Method for predicting an amount of energy harvestable by an electronic device, and associated device

Abstract

The invention relates to a method for predicting an amount of energy harvestable by an electronic device, the method comprising: determining a type of a kinetic-energy source by means of an energy-source classification model based on a signal representative of an acceleration of the electronic device (10A); and predicting, by means of an energy prediction model, an amount of kinetic energy harvestable by the electronic device (10A) depending on the type of the kinetic-energy source and on a history of amounts of kinetic energy actually obtained with the kinetic-energy source.

Description

Method for predicting a quantity of energy recoverable by an electronic device, and associated device

[0001] The present invention belongs to the general field of energy recovery systems. It relates more particularly to a method for predicting a quantity of energy recoverable by an electronic device. It also relates to an electronic device configured to implement such a method.

[0002] The invention may relate for example (but not exclusively) to any autonomous electronic device, i.e. which is not powered by a constant energy source, such as an electrical network, and / or to any electronic device implementing ambient energy harvesting technology, such as photovoltaic, kinetic, thermal, radiofrequency and / or osmotic energy.

[0003] The invention finds application in particular for applications of the "Internet of Things" type ("Internet of Things" or IoT in Anglo-Saxon literature).

[0004] Connected objects (sometimes also called "smart objects") are hardware and / or software devices characterized by their ability to interact with their immediate environment, generally through a microcontroller to control a sensor and / or an actuator, as well as by their connectivity. These objects are connected to a communication network, such as for example the public Internet network within the framework of the Internet of Things (IoT), and can therefore communicate with other systems to obtain and / or provide information.

[0005] Thus, connected objects make it possible to capture and send back to the network the current value of information specific to their environment and / or their operation, and / or to receive from the network a command whose execution can have an effect on this environment and / or this operation.

[0006] The areas in which these connected objects are used are very varied, and include in particular:

[0007] – industry – sometimes then called “industry 4.0” –, for example with the use of connected robots to enable more detailed monitoring of the different stages of production or with the use of predictive maintenance systems;

[0008] – the smart city, for example to monitor and manage a traffic and transport system;

[0009] – security, for example with the use of connected cameras and presence sensors;

[0010] – health, for example with the use of connected medical devices or fall detection devices to combat loss of autonomy;

[0011] – energy, for example with the use of communicating electricity meters with an electricity network manager;

[0012] – or household appliances, for example with the use of connected food processors or dishwashers.

[0013] The massive deployment of these new technologies in such diverse fields is, however, hampered by energy constraints, and it is now understood that the successful deployment of the next generations of connected objects will depend on their ability to achieve energy autonomy. Even if the use of electrical energy generation units – such as electric batteries, accumulators or batteries – at least partially solves this problem by ensuring relative autonomy for the connected object, it still requires regular maintenance of the connected object so that these generation units are recharged and / or replaced. In addition, their manufacture requires the use of rare resources – for example nickel, zinc, lithium or cobalt – the extraction of which is particularly polluting and the recycling of which is relatively complex and costly, particularly in terms of energy.

[0014] Several solutions have been considered that aim to use so-called "alternative" or "renewable" energy sources to power connected objects. Among these solutions, photovoltaic technology has proven its effectiveness when integrated into connected objects, but it still remains dependent on the presence of light and therefore cannot be a universal solution for powering connected objects.

[0015] There is therefore a need for a new energy harvesting solution to power electronic devices, which does not have the drawbacks of the prior art.

[0016] The present invention aims to remedy all or part of the drawbacks of the prior art, in particular those set out above, by proposing a solution which makes it possible to predict a quantity of energy likely to be recovered by an electronic device from a source of kinetic energy setting this electronic device in motion.

[0017] To this end, and according to a first aspect, the invention relates to a method for predicting a quantity of energy recoverable by an electronic device, according to claim 1. Claims 2 to 11 describe preferred embodiments of said method.

