Extracting the electrical consumption of an individual device within a set of devices connected to an electrical network

Abstract

A method for extracting the power consumption of an individual piece of equipment within a set of equipment (E1, E2, E3, …, En), comprising transforming a stream of measurements of overall power consumption into a time series, and dividing this series (S1) into a set of sequences. Then, for each sequence, the following steps are performed: generation (S4) of a positional encoding matrix (PEM); extraction (S5) of features from the sequence to form a feature matrix (FM); projection (S6) of statistical metrics into a statistics vector (SM); and concatenation (S7) of these different elements (PEM, FM, SM) to form an input matrix (IM) provided to a transformation neural network designed to infer (S8) a time sequence corresponding to the individual piece of equipment. Figure for the abstract: Fig. 1

Description

Title of the invention: Extraction of the electrical consumption of an individual device within a set of devices connected to an electrical network. SCOPE OF THE INVENTION

[0001] The present invention relates to the non-intrusive determination of the electrical consumption of electrical equipment located in a specified room.

[0002] The invention relates more particularly to a method of disaggregating the load curve of a fleet of equipment in order to determine individual consumptions and to detect periods of activation of the equipment.

[0003] It thus enables the optimization of electricity consumption for users (premises managers) and of electricity production for the supplier or distributor.

[0004] These premises, whether domestic or business premises, contain a growing number of electrical equipment whose nature and behavior are extremely diverse, both in terms of consumption and activation rate per period of time.

[0005] In a residential building, one can typically find electrical equipment such as a connected television, lighting systems, household appliances, connected sockets, water heaters, charging stations for electric vehicles, electrical panels, inverters, etc.

[0006] However, it may become increasingly important to optimize electricity consumption, or more generally energy consumption, for better management of the available power of the electrical grid, but also for economic and environmental reasons. Such optimization can only be achieved through knowledge of the individual electrical behavior of each piece of equipment.

[0007] Generally, there is no way to measure the electricity consumption of each piece of equipment. Such an approach would require installing an electricity meter for each piece of equipment, which is undesirable due to the additional cost involved. Furthermore, such a solution would likely not be acceptable to users.

[0008] Non-intrusive load monitoring (or NILM for “Non-Intrusive Load Monitoring”) approaches have been proposed.

[0009] Load curve disaggregation aims to estimate the individual energy consumption of each device or piece of equipment, and / or the activation of the on / off state, using only the aggregated total load curve (apparent power) consumed for a dwelling). This aggregated total load curve is available via a smart electricity meter.

[0010] Load curve disaggregation was initially approached as a linear combination problem. Early research investigated the use of combinatorial optimization algorithms with the aim of estimating the proportion of energy consumption at each time step and for each active device. Subsequently, hidden Markov models were the preferred approach for several years.

[0011] In recent years, proposals have been made based on networks of neurons and deep learning.

[0012] One of the first publications relating to this family of approaches is the article by Kelly, J., & Knottenbelt, W. “Neural nilm: Deep neural networks applied to energy disaggregation”, in Proceedings of the 2nd ACM international conference on embedded Systems for energy-efficient built environments (pp. 55-64), 2015.

[0013] In their study, they compared the performance of three deep learning architectures: a Denoising Auto-Encoder (DAE), a recurrent network (LSTM), and a start / stop time regressor. These three models demonstrated superior performance compared to the classical machine learning methods then in use. Furthermore, this work sparked renewed interest in the field within the community, and numerous solutions based on deep learning were subsequently proposed in the literature.

[0014] Approaches based on convolutional neural networks have emerged, such as the one described in Zhang, Chaoyun et al. “Sequence-to-point learning with neural networks for nonintrusive load monitoring” in AAAI Conference on Artificial Intelligence (2016).

[0015] Approaches based on recurrent neural networks (RNNs for “Recurrent Neural Network” in English) have also been proposed.

[0016] Unlike classical neural networks, RNNs were proposed to retain previous information in "memory" while processing current data. The key characteristic of RNNs is the use of recurrent loops that allow information to be propagated from one time T to the next. Theoretically, this helps to maintain previous states in memory, and therefore makes RNNs particularly well-suited to sequence prediction tasks, such as machine translation, text generation, or load curve disaggregation. However, basic RNNs are difficult to train in practice, mainly due to the vanishing gradient problem when training on long sequences.

[0017] To address this, variants of RNNs, such as Bidirectional Long Short-Term Memory (BiLSTM) and Bidirectional Gated Recurrent Units (GRU), have been developed. These architectures incorporate specific mechanisms to better manage long-term dependencies and mitigate the gradient explosion problem.

[0018] More recently, a new approach has emerged based on "transformer" technology. This type of algorithm was introduced in the article by Vaswani, Ashish et al., "Attention is All You Need." In Neural Information Processing Systems (2017).

[0019] An example of such an algorithm, “BERT4NILM”, is described in the article by Zhenrui Yue, Camilo Requena Witzig, Daniel Jorde, and Hans-Amo Jacobsen “BERT4NILM: A Bidirectional Transformer Model for Non-Intrusive Load Monitoring” in Proceedings of the 5th International Workshop on Non-Intrusive Load Monitoring (NILM'20). Association for Computing Machinery, New York, NY, USA, 89-93.

[0020] This mechanism allows a partial disaggregation of the load curve: the individual load (or consumption) of certain types of equipment can be determined, but this proves impossible, or too imprecise for other types of equipment.

[0021] In particular, very few studies in the literature focus on estimating "energy-intensive" uses, such as electric vehicles, heating, or water heaters. Yet these uses represent the majority of household consumption in metropolitan France, a country where electric heating is very widespread. In other parts of the world, equipment such as air conditioning can also constitute significant portions of electricity consumption.

[0022] Furthermore, very few studies are conducted on data with "low frequency" time steps, which is nevertheless the data collection time step of the majority of smart meters in France, Europe, and worldwide (e.g., Linky meters collect data at 10-30 minute intervals). Consequently, most state-of-the-art solutions are tested only on a few high-frequency public datasets (collection time steps on the order of seconds), making it impossible to evaluate their performance in concrete application scenarios.

[0023] Furthermore, the algorithms operate on a subsequence of an overall load curve for a household in order to predict the consumption curve and / or the individual activation state of a given appliance. This subdivision into subsequences is essential in order to maintain an "acceptable" number of parameters within the model and / or a computation time. In addition, some studies in the literature have shown that a length High historical consumption data given as input to the model impacts its performance.

