METHOD FOR DETECTING AN EVENT ASSOCIATED WITH A REFRIGERATED ROOM

Abstract

METHOD FOR DETECTING AN EVENT ASSOCIATED WITH A REFRIGERATED ROOM Method for detecting an event associated with a cold room comprising: First reception (REC1) of first input data; Second reception (REC2) of second input data; Selection (SLC) of a first predictive model of electrical consumption of the refrigeration unit, said model being trained from a first machine learning; Implementation (MO1) of the first predictive model of electrical consumption to determine at least one predicted electrical consumption data of the refrigeration unit characterizing an expected operating state of the refrigerated system;Detection (DTC1) of at least one event associated with the refrigeration unit by means of a characterization function automatically identifying and characterizing at least one deviation between the expected operating state of the system and an actual operating state of the system, by means of the predicted value of electrical consumption and the operating data. Figure for the abstract: Fig.2;

Description

Title of the invention: METHOD FOR DETECTING AN EVENT ASSOCIATED WITH A REFRIGERATED ROOM Field of invention

[0001] The invention relates to the field of methods for detecting events taken from classes of events.

[0002] More particularly, the invention relates to the field of methods for detecting an event from a function identifying a difference between a predicted value and an actual value.

[0003] More particularly, the invention relates to the field of methods for detecting an event associated with a refrigerated room, also commonly called a cold room, cold chamber, or even a refrigerated warehouse. State of the art

[0004] Refrigerated rooms, warehouses or rooms are known in the prior art, which contain perishable goods (meat, vegetables, fruit, medicines, to cite a few examples), and which integrate one or more refrigeration systems (also commonly called a cold unit) so as to ensure the preservation of these goods.

[0005] During storage, maintaining a set temperature, a humidity level and sometimes a precise CO2 level in rooms, chambers or warehouses is essential to avoid degradation of the stored contents. In the case of storing medicines or vaccines, a variation of a few degrees in the packaging temperature can lead to a total loss of batches.

[0006] To ensure that these storage condition parameters (set temperature, humidity level and CO2 level) are maintained, the refrigeration units must be sized to take into account several factors, including the lowering of the air temperature during loading (and sometimes the temperature of the cargo), heat exchanges with the outside by radiation and conduction through the walls of the refrigerated rooms, the heat released by the cargo to be cooled (such as the respiration heat of the fruit), the energy required to treat the outside air supplied in the event of air renewal for the purpose of cooling and / or adjusting the humidity and CO2 levels to compensate for the production of gases from the fruit (ethylene, CO2), etc.

[0007] This dimensioning in air conditioning (cooling, adaptation of the humidity level, etc.) results in a need for energy necessary to operate the refrigeration systems (compressor, evaporator, fan, but also management and communication electronics). The energy used to power these refrigeration systems is electricity, itself coming from the electrical network serving the site, from thermal generators (running on fuel, diesel, gasoline, LNG, etc.), from fuel cells (hydrogen, methanol, bio methanol, etc.), from on-board turbines (running on kerosene) or from electrical storage batteries. Also, whatever the primary energy used to operate the refrigeration systems, the measurement of the energy consumption of the refrigeration system will be made in electrical energy (kW for the total power drawn by the systems, and kWh for the energy consumed by these systems over a period of time).

[0008] Thus, the refrigerated storage of foodstuffs results in significant energy consumption, due to the power required to operate the refrigeration systems.

[0009] However, the energy consumption of a cold room, linked to the operation of the refrigeration systems, is likely to vary significantly when particular events occur. To cite just a few examples, these events may include a variation in the interior temperature, a poorly closed door, a leak in a refrigeration circuit, fouling of the evaporator, a fault in an element of a refrigeration unit such as a compressor, a fault in the electrical supply, a sealing fault, are all events which modify the efficiency of the refrigeration systems and which result in increased energy consumption to produce a determined quantity of cold. Thus, these events result in higher energy consumption of the refrigerated system to maintain its set temperature.

[0010] Therefore, there is a technical need, both preventive and curative, to deal with these events which lead to excesses in energy consumption.

[0011] The invention aims to propose a solution to this technical need, so as to, at a minimum, limit the aforementioned drawbacks. Summary of the invention

[0012] According to a first aspect, the invention relates to a method for detecting an event associated with a cold room comprising at least one cold unit, the method comprising the following steps: • First reception, by a first computer, of first input data comprising at least one temperature data item relating to an interior temperature of the cold room; • Second reception, by said first computer or by another computer, of second input data comprising at least one operating data relating to the operation of at least one piece of equipment in the room cold; • Selection of a first predictive model of electrical consumption of the refrigeration unit, said model being trained from a first machine learning; • Implementation of the first predictive electricity consumption model to determine at least one predicted electricity consumption data of at least one refrigeration unit, called predicted consumption data, and characterizing an expected operating state of the refrigeration unit; • Detection of at least one event associated with the cold room among a set of event classes, by means of a characterization function automatically identifying and characterizing at least one deviation between the expected operating state of the cold unit and an actual operating state of the cold unit, by means of the predicted consumption data and the operating data.

[0013] According to one embodiment, the operating data comprises actual electrical consumption data of at least one refrigeration unit, called actual consumption data, and the characterization function identifies and characterizes at least one difference between a value associated with said actual consumption data and a value associated with the predicted consumption data. The values associated with the actual and predicted consumption data are, for example, values of consumed electrical power.

[0014] One advantage is to identify and characterize the event based on a deviation between an expected electrical consumption value and an actual electrical consumption value.

[0015] According to one embodiment, the operating data comprises at least one piece of data relating to an operating duration and / or an operating regime of at least one piece of equipment in the cold room. This is, for example, the operating regime of a compression system in the cold room.

[0016] An advantage is to identify and characterize the event from a real parameter making it possible to deduce a consumption of the cold room. Typically, an operating regime or an operating duration of a compression system makes it possible to deduce an energy consumption of the cold room.

[0017] According to one embodiment, the characterization function comprises an inference engine implemented from a knowledge base and from at least one predefined rule.

[0018] One advantage is to automatically identify and characterize the event from a set of predefined rules and a data set. The inference engine is for example implemented by an expert system.

