FR3015792A1 - OPTIMIZED ENERGY SUPPLY OF A DISTRIBUTION NETWORK. - Google Patents

Abstract

The invention relates to a method implemented by computer means, for managing the supply of energy to a set of consumers, in which, following the obtaining of data relating to consumers and variables influencing consumption ( A): a) we build a first, coarse set of consumer classes (B), b) we then estimate an intra-class consumption model (C), for each class, c) we then build a second set, consolidated, of consumer classes (E), by grouping in the same class consumers whose models are close, according to an estimate of distances between models (D), d) defines at least one target model of consumers among the second set and an energy supply instruction (F) is generated which is recommended for consumers according to said target model.

Description

The present invention relates to the management of energy supply. It relates to a method of controlling the consumption of energy for groups of consumers or individual consumers.

Power distribution network management devices can have access to precise data on the network congestion or the consumption of the users of this network. For example, electric power providers may have accurate information on the individual consumption of each user having an electricity meter connected to central network management servers. In addition, the network management devices can transmit orders or proposals for application of energy supply instructions individually or to groups of consumers. Here, the term "energy supply instruction", instructions that can change the energy consumption of a user. Limiting instructions for the power available per consumer or location-based production plans for power plants can, for example, lead to a decrease in overall energy consumption at times when the network is usually congested. This control of the network is usually done within the framework of one or more programs of active management of the demand ("Demand Response Programs" in English). Thus, it has been proposed to process the data available at centrally managed computer servers in order to construct energy supply instructions adjusted to the state of the network. In addition to consumption information, the available data may include exogenous quantities such as weather forecasts or calendar data. The multiplication of these data leads to a significant increase in the complexity of the steps of generating the energy supply instructions. To solve these complexity problems, it has been proposed to bring together certain types of consumers in consumer classes. For example, consumers with air conditioning are grouped in a first class and consumers with a source of energy of their own (photovoltaic energy, wind, etc.) in a second class. Consumers thus collected, energy supply guidelines can be built from statistical models estimated for all consumers of a class. However, these models have weaknesses and may lack precision. In fact, the grouping of consumers is usually fixed and is carried out according to at most two criteria (type of heating and type of dwelling, for example). However, the relevance of the grouping of consumers varies according to a very large number of variables (weather forecasts, holidays, type of contracts between the consumer and the energy supplier, etc.). For example, in the days of high cold winds, the model of a class of consumers with high-performance insulation and that of a class of consumers with a personal energy source (wind or photovoltaic) have very strong similarities (low energy needs from outside). Thus, the rigidity of a treatment imposing classes of consumers a priori can be an obstacle to an optimization of consumption models.

Conversely, in another example of cold days with lots of sun but no wind, the model of a single consumer class with a source of personal energy (wind or photovoltaic) is not satisfactory. Indeed, in this situation, consumers with photovoltaic sensors do not need a lot of energy from outside, while those with wind turbines have more needs. The solution of generating a growing number of consumer classes to gain precision is also unsatisfactory. Indeed, as a power supply instruction is generated for each class, it is then necessary to adapt the energy production to each of these instructions. This adaptation is, in practice, very complex to implement. To this difficulty is added the risk of a bad statistical estimation of the models in each class. The present invention improves the situation. The invention proposes for this purpose the use of consolidated models based on a measure of distance between models generated for different classes of consumers.

The present invention thus aims at a method, implemented by computer means, of managing the supply of energy to a set of consumers, in which, following the obtaining of data relating to consumers and variables influencing the Consumption: a) a first game, rude, of classes of consumers is built; b) we then estimate an intra-class consumption model, for each class, c) we then build a second, consolidated set of consumer classes, by grouping in the same class consumers whose models are close, according to an estimate d) define a target model of consumers among the second set, and e) generate a set of energy supply, recommended for consumers according to this target model. The similarities existing between models are deduced from the estimation of distances between these models. The energy supply instructions are therefore generated for classes of consumers having similar behaviors at a given time and not according to immediate class criteria. In fact, by taking one of the examples mentioned above, estimates of distances between the coarse model generated for the class of consumers possessing efficient insulation and the coarse model of a class of consumers having a personal source of wind energy. indicate that these models are very close. Consumers in these classes will therefore have energy consumption close to usual consumption even if the cold winds would contribute to the increase in energy consumption in heating. An optimized energy supply instruction can therefore be sent to energy production means that will adapt the amount of energy to be supplied. For example, it is possible to avoid at least one emergency energy production for these consumers by using backup sources (such as thermal power plants and / or generators, in addition to the usual production from hydraulic and nuclear power plants). .

