WO2016188958A1

WO2016188958A1 - Method of identifying a data structure representative of a consumption of fluid, corresponding device and computer programme

Info

Publication number: WO2016188958A1
Application number: PCT/EP2016/061586
Authority: WO
Inventors: Jean-François HOCHEDEZ; Frédéric LEFEBVRE-NARE; Romain BENASSI
Original assignee: Wattgo
Priority date: 2015-05-22
Filing date: 2016-05-23
Publication date: 2016-12-01
Also published as: EP3298542A1; FR3036524A1; FR3036524B1

Abstract

The invention relates to a method of identifying an item of equipment carrying out a consumption of a fluid, the consumption being characterized by a unitary consumption segment called a lineament comprising a list of contextual and morphological parameters. According to the invention the method comprises: - a plurality of steps of obtaining (10) data representative of a morphological score (S_M) of said lineament (LIN) as a function of at least one value of at least one morphological descriptor (D) of the lineament (LIN) and of at least one category of equipment (C_E), delivering a list comprising at least one morphological score (LS_M); - a plurality of steps of obtaining (20) data representative of a contextual score (S_C) of said lineament (LIN) as a function of at least one value of a contextual descriptor (C) of the lineament (LIN) and of a mean quantity of fluid consumed (QF_MC) over at least one time period, delivering a list comprising at least one contextual score (LS_C); - a step of identifying (30) an equipment category (C) and/or an item of equipment giving rise to said lineament (LIN) as a function of the scores of said list of morphological scores (LS_M) and of said list of contextual scores (LS_C).

Description

A method of identifying a data structure representative of a fluid consumption, device and corresponding computer program.

1. Field of the invention

The invention relates to a technique for identifying consumption. The invention relates more particularly to a technique for identifying the consumption of a resource, such as a fluid, transported by a flow having variations, and of which repeated measurements are known. More particularly, the invention is part of an approach to control the consumption of resources, such as energy resources (gas, electricity, fuel pellets), but also natural resources (water). The proposed technique is implemented by a dedicated device comprising means for processing large amounts of data and calculation means for identifying equipment sources of fluid consumption.

2. Prior Art

The problem of the disaggregation of the consumption curve is relatively recent and the related works are not very identifiable. Among the many technical criteria that can occur in the principle and in the performance of a disintegration algorithm, a main distinction appears depending on the frequency at which the power consumed is sampled. Between ranges 1 hr - 1 min, 1 min - 1 Hz, 1-60 Hz, 60 Hz-2 kHz, 2-40 kHz, and> 1 MHz, the issues and possibilities differ greatly.

Within the framework of the proposed technique, methods based on harmonic analysis are not suitable because they are not compatible with the standard meter fleet, or even with most existing smart meters or with installed during the coming decade. These are said smart ('smart') because able to transmit information to the network manager but this information remains rudimentary and low frequency (a sample every 10 to 30 minutes, typically) so as to preserve life private users. However, in the present technique, we are interested in sampling frequencies higher than that of smart meters, although lower than the ten Hertz, and in practice, of the order of one tenth of Hertz. Although not transmitted to the network manager, this information exists inside the meters. It can be extracted using various means (external sensor, local digital output of the meter) and communicated by individuals to an "opt-in" service, to which they have voluntarily adhered. For the same reason of non-compatibility with the available sampling rates, methods based on the analysis transients are largely excluded from the field of the prior art because they generally require around 100 Hertz.

Methods comparable to the present, in terms of expected results, could be based on an analysis of quasi-stationary states: such methods, however, require steps during which fluid consumption should be sufficiently constant. Now it is known that a temporal profile of consumption can take various forms and manifest increasing or decreasing ramps. This is especially true in the case of power consumption in a dwelling. Thus, the results that could be obtained by such methods would be complex to calculate and would not necessarily be representative of the actual structure of consumption.

3. Summary of the invention

The invention does not have these disadvantages of the prior art. More particularly, the invention relates to a method of identification, implemented by an electronic device, of equipment belonging to a room connected to a fluid distribution network, said equipment realizing a consumption of said fluid, the consumption of said fluid by said equipment being represented by a unitary consumption segment, said lineament, said lineament comprising a list of parameters, said list of parameters comprising at least one contextual or morphological descriptor, a contextual descriptor being representative of a circumstance related to the consumption of said fluid and a morphological descriptor being representative of a consumption profile of said fluid by an equipment or a category of equipment. According to the invention, such a method comprises:

a plurality of steps for obtaining data representative of a morphological score of said lineament as a function of at least one value of at least one morphological descriptor of the lineament and of at least one category of equipment, delivering a list comprising at least one morphological score;

a plurality of steps for obtaining data representative of a contextual score of said lineament as a function of at least one value of a contextual descriptor of the lineament and of a quantity of average fluid consumed over at least one temporal period, delivering a list including at least one contextual score;

a step of identifying a category of equipment and / or equipment from which said lineament is derived as a function of the scores of said list of morphological scores and of said list of contextual scores. A time period may for example be the minute, the hour, the day, the month or the year.

According to a particular characteristic, prior to the identification step, at least one step of obtaining a list of morphological scores a posteriori as a function of said lineament, prior laws for each equipment or category and as a function of least a list of reference lineaments.