[0019] According to one embodiment, the step of selecting the first predictive consumption model is implemented automatically by means of a first function whose parameters depend on data taken from among the first input data and / or data taken from among the second input data.

[0020] One advantage is to allow optimized selection of the model based on the available input data. For example, in the case where little data is available, a generic model can be chosen, while a more precise model can be chosen when more data relating to the cold room is available.

[0021] According to one embodiment, the first function is a learning function trained using a second machine learning.

[0022] One advantage is to allow automatic selection of the predictive model based on the available data from a function trained on a large amount of data, which therefore allows selection of the most optimized model based on the specific case.

[0023] According to one embodiment, the temperature data comprises: • An interior temperature and / or; • A set temperature and / or; • A differential between an indoor temperature and an outdoor temperature and / or; • A differential between a set temperature and an interior temperature.

[0024] According to one embodiment, the first input data comprises at least one external environment data measured by at least one sensor positioned on the cold room or on a site in which the cold room is located, at least one external environment data being estimated from a position data received from a location device.

[0025] One advantage is to refine the choice of model according to parameters external to the cold room.

[0026] According to one embodiment, at least one external environment data item comprises at least one data item taken from: • An outside ambient temperature; • An exterior temperature of the walls; • External humidity;

[0027] According to one embodiment, at least one external environment data item is estimated from a position data item of the room, deduced from its address, or received from a location device.

[0028] One advantage is to obtain data allowing the choice of model to be refined. predictive without having to directly measure this data.

[0029] According to one embodiment, the first input data comprises at least one meteorological data item received by means of a communication interface and from at least one remote server.

[0030] One advantage is to refine the choice of the predictive consumption model from data from a remote database.

[0031] According to one embodiment, the meteorological data comprises at least one data item taken from: • Outside temperature data; • Humidity data;

[0032] According to one embodiment, the second input data comprise at least one first characteristic data item of the refrigeration unit, called first characteristic data item, and relating to an operating mode of the refrigeration room characterized by a limited interval of interior setpoint temperatures.

[0033] One advantage is being able to refine the choice of the predictive consumption model.

[0034] According to one embodiment, the second input data comprises at least one second characteristic data item of the refrigeration unit, called second characteristic data item, and relating to a model and / or a brand of the refrigeration unit characterizing technical characteristics specific to the refrigeration unit.

[0035] One advantage is being able to refine the choice of the predictive consumption model.

[0036] According to one embodiment, at least one second characteristic data item comprises data produced by a second function trained by means of a third machine learning, called second function, taking as input the real consumption data item.

[0037] One advantage is that it allows the choice of the predictive model to be refined by finding a model or brand of the refrigeration unit(s) based on their energy consumption profiles.

[0038] According to one embodiment, the first machine learning receives as input data comprising histories of the same input data relating to different cold group models or different histories of input data relating to the same cold group model.

[0039] According to one embodiment, the first machine learning training the first predictive model of electricity consumption receives as input data comprising histories of the same input data relating to different cold group models or different histories of input data relating to the same cold group model.

[0040] One advantage is to train the predictive model from varied data to optimize its performance in choosing the most suitable model.

[0041] According to one embodiment, the second machine learning training the first function receives as input data taken from at least one history of first input data and second input data.

[0042] According to one embodiment, the third machine learning driving the second function receives as input data characteristic of an electrical consumption of a plurality of cold groups of a plurality of cold rooms each having specific technical characteristics.

[0043] According to one embodiment, the event comprises at least one event taken from: • A predicted or estimated deviation from an indoor temperature; • A fault in an element of a cold group; • A door opening; • A power supply fault; • A leak of refrigerant from a refrigeration circuit - or any other technical operating problem of the refrigeration unit; • A lack of sealing in the cold room.

[0044] According to one embodiment, the event is detected when at least one deviation between the expected operating state and an actual operating state is greater than a predefined threshold value or variable according to predefined parameters.

[0045] One advantage is being able to quantify a significant deviation from the occurrence of an event.

[0046] According to one embodiment, the threshold value is calculated from an arithmetic mean carried out on values taken from among the first input data and / or taken from among the second input data and from a standard deviation associated with said values.

[0047] According to another aspect, the invention relates to a method for detecting an event associated with a cold room, comprising the implementation of a predictive model to determine at least one predicted parameter characterizing its operating state. This is for example a predictive model of an operating mode, an operating regime, an operating time, or any other parameter associated with the cold room or with equipment of the cold unit (such as its refrigeration system). Thus, the invention generally relates to a method comprising the implementation of a predictive model to predict a parameter likely to have a link on the energy consumption of the cold room.

[0048] According to another aspect, the invention relates to a system for detecting an event associated with a cold room configured to implement the steps of the method.

[0049] According to one embodiment, the system comprises: • At least one calculator for: • Receive the first input data including at least one temperature data item linked to an interior temperature of the cold room; • Receive the second input data comprising at least one operating data item relating to the operation of at least one piece of equipment in the cold room; • Select a first predictive model of electricity consumption of the refrigeration unit from a set of predictive models of electricity consumption; • Implement the first predictive electricity consumption model to determine at least one predicted electricity consumption data, called predicted consumption data, and characterizing an expected operating state of the refrigeration unit; • Detect at least one event associated with the cold room from a set of event classes, by means of a characterization function automatically identifying and characterizing at least one deviation between the expected operating state of the cold unit and an actual operating state of the cold unit, and at least by means of the predicted electricity consumption value and the operating data.

[0050] According to one embodiment, the system comprises: • The cold room including at least one cold unit; • A terminal connected to the cold room;

[0051] According to another aspect, the invention relates to a system for classifying an event associated with a cold room, the system comprising: • a plurality of cold groups each generating at least one temperature datum relating to their set temperature and at least one operating datum relating to the operation of at least one piece of equipment in the cold room; • an administration console for the operation of the plurality of cold groups; • at least one calculator configured for: • implement a first predictive electricity consumption model to determine at least one predicted electricity consumption data of the refrigeration unit characterizing an expected operating state of the cold room; • implement at least one classification function of an event characterized by at least one deviation between a state of func expected operation of the cold room and an actual operating state of the cold room, by means of the predicted value of electricity consumption and the operating data.