In addition, the class combinations contribute to reducing the number of energy supply instructions and to simplifying the generation of these instructions. Thus, more data can be provided at the input of the method of generating these instructions without, however, saturating the calculation capabilities implemented. The models and the resulting energy supply guidelines take into account more parameters and are therefore more robust in the face of changes in these parameters. Thus, with respect to an anticipated variation of one or more given parameters (such as, for example, temperature information derived from a weather report over a future period), the construction of relevant classes, based on more than one model. robust than in the state of the art, allows to deliver an effective instruction of energy supply, for a target model at least. Typically, in the above example, strong cold winds, it is then possible to generate an instruction called "erasure" (that is to say a maximum level of consumption not to be exceeded) for this particular class of consumers with wind turbines and / or isolated dwellings.

Nevertheless, the erasure instruction, for example, and more generally the energy supply proposal, if it is communicated to the targeted consumers, may have an influence on their consumption behavior (for example overheating before or after erasure). .

Thus, in an embodiment where a signal comprising information on a proposed application of the energy supply directive is transmitted to the consumers, the steps of the method according to the invention are repeated repeatedly after obtaining data. relating to consumer reactions to this signal. In particular, the information contained in this signal is injected into the estimation of intra-class models during the reiteration of the process steps. The commonly used models lack precision because they do not take into account the efficiency of delivery of the energy supply instructions. This efficiency of the execution of the instructions sent may, for example, depend on the communication problems at the level of the telecommunications networks, the relevance of the consumption models, but also the possibility for the consumer to derogate from the received instruction. The feedback of the signal comprising the proposal for applying a power supply instruction introduces this parameter of the routing efficiency of the instructions, in this embodiment. Thus, the energy supply guidelines can be refined to take into account the responsiveness of consumers to the proposals for applying these instructions. Consumer data may include items from consumption records in a habitat, number of people in the habitat, type of facility for consumption and / or energy production, type of subscription subscribed by the consumer to be supplied with energy. In addition, data on variables affecting consumption may include items among weather variables and calendar variables.

In one possible embodiment, the construction of the first set of consumption classes includes wavelet based data projection operations and this first set is constructed by classifying the coefficients in that projection. The estimate of the intra-class models can include for example an estimate of the consumptions in each class according to a base of spline functions, of type: e => fik (xi, t) +> lic (e, xi, t) + ieht ieht where: gt corresponds to a sum of consumer consumptions of a class k, and at a time t, - Jk is a set of explanatory variables, denoted xi, t, retained among the variables influencing consumption to model the class k, - and is a noise corresponding to a hazard of the class k, - pt corresponds to the information on a proposal to apply the energy supply directive, as defined above, at the instant t sent to class k, - / k (px) represents a kt elasticity function of consumers with respect to this information, - fik represents spline functions specific to explanatory variables in class k. Consumer elasticity may depend on exogenous parameters (hence the dual dependence of this function). For example, a consumer may be more in favor of erasing his consumption at certain times of the day or night than at others. In a particular embodiment, Q classes classes are obtained in the second set from Q consumer classes in the first set.

However, we can also provide many classes in the initial rough classification, and then seek to aggregate them according to the criterion of proximity of the models. To estimate the distances between consumer models, it is possible to calculate, for each variable of a model, predicted values of the variable on a partition of a set of definition of this variable, said values forming a vector reference in which are calculated said distances. In particular, these distances can be calculated by a formula of the type: ## EQU1 ## where: ## EQU1 ## mentioned above; xi with i E [[1; Aill the expected values of the variable on the ensemble partition mentioned above; f and g functions characterizing intra-class consumption models.

In addition, the vector reference can be of a dimension corresponding to an adjustable parameter according to a compromise between completeness of results and complexity of calculations.

The present invention also aims at a computer server for managing the supply of energy to a set of consumers, comprising an input interface for obtaining data relating to consumers and for obtaining variables influencing consumption, and comprising computing means (such as a processor and a working memory, for example) for: a) building a first game, coarse, classes of consumers, b) estimating a consumption model intra-class, for each class, c) the construction of a second consolidated set of consumer classes, by grouping in the same class consumers whose models are close, according to an estimation of distances between models, d) l estimating at least one target model of consumers from the second set, e) generating an energy supply instruction, recommended for consumers according to this target model. The present invention also aims at an installation for managing the supply of energy to a set of consumers, characterized in that it comprises: a computer server according to the invention, a first telecommunications infrastructure for transmission to the server above, consumer data and variables affecting consumption, and - a second telecommunications infrastructure (separate from, or common to, the first infrastructure), for transmission from the server to consumers according to the aforementioned target model of the aforementioned instruction of supply of energy. The present invention also relates to a computer program comprising instructions for implementing the method described above, when this program is executed by a processor. This program can use any programming language (for example, an object language or other), and be in the form of an interpretable source code, a partially compiled code or a fully compiled code.