According to a particular characteristic, said identification step takes into account the list of posterior morphological scores.

According to a particular characteristic, a step of obtaining a datum representative of a morphological score of said lineament, whose endogenous parameters are listed in D, comprises, for a given apparatus a _k , a calculation of:

p (D \ a _k ) = p (D

= p (D | ¾), where ¾ = (A, _m ) _1≤m≤M where:

p D \ a _k ) represents the probability of observing the morphological descriptors D knowing that it is the equipment a _k that has exhibited an activity.

p {D \ {pk, m) _{1 <m <M} ) represents the probability of observing the morphological descriptors D knowing that they are distributed according to laws parameterized by parameters (pk, m) _{1 <m <M} > called " hyper-parameters ".

i9 _{fe m} represents the M hyper-parameters of the laws describing the descriptors of the equipment a _k .

According to a particular characteristic, a step of obtaining a datum representative of a morphological score of said linearity comprises, for a category A comprising a plurality of devices a _k , when no reference linearity is available, a calculation of :

in which :

p (Z)

represents the probability of observing the morphological descriptors D knowing that it is a device of the equipment category A, which has exhibited an activity.

p (Z) | i9j) represents the probability of observing the morphological descriptors D knowing that the hyper-parameters are equal to i9j. p _{i 0} (i9 _j ) represents the distribution of the hyper-parameters of the equipment category A ,.

di9 _j represents an infinitesimal increment on the hyper-parameters

According to a particular characteristic, a step of obtaining a datum representative of a morphological score of said linearity comprises, for a category A comprising a plurality of devices a _k , when a set of reference lineaments are available, a calculation from:

p (D \ (L _r ) AA _i ) = (fp tf x pi ù d i

JJût

in which :

represents the probability of observing the morphological descriptors D knowing that it is an apparatus of the equipment category A, - which has shown an activity and knowing that the lineaments (L _ir ) have been emitted by the apparatus A ,.

p (D | i9 _j ) represents the probability of observing the morphological descriptors D knowing that the hyper-parameters are worth i9 _j .

p _ir (i9 _j ) represents the posterior distribution of the hyper-parameters of equipment category A.

di9 _j represents an infinitesimal increment on the hyper-parameters.

According to a particular characteristic, an approximation of

is calculated by preferential sampling:

p {D \ {L _r ) AA _i ) ^ (£> r, fc)

k = l ... N

where (i9 _rik ) _k correspond to particles, whose histogram approximates Vi, r (d _> constructed using a sequential Monte Carlo procedure.

According to a particular characteristic, a step of obtaining a datum representative of a contextual score of said linearity comprises a calculation of:

p {a _j

v _iik (t)

in which :

- p ( _j j | c) represents the probability of observing an activity of the apparatus at _lj of category A, - knowing the occurrence of circumstantial descriptors C.

pG4 _j | C) represents the probability of observing an activity of category A, knowing the occurrence of circumstantial descriptors C.

Ψ (^ ₍ ) is the average annual resource quantity ^v i, k (represents the product of the usage profiles according to the time of the day, the day of the week, the day of the year, etc. The integral of each of the profiles (indexed by k) is standardized to unity.

Ψ0;) is expressed in Wh [Watt hour] for the case of electricity.

In another embodiment, the invention also relates to an electronic device identification device, belonging to a local connected to a fluid distribution network, said equipment realizing a consumption of said fluid, the consumption of said fluid by said equipment being represented by a unitary consumption segment, said lineamentally, said lineament comprising a list of parameters, said list of parameters comprising at least one contextual or morphological descriptor, a contextual descriptor being representative of a circumstance related to the consumption of said fluid and a morphological descriptor being representative of a consumption profile of said fluid by an equipment or a category of consumption equipment of said fluid.

Such a device comprises:

means for obtaining data representative of a morphological score of said lineament as a function of at least one value of at least one morphological descriptor of the lineament and of at least one category of equipment, delivering a list comprising at least a morphological score;

means for obtaining data representative of a context score of said lineament as a function of at least one value of a contextual descriptor of the lineament and of an average quantity of fluid consumed over one or more periods of time, issuing a list comprising at least one contextual score.

According to a preferred implementation, the various steps of the methods according to the invention are implemented by one or more software or computer programs, comprising software instructions intended to be executed by a data processor of a relay module according to the invention. invention and being designed to control the execution of the various process steps.

Accordingly, the invention also relates to a computer program, capable of being executed by a computer or a data processor, this program comprising instructions for controlling the execution of the steps of a method as mentioned above. This program can 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 form desirable shape.

The invention also relates to a data carrier readable by a data processor, and comprising instructions of a program as mentioned above.

The information carrier may be any entity or device capable of storing the program. For example, the medium may comprise storage means, such as a ROM, for example a CD ROM or a microelectronic circuit ROM, or a magnetic recording means, for example a floppy disk, a disk Hard, SSD Solid State Disc), etc.

On the other hand, the information 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. The program according to the invention can be downloaded in particular on an Internet type network.

Alternatively, the information medium may be an integrated circuit (ASIC or FPGA type) in which the program is incorporated, the circuit being adapted to execute or to be used in the execution of the method in question.