[0052] According to another aspect, the invention relates to a computer program product comprising instructions which, when the program is executed by a computer, cause the latter to implement the steps of the method of the invention. Brief description of the figures

[0053] Other characteristics and advantages of the invention will emerge on reading the detailed description which follows, with reference to the appended figures which illustrate:

[0054] [Fig.l]: a schematic representation of a cold room comprising a cold unit

[0055] [Fig.2]: a flowchart of several steps in the process of detecting an event associated with a cold room.

[0056] [Fig.3]: a flowchart of several steps in the event detection process

[0057] [Fig.4]: A graphical representation of predicted consumption values electrical consumption of the refrigeration unit and actual electrical consumption values in the form of a curve and as a function of time.

[0058] [Fig.5]: a flowchart of a step of a method for detecting an event, in an embodiment in which the method comprises a step of implementing the first machine learning function from two data characteristics of the cold group.

[0059] [Fig.6]: a flowchart of a step of a method for detecting an event, in an embodiment in which the method comprises a step of implementing the second function.

[0060] [Fig.7]: a graphical representation of the training of the predictive consumption model, the first function and the second function using the first, second and third machine learning. Description of the invention General information

[0061] According to a first aspect, the invention relates to a method for detecting an Evm event associated with a cold room SFD.

[0062] The term "cold room" is used in this description for reasons of simplification. The term "cold room" means any type of storage location, cold room, warehouse, or room with a controlled atmosphere equipped with one or more cold units or refrigeration units.

[0063] For example, it may be a refrigerated room, also called a cold room, or a refrigerated cabinet.

[0064] In the reference, such a storage location will be referred to indifferently by the terms “cold room”, “refrigerated chamber” or “refrigerated room” to refer to any type of storage location with a controlled atmosphere equipped with one or more refrigeration units.

[0065] [Fig.l] illustrates an example of an SFD cold room comprising a GF cold unit.

[0066] The method comprises several steps making it possible to detect DTCi an event E vm associated with the cold room SFD.

[0067] An "Evm event" is a cause of a deviation between an expected operating state of the SFD cold room and an actual operating state of the SFD cold room. Such a deviation corresponds, for example, to a deviation between a predicted energy consumption value and an actual energy consumption value over time, as illustrated in [Fig. 3]. According to another example, it is a deviation between an expected operating time and / or operating regime of the SFD cold room and an actual operating time and / or operating regime of the SFD cold room, or of equipment thereof, for example its compression system.

[0068] The cold room may comprise a cold unit GF, or refrigeration unit. The expected operation of the cold unit GF is characterized by a predicted electrical consumption data item of the cold room SFD. The actual operation is characterized by an operating data item Dtfct of the cold room SFD, which comprises for example an actual electrical consumption data item of the cold unit GF, or even a data item relating to an operating duration and / or an operating regime of equipment in the cold room.

[0069] Examples of Evm events associated with the cold room include a poorly closed door, a loss of sealing of the walls, a fault in the electrical supply of the GF cold unit, a fault in the refrigeration system such as a refrigerant leak, fouling of an element or a fault in an element of a GF cold unit, for example a fault in a compressor or a variation in an interior temperature, predicted or estimated from a set of data such as historical data. All these events cause a drift in energy consumption of the SFD cold room.

[0070] In the present description, the invention is described through different embodiments illustrated by examples.

[0071] The features described in one embodiment may be directly applicable to another embodiment.

[0072] The features described in an example may be independent of one another. Thus, if an embodiment is illustrated through an example comprising a plurality of steps, the implementation of this embodiment does not not limited to the joint implementation of all the steps. For example, if an embodiment describes the training of a learning function through an example comprising an input data preprocessing step and an input data labeling step, the invention is not limited to the implementation of these two steps for this embodiment, and the training could not include an input data preprocessing step. In addition, formulations such as “according to an example” followed by “according to another example” for the same embodiment do not limit the invention to one or the other of these examples when the latter are not incompatible. First Dtinl input data: general information

[0073] With reference to figures 2 and 3, the method comprises a step of first reception RECi of first input data Dtini.

[0074] The first input data Dtini are received by a first computer CLCi. They are for example received by the first computer CLCi from a first memory MEM h which can be a local memory or a remote memory.

[0075] In the case of a remote memory, the first computer CLCi comprises, for example, means for acquiring said first input data Dtini.

[0076] According to another case, the first input data Dtini are acquired in a prior step, and recorded in a local memory to then be transmitted to the first computer CLCi which receives them.

[0077] The first input data Dtini may be transmitted in real time to a remote server by means of a telecommunications device, such as a radio or satellite device.

[0078] According to another example, the first data Dtini are extracted a posteriori from a local memory for example for a posteriori processing of this data.

[0079] The first input data Dtini comprises for example at least one piece of data which is either: - measured locally using one or more sensors; - calculated locally using measured data and / or received data, for example by means of the first CLCi calculator; - calculated remotely and received locally, for example by means of a communication interface;

[0080] According to one embodiment, the first input data Dtini comprises a mixture of measured and / or calculated data, locally and / or remotely.

[0081] According to one embodiment, the first input data Dtini comprises a mixture of measured and / or calculated data, locally and / or remotely, and recorded locally in a local memory and / or remotely on a remote server.

[0082] According to one embodiment, the first input data Dtini comprises data measured in real time.

[0083] According to one embodiment, the first input data Dtini comprises calculated and / or estimated data corresponding to a physical parameter associated with a current position of a site hosting the cold room.

[0084] According to one embodiment, the first input data Dtini comprises predicted data. This is, for example, data corresponding to predicted values associated with physical parameters. For example, it may be predicted temperature data based on the meteorological data of the site. These predicted values are, for example, estimated from other estimated values, or from measured physical parameters, or from a combination of predicted values and measured values.

[0085] First input data Dtini: temperature data Dtempi

[0086] The first input data Dtini comprises at least one temperature data item linked to an interior temperature Dtempi called temperature data item Dtempi

[0087] The temperature data Dtempi comprises for example at least one temperature value, measured, calculated or estimated. This is for example a value expressed in degrees Celsius (°C) or in Kelvin (K), or any other appropriate unit of measurement.

[0088] The temperature data Dtempi comprises, for example, at least one temperature value measured by a sensor. This is, for example, a temperature value measured inside the cold room SFD by means of a temperature probe.