Figures 2 and 3 described in detail below, can form the flow chart of the general algorithm of such a computer program. Other features and advantages of the invention will become apparent on reading the description which follows. This is purely illustrative and should be read in conjunction with the accompanying drawings in which: - Figure 1 illustrates the general context of the invention; FIG. 2 illustrates a general embodiment of the method according to a possible embodiment of the invention; FIG. 3 illustrates a distance determination between models according to a possible embodiment of the invention; FIG. 4A presents a possible installation for the implementation of the invention, according to a first embodiment, FIG. 4B presents a possible installation for the implementation of the invention, according to a second embodiment, FIG. 5 presents a possible installation for the implementation of the invention, according to a third embodiment, FIGS. 6A, 6B, 6C and 6D present curves illustrating models for classes of consumers in a mode of operation. embodiment according to the invention.

The invention is described below in its application to power supply networks to consumers. It is not limited to such an application and can for example be applied to water distribution networks, the management of air or water pollution, the management of all types of traffic (eg road, river or air example). As illustrated in FIG. 1, the invention can be implemented by a set of power consumption meters M60, M80, M100, MLO connected to a server 2 via a collection and control system 4.

Counters M60, M80, M100, MLO are present in consumers M6, M8, M10, ML. The meters can communicate with the collection and control system 4 via a local or remote telecommunication link (via line power lines or "CPL", or also via a local radio network or via a mobile cellular network, or others).

The invention can be implemented by collecting data from for example a home of an individual, or a company, an industrial, a community, or other. A consumer may also be an energy producer, as discussed further with reference to Figure 4B. The meters communicate with the collection and control system 4 which corresponds to all the telecommunication components used to communicate with the remote equipment, and in particular with the server 2. In an advanced collection and control infrastructure, the system 4 covers both the metering information system and a Wide Area Network (WAN) communication network, the intermediate gateways or hubs, or a local area network (NAN). neighborhood). This infrastructure can be based on any type of protocol and physical communication medium. The server 2 receives data, in particular those measured by the counters and transmitted via the collection and control system 4. These data are received by an input interface 14 and then transmitted to a circuit 10. This circuit, forming the means of aforementioned calculation of the server, may comprise for example: a processor capable of interpreting instructions in the form of computer program, and / or an electronic card whose steps of the method of the invention are translated in the form of instructions recorded in a memory of the card, and / or a programmable electronic chip such as an FPGA (for "Field-Programmable Gate Array" in English), or other. This circuit 10 may comprise or simply cooperate with a durable memory 12 ("hard-disk" in English) and with at least one RAM memory 16 (RAM, "random access memory" for direct access memory in French). Power supply instructions, as described below, are then generated and then transmitted to consumers M6, M8, M10, ML via an output interface 18. Referring now to FIG. 2, in step A, the server 2 receives from the collection and control system 4 consumption signals CONSO and explanatory variables VAR.

Each consumption signal CONSO [T] is specific to a consumer 1 or a group 1 of consumers. CONSO consumption signals may be physical quantities such as system status information or information directly provided by the consumer. A non-exhaustive list of the data that can constitute the CONSO consumption signals is given below: Instantaneous active power per phase [in W], total instantaneous active power three-phase [in W], instantaneous reactive power per phase [in "var , For "volt-ampere, reagent"], three-phase total instantaneous reactive power [in var], 25 instantaneous reactive power per quadrant (Q1, Q2, Q3 and Q4) [in var], instantaneous apparent power per phase [in VA] ], Three-phase total instantaneous apparent power [VA], Average active power per phase [W], Three-phase total active power [W], 30 Average reactive power per phase [in var], Three-phase total reactive power [en var], Average reactive power per quadrant (Q1, Q2, Q3 and Q4) [in var], Mean apparent power per phase [in VA], Three-phase total apparent apparent power [in VA], Active energy withdrawn (index 1 to N ) [in Wh], Power Apparent reference [in W], Apparent breaking power (cut-off device) [in W], Maximum power underwritten [in W], Available erasable power per phase [in W], Three-phase total available erasable power [in W], Switching state [open / closed], internal temperature [1/10 ° C], external temperature [1/10 ° C], power factor [without unit], current tariff [string of characters], Beginning and end of moving spikes [time stamped format], Effective Tensions (phases 1, 2 and 3) [in V], Effective Intensities (phases 1, 2 and 3) [in A], fundamental Frequency of network [in 1/10 Hz], data entered by the consumer (consumer usage data, periods when deletion is possible, desired or not, user-specified usage preferences, etc.), consumption on a set of consumer lines or appliances, and tariff or technical influence on the use of telecommunication networks mobilized for carrying out the method. The duration of integration of the instantaneous powers can take different values. For example, these powers can be integrated on 1, 2, 5, 10, 15, 30 and 60 seconds. The integration time of the average powers can also take different values. These values may for example be 1, 5, 10, 30 or 60 minutes.