[0024] Data scaling is also an essential step to apply in order to train a deep learning algorithm. Indeed, the backpropagation algorithm used to train these models is sensitive to large values ​​(the gradient explosion phenomenon seen previously). It is therefore common to normalize each data point passed as input to a network (z-normalization or max / min normalization).

[0025] However, as highlighted in the study by Kelly et al. seen previously, applying subsequence scaling is not suitable for NILM. Indeed, normalizing by "subsequences" (i.e., each subsequence given as input to a model derived from the segmentation of the overall curve of a dwelling) implies the loss of information about the reference power value, revealing only the "shape" information of the signal.

[0026] However, it can be noted that the power value is closely linked to the specific use of the devices: a similar pattern (based on the "shape" of the consumption signal) may be common to several devices, but the maximum power reached allows them to be differentiated. For example, the activation pattern of a water heater and an electric vehicle being charged may be similar, but the value reached during vehicle charging is much higher.

[0027] Consequently, the typical treatment applied to the data in the NILM literature involves scaling the overall housing consumption curve followed by a subdivision into sub-sequences, without additional normalization applied to each of them.

[0028] However, this process induces a significant change in the data distribution within each subsequence (so-called non-stationary aspects of the data), affecting model performance. It should be noted that this phenomenon has recently been studied in the context of time series forecasting using neural networks, but has not yet been taken into account in the solutions proposed for NILM.

[0029] There is therefore a major challenge in taking this phenomenon into account in order to propose more efficient solutions on cases of disaggregation, and, more generally, to improve current proposals of the state of the art. Summary of the invention

[0030] The invention therefore aims to improve the situation by offering an efficient and high-performance method for extracting the load curve of a given piece of equipment from the aggregate load curve of a set of equipment.

[0031] To this end, according to a first aspect, the present invention can be implemented by a method for extracting the electrical consumption of an individual piece of equipment within a set of equipment located in a specific room, said method comprising transforming a stream of measured values ​​of the overall electrical consumption of said set, provided by a measuring device associated with said specific room, into a time series, and dividing said time series into a set of sequences, then, for each sequence, the steps of: - determination of statistical metrics, allowing the said sequence to be normalized; - generation of a positional encoding matrix from temporal values ​​associated with the values ​​of said sequence; - extraction of features from said normalized sequence to form a feature matrix; - projection of said statistical metrics into a statistical vector; - concatenation of said positional encoding matrix, said feature matrix and said statistics vector to form an input matrix supplied to a transformation neural network intended to infer a temporal sequence corresponding to said individual equipment.

[0032] According to preferred embodiments, the invention comprises one or more of the following features which can be used separately or in partial combination with each other or in total combination with each other: - said positional encoding matrix and said feature matrix have a width equal to the number w of values ​​in said sequence and the sum d of their height is equal to the height of said statistic vector, so that said input matrix has dimension (w+l)xd; - at least one step among the extraction, projection and inference steps is based on a neural network trained on a training set corresponding to said individual equipment; - said step of generating a positional encoding matrix includes a first projection of said time values ​​into a sinusoidal basis, and a second projection into a space one of whose dimensions is substantially smaller (d / 4) than the corresponding dimension (3d / 4) of said feature matrix; - said extraction step is implemented by a dilated convolution block; - the process further includes a step of denormalizing said temporal sequence corresponding to said individual equipment using projected values ​​of said statistical vector using a learnable transformation matrix.

[0033] Another aspect of the invention relates to a method for characterizing electrical consumption comprising an implementation of the extraction method as previously described for at least a subset of said equipment set

[0034] Another aspect of the invention relates to equipment comprising at least one processor adapted to implement a process as previously described.

[0035] Another aspect of the invention relates to a computer program comprising code instructions which, when executed by a processor, carries out the steps of the process as previously defined.

[0036] Another aspect of the invention relates to a data carrier on which at least one series of program code instructions for the execution of a process as previously defined has been stored.

[0037] Other features and advantages of the invention will become apparent from the following description of a preferred embodiment of the invention, given by way of example and with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE FIGURES

[0038] [Fig. 1] illustrates a context of use of a process according to embodiments of the invention,

[0039] Figure [Fig. 2] illustrates an embodiment of the learning phase for a method of extracting electrical consumption according to an embodiment,

[0040] Figure 3 schematically illustrates, in the form of a flowchart, a method for extracting the electrical consumption of an individual piece of equipment within a set of equipment, according to an embodiment

[0041] Figure 4 illustrates an example of dimensions for the different matrices used according to embodiments of the invention,

[0042] Figure 5 schematically illustrates a dilated convolution block, according to one embodiment of the invention

[0043] Figure 6 illustrates a transformation neural network according to one embodiment of the invention,

[0044] Figure 7 schematically illustrates a multi-scale dynamic attention mechanism, according to one embodiment of the invention.

[0045] Figures 8a, 8b and 8c illustrate results from public databases of the process according to one embodiment of the invention,

[0046] [Fig.9] illustrates the extraction of the load curve for equipment such as a search terminal of an electric vehicle with a method according to an embodiment of the invention.

[0047] DETAILED DESCRIPTION OF EMBODIMENT MODES OF THE INVENTION

[0048] Figure 1 shows a set of equipment, E1, E2, E3, ..., E1, located in a specific room L.

[0049] The defined location L may correspond to a geographically limited space within which a given user can have and control the connected objects. This space may correspond to a personal dwelling (apartment, house, etc.), or to the premises of a company, a store, etc.

[0050] The determined premises may possibly have extensions outside of a limited space, when for example certain connected objects are remote: a charging station for an electric car may be located in a garage, a garden, or even open onto the road, and be outside the main space corresponding to the dwelling, while being within the user's control perimeter.

[0051] The equipment can be of different kinds. Generally speaking, it refers to any equipment connected to an electrical energy distribution network NE and consuming this energy continuously or not.

[0052] For example, one could cite: - household appliances (ovens, refrigerators, hobs) heated...), - lighting systems, - heating and thermal regulation systems, - security systems (motion detectors, surveillance cameras, etc.

[0053] A device C is associated with the specified room L, and is adapted to provide measurement data of a global load, or consumption, for all the equipment associated with the specified room L. This device can measure the energy consumed on the internal NE distribution network of the specified room L. The measured values ​​can typically represent apparent power.

[0054] This device is typically a so-called "smart" electricity meter, that is to say, one adapted to provide such data, generally in the form of a stream of numerical values. It could, in particular, be a Linky™ type electricity meter in France, for example.