[0018] The method may include:

[0019] – a determination of at least one type of a first source of kinetic energy by an energy source classification model and from a signal representative of an acceleration of said electronic device; and

[0020] – a prediction, by an energy prediction model, of a quantity of kinetic energy recoverable by the electronic device as a function of the type of the first source of kinetic energy and a history of quantities of kinetic energy actually obtained with said first source of kinetic energy.

[0021] By "kinetic energy source" is meant here a physical phenomenon generating kinetic energy due to a movement (for example, due to this sole fact). As mentioned below, a motor vehicle, a road, a human activity (and corresponding for example to an action of daily life) are different examples of types of kinetic energy sources. It is important to note that no limitation is attached to the types of kinetic energy sources considered (i.e., the types of kinetic energy sources considered are not a limiting factor of the invention).

[0022] Acceleration is a physical quantity representing a change in the speed of a movement as a function of time. This quantity can be positive or negative, and is also called "deceleration" when it is negative. In certain implementations, acceleration corresponds to a vibratory movement of the electronic device which then oscillates around a first position (such as a reference position).

[0023] Furthermore, no limitation is attached to the considered "energy source classification model". Any model implementing a learning algorithm ("machine learning") and providing, as output, a designation of a kinetic energy source having generated this acceleration (for example the most probable energy source) can be considered in the context of the invention. Furthermore, the energy source classification model can be, at least in certain embodiments, independent of the training method considered to train this model, and any training criterion can be considered during the training phase of the energy source classification model, such as the least squares method or the minimization of cross entropy.

[0024] No limitation is otherwise attached to the "energy prediction model" considered. Thus, any model analyzing quantities of energy previously obtained by the electronic device to predict a quantity of recoverable energy can be considered in the context of the invention.

[0025] Finally, the "quantity of recoverable energy" is expressed for example in Watt-hours (or Joules). However, it is also possible to represent it by other quantities, such as a power expressed in Watts, or an electrical voltage expressed in Volts. Thus, in the remainder of the description, the "quantity of recoverable kinetic energy" considered as an example is represented by a generated electrical voltage usable by the electronic device.

[0026] Generally speaking, it is considered that the steps of a process should not be interpreted as being linked to a notion of temporal succession.

[0027] In certain embodiments, the method for predicting a quantity of recoverable energy may further comprise one or more of the following characteristics, taken individually or in all technically possible combinations.

[0028] In some implementations, the energy source classification model comprises a support vector machine (SVM) or a multi-layer perceptron (MLP) neural network.

[0029] In some implementations, the energy prediction model includes a statistical model, such as an exponential moving average (EMA) or an autoregressive integrated moving average (ARIMA).

[0030] In the present application, a "moving average", also called a "rolling average", is used. This type of statistical average can be suitable, for example, for the analysis of time series, because it removes transient fluctuations so as to highlight longer-term trends. This average is called a moving average because it is recalculated continuously, using a subset of elements in which a new element replaces the oldest or is added to the subset.

[0031] A moving average is called "exponential" when it uses exponentially decreasing weighting of terms. The weight of each value participating in the average is greater than the value preceding it in the series, which gives more importance to the most recent observations, without completely eliminating the effect of the oldest values.

[0032] In some embodiments, an autoregressive integrated moving average may be used, as it is a statistical model that may be particularly suitable for analyzing time series for making predictions. Since this model is autoregressive, the values ​​integrated into the time series are then determined based on previous observed values. In addition, the use of this statistical model may help to track and anticipate the evolution of a phenomenon.

[0033] In some embodiments, the method further comprises training the energy source classification model from labeled input data, the data being derived from a plurality of signals representative of an acceleration of said electronic device due to a plurality of different kinetic energy sources, a label identifying the kinetic energy source among the plurality of kinetic energy sources.

[0034] In some embodiments, the method further comprises adapting the functionalities and / or performances of the electronic device, based on the predicted amount of recoverable kinetic energy.

[0035] Such implementation methods can help to adapt the electrical consumption of the electronic device based on a prediction of an amount of electrical energy likely to be obtained.