According to one embodiment, the invention is implemented by means of software and / or hardware components. In this context, the term "module" may correspond in this document as well to a software component, a hardware component or a set of hardware and software components.

A software component corresponds to one or more computer programs, one or more subroutines of a program, or more generally to any element of a program or software capable of implementing a function or a program. set of functions, as described above and below for the module concerned. Such a software component is executed by a data processor of a physical entity (terminal, server, gateway, router, etc.) and is capable of accessing the hardware resources of this physical entity (memories, recording media, bus communication cards, input / output electronic cards, user interfaces, etc.).

In the same way, a hardware component corresponds to any element of a hardware set (or hardware) able to implement a function or a set of functions, as described above and below for the module concerned. . It can be a component programmable hardware or with an integrated processor for executing software, for example an integrated circuit, a smart card, a memory card, an electronic card for executing a firmware, etc.

4. Figures

Other characteristics and advantages of the invention will appear more clearly on reading the following description of a preferred embodiment, given as a simple illustrative and nonlimiting example, and the appended drawings, among which:

Figure 1 shows a block diagram of the proposed technique for the treatment of fluid consumption data;

- Figure 2 shows a device for implementing the proposed method.

5. Description

5.1. Recall of the principle

The general principle of the technique described consists in characterizing data structures previously extracted from a consumption curve (data structures called arbitrarily linear) with the help of notations ("scores") which can make it possible to identify equipment that physically performed the said consumption in a given room. At the very least, according to the described technique, when it is not possible or possible to precisely identify a particular physical item of equipment, the technique described can be arranged in ascending (or decreasing) order of likelihood the classes of equipment to which the lineament can belong and later exploit this ordered list for different purposes. As an indication, one can for example define the following classes, when the fluid consumed is an electric fluid: refrigerator, electric heater, oven, dryer, etc. When the fluid is for example the gas, we can define the classes: boiler, water heater, hob. When the fluid is running water, we can define the classes: shower, toilet, bath, washing machine, dishwasher, etc.

It is recalled, for all intents and purposes, that the data structures on the basis of which the present technique is implemented are distinct segments of consumption called lineaments. These segments are previously obtained by a method that is not the subject of the present. The lineaments are for example obtained by identifying and bringing together at least two transitions (upwards and downwards) resulting from an initial consumption curve: that is, changes in the consumption flow that exceed ( in absolute value) a threshold parameter whose value can evolve along a process of cutting the curve of consumption. A lineament is in the form of a data structure, for example stored within a file or database as a record comprising several fields. In a particular implementation, a lineament includes exogenous data, dependent on external variables, usually circumstantial and usually having a temporal dependence, and endogenous data, describing the temporal profile of the consumption, generally of morphological type. , usually not having a time dependency. For example, a lineament may comprise, in a specific embodiment, a dozen different fields, corresponding to as many dimensions making it possible to identify this lineament in a unique manner. One of the main objects of the present technique is to assign the lineaments or any other episode of consumption representing a distinct event to physical equipment of the local where the initial consumption curve (for extracting episodes) has been generated. . This equipment is for example a boiler, a water heater, a refrigerator, a washing machine, a coffee maker, a television, a hob (gas or electric), etc. This assignment has several objectives: (1) to be able to determine profiles and typologies of consumption, (2) to be able to detect malfunctions in the equipment used, (3) to be able to advise or inform actors - human or not - about measures to take in order to optimize fluid consumption, etc. The proposed technique therefore has quite practical and useful applications.

More precisely, a lineament is a finite sequence ([/? ", T _n ]) of pairs of scalars, which, in this" raw "state, can not be processed in order to perform a classification. Thus, in order to accomplish this classification, a lineament is subsequently characterized using standardized descriptors. These descriptors constitute two vectors (viz, two ordered lists of scalars) which characterize, on the one hand, the endogenous morphology of the signature (in an imaginary way, the shape of the lineament), and on the other hand the exogenous circumstances under which it is happened (under what conditions this form appeared). We will also say its context.

The morphological descriptors are denoted D = [D _k ] ₁ _{k k} _{0 0} and the contextual descriptors C = [C _k ] ₁ _{k k} c _c . The dimension D of the vector D can be of the order of a few units but it can reach a dozen or more. The integrated consumption of L, its maximum consumption, its total variation (TV), or its global slope represent some examples of parameters likely to belong to D. The timestamp and its derived values (time slot, day of the week, working day / holiday), the inside and outside temperatures are indicated, when they are wanted and available, in C.

In the following, we consider that a lineament L has occurred, it is completely characterized by the pair [D, C].