[0055] A suitable GTW gateway may also be provided to communicate with the measuring device C, in order to acquire a stream of measurement values ​​of overall electrical consumption provided by the measuring device C.

[0056] The GTW gateway may include data processing means. It may also have interfaces with the measuring device C and a telecommunications network N.

[0057] The interface with the measuring device C can be a radio interface, in order to minimize wired connections within the room L and for ergonomic reasons. To achieve this, a radio module may need to be connected to the measuring device C to ensure radio connectivity between it and the GTW gateway. This module can be a local radio transmitter (LRT) using the Zigbee™ protocol, for example.

[0058] According to a particular embodiment, the GTW gateway may include a Raspberry Pi type microprocessor, RAM memory (between 4 GB and 8 GB, for example), and mass storage, for example of the eMMC type, of about 16 GB.

[0059] The interface to the N telecommunications network can be wired (Ethernet) or wireless (Wi-Fi, etc.). More precisely, this telecommunications network can be composed of a local network to which a telecommunications gateway is connected in order to allow access to the public telecommunications network (the Internet). The N telecommunications network can therefore be viewed as a set of subnets.

[0060] The telecommunications network N allows the GTW gateway to communicate with a service platform S.

[0061] Different role sharing arrangements between the GTW gateway and the S services platform can be envisaged.

[0062] The S service platform can be in charge of managing the GTW gateway, including software updates, fault management, etc.

[0063] Data and processing can also be transmitted from the GTW gateway to the S service platform. This processing and data may correspond to tasks that cannot be performed locally, particularly because the computing power of an S service platform is required. The S service platform can be deployed on a server farm or, abstracted as a cloud computing server. Its computing power can be adapted according to the task to be performed, as submitted by the GTW gateways.

[0064] These treatments may correspond to the training of predictive models on training sets.

[0065] Indeed, training is costly in terms of computing power and execution memory. It is therefore advantageously executed on a service platform S which can be adequately sized for this type of processing, whereas the GTW gateway can only have limited resources since it is a device intended to be deployed at the user's premises: it therefore meets strong cost and sizing constraints.

[0066] The GTW gateway can advantageously implement the inference process allowing the extraction of the electrical consumption of an individual piece of equipment within a set of equipment.

[0067] In such an implementation, the model(s) trained by the service platform S can be uploaded to the GTW gateway.

[0068] Other arrangements and other distributions of processing between a local GTW gateway (possibly subdivided into several separate devices) and a remote S service platform can also be implemented.

[0069] In particular, in inference, part of the processing can be moved to the S service platform, in particular to aggregate data from different premises (in which case the data can be anonymized beforehand).

[0070] Figure [Fig.2] illustrates an embodiment of the learning phase of the different predictive model(s) used in the proposed process for extracting the electrical consumption of an individual piece of equipment.

[0071] In this figure, the GTW gateway and the S service platform are represented as a single GTW / S functional block, since the focus here is on the functional and algorithmic aspects and not on the implementation of the functions on particular equipment.

[0072] As described previously, the measuring device C collects load curves corresponding to the electrical consumption of a set of (electrical) equipment associated with a given room, i.e., a stream of measured values ​​of overall electrical consumption. This data is provided to the GTW / S functional block.

[0073] Furthermore, a measurement device C; can be provided to collect the particular load curve (flow of measurement values) of a given individual equipment Ei. This load curve is also provided to the GTW / S functional block and corresponds to the label, or truth (or "ground truth" in English), to which, in a manner known per se, the predictions of the predictive models can be compared during the learning phase.

[0074] This learning phase consists of evolving the internal parameters of the predictive model(s) to minimize the errors between a prediction made from a load curve provided by the measuring device C and the particular load curve provided by the measuring device C;, or "label" Thus, the predictive models learn to extract the particular load curve of the equipment E; from a load curve of a set of equipment.

[0075] Several training sentences can be provided in order to train several predictive models, each corresponding to a distinct individual piece of equipment: water heater, electric vehicle search station, etc.

[0076] Public training databases exist and can be used, such as the REDD, UK-DALE, or REFIT databases. These three databases present Consumption values ​​at fine time steps (5-10 seconds) for overall load curves and household appliances. REDD offers data from 6 US homes collected in 2011, UK-DALE from 5 English homes collected from 2012 to 2015, and REFIT from 23 English homes collected from 2015 to 2017.

[0077] It is important to note, however, that these datasets do not allow predictive models to be trained in all application cases of the proposed method.

[0078] Indeed, the data collected in these studies were gathered at "high-frequency" time steps (one data point collected per second), and the instrumented devices were small, low-energy-consuming appliances such as washing machines, dishwashers, or kettles. These training sets therefore do not allow for sufficient training of a predictive model to perform load curve extraction for equipment such as water heaters, electric vehicle charging stations, etc.

[0079] However, even within this restricted scope of application, it turns out that the performance of the proposed process exceeds that of the prior art processes.

[0080] In order to take into account other equipment, the mechanism previously described in connection with [Fig. 2] can be implemented. It can, for example, be implemented to train one or more predictive models that have already undergone initial training on another training set, for example a public dataset as previously described.

[0081] Fig. 3 schematically represents, in the form of a flowchart, a method for extracting the electrical consumption of an individual piece of equipment within a set of equipment, according to one embodiment.

[0082] As previously stated, the proposed method is an algorithmic process that aims to disaggregate (or extract) the aggregated load curve of a location collected by a communicating meter (also called a Delivery Point, PDL). This location may correspond to a home or dwelling, or to an industrial (SME...) or commercial premises, for example.

[0083] The process makes it possible to extract the individual load curve of several large devices in particular, such as the electric vehicle, the heating and the water heater.

[0084] The method takes as input a stream of measured values ​​for overall electrical consumption provided by a measuring device such as a PDL (Power Delivery Line). This overall consumption represents the load curve of all the equipment associated with room L corresponding to the PDL. In general, the measurements are average power measurements for 30-minute time steps.

[0085] This load curve can correspond to a variable duration. It can represent historical data or a current flow for near real-time processing.

[0086] However, since the purpose of the process is primarily to diagnose the average energy consumption of room L and to draw conclusions about the behavior of the equipment with a view to potential optimization, it is important to have sufficient historical data to detect the emergence of patterns. A history of approximately 3 days is recommended.