[0089] According to one embodiment, the temperature data Dtempi comprises at least one temperature value calculated, estimated or deduced from other parameters.

[0090] These parameters include, for example, other values of measured, calculated or estimated physical parameters, such as other temperature values (for example, an outside ambient temperature value).

[0091] According to one example, the temperature data Dtempi comprises a calculated value obtained from a differential between an interior temperature value of the cold room SFD and at least one other temperature value, such as an exterior temperature value of the cold room SFD. Such a value is for example calculated by means of the first calculator CLCi. It can also be calculated by means of another calculator, locally or remotely. If the value is calculated from a remote entity, it is for example transmitted by this remote entity and received by the first calculator CLCi by means of a communication interface.

[0092] According to various examples, the temperature data comprises at least one data item taken from the following data and / or calculated and / or estimated from the following data: - an interior set temperature for the SFD cold room; - an actual interior temperature of the SFD cold room; - an outside temperature in the SFD cold room, such as a temperature ambient;

[0093] First input data Dtini: external environment data Dtext

[0094] According to one embodiment, the first input data Dtini comprises at least one external environment data Dtext.

[0095] The external environment data Dtext characterizes an external physical parameter.

[0096] It includes, for example, data measured by means of one or more sensors. These are, for example, sensors located on the building housing the SFD cold room or near it.

[0097] According to another example, the external environmental data Dtext comprises data calculated or estimated from one or more physical parameters, external or not to the cold room SFD, themselves measured, calculated or estimated.

[0098] According to one embodiment, the external environmental data Dtext comprises at least one piece of data received from a remote device. This is, for example, a device connected to a server or a database comprising data relating to physical parameters surrounding the site, such as a meteorological database.

[0099] According to one embodiment, the external environmental data Dtext comprises at least one data item calculated from at least one data item from a remote server or a remote database, such as a weather database.

[0100] According to various examples, the external environment data Dtext comprises at least one data item taken from: - temperature data, for example an external temperature value of a cold room SFD such as an ambient air temperature value, a temperature value of a wall of the cold room SFD; - humidity data, for example a relative humidity level;

[0101] First input data Dtini: position data Dtpos

[0102] According to one embodiment, the first input data Dtini comprises a position data Dtpos of the site.

[0103] The position data Dtpos includes, for example, an address or the GPS coordinates of the site.

[0104] The position data Dtpos is for example received by the first calculator CLCi. It can also be received by another computer.

[0105] It is for example recorded in a memory, for example the first memory MEMi, or in a different memory.

[0106] For example, ambient temperature data can be estimated based on the position of the site at a given time, based on a forecast weather database.

[0107] According to one embodiment, at least one of the first Dtini data received relates to a DtpOS position data item of a site in which the SFD cold room is located. By "relating to the DtpOS position data item" is meant a data item comprising a measured, calculated or predicted value for the position. For example, if the first Dtini input data includes a plurality of ambient temperature values and the DtpOS position data item includes a set of geographic coordinates over a given period, then the first Dtini input data includes, for example, ambient temperature values measured at said coordinates and over the given period. It may also be an average temperature over said period.

[0108] According to one embodiment, at least one first input data Dtini is obtained by means of observations from meteorological and / or satellite stations.

[0109] First input data Dtini: meteorological data Dtmt

[0110] According to one embodiment, the first input data Dtini comprises at least one meteorological data Dtmt.

[0111] The meteorological data comes, for example, from a database. This is, for example, a database of a meteorological service capable of providing meteorological data from position data associated with temporal data (for example, a time-stamped geographical position).

[0112] According to several examples, the meteorological data Dtmt may comprise: - temperature data, for example an ambient air temperature or a dew point; - humidity data, for example a relative humidity value;

[0113] In one embodiment, the method comprises implementing a function for determining at least one external environment data item Dtext and / or at least one meteorological data item Dtmt, or for determining a variation of at least one of these data items from at least one other external environment data item Dtextet / or from at least one meteorological data item Dtmt, and / or from a variation of one of these data items.

[0114] More generally, the first input data Dtini are likely to include any type of data likely to have an impact on the interior temperature of the cold room SFD.

[0115] According to one embodiment, the first input data Dtini comprise data estimated, calculated or predicted by means of a mathematical model taking as input at least one position data Dtpos of the site received, measured, estimated, calculated and / or predicted.

[0116] According to one embodiment, the first input data Dtini comprises position data DtpOS of the site estimated, calculated or predicted by means of a mathematical model taking as input meteorological data Dtmt received, measured, calculated, estimated and / or predicted. Second input data Dtin2: general information

[0117] The method comprises a second step of receiving REC2 second input data Dtin2.

[0118] The second input data Dtin2 are received by a computer, which may be the first computer CLCi or another computer.

[0119] The second input data Dtin2 are for example received by the computer from the first memory MEMi or from another memory, which may be a local memory or a remote memory.

[0120] When the data are received from a remote memory, the computer comprises, for example, means for acquiring said second input data Dtin2.

[0121] According to another case, the second input data Dtin2 are acquired in a prior step and recorded in a local memory, to then be transmitted to the computer which receives them.

[0122] Second input data Dtin2; operating data Dtfct

[0123] The second input data Dtin2 comprises at least one operating data Dfct relating to an operation of at least one piece of equipment in the cold room SFD.

[0124] They comprise, for example, at least one piece of actual energy consumption data for the cold room SFD or for equipment in the cold group GF, for example a compression system for the cold room SFD. This is, for example, actual electrical consumption data Cre for the cold room SFD. The operating data Dtfct comprises, for example, at least one piece of data relating to energy consumption over time, such as power data, for example total power drawn expressed in kW, or current or voltage data expressed respectively in Amperes and Volts, for example drawn current data, or energy consumption data expressed in kWh or Joules.

[0125] According to another example, the operating data Dtfct comprises at least one piece of data measured by a piece of equipment in the cold room SFD, or even one piece of data measured or calculated relating to a piece of equipment in the cold room SFD. This is, for example, data making it possible to determine an operating time and / or an operating regime of one or more pieces of equipment in the cold room SFD, for example example data relating to one or more compression / expansion cycles of a refrigeration system, data relating to air ventilation inside the SFD cold room, or data characteristic of events occurring during operation or of equipment thereof, such as data recorded in a memory in the form of a "log". Such data characterizes, for example, a history of events occurring over time during the operation of at least one piece of equipment in the SFD cold room. This data makes it possible, for example, to deduce energy consumption over time of the SFD cold room.