L = [D, C]

With reference to FIG. 1, the proposed technique implemented by an electronic device is described. This technique identifies an equipment, belonging to a local connected to a fluid distribution network, said equipment realizing a consumption of said fluid, the consumption of said fluid by said equipment being represented by a unit consumption segment, called linearity (LI N), said linearity comprising a list of parameters, said list of parameters comprising at least one contextual or morphological descriptor, a contextual descriptor being representative of a circumstance related to the consumption of said fluid and a morphological descriptor being representative of a device or a device; category of consumption equipment of said fluid. In general, for this given lineament (LI N), one implements:

a plurality of steps of obtaining (10) data representative of a morphological score (S _M ) of the lineament as a function of the values of the morphological descriptors D of the lineament and of at least one equipment category (C _E ), providing a list (LS _M ) comprising at least one morphological score (S _M );

a plurality of steps (20) for obtaining data representative of a contextual score (S _c ) of the lineament as a function of the values of the contextual descriptors (C) of the lineament and of at least a quantity (QF _MC ) of fluid consumed on average over one or more characteristic duration (for example: hour, day, week, month, year), delivering a list (LS _C ) comprising at least one contextual score (S _c );

an identification step (30) of a category of equipment (Ca) and / or equipment (E) from which said lineament (LI N) is derived as a function of the scores of said list of morphological scores (LS _M ) and said list of contextual scores (LS _C ). Alternatively, or additionally, the categories may for example be ranked in the order of the scores.

This method is implemented for at least a portion of the lineaments of a load curve, for example for the lineaments considered significant (for the lineaments representative of a fairly high consumption, for example). More particularly, one of the main objectives of the proposed technique is to know whether a given linearity L = DAC (form D emitted in a context C) corresponds to the fluid consumption of a given device _k . To obtain this information, it is for example possible to calculate a probability that a lineament L comes from the use of the apparatus a _k .

From a theoretical point of view, according to Bayes' theorem, this probability is written:

Pk = p .a _k \ L) = p {a _k \ DAC) = p (D \ C) in which:

p _k represents the probability that the lineament L comes from the apparatus a _k ,

corresponding to the probability of a _k knowing D and C;

C represents the different values of the exogenous parameters;

D represents the different values of the endogenous parameters;

p (D \ a _k AC) represents the probability of issuing the morphological descriptors D knowing that it is the apparatus a _k which indicates an activity and that the circumstances are those described by the descriptors C.

pa _k \ C) represents the probability that the device has _k shows an activity knowing the circumstances are those described by the descriptors C.

p D \ C) represents the probability of issuing the morphological descriptors D knowing that the circumstances are those described by the descriptors C. In a process of assigning the lineament L to an apparatus a _k , for example by an estimator of the maximum a posteriori (MAP), we search the probability p _k the highest (among the probabilities p _k calculated for different devices a _k ), D and C being fixed. However, it is not necessary to calculate all the terms mentioned above giving the expression of this probability. Indeed, as explained above, the fact of dividing in two components of the lineament (morphology versus context) highlights the value of p (D \ C) at the denominator of the expression. However, the inventors have estimated that the morphology D of the lineaments (that is to say the shape of a lineament) could, for the most part, be considered as independent of the context C in which the episode of consumption appears: the morphology of the consumption of a device (for example a washing machine or a flush of water), does not vary according to for example the time of activation of the apparatus: the consumption of water of a flush or the The power consumption of a washing machine is practically the same regardless of, for example, the time of day or the outside temperature.

Thus, a score or score S corresponding to the stated specifications (ie having a relative value, viz., Suitable for the comparison between devices) is defined by keeping the expression of the probability p _k only as follows:

S _L ^fc = S ^ak (D, C) = p (D \ _ak ) xp (a _k \ C) = S x S

This gives the score of morphological descriptors D issued in context C.

Since D and C are fixed, the score S ^k is proportional to the probability that the pair (D, C) comes from the apparatus a _k . The score appears as the product of the probability of emission of the morphological descriptors of the lineament (or morphological score S _M ) and of the prior probability of the apparatus in the context (or contextual score S _c ).

The score is thus defined as the product of two factors. In another implementation, it could be a more sophisticated function of these two factors. The first factor accounts for the adequacy of the morphological descriptors of the lineament to the generic signature of the apparatus a _k . This is the probability of emission or morphological score S _M. The second estimates the probability that a unit's activity happens _k under the circumstances. This is the contextual score S _c .

These two scores are calculated independently of each other, for a device a _k given. They are for example recorded in a list (or a table or a table) in which the lineament L is associated with the apparatus a _k and comprising the two previously calculated scores.

When the scores of the L-lineament are calculated for all the devices one can perform the selection of the highest score (ie the highest score Si) in the list of scores s and select the lineament L as coming from the apparatus associated with the highest score. Alternatively, the list of scores may be used by a downstream module which proceeds to the assignment using this list together with other information, in a framework for example fuzzy logic.

These steps are implemented by means of modules and / or electronic devices for performing a calculation, for example a distributed calculation.

In the following embodiment, an implementation of the technique proposed for realizing an assignment of lineations resulting from consumption curves is described. electric. This embodiment is accompanied by theoretical explanations to validate the approach of the proposed technique.