[0087] The method can provide as output a set of measured values ​​corresponding to an individual piece of equipment (electric vehicle, water heater, heating system, etc.). As many sets of values ​​(or load curves) can be provided as there are individual pieces of equipment considered. The individual pieces of equipment considered can be defined by configuration.

[0088] The process includes transforming the flow of measurement values ​​of overall electrical consumption, provided by the measurement device C associated with said determined local L, into a time series.

[0089] To achieve this, in one embodiment, the GTW gateway may include a preprocessing module adapted to transform this stream of measurement values ​​into a time series.

[0090] This time series is constructed to be provided as input to predictive models of the multilayer neural network type.

[0091] This time series can represent a time window in which each value represents a measurement value for an interval of this time window.

[0092] For example, we can have a time interval, or no time, of 15 seconds, and a time series of 480 values, which represents in total a time window of 480x15 = 2 hours.

[0093] Within the time series, the value assigned to a time interval can correspond to a resampling of the measurement stream values. For example, if more than one measurement stream value is available for the same time interval, the MT preprocessing module can establish an average value (for example) in order to construct the output time series.

[0094] The process then includes, in a step SI, the division of the time series into a set of sequences.

[0095] As previously mentioned, the time series can correspond to a load curve extending over several hours, or even several days. For subsequent processing, this large volume of data is subdivided into sequences corresponding to shorter time periods. Preferably, the sequences are of constant duration.

[0096] We denote w the number of values ​​in a sequence.

[0097] A sequence can be considered as comprising two series of values: - a series of w discrete values ​​of power (or load) readings such that xwl = (X], x2, ... xw) - a series of w respective timestamp values, each associated with a discrete power reading value, such as t2, ...

[0098] The timestamp values ​​are the discrete calendar variables extracted from the time steps (minute, hour, day of the week, month).

[0099] Several steps are then applied to each of these sequences.

[0100] Each of these steps S4, S5, S6 relates to an element of information contained in the input sequence: - the values ​​of the input sequence, - statistical metrics on this input sequence (mean, standard deviation) to manage the non-stationary phenomena inherent in load curves, - the time values ​​associated with the values ​​of the input sequence, in order to take into account the time data related to the uses of electrical equipment in a room.

[0101] The separate processing of these three elements present in the input sequence makes it possible to retain the information in the most relevant way possible by processing each element appropriately.

[0102] In general, these steps are implemented by an architecture comprising functional blocks of the "neural network" type, taking as input an aggregate consumption sequence of size w and generating a sequence of equivalent size. This is a sequence-to-sequence model (also called "seq2seq").

[0103] These steps can be carried out in parallel, or in parallel sequences. For example, steps S4, S5, S6 are independent and can be carried out in parallel. If they are executed sequentially, the order is then irrelevant.

[0104] A step S2 consists of determining statistical metrics from the sequence under consideration. Typically, these statistical metrics are the mean, p, and the standard deviation, 0, of the values ​​in the sequence. Thus:

[0105] u-±ywx / । ~ “yw-1 f*)

[0106] In an S3 step, these statistical metrics then allow the input sequence to be normalized.

[0107] This normalization can be z-normalization. Z-normalization, also called standardization, is a commonly used data normalization technique used in statistics and machine learning to transform data so that it has a mean of 0 and a standard deviation of 1. This transformation is useful for making the different characteristics of a dataset comparable and for improving the convergence of learning algorithms.

[0108] In this embodiment, we therefore obtain a standardized sequence Z such that [°109l Vieil.

[0110] It is this normalization step which makes it possible to take into account the non-stationary aspect of the data of the input sequence X.

[0111] In a step S4, a positional encoding matrix, MEP, is generated from the time values ​​Twxlas associated with the values ​​Xwxl of the sequence under consideration.

[0112] As will be seen below, a later step S8 uses a transform neural network. However, transformer-based models are by definition permutation-invariant (i.e., by default the mechanism does not take into account the position of the time steps in a sequence). It is therefore essential to provide the model with this information.

[0113] To achieve this, it is common to add a positional encoding matrix. However, this approach does not take into account the intrinsic discrete calendar values ​​of the time steps of the sequence (F timestamping).

[0114] In general, in the prior art, positional encoding matrices (PEMs) are weight matrices (learnable or static) that are added to the encoding (or "embedding") before a Transformer block. These matrices only take into account the relative position of each token in the given input sequence.

[0115] In our case, we propose to generate the matrix based on these Twxlas values ​​associated with the Xwxl values ​​of the considered sequence.

[0116] In particular, the proposed MEP matrix is ​​a projection of discrete calendar information (time series timestamp) into a higher-dimensional space, using a convolution layer. This matrix then contains information relating to the position of each time step in the subsequence given as input to the model, as well as global position information (time of year, for example). This information is essential in order to introduce more global information into the model (a device used primarily at a specific time of year or hour).

[0117] Unlike prior art MEP matrices, the positional encoding matrix, MEP, is concatenated to the encoding. Thus, moreover, the encoding information does not alter the characteristics relating to the shape of the signal.

[0118]

[0119]

[0120]

[0121]

[0122]

[0123]

[0124]

[0125]

[0126] It has been empirically demonstrated that this positional encoding matrix outperforms the use of conventional MEPs. More specifically, according to some embodiments, the values ​​... tw) extracted from the considered sequence are subjected to a first projection onto a sinusoidal basis such that: ^4) = ^” / With ie [ 1; p] corresponding to the set of p possible values ​​for each discrete timestamp variable. For example p=24 for time encoding, p=7 for encoding based on days of the week, etc. According to one embodiment, the generated encoding is then subjected to a second projection into a two-dimensional space, one dimension of which is equal to the number w of values ​​in the sequence under consideration, and the other dimension is a given number substantially less than the corresponding dimension of a matrix of features extracted from this same sequence, in a step S5. In the example illustrated in [Fig.4], the dimension of this space corresponding to the positional encoding matrix, MEP, is d / 4, while the feature matrix MF has a corresponding dimension of 3d / 4. The sum of these two dimensions equals a value d which corresponds to the height of the vector, or matrix, of statistics MS. The value of the internal dimension d is a hyperparameter that can be chosen or determined in various ways. In particular, the value d was chosen empirically to achieve the best average performance across a set of metrics. This value may vary for each device and dataset. Values ​​of d = {64, 96, 128, 256} are generally considered acceptable for deep learning models applied to univariate time series in the scientific literature. In this case, d = 96 can be used. The proportion d / 4 for the positional encoding matrix, MEP, is a ratio used to balance (or "balance") the amount of information from the extraction of features related to the shape of the curve, in step S5, which will be explained below, and the amount of information from the positional encoding matrix. This value of 4 was determined empirically through experiments and is based on the observation that information about the shape of the curve is less redundant and should therefore be used more frequently. Other values ​​are possible, however: it is by A possible example of varying the denominator between 2 and 8 (i.e., an interval {d / 2, d / 4, d / 8} depending on the value of d and the importance to be given to positional encoding.

