[0290] In S202, the instant h of deactivation of RRU4 is then calculated as follows, from the aforementioned relation (13):

[0291] ^ = ^ + 4

[0292] Although in relation to Figures 7A, 7B and 8, the step of calculating the K deactivation instants has been described in the context of an RRU-by-RRU deactivation strategy, it goes without saying that such a step can also be implemented, correspondingly, in the context of a deactivation strategy, and without departing from the scope of the invention:

[0293] - frequency band by frequency band,

[0294] - generation of cellular radiocommunication network by generation of network of cellular radio communication,

[0295] - set of RRUs per set of RRUs, a set possibly containing one or several RRUs.

[0296] The steps complementary to steps 308 and 309, described with reference to figures 4, 5, 6, 7A, 7B and 8 make it possible to ensure the continuity of the services provided by the radio site S, when a power outage of a duration greater than the autonomy of the storage device STO is accepted. They also make it possible to optimize the use of energy, thus contributing to more efficient and sustainable management of the communication infrastructures.

[0297] The CAL calculation module operates according to an ALGO algorithm, an example of which is shown above. In this example, the algorithm is written in a generic algorithmic language, adaptable to any programming language, such as for example Python, R, C++, Java, Scala, MATLAB, etc.

[0298] ALGO algorithm:

[0299] (1) Inputs: E'o, Cutoff plan, Th, À, (l^k^K)

[0300] (2) Inputs: KPIs and Ek (l <k<K) at the event of power outage and co

[0301] (3) Initialization: p© _ G = 0

[0302] (4) Calculate Ao based on Eq. (11)

[0303] (5) If Th>A0, thenTh^ÀA^

[0304] (6) For i from m 0 to K - 2

[0305] (7) Use Eq. (10) to calculate ^+i

[0306] (8) Use Eq. (11) to calculate

[0307] (9) ^Comment: Strategy of mobility

[0308] (10) Calculate o _ L+i Pq ~ K4A

[0309] (11) Calculate the vector SV

[0310] (12) Find the set Si=find (5V <pQ)

[0311] (13) Vk=n / M for ail

[0312] (14) Find the set S2=find (<5V >pQ)

[0313] (15) Vk=Vk+^0 for ail keS2

[0314] (16) u=sum[dV (k)] for ail keSi

[0315] (17) c=cardinal(S2) #Comment: Length S 2

[0316] ma=VM-u-cpQ;p = Z

[0317] (19) While [(a>0) & (c>l)]

[0318] (20) Calculate the vector 5V ^Comment: of size c

[0319] (21) Find the set S^find (ôV < / ?)

[0320] (22) Vk=V^for ail teSi [0321 ] (23) Find the set S2=find (ôV > / i)

[0322] (24) Vk=Vk+ / ?for ail k&S2

[0323] (25) u =sum[<5V(£)] for ail fceSi

[0324] (26) c =cardinal(S2)

[0325] (27) a = a - u -C =

[0326] (28) End While

[0327] (29) Update the other KPIs for RRU i+2 to K

[0328] (30) ^Comment: end of the strategy of mobility

[0329] (31) Calculate the new expressions of Ek based on Eq. (5)

[0330] (32) Calculate A / +] based on Eq. (12)

[0331] (33) End For

[0332] (34) Calculate t K based on Eq. (13)

[0333] In line (10), is the value of the traffic volume of the RRU (i + 1) which is deactivated at ti+{, this value being averaged over all the remaining RRUs.

[0334] In line (11), <5V is a vector that calculates the remaining capacity (in terms of traffic volume) of the RRUs ( ï + 2) at K. Then, ÔV = - V for i+2 < k < K.

[0335] As indicated in line (12), if ÔV (k) < then RRUk belongs to a set S]. This means that the RRU is not able to accept the entire quantity because it exceeds its capacity. In this case, as indicated in line (13), Vk — VA the opposite, line (14), if the remaining capacity of RRUk is greater than / ?0 (AV(k) > then RRUk belongs to a set S2. In this case, as indicated in line (15), v^v^.

[0336] In lines (16) and (17), an update u of the traffic KPIs of the RRUs belonging to Sj and an update c of the traffic KPIs of the RRUs belonging to S2 are implemented.

[0337] The rest of the traffic of the RRUs (i + 1) not yet distributed is given in line (18): a - Vï+1 - u - , where “ etc are given respectively in lines (16) and (17). In line (18), is redefined, such that — as the remaining traffic of RRU (i + 1) averaged over the remaining RRUs that still have the capacity to absorb the traffic (RRUs of the set^ j

[0338] In line (19), as long as a > 0, which means there is still undelivered traffic, and c > 1, which means there is still one or more RRUs that still have capacity, the While loop between lines (19) and (28) is activated to distribute the remaining traffic of the RRU ( i + 1 ).