It is understood that the object of the present is not to protect a theoretical approach to solve the problem, but a calculation technique. More particularly, as explained above, an interesting point of the described technique is to independently perform on the one hand a morphological calculation and secondly a contextual calculation, these two calculations can be conducted in parallel to accelerate the treatment speed. Furthermore, an object of the present technique is to propose a double level of calculation. Thus, a morphological score (basic when the calculation is made for a device, or categorical when the calculation is performed for a device category) can be calculated locally (that is, for a particular room or dwelling) and "reassembled" within a global database, in order to refine calculation results on other premises or housing (later local reuse). In particular, when an acceptable recognition rate is reached for a local or a dwelling, there is a mechanism for reporting this information within a so-called "global" database. This feedback is achieved through so-called "reference" lineaments. A reference lineament is a lineament that can be assigned to a category of equipment by virtue of its morphological and circumstantial characteristics which can then be associated with said category. The reference lineaments can be used to construct the probability densities that will subsequently be used to (1) refine the local allocation, but also to (2) evaluate the consumption of new housing or dwelling (ie new in that they have not yet been analyzed). We thus have a virtuous circle in which the identification of the lineaments of a load curve of a dwelling uses the consolidated results obtained previously in this same dwelling, and also the consolidated results obtained in other dwellings. The new results obtained are also used to update the old ones. We thus maintain a global database that is constantly updated.

In practice, the local databases and the global database are not necessarily separate, a local database can be an extract from the global database.

Thus, in general, the method that is the subject of the present technique comprises: initial notations, a priori, lineaments, to a given apparatus or category; Consolidated post-clearance ratings of these allocations, including the use of reference lineaments;

updates of reference lineaments locally and globally.

5.2. Description of an embodiment

This embodiment is presented for assigning a single lineament L to a category A. It is understood, however, that the assignment of multiple lineaments is performed in a substantially identical manner to a unitary assignment: it suffices to repeat calculations to determine the assignment of other lineaments.

In this embodiment of the proposed technique, it is also desired, in addition, to calculate a score by category A, and not only by apparatus. In this case, such a calculation begins with the use of the Bayes theorem, as for that carried out for a single apparatus, and is mathematically translated by the following expression:

_Vt = vi _t \ DAC) = -

Where p is the probability that a lineament L (summarized by the descriptors D issued in a context C) comes from the category of apparatus A ,. In the same way as for the "elementary" definition (view by device) previously posed above, the categorical score is defined by:

S ^Al (D, C) = S [= _P (D \ AX p (Al \ C) = S _M ^l x S _c ^l

The categorical score is a quantity proportional to the probability that a vector of descriptors D emitted in a context C comes from the category of devices A, (D and C being fixed). As for the elementary score, the categorical score is a function of two factors (eg their product): the morphological score (category) and the context score.

5.2.1. Calculation of the probability of emission of the descriptors of a lineament (morphological score) The first factor S _M appearing in the formulation of the score, p (Lla _k ) or p (LlA _; ), represents the probability of emission of the descriptors of the linearity L (by the apparatus a _k or by category A.

The inventors have noted that, for an apparatus given _k, its emission probability ordinarily follows a parametric law entirely determined by M parameters (i which are hereinafter referred to as the "hyper-parameters of a _k ". is usually equal to 2. This is the case for a normal law where the hyper-parameters encode its mean and its standard deviation, or, for example, for the gamma and inverse gamma laws, but M can be worth only 1 as in the case of an exponential law, or more than 2 (eg for the asymmetric normal law or for the generalized normal law).

For the sake of simplification, in a particular implementation, the inventors consider that all the descriptors of all the devices can be described by parametric laws, and that these laws are of the same nature (eg normal, exponential, etc.) within each category A, appliances. As a result, at the level of an individual apparatus:

V (P

= p (D | ¾), with ¾ = (, m) _1≤m≤M

This represents the expression of the probability of emission of the D-shaped lineament as a function of the hyper-parameters of the apparatus a _k for which it is desired to calculate a score.

For

This development represents the expression of the probability of emission of the shape D lineament by category A, for which a score is desired. By passing to the limit for a large number of elements with _u of representative A _h for example of a national set of equipment of a category:

This is therefore the expression of the probability of emission of the lineament of endogenous parameters D as a function of the hyper-parameters of the category A ,. This expression is used to perform the category calculations.

In a complementary way, one finds, in this expression, that one finds the probability of emission of the individual devices p (Z) | i9j) -which follow parametric laws, as indicated above-, but these are now weighted by < _, ο (¾ which must be interpreted as the a priori probability density of the hyper-parameters since it proceeds from the accumulation of probability of appearance of different devices at _u relative to the whole category A, to which they belong.

Thus, in order to evaluate the probability of D being emitted with respect to a fixed category, a calculation of the expression (J) \ A) is carried out via an estimation of the integral above. Despite the parametric character of the elementary emission probabilities p (Z) | i9j), the computation can not be carried out analytically because the product of a probability of elementary emission with Γ a priori ρι ₀ (βι) does not ordinarily admit an analytic form, especially since Ρι _ι0 { _ϋ ) can take the form of crenellations, combs, normal laws, bounded uniform laws, or other forms parametric. The a priori ρ _{ι ο} (βι) is coded as N designated particles (i9 _j j) _{eW (i)} (NB: two indices for particles but only one for hyper-parameters) whose histogram approximates the laws _{; 0} (i9 _j ). In concrete terms, the probability of issuance is estimated as follows:

Generally, I p (f) i) have a sufficiently simple form that the generation of the particles is trivial using a computer medium (for example, if the form chosen for P; ₀ (ui) is that of a normal law, any random generator of observations derived from a normal law can be used). Nevertheless, if the P; ₀ (ui) have a complex shape, the particles can then be obtained, for example, from several iterations of a Metropolis-Hastings algorithm.