[0127] This projection into a space of dimensions X] can, for example, be carried out by means of a one-dimensional (1D) convolution layer with a kernel of size 1.

[0128] 1D convolution is used to apply filters to one-dimensional data. A kernel (or filter) of size 1 means that the filter covers exactly one unit of the dimension to which it is applied. Since the kernel size is 1, it encompasses only one element at a time in the input sequence.

[0129] When the filter is applied with a kernel size of 1, each element of the input sequence is multiplied by a weight and can then have a bias added, which essentially amounts to a linear transformation of each individual element.

[0130] This convolution layer is parameterized to obtain an output in a space of size x j. In particular, the step (or "stride" in English) and padding (or "padding" in English) parameters can be fixed.

[0131] The convolutional layers use a kernel k=3 (an empirically determined value that may be a parameter to be optimized depending on the application, i.e., the device in question). Consequently, the fill value used in the convolutional layer is equal to k / 2 in order to obtain an output sequence of a size equivalent to that given as input. However, in general, the kernel size and the fill value can be empirically optimized by considering performance metrics for different tested configurations. Kernel sizes k of 3, 5, or 7 can, for example, be used.

[0132] Similarly, in order to obtain a sequence of size equivalent to that given as input to the convolution layer (we want to keep the same number of time steps) the step value used is equal to 1.

[0133] A feature extraction step S5 is also implemented. This step takes as input the previously normalized sequence, Z, and generates as output a feature matrix MF.

[0134] This feature extraction can be based on a neural network trained on a training set corresponding to the individual equipment considered: as detailed previously, predictive models are trained equipment by equipment during a learning phase and are therefore specialized.

[0135] In particular, according to one embodiment, the extraction step S5 can be implemented by a dilated convolution block.

[0136] A dilated convolution block, also called a "hole convolution" or "dilated convolution" in English, is a variation of the standard convolution layer used in convolutional neural networks (CNNs). It introduces spaces (or holes) between the elements of the convolution kernel, which makes it possible to increase the receptive field without increasing the number of parameters or the size of the filter.

[0137] This technique was notably described in the article YU, Fisher and KOLTUN, Vladlen. “Multi-scale context aggregation by dilated convolutions”. arXiv preprint arXiv: 1511.07122, 2015.

[0138] See also the article by YU, Fisher, KOLTUN, Vladlen, and FUNKHOUSER, Thomas. “Dilated residual networks”. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2017. p. 472-480.

[0139] The approach of this latter article combines residual networks with dilated convolutions.

[0140] According to a particular embodiment, such a dilated residual network can be used to extract, for a given normalized sequence Z, several representations via a set of learnable convolution filters.

[0141] As seen previously, the number of convolution filters, which determines the size of the space corresponding to the MF feature matrix, is substantially greater than the corresponding dimension of the positional encoding matrix (see [Fig. 4]). In the numerical example given above, we can therefore expect a dilated convolution block with 3d / 4 filters.

[0142] Figure [Fig. 5] schematically represents such a dilated convolution block, according to one possible embodiment.

[0143] This figure presents the dilated convolution block as a succession of steps. However, it can also be considered as a succession of layers or blocks of neurons, each of which represents a step in the data traversal from an input to an output.

[0144] In this embodiment, this expanded convolution block is composed of four stacked residual units using convolution filters. In the figure, one residual unit is shown; the other three follow one another until the final output and can be identical. The number of residual units (for example, four) is determined empirically by conducting experiments and performing performance metric measurements. This number can vary depending on the datasets and the type of device. The model can therefore differ, in structural terms, depending on the equipment from which the curve is to be extracted.

[0145] A residual unit, or "ResUnit," is a fundamental block used in residual neural networks, known as ResNets. The main idea behind residual units is to introduce short-circuit connections (or residual connections) that allow gradients to propagate directly through the layers without dissipating, which helps to mitigate the problem of gradient degradation observed in very deep networks.

[0146] Generally speaking, a typical residual unit consists of two or three convolution layers with added short-circuit connections.

[0147] In the embodiment described in relation to [Fig.5], each unit is composed of a convolution layer S51, an activation function S52 (GeLU activation), and a normalization layer S53 (BatchNorm).

[0148] Note that a residual connection is used between each ResUnit in order to maintain stable training of the model.

[0149] For the S51 convolution layer, or step, a step parameter (or "stride") set to 1 is used in order to keep the time dimension unchanged.

[0150] For each residual unit ie[l,.. .,4], a dilation parameter D=2' is used, increasing exponentially with the depth i of the residual unit.

[0151] This expansion rate determines the number of spaces between the elements of the convolution kernel. Thus, an expansion rate of D means that there are Dl zeros inserted between each element of the kernel.

[0152] The use of dilation allows the filter to cover a larger region of the input without increasing the kernel size or the number of parameters, which is useful for capturing information at different scales. A dilated convolution block thus makes it possible to extend the receptive field of convolutional neural networks without increasing computational complexity.

[0153] The S52 activation layer, or step, can be a GELU activation layer (for "Gaussian Error Linear Unit" in English, or "Unité Linéaire d'Erreur Gaussienne" in French).

[0154] The GELU function applies probabilistic activation while taking into account the normal (or Gaussian) distribution of the inputs. It combines properties of linear and nonlinear activation functions in such a way as to retain the advantages of each.

[0155] Layer, or step, S53 is a batch normalization layer. This normalization technique was proposed in the article by Loffe, S., & Szegedy, C. (2015), “Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift”. In Proceedings of the 32nd International Conference on Machine Learning (ICML-15) (pp. 448-456).

[0156] This layer allows for the standardization of activations (S52) by recentering and resizing them. This reduces the problem of internal covariant shift, which occurs when the distribution of activations in a layer changes during training. By stabilizing the activations, batch normalization allows for higher learning rates and accelerates convergence.