[0036] In certain embodiments, the method further comprises recording a value representative of a quantity of energy actually obtained by the electronic device during a first time interval (prior to the current instant) in the history of quantities of kinetic energy.

[0037] In some implementations, the predicted amount of kinetic energy corresponds to an amount of kinetic energy recoverable during a second time interval (later than the current time).

[0038] The use of time intervals to determine values ​​of amounts of kinetic energy actually obtained and to predict an amount of recoverable kinetic energy is based on the assumption that a source of kinetic energy is potentially volatile, but generates, on average, approximately the same amount of kinetic energy during a given time interval.

[0039] In some implementations, the second time interval has a similar duration (denoted below) to that of the first time interval.

[0040] In certain embodiments, the value representative of a quantity of energy actually obtained by the electronic device during a first time interval corresponds to an average of values ​​of quantity of energy obtained by the electronic device and with said first source of kinetic energy during the first time interval.

[0041] In some embodiments, the determination of the type of the first kinetic energy source is carried out regularly at a frequency , and the prediction of a quantity of recoverable kinetic energy is implemented regularly at a frequency , with for example .

[0042] Such implementation modes can help to adapt the prediction in real time, and, if necessary, to take into account a change in the source of kinetic energy.

[0043] In certain implementation modes, the first and second time intervals may correspond to sliding time intervals, i.e. having a certain duration, and of which the upper limit (eg, the end date) of the first time interval and the lower limit (eg, the start date) of the second time interval correspond to the same instant during which the quantity of recoverable kinetic energy is predicted.

[0044] In some embodiments, the method further comprises:

[0045] – detection of a change in the source of kinetic energy, the acceleration of said electronic device then being generated by a second source of kinetic energy; and

[0046] – a correction of at least one prediction of the quantity of recoverable kinetic energy subsequent to the detection time, considering a history of quantities of kinetic energy actually obtained by said electronic device with the second source of kinetic energy.

[0047] In certain embodiments, the method further comprises, prior to a correction of a prediction of the quantity of recoverable kinetic energy, a synchronization of the lower bound of a time interval (of duration ) between two predictions of kinetic energy quantity with the instant of detection of a change in kinetic energy source.

[0048] In some embodiments, the method further comprises:

[0049] – obtaining the signal representative of an acceleration of said electronic device;

[0050] – a determination of a power spectral density (or "Power Spectral Density", PSD, according to Anglo-Saxon terminology) of the signal obtained;

[0051] – a decomposition of the signal power spectral density into frequency bands; and

[0052] – a determination, for each frequency band, of a quadratic mean, so as to obtain a plurality of characteristics representative of the acceleration signal.

[0053] Thus, the acceleration signal at the input of the module for determining at least one type of kinetic energy source can be processed so as to generate frequency characteristics representative of the acceleration signal, for a given source. In certain implementations, these steps for processing the input signal can be implemented during the training phase, to identify a set of frequency characteristics (called "reference") specific to a given kinetic energy source. In certain implementations, these steps for processing the input signal can be implemented during the inference phase to determine the characteristics of an acceleration signal whose source must be determined.

[0054] In certain embodiments, the first and / or second determined source of kinetic energy corresponds to a motor vehicle – for example, an automobile, a bus, a four-wheeler, a tram, a metro, a train – selected from a plurality of motor vehicles, a road of a certain type – for example, a motorway, a national road, a departmental road, a metropolitan road, a municipal or rural road – selected from a plurality of road types, or to a human activity – for example, ironing, vacuuming, climbing and / or descending stairs, walking, Nordic walking, cycling, running, etc. – selected from a plurality of human activities.

[0055] According to a second aspect, the invention relates to a computer program comprising instructions for implementing a method for predicting a quantity of recoverable energy, when said program is executed by a processor.

[0056] This program may use any programming language, and may be in the form of source code, object code, or code intermediate between source code and object code, such as in a partially compiled form, or in any other desirable form.