These laws are initialized "by hand" from databases (manufacturers' catalogs for example), then supplemented and enriched by a feedback mechanism described below.

5.2.2. Calculation of the prior contextual probability (contextual score)

The "contextual" score is the prior probability of activity of an apparatus with _k or of category Ai as a function of the context C. The calculation of this score is simpler than that of the morphological score.

To account for its dependence on time of day, day of the year and time cycles in general, we ask:

p {a _j

oc Ψ (Α x ¾ (t)

k

Ψ (Α,) is the average annual resource quantity (expressed in Wh) consumed by the device or its category. v _ik (t) characterize usage patterns by time of day, day of the week, day of the year, and so on. Their integrals over the period considered are reduced to unity. The purpose of this modeling is to take into account (thanks to v _jk ) the fact that certain hours of the day and / or certain days of the week are more likely than others for the use of a given device, but to be nevertheless conservative in energy at the moment of the attribution thanks to Ψ (Α,). 5.3. Reference lineaments

One of the main objectives of the technique described is to allow recognition and optimized assignment of lineaments to devices or categories of devices. To do this, the inventors had the idea to introduce the notion of "Reference Linear". In a simple way, a reference lineament is a lineament that is known to be transmitted by a given device and whose descriptors will be able, as such, to be exploited later. When examining a lineament from the consumption curve, reference lineaments are used to increase the recognition rate. They allow an adaptation to local specificities (laws a priori (global / national) can indeed present mutual recoveries and thus induce indeterminations that housing devices do not manifest), increasing the recognition rate when the conditions for obtaining lineaments are less favorable. Secondly, they are exploited to enrich a national (global) database that is gradually becoming more diversified and more representative of the equipment actually owned by a large population.

The probability densities, - _{, 0} (ϋ _ί ) are initialized from datasheets and semi-quantitative experiments. The reference lineaments improve these probability densities _i0 fi? _/ J indirectly at the global level and compensates locally more directly.

5.3.1. Global initialization of the reference lineaments.

At the origin of the operation of a disintegration service, no lineament is available. But at this stage, simple and tolerant parametric laws (for example normal or uniform and bounded) can be introduced from the outset in the system on the basis of technical instructions, for example, so that morphological and contextual first scores can be calculated.

Therefore, for a given local site w (eg a dwelling), it is possible to associate scores with the lineaments (produced by the disaggregation of the consumption curve), and this for each of the devices a _k (falling within a range). category A.

At any moment, and for each category A "we consider the set £ ^ (t) of cardinal i. (T) of all pairs of type [lineament; its score for category A, and the subset ^ i ^ i) of cardinal r, (t) for which the scores exceed an empirically fixed threshold.

£ Γ (= { _u ] IZ e N _L - _(t) ] The set £ (t) is the set of all the lineaments extracted from a local load curve associated with their score for a given category A -. Cardinal L ^w (t) of this set depends on the w site and the time, but not in category A ,. It is a growing function of t.

-rf (t) represents the set of pairs of type [lineaments; local scores], for which the score exceeds an empirical threshold allowing to assign these lineaments to their category unambiguously. The cardinal r, ^w (t) of this set depends on the site w, the category A _h of the threshold and the time. It is a growing function of t.

r _{j 0} (t) is the expected number of lineaments for category A, and for duration t. This number can be derived from scientific literature or experience. We define the list of reference lineaments of local site w for category A, by:

Xf (t) = {[L ^] e rf (t) | Z e N _R AV [L _m , S _{i> m} ] 6 Am <¾}

with /? r (= min (rr (.n, o ()

It is therefore the Rl "(t) lineaments of the considered period having the best scores, and their number can not exceed the expected number.

The threshold is set empirically. This assumes a visual follow-up of the affectibility of the first lineaments, which helps to prevent the scoring system from forking into an unstable state. This humanly supervised phase is temporary and its time-consuming nature remains limited. Thus, at this stage, there are local lists of reference lineaments for each site and category.

5.3.2. Feedback from local lists to the global preconception

When a list of reference lineaments for a site w appears reliable (by the performance of the results or by a human supervision) it is useful to exploit it not only locally for the site w, but also globally. Taking this into account at global level also makes it possible, ultimately, to increase the representativeness of Va priori ρ _{ί 0} (ϋ {).

This feedback loop (local initial computation, global update and then local reinjection), which is an important feature of this embodiment of the proposed technique, makes it possible to take into account locally ( ^' .e. 'a particular dwelling or place), aggregations and global syntheses.

5.3.3. Enrichment of local lists and limitation of their cardinal

The cardinal of each set of reference lineaments (for any site and for any category) increases over time. This phenomenon is first of all beneficial since it enriches the list and the makes it more representative of the local device. However, it is necessary to set limits so that the system does not become too heavy. For this, the threshold is raised (very) gradually and thus becomes a function of time: = M fit). Simultaneously, Rfit) is bounded by a value,, (independent of the site and time), which forces any new lineament to take the place of one of the past reference lineaments. Different strategies are possible but it may seem stabilizing to replace randomly the reference lineaments passed while preserving the first lineaments to have been labeled as "reference" (those added during the first week for example), especially if have been validated visually by a human operator.