[0157]

[0158]

[0159]

[0160] In an S6 step, the statistical metrics are projected into an MS statistics vector. This projection can be based on a neural network trained on a training set corresponding to the individual equipment considered: as detailed previously, predictive models are trained equipment by equipment during a learning phase and are therefore specialized. In particular, according to one embodiment, the projection step S6 can be implemented using a learnable linear transformation matrix (the value 2 arising from the fact that, according to one embodiment, 2 statistical metrics are used: the mean and the standard deviation). This yields, at the output of this step, or block, a vector of size d, as illustrated in [Fig.4]. In step S7, the results of steps S4, S5 and S6 are concatenated to provide input data to a transformation neural network.

[0161] In other words, we concatenate the positional encoding matrix MEP, the feature matrix MF and the statistics vector MS.

[0162] This concatenation is performed so that the values ​​of the positional encoding matrix MEP are positioned in the same column as the corresponding values ​​of the feature matrix MF (i.e., corresponding to the same position i of the input data X). The concatenation of these two matrices therefore forms a matrix of dimensions wxd.

[0163] Furthermore, the MS statistics vector is concatenated as an additional column to this MFIMEP matrix.

[0164] The final ME matrix is ​​therefore a matrix of dimensions (w+l)xd.

[0165] This final matrix, or ME input matrix, is then provided to a transformation neural network. This transformation neural network, or "transformer," is designed to infer, in an S8 step, a temporal sequence corresponding to the individual equipment under consideration.

[0166] To do this, it may have been previously trained on a training set corresponding to that individual equipment in question.

[0167] As mentioned previously, in machine learning, a transformer is a neural network architecture introduced by Vaswani et al. in their 2017 paper entitled "Attention is AU You Need". Transformers allow for the efficient capture of long-range dependencies in data sequences through a self-attention mechanism.

[0168] According to one embodiment, the transformation neural network consists of several blocks, each made up of the same elements. For example, it may consist of 3 successive blocks.

[0169] The number of blocks (for example, 3) is determined empirically by conducting experiments and measuring performance metrics. This number may vary depending on the dataset and the type of device. The model may therefore differ, in structural terms, depending on the equipment from which the curve is to be extracted.

[0170] According to an embodiment illustrated by [Fig.6], each block comprises a normalization element S81, a self-attention element S82, a new normalization element S83 and a dense network element S84.

[0171] This [Fig. 6] illustrates the transformation neural network as a succession of elements (i.e., a set of neuronal layers), but (as with [Fig. 5]) each element can also be considered as a step in a process. This figure can therefore be viewed both as a functional architecture and as a flowchart.

[0172] The S81 element therefore represents a layer normalization element, "LayerNorm".

[0173] Layer normalization (“LayerNorm”) was introduced in the article by Ba, Jimmy Lei, Jamie Ryan Kiros, and Geoffrey E. Hinton, “Layer Normalization”, arXiv preprint arXiv: 1607.06450 (2016). Such an element notably allows for the stabilization and acceleration of neural network training.

[0174] The second element S82 is a self-attention element.

[0175] According to one embodiment, the self-attention element is a multi-head attention mechanism implementing multi-scale dynamic attention (DMSA for "Diagonally Masked Self Attention," introduced by S. Lee, S. Lee and BC Song, in "Improving Vision Transformers to Learn Small-Size Dataset From Scratch," in IEEE Access, vol. 10, pp. 123212-123224, 2022). This mechanism can be used in place of the original attention mechanism.

[0176] The difference between the original attention mechanism and the attention mechanism lies in the application of a diagonal "mask" in the attention matrix. The diagonal mask is used to restrict attention primarily to "other" elements, which is particularly useful in tasks where interdependence (i.e., an element's attention to itself) does not need to be taken into account.

[0177] This type of mechanism has been shown to improve the performance of models using Transformer layers on smaller datasets. Furthermore, its effectiveness has been demonstrated on electricity consumption time series.

[0178] Figure 7 schematically illustrates such a dynamic attention mechanism multi-scale. It outlines a possible implementation of step / element S82.

[0179] Step S821 corresponds to a dot product and scaling step. It takes as input queries (Q for "query") and keys (K for "key").

[0180] Step S822 corresponds to the application of the diagonal mask (“DiagMask”).

[0181] Step S823 corresponds to the application of a Softmax function.

[0182] Step S824 corresponds to a final dot product step between the values ​​resulting steps S821-S823 providing attention and V values ​​(for "Value" in English).

[0183] This mechanism is also described in the article "Improving Vision Transformers to Learn Small-Size Dataset From Scratch" previously presented.

[0184] The linear projections Q, K, V may correspond to the prior art in the field of transformers, in particular as described in the aforementioned article.

[0185] In an S83 step, a new normalization element is applied.

[0186] Finally, in an S84 step, a dense S84 network is applied. This dense network can be a positional fully-connected network. It can be a PFFN (Position-wise Feed-Forward Network). This type of network was described in the seminal paper on attention mechanisms such as transformers, "Attention is All Your Need."

[0187] By S8 inference, we obtain at the output of this transformation neural network a representation (a matrix) corresponding to the individual equipment considered.

[0188] The representation obtained at the output of the transformation neural network is of the same size as the input, i.e. a matrix of size (w+l)xd.

[0189] However, the data corresponding to the MS statistic vector can be removed, retaining only those corresponding to the MF feature matrix and the MEP time encoding matrix, in order to obtain measurement values ​​corresponding to the equipment under consideration with the associated time information. This results in a matrix of size wxd.

[0190] In addition, a convolution step can be implemented in order to obtain a time sequence of the same dimension as the time sequence provided as input, i.e. wxl (a one-dimensional time series).

[0191] This convolution layer can be parameterized with a kernel of size 3, a padding of 1 and a stride of 1.

[0192] Finally, according to one embodiment, a denormalization step S9 can be performed on the temporal sequence corresponding to the individual equipment under consideration, in using projected values ​​of the MS statistical vector with a learnable transformation matrix.

[0193] This learnable transformation matrix can, in particular, be a linear transformation matrix of dimension dx2, where 2 is the number of statistical metrics. This yields two projections, proj(q) for the mean and proj(o) for the standard deviation. This step allows us to estimate, during model training, the mean and variance to be applied in order to denormalize the output signal and to more accurately estimate the individual consumption curve of the device we are trying to predict.

[0194] The S9 denormalization can then use these proj(q) and proj(o) values ​​by performing the inverse transformation of that performed in step S3 in order to obtain output values ​​consistent with those recovered from the measuring device.

[0195] In the case where a Z-normalization has been performed in step S3, the inverse transformation can therefore be applied:

[0196] [i,w]? Xi-Zix.Proj(ji) +Proj(a}

[0197] We then obtain a series of X values, which forms the sequence of the disaggregated load curve for the individual equipment concerned and for a sequence provided as input.