More generally, the periodicity of the readings can increase to improve the precision of the computations or on the contrary to be reduced to accelerate these computations. This periodicity can be fixed or variable. If it is fixed, it can for example be monthly, weekly, daily, hourly or almost instantaneous (minutes, seconds, milliseconds). For example, a daily frequency for residential consumers and an hourly frequency for industrial consumers is conventional. In addition, the consumption signals can be sent live or delayed.

The existence of latencies (for example generated by telecommunications systems) can condition the temporal management of CONSO consumption signal transmissions. The precision implemented for the acquisition of these data depends on the calculation capacities available at the server 2. For example, the accuracy of the temperature measurement can be 0.5 ° K. Each explanatory variable VAR [m; n] is specific to a consumer m or group m of consumers and to a phenomenon n having an impact on energy consumption. Classically, these are weather variables (temperature, cloudiness, solar radiation, wind, etc.), calendar variables (type of day, time, dates of school holidays, holidays, etc.), price variables (supply tariffs subscribed by the consumer, tariff events such as erasures, etc.), socio-economic variables (type of housing, number of people in housing, economic growth, etc.), information on the electrical installation of the consumer (decentralized production, heat pump, etc.) or its uses. The number N of these phenomena can therefore be relatively large (of the order of several tens). Moreover, the number of explanatory variables may vary for each consumer m or group m consumers. The CONSO consumption signals and the VAR explanatory variables are collected in step A of FIG. 2 in order to constitute the DATA data. In step B, a first, rude game of consumer classes is built. This construction can be done by projecting DATA data on a wavelet basis and then classifying the coefficients resulting from this projection. Alternatively, this first classification can be fixed and provided according to simple criteria such as the type of heating, geographical location, etc. This classification leads to Q classes CLASSk (DATA) consisting of a variable number of consumers. Q may be important but this number is chosen so that the average number of consumers per class is approximately equal to 100, in one embodiment having given good results. Indeed, a correct estimate of an intra-class consumption model as implemented in the following steps requires about 100 samples. In step C, an intra-class consumption model is estimated for each class CLASSk (DATA). This model is based on the sum of the consumptions in each class. In particular, semi-parametric models are proposed to estimate aggregated consumptions in each class of consumers: e = fik (xi, t) + (p'tc, xi, t) + Et (1) jEJk jElk where: gt corresponds to a sum of consumer consumptions of a class k 10 constructed in step B, and at a time t, - Jk is a set of explanatory variables, denoted xy, retained among the variables VAR [m; n] influencing the consumption to model the class k, - and is a noise corresponding to a randomness of the class k, - / k (px) represents a function of elasticity of the consumers of the class kt 15 relative to a setpoint SIGk of supply of energy constructed in a prior iteration of the present process (in step F described hereinafter). This elasticity function is therefore available at this stage if the CONSO consumption signals provided in step A comprise consumer responses to the SIGk instructions application previously sent. For example, this elasticity function can take into account all the responses of the consumers of the class k over the last 3 to the last years. Fik represents spline functions specific to explanatory variables in the class k. Such functions reflect the evolution of consumption in relation to the explanatory variables xy.

It is assumed here that the elasticity li-c (ie, xi, t) is different according to the values taken by the explanatory variables xy. As a variant, it can be considered that this elasticity does not depend on such explanatory variables and we will then note II '(e).

Time-adaptive models can then be used to speed up computations and provide finer time analysis. Typically, this fine temporal analysis makes it possible to take better account of slow or rapid deformation of consumption. Thus, step C returns Q models gk (CLASSk) for each class CLASSk.