5.3.4. Redefinition of the score considering the reference lineaments

We then redefine the score (elementary or categorical) so as to take into consideration the reference lineaments of all equipment or all categories. By using Bayes' theorem again and noting that D, C and (L _jr ) _jr are not conditioned to each other, we obtain:

Pi.r = Pi, r ( ^L ) = PA L _{r Jr} .D, C) = P ( ^A i \ ^D > ^C > ( ^L )) p _ir is the probability of issuing the morphological descriptors D in the context C knowing that it is the apparatus A _t which testified of an activity and in view of the assignment of all reference lineaments Lj _tr . By swapping the elements of the last term, reapplying

Bayes and using the mutual independence of D and L _jjr versus C:

_Ρ (φ _{ί Λ} () AC xp ^ AC)

Pi, r = - _ΓΎ τ ^"κ V _\ P Λ (L _r )) X viAAC),

• the inventors having indeed found that the probability p (Σ ⁾ AI A {L _j ^ A (where the index

']' lists all the equipments) only depends on the reference lineaments of the apparatus A _it and is thus worth p (Z)

A (L _ir ) _r AC) (where 'i' is fixed),

• the activity of A, being assumed to be independent of {L _j, r). _r ,

· And the value pi) | (L _jjr.) Acj being identical for all the _j, it is not necessary to calculate it.

By keeping p _ir its numerator can therefore ask:

S _ir = S x _r _M ^l SC, noting:

¾, ⁼

the morphological score,

S _c ^l p = Ai \ C) \ e pop score.

It is the expression of the categorical score S _jjr of L in view of the reference lineaments L _jjr . By reintroducing hyper-parameters, and resuming the presented development

previously (see §5.2.1) in the absence of reference lineaments, the following results:

This is the expression of the posterior morphological score of L, ie in view of the reference lineaments L _ir . for category A, characterized by the function p, - ₀ of its hyper-parameters. With the notation P _jr (i9 _j ) = p (i9 a posteriori of the set of L _ir ,

This is the expression used for the posterior morphological score. The value of this morphological score is evaluated via an estimate of the integral above. Specifically, this score is estimated from N _K noted particles (û _irk ) _keNK whose histogram approximates the posterior laws p _ir di) via the following calculation:

In practice, the particles (û _rik ) _k are generated sequentially from the preceding particles (ι9 _{ί 7} -_ _ΐ ) - or, if the latter are not defined, from the particles (Pi _,] ) _] defined more high. This sequential generation of particles can be performed using a sequential Monte Carlo approach (also known as particle filtering). The use of such an approach is made possible by the fact that the posterior distribution Pi _r (i9j) can be computed to a constant. Indeed, we have with

known to the user.

5.3.5. Gradual improvement of the law a priori (reinjection of acquired knowledge)

As seen in the section "Feedback from local lists to the global a priori", we want to trace the local information back to the global database. To do this, it is necessary to wait until the posterior law of a site w is sufficiently informed. Since p (L | i9 _É ) is parametric, the posterior distribution p _ir {d) = p (i9j | 32f (£)) converges to a Dirac distribution δ _Θ γ, where 0f can be estimated by the MAP method .

0 = arg max p _iir (ui)

Thus, globally, for each category A, a collection is formed

{Qf \ we N _w ] from which a density Ρι _ι0 (ί) is constructed, either non-parametrically (for example by kernel methods), or parametrically by regression. The prior laws are therefore constantly updated to ensure the best representation possible. This feedback increases the performance of line recognition, both for new sites that do not yet have reference lineaments, and for sites whose reference lineaments are not yet very numerous.

5.4. Implementing device

With reference to FIG. 2, a device implemented to assign a lineament to a piece of equipment responsible for consumption or to a category to which this equipment belongs is described, according to the method described previously. For example, the device comprises a memory 21 constituted by a buffer memory, a processing unit 22, equipped for example with a microprocessor, and driven by the computer program 23, implementing a method of creating a structure of consumption data.

At initialization, the code instructions of the computer program 23 are for example loaded into a memory before being executed by the processor of the processing unit 22. The processing unit 22 receives as input representative data. consumption, in the form of lineaments comprising morphological characteristics and circumstantial / contextual characteristics (having a temporal dependence). The microprocessor of the processing unit 22 implements the steps of the creation method according to the instructions of the computer program 23 to associate these lineaments with predetermined equipment. For this, the device comprises, in addition to the buffer memory 21, communication means, such as network communication modules, data transmission means and possibly an encryption processor.

These means may be in the form of a particular processor implemented within the device. According to a particular embodiment, this device implements a particular application which is in charge of the calculations.

These means are also presented as communication interfaces for exchanging data on communication networks, interrogation means and database update, ...