[0126] According to one embodiment, the second input data Dtin2 comprises air temperature data from a refrigeration system of the cold room SFD. This is for example a setpoint temperature, a supply air temperature, also referred to as Supply Air Temperature in the English literature, or an extract air temperature, also referred to as Return Air Temperature in the English literature.

[0127] In one embodiment, the second input data Dtin2 comprise at least one piece of data characteristic of an operating mode of the cold room Dc, called first characteristic data Dci. The first characteristic data Dci comprises, for example, data characteristic of an operating temperature range SFD, for example, characteristic data relating to a limited range of setpoint temperatures that can be applied to the cold room SFD.

[0128] In one embodiment, the second input data Dtin2 comprises at least one characteristic data item of a model and / or a brand of the GF refrigeration unit, called second characteristic data item Dc2. This is for example data relating to a brand of GF refrigeration unit with which specific technical characteristics are associated (energy consumption, operating temperature range, etc.)

[0129] According to one embodiment, the second input data Dtin2 comprises data from a cold room management system SFD.

[0130] This is for example the first characteristic data Dc[, of techniques of the cold group GF, or even data relating to refrigeration characteristics of the cold room SFD, histories of events occurring during the operation of equipment of a cold room, such as its refrigeration system.

[0131] Link between the first and second input data

[0132] According to one embodiment, the second input data Dtin2 comprises the first input data Dtini.

[0133] According to one embodiment, the first input data Dtini comprises the second input data Dtin2. Selecting the model from a set of models

[0134] The method comprises a step of selecting SLCi a first predictive model of MODP electricity consumption of the SFD cold room, called first predictive model MODP.

[0135] The first predictive MODP model is selected from a set of predictive models of electrical consumption.

[0136] The choice of the first MODP predictive model from the set of models depends for example on the first input data Dtini and / or the second input data Dtin2. Thus, the first MODP predictive model is for example selected as a function of data taken from the first input data Dtini and / or taken from the second input data Dfn2.

[0137] For example, when the second input data Dtin2 does not include the second characteristic data Dc2, a generic predictive model suitable for any brand of GF refrigeration unit can be chosen.

[0138] According to another example, when the second input data Dtin2 includes the second characteristic data Dc2jon will be able to choose the first predictive model from a smaller subset of predictive models.

[0139] In one embodiment, the first MODP predictive model is manually chosen by a user from the set of models.

[0140] Training the MODP consumption predictive model: machine learning

[0141] With reference to [Fig.7], the first predictive model MODP is trained TRi from a first machine learning model AMb

[0142] The first machine learning model AMi comprises, for example, at least one model taken from various models such as a neural network, for example a recurrent neural network RNN, designating "recurrent neural network" in English terminology, a CNN type neural network, designating "convolutional neural network" in English terminology, or a random forest, also called a decision tree forest, or a linear regression, or an ARIMA (AutoRegressive Integrated Moving Average) model.

[0143] In one embodiment, the first machine learning model AMi receives as input at least one piece of data taken from the first input data Dtiniet / or taken from the second input data Dtin2.

[0144] This is for example data taken from the first temperature data Dtempi, the position data Dtpos, the meteorological data Dtmt, the external environment data Dtext and / or the operating data Dtfct. This is for example historical data.

[0145] The training data are preferably vectorized and normalized. Each input vector thus defined comprises, for example, a set of data training.

[0146] According to one embodiment, the implementation of the first machine learning model AMi comprises a labeling of each input vector. The labels define for example characteristic states of the cold group GF, data relating to an operation carried out on the cold room or any other context data, for example a loading state of a payload inside the cold room SFD.

[0147] According to one embodiment, output data of the first machine learning model AMi comprises classified data. The output data of the first machine learning model AMi comprises, for example, labeled data.

[0148] According to one embodiment, the implementation of the first machine learning model AMi includes a step aimed at correcting data classified by the first machine learning AMp. This step includes, for example, a modification of a label, an addition of a label, an enrichment of a label, etc.

[0149] According to one embodiment, the method comprises a step of preprocessing the input data of the first AMp machine learning model. The input data is for example collected, cleaned and preprocessed.

[0150] According to one embodiment, the method comprises a step of defining the learning model for the first machine learning model AM[.

[0151] This is for example a supervised model such as a classification model or a regression model.

[0152] According to another example, it is an unsupervised model taking unlabeled data as input, such as a grouping model, often called a “clustering” model.

[0153] According to one embodiment, data taken from the input data of the first machine learning model AMi are divided into several subsets. These are, for example, a training data set and a validation data set. The model is then, for example, trained on the training data set.

[0154] According to one embodiment, parameters of the first machine learning model AMi are adjusted to reduce an error function (or cost function). Such a step is for example implemented by means of an optimization algorithm such as gradient descent, mean square error or binary cross entropy.

[0155] According to one embodiment, the first machine learning model AM[ is evaluated using data taken from the validation data.

[0156] An advantage is to estimate the capacity of the predictive consumption model MODP electric to generalize its predictions to new input data.

[0157] According to one embodiment, the method comprises several training and evaluation cycles to optimize the predictive MODP electricity consumption model.

[0158] According to one embodiment, the training of the MODP electricity consumption model comprises a parameter adjustment step. These are, for example, adjustable parameters defined before the start of training. The parameters comprise, for example, hyperparameters. Examples of hyperparameters comprise a learning rate, a number of epochs (or iterations), a size of the training data batch, or an activation function. This advantageously makes it possible to optimize the performance of the model.

[0159] According to another example, a neural network of the RNN type, designating “recurrent neural network” in Anglo-Saxon terminology, is implemented.

[0160] The output values of the predictive electricity consumption model MODP include, for example, continuous values, for example a continuous value of electricity consumption in kilowatt-hours, or a discrete value of electricity consumption, for example a rounded value.

[0161] According to another example, the output values of the predictive electricity consumption model MODP comprise output vectors comprising a classification of the electricity consumption of the cold room SFD.