In step D, a second set of consumer classes is constructed by grouping, in a same class, consumers whose models are close. This grouping of models is performed by estimating the distances between the Q models gk (CLASSk) in order to obtain a similarity measure between models. Thus, the consumer classes obtained are the most homogeneous possible in terms of consumption behavior. This grouping of models gk is now described with reference to FIG. 3. Starting from the consumption gk of a class theoretically modeled by: e = fik (xi, t) + (p'tc, xi, t) + Et ( 1) we obtain (after estimation of the semi-parametric model on the data): One can then consider a simplification of this equation in: gk = fi <(x) îk (z) with: One can consider finally a simplification (2) of the equation (1) explained above such that: gk = k (x) ik (z) (2) where we note that: - the sum Ei, jk tic (xi, t) is simplified in fk (X), iE jk / 11 (pitc, in Îk (z) Thus, in step DB of figure 3, a partition of this model is carried out, because for the calculations of distances estimation between two Since models gk and gq can be performed by the hardware means 10 defined with reference to FIG. 1, the values of the functions fk (x) and kk (Z) must first be measured in a finite number of values. e X and Z included in the simplification mentioned in equation (2), we calculate the values provided by the simplification on a partition x1, x2, ..., xN and z1, z2, zN. This can be done on a grid of values (equidistant partition or equiquantous partition for example). The number N of measuring points and their position have an influence on the speed of calculations and the accuracy of these calculations. Thus, a significant number of points can be chosen to obtain accurate results and a lower number of points can be chosen to obtain results more quickly.

In step DC, an estimation of the distance between two models gk and gq is performed. We denote by d (r, s) the distance between two functions r and s defined and integrable on the set B, with for example d (r, s) = f (r (x) - s (x)) 2dx. As mentioned above in step DB, for such a calculation to be implemented, a grid x1, x2, ..., xN is used in order to approach d (r, s) by: dN (r, s ) = EN l (r (xi) - s (xi)) 2, where N is the size of the grid.

Thus, by taking again the equation (2) mentioned above, we obtain: d (e, gq) = duk (x), fq (x)) dosk (z), isq (z)) (3) Or , taking again the equation (1) and neglecting the influence of Et: d (gk gq) = d (f jk (X), fq (x)) + d (Ill (x), Here (x)) (4) jelkU lq jelkU lq This distance can be adapted by a weight W representing the importance of a variable j in the measurement of similarities between classes. We obtain then: dkg => Wi d (fi; (x), ri (x)) + Wi d (Ill (x), Here (x)) (4 ') jelkU lq jelkU lq For example, if we wants to group individuals close to each other in terms of the effect of temperature on their consumption, the weight associated with this variable may be more important. Based on the similarity measures obtained in step DC, new classes are constructed in step DD by performing class groupings. For example, a k-means classification algorithm can be implemented to perform these groupings. We can fix at Q the number of consolidated classes obtained. New CLASSp [e] consolidated classes are thus obtained at the end of step D.

In step E, new Gp [CLASSp] models are estimated for the CLASSp [e] consolidated classes. In step F, SIGp instructions for supply of energy are generated from the consolidated models Gp [CLASSp]. Each instruction corresponds to a consolidated model and a consolidated class of consumers. An instruction is therefore recommended for all consumers belonging to a consolidated class. In an optional step G, the power supply instructions SIGp can be optimized from the models Gp [CLASSp] calculated in step E. In fact, these models show a future global or partial consumption depending on the different exogenous variables (eg weather forecasts, future calendar events) and future signals generated from energy supply guidelines that could be sent to each class of consumers. Thus, at a time t, it is possible to obtain an aggregate future consumption curve over a future time interval [t + 1, t + h], h is a calculation parameter that can be fixed or variable. For example, a high value of h will reduce the overall relevance of the future consumption curve. We note this future consumption curve fv + i, t + h]. Only the dependence of the signal (SIG) to send to the consumers is taken into account insofar as the optimization is done only on this signal. It is indeed difficult to claim to have an influence on exogenous variables such as those resulting in meteorological events, for example. The optimization phase then consists in determining the signal to be sent to consumers which minimizes an indicator calculated from the consumption curve fi - [t + i, t + h] (for example, the consumption itself, or a gap between consumption and production expected between t + 1 and t + h). Different optimization techniques (Monte Carlo simulation, stochastic optimization, Newton method, etc.) can be used to solve this problem. In a variant, a production cost associated with the consumption curve fv + i, t + h] is taken into account in the optimization. In another variant, a cost of routing the sending signals to consumers is also taken into account. In another variant, the dependence of other variables is taken into account in the optimization. For example, the optimization is made from models gk, and / or from the consolidated models Gk, and / or from the energy supply instructions SIGp. Several possibilities are possible for the implementation of these energy supply instructions. These possibilities are described below with reference to FIGS. 4A and 4B. These figures show two consumers M6 and M8, respectively belonging to the CLASS and CLASS8 consolidated classes. This is of course an example and many consumers can belong to many classes. In FIGS. 4A and 4B, consumption signals CONSO are generated from measurements made by sensors M61 and M81 respectively included in the counter M60 at the consumer M6 and at the counter M80 at the consumer M8. These signals are then transmitted to the server 2 where they are treated as described above with reference to FIGS. 2 and 3. Considering in this example that the consumers M6 and M8 belong to different CLASS6 and CLASS consolidated classes, two supply instructions SIG6 and SIGE are then generated for these classes. In a first embodiment, a signal SIG UEp comprising a proposal for applying the setpoint SIGp is directly transmitted to a consumer belonging to a consolidated class p. This is called "multicast" sending. Thus, with reference to FIG. 4A, the set of signals (SIG_UE) p = is transmitted to the collection and control system 4. The appropriate signals SIG UE6 and SIG UEE are then respectively transmitted to the consumers included in the consolidated classes CLASS6 and CLASS (here M6 and M8, respectively). These signals are then used by M62 and M82 control modules that can for example offer consumers to directly apply the SIG6 and SIGE energy supply instructions (we are talking about a direct charge management proposal). In a variant, these signals may comprise an order of application of the instructions. The sending of a direct control order (eg switching off or switching on the electric heater, controlling the hot water tank, etc.) guarantees high reactivity but can generate discontinuities and discomfort for the consumer. This solution may be useful, for example, in the case of a forecast of a local or regional constraint on the power generation network (unavailability of a production resource or local network congestion). These signals can also indicate a price and / or a range of tariffs in order to encourage the consumer to adapt his energy consumption (for example in the context of "price control"). For the consumer, price control has the advantage of being a recommendation for consumption and not an obligation. From there, it can be analyzed a reaction to this signal. Other examples of signals generated from the instructions SIGp are conceivable.