[0339] As indicated in line (24), the algorithm finds the new value of Vk for an RRU such that (l <k<k). vk est composé du trafic des couches 2g, 3g et 4g. de la même manière que l’algorithme qui vient d’être décrit, distribution de sur toutes les également effectuée itérative sans dépasser le maximum chaque couche. par souci simplicité, nous n'ajoutons pas cette étape dans cet algorithme.

[0340] At line (29), an update of the KPIs of the remaining RRUs, RRUs ( i + 2 ) to K is implemented.

[0341] In line (31), Ek is again recalculated from the above equation (5).

[0342] In line (32), the autonomy A;+1 of the storage device is calculated from expression (12).

[0343] At line (34), the algorithm ends by calculating the time (r) of deactivation of the last RRU, RRUK, from equation (13).

Claims

Claims

1. Method for managing a power outage in a site (S) of a cellular radiocommunication network (RC), said site being provided with an energy storage device (STO), the method comprising: - a reception (300) of a request for power outage from the site, the request indicating a required outage duration; - a prediction (302) of an energy consumption of the site for a future prediction horizon, as a function of indicators representative of an operation of the site; - a calculation (303) of an autonomy of the storage device of the site as a function of the prediction of energy consumption of the site; - a comparison (304) between the calculated autonomy of the storage device and the required outage duration; - a decision (305-308) on the request, as a function of a result of the comparison.

2. Method according to claim 1, in which the indicators representative of the operation of the site (S) comprise a history of past values ​​of an energy consumption indicator of the site.

3. Method according to claim 2, in which the indicators representative of the operation of the site (S) further comprise at least one current value of at least one other indicator of operation of the site than the indicator of energy consumption of the site.

4. Method according to one of claims 2 and 3, in which the prediction (302) of the energy consumption of the site (S) is based on a predictive model defined by parameters associated respectively with past values ​​of the historical energy consumption indicator and / or at least one current value of at least one other operating indicator of the site.

5. The method of claim 4, wherein the parameters are defined by machine learning, from a training data set and a test data set.

6. Method according to one of the preceding claims, in which the operating indicators of the site (S) are global operating indicators of an entire site.

7. Method according to one of claims 1 to 5, in which the site (S) operates according to at least one frequency band and / or at least one generation of cellular communication network, and in which the operating indicators of the site are indicators associated with the at least one frequency band and / or the at least one generation of cellular communication network.

8. Method according to one of the preceding claims, in which, if the autonomy of the storage device (STO) is less than the required cut-off duration, the decision is a refusal (306) of the power cut.

9. Method according to one of claims 1 to 7, in which, if the autonomy of the storage device (STO) is less than the required cut-off duration, the decision is a transmission (307) of a power cut proposal indicating a new cut-off duration less than the autonomy of the storage device.

10. Method according to one of claims 1 to 7, wherein the site (S) operates according to at least one frequency band and / or at least one generation of cellular communication network, wherein, if the autonomy of the storage device is less than the required outage duration, the decision comprises an acceptance (308) of the power outage for the required outage duration, and further comprising, at the time of the power outage: - an activation (SI) of the storage device, - a calculation (S2) of a time of deactivation of said at least one frequency band or of said at least one generation of network, said calculation being implemented according to a criterion for optimizing the energy stored in the storage device, - a deactivation (S3), at said calculated deactivation time, of said at least one frequency band or of said at least one generation of network.

11. Method according to claim 10, wherein when said site (S) operates according to at least two frequency bands and / or at least two generations of cellular radiocommunication network, said method comprises, at the time of the power cut, once the storage device (STO) is activated: - a calculation, according to said optimization criterion, of at least two successive instants of deactivation of respectively said at

12.

13.

14. at least two frequency bands or respectively said at least two network generations, - a deactivation, at said at least two calculated deactivation times, of respectively said at least two frequency bands or of respectively said at least two network generations. A power outage management device (DIS) in a cellular radiocommunication (RC) site (S), the power outage management device being configured to: - receive a request for a power outage from the site, the request indicating a required outage duration; - predict the site's energy consumption for a future prediction horizon, based on indicators representative of the site's operation; - calculate the autonomy of the site's storage device based on the site's energy consumption prediction; - compare the calculated autonomy of the storage device and the required cut-off time; - make a decision on the request, based on a result of the comparison. Computer program comprising program code instructions for implementing the method for managing a power outage according to any one of claims 1 to 11, when executed on a computer. A computer-readable information medium comprising instructions of a computer program according to claim 13.

Images