[0162] The output values of the predictive electricity consumption model MODP include, for example, a time component. For example, the output values each include an electricity consumption value of the cold room SFD per hour, or an electricity consumption value SFD per day.

[0163] According to another example, a random forest is implemented and comprises several decision trees, which are created from subsets of training data each trained on a different sample of data, which are for example randomly selected from a set of training data. Each decision tree calculates for example a predicted value as a function of the input variables, which are for example continuous variables such as values of pressure, humidity, temperature, cloudiness, wind speed etc. to output a predicted value of electricity consumption. First function Fl: general information

[0164] In one embodiment, with reference to [Fig.4], the first predictive model MODp is chosen by means of a first function Fh

[0165] The first function Fi comprises for example a function dependent on the first input data Dtini and / or the second input data Dtin2. This is for example an expert system configured to implement an inference engine based on a knowledge base and a rule base. Examples of inference engines include a forward, backward, or mixed chaining engine.

[0166] In one embodiment, the first function Fi comprises a learning function, or machine learning function.

[0167] The first function Fi receives for example as input at least one data item taken from the first input data Dtiniet / or taken from the second input data Dtin2. This is for example a temperature data item, for example the first temperature data item Dtempi. According to another example, this is the second characteristic data item Dc2.

[0168] According to another example, the first function Fire receives as input a plurality of data taken from the first input data Dtiniet / or taken from the second input data Dtin2.

[0169] One advantage is to choose as precisely as possible the most suitable predictive model based on the available data.

[0170] The first function Fi for example implements at least one model taken from: a neural network, for example a recurrent neural network RNN, a random forest, also called a decision tree forest, or a linear regression, or an ARIMA (AutoRegressive Integrated Moving Average) model. First function Fl: training

[0171] According to one embodiment, with reference to [Fig.6], the first function Fi is trained TR2 by means of a second machine learning AM2.

[0172] According to one embodiment, the first function Fi is trained by means of the second machine learning AM2 from data taken from the first input data Dtinl and / or taken from the second input data Dtin2 and relating to the same model or the same brand of refrigeration unit GF.

[0173] This is for example a history of first input data Dtini and / or second input data Dtin2 relating to the same model or the same brand of GF refrigeration unit.

[0174] According to one embodiment, the first function Fi is trained from data taken from the first input data Dtini and / or taken from the second input data Dtin2 and relating to different models or different brands of GF refrigeration unit.

[0175] This is for example a history of first input data Dtini and / or second input data Dtin2 relating to different models or brands of GF refrigeration unit.

[0176] The training data are preferably vectorized and normalized. Each input vector thus defined comprises, for example, a set of training data.

[0177] According to one embodiment, the implementation of the second machine learning model AM2 comprises a labeling of each input vector. The labels define for example characteristic states of the cold group GF, data relating to an operation carried out on the cold room or any other context data, for example a loading state of a payload inside the cold room SFD.

[0178] According to one embodiment, output data of the first machine learning model AMi comprises classified data. The output data of the first machine learning model AMi comprises, for example, labeled data.

[0179] According to one embodiment, the implementation of the first machine learning model AMi includes a step aimed at correcting data classified by the first machine learning AMb. This step includes, for example, a modification of a label, an addition of a label, an enrichment of a label, etc.

[0180] According to one embodiment, the method comprises a step of preprocessing the input data of the first AMp machine learning model. The input data is for example collected, cleaned and preprocessed.

[0181] According to one embodiment, the method comprises a step of defining the learning model for the first machine learning model AM[.

[0182] This is for example a supervised model such as a classification model or a regression model.

[0183] According to another example, it is an unsupervised model taking unlabeled data as input, such as a grouping model, often called a “clustering” model.

[0184] According to one embodiment, data taken from the input data of the second machine learning model AM2 are divided into several subsets. These are, for example, a training data set and a validation data set. The model is then, for example, trained on the training data set.

[0185] According to one embodiment, parameters of the second machine learning model AM2 are adjusted to reduce an error function (or cost function). Such a step is for example implemented by means of an optimization algorithm such as gradient descent, mean square error or binary cross entropy.

[0186] According to one embodiment, the second machine learning model AM2 is evaluated using data taken from the validation data.

[0187] One advantage is to estimate the ability of the first function Fi to generalize predictions to new input data.

[0188] According to one embodiment, the method comprises several training and evaluation cycles to optimize the first function Fh

[0189] According to one embodiment, the training of the first function Fi comprises a parameter adjustment step. These are, for example, adjustable parameters defined before the start of training. The parameters comprise, for example, hyperparameters. Examples of hyperparameters include a learning rate, a number of epochs (or iterations), a size of the training data batch, or an activation function. This advantageously makes it possible to optimize the performance of the model.

[0190] According to an example, the output values of the first function Fi comprise output vectors comprising a classification of the electrical consumption of the refrigeration unit GF.

[0191] The output values of the first function Fi include, for example, a time component.

[0192] According to another example, a random forest is implemented and comprises several decision trees, which are created from subsets of training data each trained on a different sample of data, which are for example randomly selected from a set of training data. Each decision tree calculates for example a predicted value as a function of the input variables.

[0193] Event detection / characterization function

[0194] The method comprises a step of detecting DTCi at least one event associated with the cold room SFD.

[0195] Examples of events that can be associated with the cold room are: - A predicted deviation, for example by means of a history of evolution of historical data, for example of internal temperature data of the cold room SFD. In the latter case, portions of temperature evolutions of a history can be characteristic of singular events, such as a poorly closed door, a temporary opening of the door, evolutions linked to micro-cuts linked to excessive energy consumption of a set of equipment, a power cut, etc. A database comprising a characterization of the events by families of curves can be used. The historical data can include curve profiles or characteristic points of a curve, such as thresholds, or even a particular dynamic of the curve. - A predicted deviation, for example by means of a mathematical model and historical data, of an interior temperature of the SFD cold room; The mathematical model can result from a function learned from a machine learning model. The mathematical model can correspond to a function characterizing the evolution of a curve of a physical parameter such as the evolution of temperature in a given context. For this purpose, the coefficients of the function can be learned from historical data. - an open door; - loss of watertightness; - a fault in a refrigeration circuit, for example a compressor or evaporator fault, or a refrigerant leak; - a fault in the power supply to the SFD cold room;

[0196] The event is detected among a set of event classes by means of a characterization function Fc.