[0057] According to a third aspect, the invention relates to a computer-readable recording medium on which the computer program according to the invention is recorded.

[0058] The information carrier may be any entity or device capable of storing the program. For example, the carrier may include a storage medium, such as a rewritable non-volatile memory or ROM, for example a CD-ROM or a microelectronic circuit ROM, or a magnetic recording medium, for example a hard disk.

[0059] On the other hand, the information carrier may be a transmissible carrier such as an electrical or optical signal, which may be conveyed via an electrical or optical cable, by radio or by other means. The program according to the invention may in particular be downloaded from a network such as the Internet.

[0060] Alternatively, the information carrier may be an integrated circuit in which the program is incorporated, the circuit being adapted to perform or to be used in the performance of the method in question.

[0061] According to a fourth aspect, the invention relates to an electronic device comprising at least one processor configured (or configured together) to implement the method for predicting a quantity of recoverable energy previously mentioned in any of its embodiments.

[0062] According to a fifth aspect, the invention relates to an electronic device configured to predict a quantity of recoverable energy according to claim 12. Claim 13 describes a preferred embodiment of said electronic device.

[0063] The electronic device may include:

[0064] – a module for determining at least one type of a kinetic energy source by an energy source classification model from a signal representative of an acceleration of said electronic device; and

[0065] – a prediction module, by an energy prediction model, of a quantity of kinetic energy recoverable by the electronic device as a function of the type of the kinetic energy source and a history of quantities of kinetic energy actually obtained with said kinetic energy source.

[0066] For each step of the method of predicting a quantity of recoverable energy, in any of its embodiments, the electronic device of the present application may comprise a corresponding module configured to carry out said step.

[0067] According to a sixth aspect, the invention relates to a system for predicting a quantity of recoverable energy including a transport vehicle (land, sea, air and / or space) in which the electronic device mentioned above is embarked, in any of its modes of implementation.

[0068] Other characteristics and advantages of the present invention will emerge from the description given below, with reference to the appended drawings which illustrate an exemplary embodiment thereof without any limiting character. In the figures: la is a first example of an environment in which a method for predicting a quantity of recoverable energy can be implemented; la is a second example of an environment in which a method for predicting a quantity of recoverable energy can be implemented; la represents modules embedded in an electronic device for predicting a quantity of recoverable energy, according to an exemplary implementation of the invention; la represents an example of hardware architecture of an electronic device for predicting a quantity of recoverable energy;illustrates, in the form of a flowchart, the main steps of a method for predicting a quantity of recoverable energy of the present application, according to an exemplary implementation;illustrates the temporal evolution of the relative standard deviation of several human activities;[Figure 5B illustrates the temporal evolution of the relative standard error of several human activities;illustrates the experimental results of changes in real human activities and as detected by the electronic device of the present application; andillustrates the experimental results of actual quantities of energy and as predicted by the electronic device of the present application.;

[0069] This is a first example of an environment in which a method for predicting a quantity of recoverable energy can be implemented.

[0070] As illustrated by the, an electronic device 10A for predicting a quantity of recoverable energy according to the invention is located inside a train. This mode of transport can generate elastic waves. These waves, commonly called "vibrations", are generated by the interaction between the rail and the train, and can in particular be induced by the irregularities of the surfaces of the wheels of the train and the rail with which they are in contact. The condition of the rails, the type of train – for example a high-speed train, TGV, or a regional train –, the type of trains – for example cars dedicated to the transport of passengers or wagons dedicated to the transport of animals or goods –, the speed of travel of the train, the composition of the underlying land and subsoil are all factors that can influence these vibrations.

[0071] These vibrations propagate in the air, the ground of the railway track, but also in the train, and are detected by the electronic device 10A for predicting a quantity of recoverable energy via the reception, by the electronic device 10A, of a vibration signal.

[0072] The electronic device 10A is, for example, a user terminal such as a laptop, a personal assistant, a connected watch or a mobile phone of the "smartphone" type – a router, etc. Generally speaking, no limitation is attached to the structural form that can be taken by this electronic device 10A.