Claims

A method of identification, implemented by an electronic device, of equipment, belonging to a room connected to a fluid distribution network, said equipment realizing a consumption of said fluid, the consumption of said fluid by said equipment being represented by a segment of unitary consumption, said lineament (LIN), said lineament comprising a list of parameters, said list of parameters comprising at least one contextual or morphological descriptor, a contextual descriptor being representative of a circumstance related to the consumption of said fluid and a descriptor morphological being representative of a consumption profile of said fluid by an equipment or a category of equipment,

said method comprising:

a plurality of steps for obtaining (10) data representing a morphological score (S _M ) of said lineament (LIN) as a function of at least one value of at least one morphological descriptor (D) of the lineament (LIN) ) and at least one equipment category (C _E ), delivering a list (LS _M ) comprising at least one morphological score;

a plurality of steps of obtaining (20) data representative of a context score (S _c ) of said lineament (LIN) as a function of at least one value of a linearity (C) contextual descriptor (C) and a quantity of average fluid consumed (QF _MC ) over at least one time period, delivering a list (LS _C ) comprising at least one contextual score;

an identification step (30) of a category of equipment (C _E ) and / or equipment (E) from which said lineament (LIN) is derived as a function of the scores of said list of morphological scores (LS _M ) and said list of contextual scores (LS _C ).

Identification method according to claim 1 comprising, prior to the identification step, at least one step of obtaining a list of posterior morphological scores as a function of said lineament (LIN), a priori laws for each equipment or category and according to at least one list of reference lineaments. 3. Identification method according to claim 2, characterized in that said identification step takes into account the list of morphological scores a posteriori.

4. Identification method according to claim 1, characterized in that a step of obtaining a datum representative of a morphological score of said lineament, the endogenous parameters of which are listed in D, comprises, for an apparatus a _k given, a calculation of:

p (D \ a _k ) = p (D

= p (D | ¾), where ¾ = (A, _m ) _1≤m≤M where:

p D \ a _k ) represents the probability of observing the morphological descriptors D knowing that it is the equipment a _k that has exhibited an activity;

p {p \ {pk, m) _{1 <m <M} ) represents the probability of observing the morphological descriptors D knowing that they are distributed according to laws parameterized by parameters {k, m) _{1 <m <M} > called "hyper-parameters';

5. Identification method according to claim 1, characterized in that a step of obtaining a data representative of a morphological score of said lineament comprises, for a category A comprising a plurality of devices a _k , when no reference lineament is not available, a calculation of:

p D \ A _i ) = [f pWâ x pi _fi

JJût

in which :

p D represents the probability of observing the morphological descriptors D knowing that it is a device of the equipment category A, which has exhibited an activity;

p (D | i9j) represents the probability of observing the morphological descriptors D knowing that the hyper-parameters are equal to i9j;

p _{i 0} (i9j) represents the distribution of hyper-parameters of equipment category A;

di9j represents an infinitesimal increment on the hyper-parameters. 6. Identification method according to claim 1, characterized in that a step of obtaining a data representative of a morphological score of said lineament comprises, for a category A comprising a plurality of devices a _k , when a set of reference lineaments are available, a calculation of:

p (D \ (L _r ) A Ai) = [f piDWd pi & d d i

J ^J

in which :

represents the probability of observing the morphological descriptors D knowing that it is an apparatus of the equipment category A, which has shown an activity and knowing that the lineaments (L _ir ) have been emitted by the apparatus A ,.

p _ir (i9 _j ) represents the posterior distribution of the hyper-parameters of equipment category A ,.

di9 _j represents an infinitesimal increment on the hyper-parameters.

7. Identification method according to claim 6, characterized in that an approximation of

is calculated by preferential sampling:

p {D \ {L _r ) AA _i ) ^ (£> r, fc)

k = l ... N

where (i9 _rik ) _k correspond to particles, whose histogram approximates j _r (i9 _j ), constructed using a sequential Monte-Carlo procedure.

8. Identification method according to claim 1, characterized in that a step of obtaining a data representative of a contextual score of said lineament comprises a calculation of:

p {a _j

v _iik (t)

in which :

p _(j j | c) represents the probability of observing an activity of the c¾ unit in the class _A, knowing the befallen circumstantial C. descriptors

Ψ (Α) is the average annual resource quantity ^v i, k (t) represents the product of the usage patterns according to the time of day and / or day of the week and / or day of the year.

9. Computer program product downloadable from a communication network and / or stored on a computer readable medium and / or executable by a microprocessor, characterized in that it comprises program code instructions for the execution of an identification method according to claim 1, when executed by a processor.

10. An electronic device identification device, belonging to a local connected to a fluid distribution network, said equipment performing a consumption of said fluid, the consumption of said fluid by said equipment being represented by a unit consumption segment, said linearly, said linearity comprising a list of parameters, said list of parameters comprising at least one contextual or morphological descriptor, a contextual descriptor being representative of a circumstance related to the consumption of said fluid and a morphological descriptor being representative of a consumption profile said fluid by an equipment or a category of consumption equipment of said fluid,

said device comprising:

means for obtaining data representative of a morphological score of said lineament as a function of at least one value of at least one morphological descriptor of the lineament and of at least one category of equipment, delivering a list comprising at least one morphological score;

means for obtaining data representative of a context score of said lineament as a function of at least one value of a contextual descriptor of the lineament and of an average quantity of fluid consumed over one or more periods of time, issuing a list comprising at least one contextual score;

means for identifying a category of equipment and / or equipment from which said lineament is derived as a function of the scores of said list of morphological scores and said list of contextual scores.