[0197] The characterization function Fc automatically identifies and characterizes at least one deviation between an expected operating state of the SFD cold room and an actual operating state of the SFD cold room. An expected operating state corresponds, for example, to an expected energy consumption for the SFD cold room over a given duration.

[0198] The characterization function Fc identifies and characterizes a difference between an expected operating state of the cold room SFD and an actual operating state of the cold room SFD, for example by comparing the operating data Dfct with the predicted consumption data Cpr. This is, for example, a comparison between an actual consumption data Cre and the predicted consumption data Cpr. In another case, it is a comparison between data relating to an operating mode of the cold unit GF or of equipment in the cold room, such as data relating to an operating regime of a compression system, and the predicted consumption data Cpr. Such a comparison is, for example, possible when the operating data Dfct makes it possible to deduce an electrical consumption from another parameter, for example an operating time or an operating regime of equipment in the cold unit GF.

[0199] In one embodiment, an event Evm is identified from a set of event classes if a deviation between a predicted value and an actual value exceeds a predefined threshold. The threshold may also be a variable value depending on one or more parameters.

[0200] For example, the parameters may include the position of the site or any other physical parameter characterizing the state of the cold room.

[0201] The characterization of the deviation can be linked to a punctual variation of one or more parameters compared to one or more predicted values.

[0202] According to one case, a recurrence of a variation of a parameter identified by means of the characterization function Fc allows to characterize the event Evm. For example, a deviation below a threshold value could be considered acceptable taken individually, but the repetition of this variation over a given period allows to identify and characterize an event.

[0203] According to one embodiment, the characterization function Fc comprises an expert system configured to implement an inference engine. This is for example a software component allowing a system to reason on a data set and to deduce information from this data set using a set of rules.

[0204] The inference engine is for example implemented from a knowledge base and from at least one predefined rule. The knowledge base comprises for example predicted values of electricity consumption and associated actual values of electricity consumption. An example of an implemented rule may be that any deviation greater than a value X between the predicted value of electricity consumption Cpr and the actual value of electricity consumption Cre is associated with a problematic event. Another example of a rule is that any average deviation greater than a value Y between the predicted value of electricity consumption Cpr and the actual value of electricity consumption Cre over a time range greater than or equal to Z minutes characterizes an opening of the door of the cold room SFD.

[0205] According to one embodiment, the detection and characterization of an event Evm by means of the characterization function Fc comprises the association of said event Evm with an event class.

[0206] In one embodiment, an event is automatically relabeled if a recurrence greater than a threshold value of this event is observed over a given time interval. This is for example a variation in energy consumption which, taken punctually, does not characterize an event in itself, but the recurrence of which makes it possible to determine that an event has occurred.

[0207] Characteristic data of a model Dc2 and second trained machine learning function F2

[0208] In one embodiment, with reference to [Fig.6], the second characteristic data Dc2 comprises data produced by a second machine learning function F2, called second function F2.

[0209] The second function F2 takes for example as input an electrical consumption of the cold room SFD.

[0210] The second function F2 makes it possible, for example, to determine a classification, for example to associate an electrical consumption value, a set of electrical consumption values or even an electrical consumption profile with a given GF refrigeration unit model taken from a set of refrigeration unit models.

[0211] According to one embodiment, with reference to [Fig.6], the second function F2 is trained TR3 by means of a third machine learning AM3.

[0212] According to an example of training the second function F2, the method comprises a step of collecting data, for example characteristic data of a cold room relating to specific models of GF cold groups, such as electrical consumption data associated with given models of GF cold groups. The method then comprises for example cleaning and preprocessing this data. The data is for example divided into several sets, such as a training data set and a validation data set. The model is then trained on the training data set. The second function F2 receives for example as input, during training, electrical consumption data, such as consumption values over time and as a function of given parameters, associated with models of GF cold groups.The method then comprises a step of defining the learning model, for example a classification model for associating electricity consumption values or electricity consumption profiles with GF refrigeration unit models. The parameters of the second function F2 are for example adjusted to reduce an error function (or cost function), for example by means of an optimization algorithm such as gradient descent, mean square error or binary cross entropy. The model is for example evaluated using data taken from the validation data. Such an evaluation makes it possible in particular to estimate the capacity of the first function F2 to accurately determine a GF refrigeration unit model based on electricity consumption data. The training comprises for example several training and evaluation cycles to optimize the second function F2.

[0213] According to one embodiment, any of the training steps of any of the first, second or third machine learning Ami, Am2, Am3 is directly applicable to another machine learning.

[0214] According to one embodiment, any of the input or output data of any of the first, second or third machine learning AMi, AM2, Am3 can be used as input data of another machine learning. System

[0215] According to another aspect, the invention relates to a system configured to implement the steps of the method according to the invention.

[0216] The system further comprises at least one first CLCi computer configured to receive the first input data Dtini. The first input data Dtini are for example received by the first CLCi computer following their transmission from a local memory or from a remote memory such as a remote server.

[0217] The first calculator CLCi is for example configured to generate, by means of a first predictive model of electrical consumption of the cooling unit MODP trained from the first machine learning AMb at least one predicted value of electrical consumption of the cooling unit. According to another example, it is another calculator.

[0218] The system further comprises means for acquiring at least one operating data item of the cold group Dtfct, such as an acquisition device.

[0219] The system further comprises a calculator for detecting, by means of a characterization function Fc, at least one event associated with the cold room SFD from a set of event classes.

[0220] In one embodiment, the system comprises the SFD cold room.

[0221] In one embodiment, the system comprises a terminal connected to the group refrigeration unit to receive operating data Dtfct from said SFD refrigeration unit. The refrigeration unit and the terminal are connected, for example, by wired means or by a wireless link.

[0222] The system comprises, for example, a plurality of refrigeration units each generating at least one piece of operating data, such as actual electricity consumption data Cre.

[0223] The system comprises, for example, an administration console for the operation of said GF refrigeration units.

[0224] The administration console communicates, for example, with the SFD cold rooms locally by means of a wired connection, or remotely by means of a wireless connection.

[0225] The system comprises for example at least one calculator configured to implement at least one function for classifying the event Evm.