A non-exhaustive list of these signals is given below: Upper limit of power consumption [in W] Lower threshold of power consumption [in W]. The lower threshold can for example apply in the case of high production of renewable energy where it is desired to increase the consumption via for example the operation of hot water tanks. Erase signal of a consumption type [unload / reset / refill]. The type of consumption is for example the heating, the operation of hot water balloons, lighting, air conditioning, etc. Price of current tariff [in monetary unit, ex. cents). This may be a price matrix, for example for all tariff items for the next 24 hours. Price of the future tariff [in monetary unit, ex. cents] Upper limit of power consumption [in W] by tariff range [in monetary unit, for example Euros cents] Upper limit of power consumption [in W] by time zone Signals may be at instant or deferred execution. In the latter case, the signal also includes a part indicating the activation date and validity date in time. In the example of an electrical system where the meter is an electricity meter, control can be done, either directly via a switching device integrated in the meter (eg relay or dry contact), or by an order sent via a wired or wireless digital communication link downstream of the meter to an external control device. In an alternative embodiment, a signal SIG ELECp adjustment to the energy supply directive SIGp is transmitted to the entity in charge of the supply of energy. Thus, with reference to FIG. 4B, an ELEC entity is in charge of supplying energy for several consumers M9 and Mll respectively belonging to the consolidated classes CLASS9 and CLASSii. The set of adjustment signals (SIG_ELEC) p = i ..... Q is transmitted to the collection and control system 4. The adjustment signals (S / G_ELEC) p = 9, ii which correspond to the classes consolidated CLASS9 and CLASSii are then transmitted to the ELEC entity. The control module ELEC1 then applies these signals to the power generation means so that the production for the consumers included in the CLASS9 and CLASSii classes are respectively adjusted to the instructions SIG9 and SIGH. For example, the energy transmission line LGN has been shown two flashes for the consumer M9 and a single flash for the consumer M11 to illustrate a different fit in the power generation for these consumers.

Other variants are also conceivable. In a broadcast system, all signals are sent to all consumers. The meters of these consumers then incorporate a device for identifying the signals intended for them. In a point-to-point system or "PtP", a signal is intended for a single consumer. This situation can, for example, emerge when a class includes a single consumer (typically a large consumer of energy, such as an industrialist). In time, these signals can be sent continuously, on a fixed date or in advance. These latter possibilities make it easier to apply the signals simultaneously and at the right time to a large number of consumers.

The telecommunication architecture implemented for the implementation of the method can be based on a single and bidirectional infrastructure, as illustrated in FIGS. 1, 4A and 4B, or on a double infrastructure separating the upstream and downstream flows, such as illustrated in Figure 5.

A single, bidirectional infrastructure has the advantage of simplicity, but it is not possible in states where regulations require delegation of the metering data collection system to a regulated operator. Similarly, energy services, including pilotage, can be reserved for the deregulated competitive market. In this situation, a double infrastructure is required and the collection of the counting data 6 is separate from the control 8 of the energy distribution network. The technologies used for these two infrastructures may be identical or different. For example, the metering data collection system 6 may use PLC technology for the NAN network and the mobile network for the WAN. The control system 8 can itself use only an Internet connection via a cable operator.