[0073] In the present embodiment, and for the purpose of simplifying the description, it is considered that the environment comprises a single electronic device 10A, as well as a single source of kinetic energy corresponding, in this example, to a train. It should however be noted that no assumption is made as to the number of electronic devices 10A and sources of kinetic energy. The following developments can in fact be generalized without difficulty by the person skilled in the art to the case where more than one electronic device and / or more than one source of kinetic energy are considered.

[0074] The electronic device 10A analyzes the detected vibrations, and identifies a source of kinetic energy likely to have generated these vibrations, from a frequency analysis of the vibration signal. In this regard, the electronic device 10A uses a (local or remote) energy source classification model configured to determine at least one type of kinetic energy source (for example the most probable) from the vibration signal. In certain embodiments, the energy source classification model designates at output at least the most probable type of energy source from among a plurality of types of potential kinetic energy sources. Optionally, it can also generate at output a probability associated with one or more (for example each) of the types of sources from this plurality of types of potential kinetic energy sources.

[0075] In this example, the electronic device 10A is configured to determine the most likely type of kinetic energy source from a plurality of potential kinetic energy source types including train, car, bicycle, airplane, and running and to derive therefrom an amount of kinetic energy that can be recovered by the same electronic device 10A (or by another device subjected to vibrations similar or correlated to those received by the device 10A). For example, after determining that the most likely source of kinetic energy is a train, the electronic device 10A predicts, using an energy prediction model, an amount of kinetic energy that can be recovered by the same electronic device 10A from the vibrations generated by the train.

[0076] More precisely, the knowledge of the source of kinetic energy producing these vibrations is used by the electronic device 10A to predict, using the energy prediction model, a quantity of kinetic energy that it is likely to recover from the vibrations emitted by this train.

[0077] Based on the predicted amount of kinetic energy, the electronic device 10 may decide to adapt its own functionalities. In this example, the emitted vibrations are significant in terms of amplitude and repetitive nature, and the predicted amount of kinetic energy may prompt the electronic device 10 to activate functionalities that were previously deactivated because they were too energy-intensive.

[0078] This is a second example of an environment in which a method for predicting a quantity of recoverable energy can be implemented.

[0079] In this second example, an electronic device 10B for predicting a quantity of recoverable energy according to the invention is located inside a car traveling on a deformed road surface comprising bumps and / or potholes.

[0080] As previously, the electronic device 10B analyzes the detected vibrations, and identifies the type(s) of kinetic energy source that likely generated these vibrations from a frequency analysis of the vibration signal. In this example, the electronic device 10B is, however, configured to determine an energy source corresponding to a probable type of road on which this car is traveling from among a plurality of types of roads such as a highway; a national road; a departmental road and a municipal road.

[0081] La represents modules embedded in an electronic device 10 for predicting a quantity of recoverable energy, according to an exemplary implementation of the invention.

[0082] This electronic device 10 corresponds for example to one of the electronic devices 10A and 10B of figures 1A and 1B, and comprises in particular:

[0083] – a MOD_SRC module for determining at least one type of kinetic energy source SRC from a SIG signal representative of an acceleration of this electronic device 10.

[0084] – a MOD_PRED module for predicting PRED a quantity of kinetic energy likely to be recovered from accelerations (or, in a particular case, from vibrations) produced by a kinetic energy source. As illustrated by the, this MOD_PRED module takes as input the type of kinetic energy source SRC determined by the SRC module, and can also access an association table (referenced BDD) each entry of which associates an SRC source with a history of quantities of kinetic energy actually obtained with said SRC source. This history can take the form of a list of quantities of kinetic energy, this list being ordered according to the times at which the quantities of kinetic energy were obtained.