[0226] According to another aspect, the invention relates to a computer program product comprising instructions which, when the program is executed by a computer, cause the latter to implement steps of the method of the invention.

Claims

Claims

1. Method for detecting an event (Evm) associated with a cold room (SFD) comprising at least one cold group (GF), the method comprising the following steps: • First reception (RECi), by a first calculator (CLCi), of first input data (Dtini) comprising at least one temperature data item (Dtempi) relating to an interior temperature of the cold room (SFD); • Second reception (REC2), by said first computer (CLCi) or by another computer, of second input data (Dtin2) comprising at least one operating data (Dfct) relating to the operation of at least one piece of equipment in the cold room (SFD); • Selection (SLCJ) of a first predictive model of electricity consumption (MODP) of the refrigeration unit (GF), said model being trained from a first machine learning (AMi); • Implementation (MOJ) of the first predictive electricity consumption model (MODP) to determine at least one predicted electricity consumption data of the cold group (Cpr), called predicted consumption data (Cpr), and characterizing an expected operating state of the cold room (SFD); • Detection (DTCi) of at least one event associated with the cold room (SFD) among a set of event classes, by means of a characterization function (Fc) automatically identifying and characterizing at least one deviation between the expected operating state of the cold room (SFD) and an actual operating state of the cold room (SFD), by means of the predicted consumption data (Cpr) and the operating data (Dtfct).

2. Method according to claim 1, in which the operating data (Dfct) comprises actual electrical consumption data of at least one refrigeration unit (Cre), called actual consumption data (Cre), and in which the characterization function (Fc) identifies and characterizes at least one difference between a value associated with said actual consumption data (Cre) and a value associated with the predicted consumption data (Cpr).

3. Method according to any one of the preceding claims, in which the operating data (Dtfct) comprises at least one data item relating to an operating duration and / or an operating regime of at least one piece of equipment in the cold room (SFD).

4. Method according to any one of the preceding claims, in which the characterization function (Fc) comprises an inference engine implemented from a knowledge base and from at least one predefined rule.

5. Method according to any one of the preceding claims, in which the step of selecting (SLCi) the first predictive consumption model (MODP) is implemented automatically by means of a first function (Fi) whose parameters depend on data taken from among the first input data (Dtini) and / or data taken from among the second input data (Dtin2).

6. Method according to claim 5, in which the first function (Fi) is a learning function trained by means of a second machine learning (AM2).

7. Method according to any one of the preceding claims, in which the first input data (Dtini) comprises at least one external environment data (Dtext) measured by at least one sensor positioned on the cold room (SFD) or on a site in which the cold room (SFD) is located, at least one external environment data (Dtexl) being estimated from a position data (DtpOS) received from a location device.

8. Method according to any one of the preceding claims, in which the first input data (Dfni) comprises at least one meteorological data (Dtmt) received by means of a communication interface and from at least one remote server.

9. Method according to any one of the preceding claims, in which the second input data (Dtin2) comprise at least one first characteristic data item of the refrigeration unit (Dtci), called first characteristic data item (Dtci), and relating to an operating mode of the refrigeration unit (GF) characterized by a limited interval of interior setpoint temperatures (SFD).

10. A method according to any preceding claim, in in which the second input data (Dtin2) comprise at least one second characteristic data item of the refrigeration unit (Dc2), called second characteristic data item (Dc2), and relating to a model and / or a brand of the refrigeration unit (GF) characterizing technical characteristics specific to the refrigeration unit (GF).

11. Method according to claim 10, in which at least one second characteristic data (Dc2) comprises data produced by a second function (F2) trained by means of a third machine learning (AM3), called second function (F2), taking as input the real consumption data (Cre).

12. Method according to any one of the preceding claims, in which the first machine learning (AMi) receives as input data comprising histories of the same input data (Dfni, Dtin2) relating to different cold group models (GF) or different histories of input data (Dtini, Dtin2) relating to the same cold group model (GF).

13. System for detecting an event associated with a cold room (SFD) comprising at least one cold group (GF) configured to implement the steps of the method according to any one of claims 1 to 12.

14. System according to claim 13, comprising: • At least one calculator for: i. Receiving (RECi) the first input data (Dtini) comprising at least one temperature data item (D tempi) related to an interior temperature of the cold room (SFD); ii. Receiving (REC2) the second input data (Dt in2) comprising at least one operating data item (Dtfct) relating to an operation of at least one piece of equipment of a cold group (GF); iii. Selecting (SLCJ) a first predictive model of electrical consumption (MODp) of the cold group (GF) from a set of predictive models of electrical consumption (ENSM); iv. Implementing (MOi) the first predictive model of electrical consumption (MODp) to determine at least one piece of electrical consumption data predicted cold group (Cpr), called predicted consumption data (Cpr), and characterizing an expected operating state of the cold room (SFD); v. Detect (DTCi) at least one event (Evm) associated with the cold room (SFD) among a set of event classes, by means of a characterization function (Fc) automatically identifying and characterizing at least one deviation between the expected operating state of the cold room (SFD) and an actual operating state of the cold room (SFD), and at least by means of the predicted electrical consumption value (Cpr) and the operating data (Dfct).

15. A system according to any one of claims 13 to 14, comprising: • The cold room (SFD) comprising at least one cold group (GF); • A terminal connected to the cold room (SFD).

16. System for classifying an event (Evm) associated with a cold room (SFD), said system comprising: • a plurality of cold groups (GF) each generating at least one temperature datum (Dttempi) relating to their set temperature and at least one operating datum (Dtfct) relating to the operation of at least one piece of equipment in the cold room (SFD); • an administration console for the operation of the plurality of cold groups (GF); • at least one calculator configured for: • implement a first predictive electricity consumption model (MODP) to determine at least one predicted electricity consumption data of the cold unit (Cpr) characterizing an expected operating state of the cold room (SFD); • implement at least one event classification function (Evm) characterized by at least one deviation between an expected operating state of the cold room (SFD) and an actual operating state of the cold room (SFD), by means of the predicted electrical consumption value (Cpr) and the operating data (Dfct).

17. A computer program product comprising instructions which, when the program is executed by a computer, cause the latter to implement the steps of the method according to any one of claims 1 to 12.