[0198] It is sufficient to concatenate the Xi sequences for a set of sequences in which the initial time series has been cut to obtain the disaggregated load curve.

[0199] According to one embodiment, this same process can be carried out for a subset (or for all) of the equipment located in a given room L. The electrical consumption of this room can then be characterized by obtaining the load curves for each piece of equipment in this subset. For example, a dwelling can thus be characterized by determining the load curves of the most energy-consuming equipment: electric vehicle charging station, heating or air conditioning units, water heater, etc.

[0200] The proposed solution has been tested and proven on several real-world disaggregation cases for the purpose of estimating the quantity of energy and detecting activation periods.

[0201] It should be noted that several metrics exist in the literature to compare the performance of disaggregation methods; we detail them initially below.

[0202] For each measurement T represents the total number of intervals (number of time steps or number of days / weeks / months), with yt the actual consumption value (measured or obtained from a database) and y is the value predicted by the described process.

[0203]

[0204]

[0205]

[0206]

[0207]

[0208]

[0209]

[0210]

[0211]

[0212]

[0213]

[0214]

[0215]

[0216]

[0217]

[0218]

[0219] We can first define “classic” regression metrics MAE (for “Mean Average Error” in English, or mean absolute error) and RMSE (for “Root Mean Squared Error” in English, or mean squared error): RMSE = li; v |' We can also define the precision P (or "accuracy" in English) from the true positive (TP), true negative (TN), false negative (FN), and false positive (FP) rates as: r = (TP + TN) / (TP + TN + FP + FN) An Fl score can also be defined as follows: 171 _ o x----TP---- 2 1 - X 2TP+FP+FN Total Energy Correctly Assigned (TECA) quantifies the accuracy of the predicted energy consumption for a piece of equipment. It can be defined as follows: TECA = 1 - 4 The MR (Matching Ratio) is defined such as the ratio of overall accuracy between actual and estimated consumption for a piece of equipment. It can be defined as follows: MR = The normalized disaggregation error (NDE) is close to the RMSE and is defined as the normalized disaggregation error committed: ^.yt2 The Signal Aggregate Error (SAE) quantifies the variance of the energy prediction. It can be defined as: SAE = -^r.....— Figures 8A, 8B and 8C illustrate results from public databases. They compare results using different methods of extracting the electrical consumption of an individual piece of equipment within a set of equipment according to the metrics mentioned above. The different processes being compared are: - NILMFormer: the process proposed in this patent application, - BERT4NILM, BILSTM: previously presented, - The "BIGRU" network, presented in the article by Cho, Kyunghyun, et al. "Leaming Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation." Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2014, pp. 1724-1734. - FCN (for "Fully Connected Network": a model based on a succession of convolutional layers, - UNet NILM, proposed in the article by Anthony Faustine, Lucas Pereira, Hafsa Bousbiat, and Shridhar Kulkarni. 2020, “UNet -NILM: A Deep Neural Network for Multi-tasks Appliances State Detection and Power Estimation” in NILM. In Proceedings of the 5th International Workshop on Non-Intrusive Load Monitoring (NILM’20). Association for Computing Machinery, New York, NY, USA, 84-88. https: / / doi.org / 10.1145 / 3427771.3427859.

[0220] Other processes exist, but this corpus forms a representative sample allowing us to highlight certain advantages of the proposed process in terms of performance in extracting a load curve for a given piece of equipment.

[0221] Figures 8A-8C show that the performance of the proposed process, NILMFormer, surpasses that of other processes according to all metrics: - it obtains the lowest (and most favorable) values ​​for the MAE, RMSE, NDE and SAE metrics, and, - the highest (and most favorable) values ​​for the P, Fl, TECA and MR metrics

[0222] As mentioned previously, these results demonstrate the efficiency of the proposed process (“NILMEormer”) compared to other proposals, including for equipment already considered by the state of the art and on the basis of training from public learning bases (UKDALE, REFIT...).

[0223] Other experiments and tests were carried out with other equipment such as electric vehicle charging stations, water heaters and heating appliances. To do this, the different predictive models (neural networks) were trained on training sets constructed as indicated above.

[0224] Fig. 9 illustrates the extraction of the load curve for equipment such as an electric vehicle search station.

[0225] The upper part shows the evolution of the aggregate power consumed (in watts) over time (in 30-second increments) for a set of equipment. This is a load curve that can be measured by a measuring device corresponding to a room, or PDL (delivery point).

[0226] Below are represented on the one hand (in solid line) the load curve of the charging terminal as measured by a measuring device associated with this terminal (which therefore represents the "ground truth"), and in dotted line a load curve extracted by the process which has just been described.

[0227] We observe a very strong correlation between these two curves which are almost superimposed.

[0228] Similar results are obtained for other types of equipment tested, including water heaters and heating equipment.

[0229] By determining the TECA performance for tests based on the three types of equipment, the performance obtained also appears very convincing, with an average of 85% of the monthly energy quantity correctly estimated (up to 90% for domestic hot water) and more than 70% of the energy quantity correctly estimated for each time step. These results confirm the viability of using the proposed method in a real-world context for estimating the energy quantity for at least these three different devices.

[0230] Extracting the electrical consumption of an individual piece of equipment within a set of equipment allows for various practical uses. As mentioned previously, consumption curves can be extracted for several pieces of equipment, thus enabling the characterization of the electrical consumption of a room.

[0231] This extraction (or characterization) for a local L is important for various reasons, and can be exploited in various ways.

[0232] From a general point of view, this extraction has great interest, both from a purely scientific point of view and from a technological point of view.

[0233] In particular, it helps to manage energy consumption from the perspective of the consumer, the energy supplier, and the environment as a whole. This last point addresses current and future requirements regarding the environmental impact of users' energy consumption (both residential and industrial).

[0234] Furthermore, in order to design smart buildings, more efficient distribution systems, etc., it is important for the scientific community to have more precise and numerous metrics. Such a mechanism for extracting electricity consumption data from a household or business can contribute to building a knowledge base to facilitate research and development in these areas.

[0235] The extraction and / or characterization of electricity consumption can be used for the optimization of energy production by the electricity energy supplier.

[0236] The characterizations of a large number of end consumers can be transmitted, in anonymized form, to a concentrator which can compile consumption statistics. This concentrator can be an application module on a service platform accessible to the energy supplier.