[0085] – a TDC transducer configured to convert acceleration (or vibration) into an electrical signal. This TDC transducer corresponds, for example, to a piezoelectric pressure transducer, also called a "piezoelectric transducer" or "piezoelectric sensor". Such a piezoelectric transducer is, for example, composed of a piezoelectric material, such as quartz or ceramic, positioned between two electrodes. When the material is subjected to mechanical stresses, such as accelerations or vibrations, it deforms and generates an electric field. This electric field is then converted into electrical energy by the piezoelectric transducer.

[0086] – an STK energy storage unit for storing the energy produced by the TDC transducer and taking the form, for example, of a battery, a battery or a capacitor.

[0087] – and an energy management module MOD_MNG configured to adapt the functionalities and / or performances of the electronic device 10, according to the predicted quantity of recoverable kinetic energy.

[0088] The invention has so far been described in the case where the energy source classification model and the energy prediction model are embedded in the electronic device 10, respectively in the MOD_SRC and MOD_PRED modules. These provisions are however not limiting of the invention, and nothing excludes the possibility of the energy source classification model and / or the energy prediction model being executed by one or more electrical devices distinct from the electronic device 10. In this case, the MOD_SRC module and / or the MOD_PRED module are for example configured to interact with these electrical devices distinct from the electronic device 10 through a telecommunications network.

[0089] Furthermore, it is also important to note that the TDC transducer, the BDD association table, the STK energy storage unit and / or the MOD_MNG energy management module may also be embedded on one or more electronic devices distinct from the electronic device 10.

[0090] Thus, in the case where the energy management module MOD_MNG is embedded on an electronic device separate from the electronic device 10, the module MOD_PRED for predicting a quantity of recoverable kinetic energy is for example configured to interact with this remote MOD_MNG module via a telecommunications network.

[0091] La represents an example of hardware architecture of an electronic device 10 for predicting a quantity of recoverable energy.

[0092] As illustrated by the, the electronic device 10 for predicting a quantity of recoverable energy has the hardware architecture of a computer. Thus, the electronic device 10 for predicting a quantity of recoverable energy comprises in particular a processor 1, a random access memory 2, a read-only memory 3 and a non-volatile memory 4. It further comprises a communication module 5.

[0093] The read-only memory 3 of the electronic device 10 for predicting a quantity of recoverable energy constitutes a recording medium as proposed, readable by the processor 1 and on which is recorded a computer program PROG in accordance with the invention, comprising instructions for executing steps of the method for predicting a quantity of recoverable energy as proposed in the present application. The program PROG defines one or more functional modules of the electronic device 10 for predicting a quantity of recoverable energy, which rely on or control the hardware elements 1 to 5 cited above, and which include in particular:

[0094] – a MOD_SRC module for determining at least one type of kinetic energy source SRC from a SIG signal representative of an acceleration of this electronic device 10; and

[0095] – a MOD_PRED module for predicting a quantity of kinetic energy likely to be recovered from vibrations or more generally from accelerations produced by a source of kinetic energy.

[0096] Furthermore, the electronic device 10 for predicting a quantity of recoverable energy may also comprise other modules, in particular for implementing certain embodiments of the method for predicting a quantity of recoverable energy, as described in more detail later.

[0097] Illustrates, in the form of a flowchart, the main steps of a method for predicting a quantity of recoverable energy of the present application, according to an exemplary implementation. This method is for example implemented by the electronic device 10 of Figures 2 and 3.

[0098] As illustrated by the, the prediction method comprises a step S100 during which at least one type of kinetic energy source SRC causing accelerations or vibrations of the electronic device 10 is determined using an energy source classification model from a signal representative of these accelerations. This step can for example be implemented by the module MOD_SRC for determining at least one type of kinetic energy source described with reference to FIGS. 2 and 3. In certain implementations, the energy source classification model can comprise a support vector machine or a neural network of the "multilayer perceptron" type.

[0099] This SRC source of kinetic energy is then used during a step S110 to predict, using an energy prediction model, a quantity of recoverable energy likely to be obtained from an acceleration generated by this SRC source of kinetic energy. This step is for example implemented by the MOD_PRED module for predicting a quantity of kinetic energy described with reference to FIGS. 2 and 3.