A concrete example of application of the invention as presented above is now described with reference to Figures 6A, 6B, 6C and 6D. At time t = 0, we consider four previously defined consumer classes (for example by means of a projection in a wavelet basis) and corresponding here to the type of consumption of a consumer and the tariff option chosen. by the latter: - class 1: consumers at basic residential rate with electric heating and no air conditioner; - class 2: consumers at the off-peak rate with electric heating and no air conditioner; - class 3: basic residential rate consumers with electric heating, air conditioner; and - class 4: basic residential rate consumers with no electric heating, no air conditioner but an electric auxiliary heater.

In accordance with step C previously described, an intra-class consumption model for each class is estimated. We choose here a model such that for the classes k E {1, we have: 7 gitc (1) .s. / C (ht) 11Dayt-ci + sic (ht, Tti) + fk (Toyt) q = 1 The estimated consumption of a consumer t belonging to the class k is thus defined by: - Eq7 = 1 Sqk (ht) 11Dayt = q the form of the consumption according to the time of the day (denoted ht: the time of observation t), Dayt worth 1 if the observation t is a Monday, 2 for a Tuesday, ..., 7 for a Sunday. - sk (ht, Tt) the consumption as a function of the time of day and the temperature T / associated with the consumer j (measured at the nearest weather station, at the consumer himself, etc.). The effects of sk (ht, Tt) are shown in Figures 6A, 6B, 6C and 6D respectively for classes 1, 2, 3 and 4. - r (Toyt) the annual cycle of power consumption, where Toyt is a variable continuing 0 and January of each year and 1 on 31 December.

This effect models, for example, the decrease in holiday-related consumption, but also the variations in consumption related to lighting during the year. Figure 6A shows that the effect of temperature on consumers of class 1 is zero for warm temperatures, but significant for cold temperatures (for example below a threshold of 12 degrees C). This effect varies according to the time of the day and is highest in the evening around 19h. As shown in Figure 6C, the effect of temperature for consumers with air conditioning is significant in the afternoons for temperatures above 20 degrees Celsius. Consumer consumption at base rate is higher around noon and in the evening 10 and this phenomenon is accentuated on Sunday. Off-peak consumers consume more when off-peak hours are effective: on average between 12pm and 2pm and at night. This analysis of the similarities between curves is here done with the only visual study of the curves. Of course, in practice, this analysis is done by distance calculations between the models that made it possible to generate these curves as described above in step D. The energy supplier having these 4 types of consumers in his portfolio may wish to control their consumption so as to lower the peak consumption at 19h by smoothing this peak between 17h and 21h when it is very cold (temperatures below 5 degrees for example). Thus, following the distance measurements as described in step D above, new consolidated classes of consumers can be constructed: - a consolidated class comprising the consumers of the old classes 1 and 3, - a consolidated class comprising the former class 2 consumers; and 25 - a consolidated class comprising the consumers of the old class 4. New models are then determined for these consolidated classes and energy supply instructions are then constructed from these models. Thus, in the example described above, where SIG signals UE comprise information on a setpoint application proposal, these signals can propose: for the consumers of the first consolidated class: normal central heating from 17 to 18h and from 8:00 pm to 9:00 pm, when it is very cold and the central heating is closed from 6:00 pm to 8:00 pm when it is cold but the overall consumption is high; - for Consumers of the second consolidated class: central heating flanged during peak hours; and - for Consumers of the Third Consolidated Class: the same instruction as for first class consumers by reducing the reduction of central heating to restricted hours. These signals therefore make it possible, for the first consolidated class, to reduce consumption around 19h, however compensating for this drop in neighboring hours. The influence of these signals is stronger for consumers in the third consolidated class in order to avoid a typical "rebound" effect after erasure, for example linked to the use of an auxiliary heating system when it is very cold. . At a time t + At, with for example At = 1 day, the actual consumption of the consumers is measured and then repatriated at night to the server 2. The optimization method within the meaning of the invention then estimates the effect of the signals sent. the day before (for example via the calculation of the elasticity 11 (pitc, x, t)) and carries out a new classification of the consumers. Consumers are now classified according to their uses but also their potential acceptance of instructions. Typically, a new class can be created to include consumers using heavily backup heating when the central heating is flanged. The optimization process can therefore optimize the sending of new signals according to these new characteristics by targeting reactive consumers for example. Another example of application relates to the integration of renewable energies in the production fleet. These energies are inexpensive and "clean" but have an intermittency (dependence on wind, sunlight for example). The development of these renewable energies in the production fleet leads to greater volatility in production and therefore requires flexible demand. In the case of, for example, a strong wind with high wind output, an incentive to consume may be sent to consumers for whom a high incentive elasticity has been measured in this type of situation (inferring this elasticity on the data also allows for quick reaction by sending this signal of encouragement to consumers most likely to be sensitive to it). The invention can also be implemented locally (by region, department, etc.) to facilitate the management of the network. For example, the action of pumps for filtering pools in the south of France can be implemented when there is wind in this region or when sunshine generates a strong photovoltaic production.