[0237] As an example, it is possible to establish an average consumption, or activation, curve per type of equipment.

[0238] Based on these statistics, the energy supplier can adapt its production network. In particular, it can build a reliable prediction of customer usage and thus anticipate increases in load at its production points or, conversely, decreases in load, thereby allowing it to reduce its exposure to speculative risks in the inter-producer electricity market.

[0239] For example, it can characterize a peak load related to electric vehicle charging stations at the end of the day, and search for mechanisms to respond to it.

[0240] The characterization process can therefore enable energy suppliers to optimize energy production, in particular to avoid waste through overproduction, thanks to the transmission of characterization data to a service platform accessible to the energy supplier.

[0241] Also, an optimization module can be provided in the GTW gateway or in the S service platform, in order to allow the optimization of the equipment fleet El, E2, E3, ..., En located in a given room L according to the consumption data obtained from the predictive models.

[0242] Preferably, the optimization module is embedded (as a software or application module) within the GTW gateway. This allows optimization to be performed based on locally available information, thereby minimizing data transmission to the Internet, which is beneficial for managing personal data and conserving bandwidth and network resources. Furthermore, optimization can continue even if the link to the telecommunications network and the S service platform is interrupted.

[0243] The optimization module can determine control orders for one or more pieces of equipment among those connected to the network in order to meet these internal or external constraints of the specified room L, or of the building to which the specified room L belongs.

[0244] In particular, a possible constraint may be to balance consumption, or at least to constrain fluctuations in consumption beyond a given threshold.

[0245] Thus, in the event of a consumption peak (detectable from the consumption curve provided by the measuring device C), the characterization of the electrical consumption across the equipment fleet can make it possible to determine a The likely cause of this spike (either consumption data or, failing that, activation data) can then be determined. Control commands can then be established to smooth out electricity consumption.

[0246] For example, some equipment can be momentarily switched off or forced into a less power-consuming state, for a duration corresponding to an activity of the equipment that is the source of the peak consumption.

[0247] By way of illustrative example, a kitchen heating system can be lowered when an oven is switched on: its operating time is limited so that lowering the heating is not very detrimental to the inhabitants.

[0248] Furthermore, characterizing consumption makes it possible to identify the share of energy consumption per piece of equipment and thus provide an energy cost for each. Specific actions for reducing energy consumption per piece of equipment can then be proposed to the user, and a precise return on investment time can be provided, since it is possible to know the energy and financial share associated with each piece of equipment: it is therefore possible to indicate a potential saving on their energy bill by lowering the heating by one degree.

[0249] Similarly, the characterization of electrical equipment over a given time window can be one of the inputs to a load control model based on energy predictions that take this characterization as input. For example, this control model could send commands to switch electrical devices on or off based on past characterization.

[0250] Furthermore, the optimization module can provide the user with a representative overview of consumption and activation data for all equipment. This data can be aggregated and consolidated to provide a comprehensive view of their electricity usage.

[0251] The optimization module can detect optimization opportunities and, possibly after validation by a user from local L, transmit suitable command orders.

[0252] For example, it can detect standby equipment and electricity consumers: plugged-in household appliances, for instance. It can also identify heating systems used at night. In some cases, it may be beneficial to limit these uses, and the MO optimization module can suggest control commands aimed at reducing this consumption, which should not impact the comfort of the occupants (or at least present an energy savings / comfort loss ratio greater than one).

[0253] Obviously, other optimization scenarios can be implemented by the optimization module, taking advantage of the detailed knowledge of the electrical uses of the premises obtained by the characterization process described.

[0254] Various other use cases are also described in the scientific or commercial literature related to non-intrusive load characterization (NILM).

[0255] It is clear that the characterization obtained by the GTW gateway allows the optimization of the electricity consumption of consumers and the optimization of the electricity production of electricity suppliers.

[0256] Of course, the present invention is not limited to the examples and embodiment described and illustrated, but is defined by the claims. In particular, it is susceptible of numerous variations accessible to those skilled in the art.

Claims

Demands

1. A method for extracting the electrical consumption of an individual piece of equipment within a set of equipment (Eb E2, E3, ..., En) located in a specific room (L), said method comprising transforming a stream of measured values ​​of the overall electrical consumption of said set, provided by a measuring device (C) associated with said specific room (L), into a time series, and dividing said time series into a set of sequences, then, for each sequence, the steps of: - determining (S2) statistical metrics, allowing the normalization (S3) of said sequence; - generating (S4) a positional encoding matrix (PEM) from time values ​​associated with the values ​​of said sequence; - extracting (S5) features from said normalized sequence to form a feature matrix (FM);- projection (S6) of said statistical metrics into a statistics vector (MS); - concatenation (S7) of said positional encoding matrix (MEP), of said feature matrix (MF) and of said statistics vector (MS) to form an input matrix (ME) supplied to a transformation neural network intended to infer (S8) a temporal sequence corresponding to said individual equipment.

2. A method according to the preceding claim, wherein said positional encoding matrix and said feature matrix have a width equal to the number (w) of values ​​in said sequence and the sum (d) of their height is equal to the height of said statistic vector (MS), so that said input matrix (ME) has dimension (w+l)xd.

3. A method according to any one of the preceding claims, wherein at least one step among the extraction (S5), projection (S6) and inference (S8) steps is based on a neural network trained on a training set corresponding to said individual equipment.

4. A method according to any one of the preceding claims, wherein said generation step (S4) of a positional encoding matrix (PEM) comprises a first projection of said time values ​​into a sinusoidal basis, and a second projection into a space one of whose dimensions is substantially smaller (d / 4) than the corresponding dimension (3d / 4) of said feature matrix (FM).

5. A method according to any one of the preceding claims, wherein said extraction step (S5) is implemented by a dilated convolution block.

6. A method according to any one of the preceding claims, further comprising a denormalization step (S9) of said time sequence corresponding to said individual equipment using projected values ​​of said statistical vector (MS) using a learnable transformation matrix.

7. A method for characterizing electrical consumption includes an implementation of the extraction method according to one of the preceding claims for at least a subset of said equipment set.

8. Equipment (S), comprising at least one processor adapted to implement a method according to one of the preceding claims.

9. A computer program capable of being implemented on a server (S), the program comprising code instructions which, when executed by a processor, carries out the steps of the process defined in claims 1 to 6.

10. Data carrier on which at least one series of program code instructions for the execution of a method according to any one of claims 1 to 6 has been stored.

Images