Of course, the present invention is not limited to the embodiment described above; it extends to other variants. For example, an embodiment has been described above in which the consumption measurements and the control signals were processed by a counter. Of course, these measurements and these signals can also be processed by other devices such as a box connected to the internet. This box can for example be dedicated to energy management (Energy-box type) or be directly an Internet access box (integrating the xDSL modem / cable). In this configuration, the measured quantities can reach the box by an external communicating meter via a wired or wireless digital communication interface and the control can be performed via another local interface of the box.

Claims (15)

REVENDICATIONS1. Method implemented by computer means for managing the supply of energy to a set of consumers, wherein, after obtaining data relating to consumers and variables influencing consumption (A): we build a first game, coarse, of classes of consumers (B), b) we then estimate an intra-class consumption model (C), for each class, c) we build a second, consolidated set of classes of consumer (E), by grouping in the same class consumers whose models are close, according to an estimation of distances between models (D), d) defining at least one target model of consumers among the second set, and e) on generates an energy supply instruction (F), recommended for consumers according to said target model. 2. Method according to claim 1, wherein a signal comprising information on a proposal for applying said instruction is transmitted to consumers, and wherein the process steps are repetitiously repeated after obtaining data relating to at least consumer reactions to said signal (A). 3. Method according to claim 2, wherein the information included in said signal is injected into the estimation of intra-class models during the reiteration of the process steps. The method according to one of the preceding claims, wherein the consumer data includes items from consumption records in a habitat, number of people in the habitat, type of facility for consumption and / or energy generation, a type of subscription subscribed by the consumer to be supplied with energy, and in which the data relating to the variables influencing consumption include elements among meteorological variables and calendar variables. 5. Method according to one of the preceding claims, wherein the construction of the first set comprises the data projection operations on a wavelet basis and the construction of the first set of class by classification of coefficients of said projection. 6. The method according to claim 3, wherein the estimation of the intra-class models comprises an estimation of the consumptions in each class according to a spline basis, of type: e => 1 -., K (,, t) + 11c (exi, t) + iehc iehc where: e corresponds to a sum of consumer consumptions of a class k, and at a time t, Jk is a set of explanatory variables, denoted xi, t, retained among the variables influencing the consumption to model the class k, - and is a noise corresponding to a randomness of the class k, - pt corresponds to the said information of the signal at the instant t sent to the class k, - / k (pk x) represents a k-class consumer elasticity function with respect to said signal information, - fik represents spline functions specific to explanatory variables in class k. 7. Method according to one of the preceding claims, wherein, from Q consumer classes in the first set, we obtain Q model classes in the second day. 8. Method according to one of the preceding claims, wherein, to estimate distances between consumer models, for each variable of a model, it calculates expected values of the variable on a partition of a set of definition of this model. variable, said values forming a vector coordinate system in which said distances are calculated. 9. The method according to claim 8, wherein said distances are calculated by a formula of the following type: ## EQU1 ## where: (1) (x,) - g (x,)) 2 i = 1 with: N is the dimension of said vector coordinate system; xi with i E [[1; Aill said predicted values of the variable for a model; and f and g functions characterizing intra-class consumption models. 10. Method according to one of claims 8 or 9, wherein said vector coordinate is of dimension corresponding to an adjustable parameter according to a compromise between completeness of results and complexity of calculations. 11. Computer server (2) for managing the supply of energy to a set of consumers, comprising an input interface (14) for obtaining data relating to consumers and for obtaining variables influencing consumption , and having a processor (10) for: a) constructing a first game, coarse, classes of consumers, b) estimating an intra-class consumption model, for each class, c) building of a second, consolidated game of consumer classes, by grouping in the same class consumers whose models are close, according to an estimation of distances between models, d) the estimation of at least one target model of consumers among the second set, e) the generation of an energy supply directive, recommended for consumers according to said target model. 12. Installation for managing the supply of energy to a set of consumers, characterized in that it comprises: - a computer server (2) according to claim 11, - a first telecommunications infrastructure for transmission to the server ( 2), consumer data and variables influencing consumption, and - a second telecommunications infrastructure, for transmission from said server (2) to the consumers according to said target model, of said energy supply directive. 13. Installation according to claim 12, characterized in that the second infrastructure is distinct from said first infrastructure. 14. Installation according to claim 12, characterized in that the second infrastructure is common to said first infrastructure. 15. Computer program product comprising instructions for implementing the method according to one of claims 1 to 10, when the program is executed by a processor (10).