[0100] In some implementations, the energy prediction model may be a statistical model, such as an exponential moving average or an autoregressive integrated moving average.

[0101] During this step S110, a first "timer" elapses during a time interval of duration between at least two determinations of a source of kinetic energy and a second "timer" running over a time interval of duration between at least two predictions of an amount of kinetic energy likely to be recovered from accelerations are triggered. In certain embodiments, the values ​​of these "timers" may for example be such that .

[0102] As mentioned previously, this feature is advantageous in that it allows the prediction to be adapted in real time, particularly in the case of a change in the kinetic energy source, with greater reactivity than when , which can help to potentially obtain a more reliable prediction.

[0103] A "timer" here corresponds to a counter register which increments or decrements according to the pulses of a clock which could be that of processor 1 (for example at each clock pulse).

[0104] The prediction method further comprises a step S120 during which the electronic device 10 checks whether the duration between at least two determinations (for example between two determinations) of a type of kinetic energy source has elapsed or not. If this is the case (choice "Y"), a step S130 is implemented during which the electronic device 10 determines the current source of kinetic energy causing accelerations of the electronic device 10. This step is for example implemented by the module MOD_SRC. During this step 130, the electronic device 10 also resets the first "timer" elapsing during the time interval of duration between at least two determinations of a type of kinetic energy source.

[0105] Back to step S120, if on the other hand the duration between at least two determinations of a type of kinetic energy source has not elapsed (choice "N"), the electronic device implements a step S160 during which it checks whether the duration between two predictions of a quantity of kinetic energy likely to be recovered from accelerations has elapsed or not. If this duration has not yet elapsed (choice "N"), the prediction method loops back to step S120. If, on the other hand, this duration has elapsed (choice "Y"), the electronic device 10 implements a step S170 of determining a value representative of a quantity of energy actually obtained by the electronic device during a time interval, called the first interval, of duration and ending at the current time (i.e., the time interval of duration immediately prior to the current time.

[0106] In certain embodiments, the value representative of a quantity of energy actually obtained by the electronic device during this first time interval corresponds to an average of values ​​of quantity of energy actually obtained by the electronic device 10 and with this source SRC of kinetic energy during this first time interval.

[0107] The prediction method further comprises a step S180 during which the value determined during step S170 is recorded in an association table, each entry of which associates a source SRC with a history of quantities of kinetic energy actually obtained with said source of kinetic energy SRC. In certain embodiments, this history corresponds to a list ordered according to the instant at which a quantity of kinetic energy was actually determined (i.e., the instant at which the interval considered for the entry ended).

[0108] During a step S190, the electronic device 10 predicts, using an energy prediction model, a quantity of kinetic energy recoverable by the electronic device 10 as a function of said kinetic energy source and the history of quantities of kinetic energy actually obtained with said kinetic energy source SRC. This step S190 is for example implemented by the MOD_PRED module of the electronic device 10.

[0109] In some implementations, this prediction is stored in a list accessible by the MOD_MNG energy management module associating predictions and times to which the predictions refer. This may be, for example, a list ordered according to the times associated with the predictions (for example chronologically).

[0110] In certain embodiments, step S190 further comprises a transmission, to the energy management module MOD_MNG, of said prediction, so as to allow the management module MOD_MNG to adapt the functionalities of the electronic device 10 accordingly. This adaptation comprises, for example, the activation of functionalities previously deactivated because they are deemed to be too energy-intensive or, on the contrary, the deactivation of functionalities previously activated, as well as the modification of parameters (e.g., the brightness of the screen, the location of the electronic device 10, etc.).

[0111] In certain embodiments, the energy management module MOD_MNG is embedded in an electronic device separate from the electronic device 10, for example an electronic device for supervising the energy management of the electronic device 10 (or of a plurality of electronic devices including the electronic device 10). In this case, step S190 may comprise a transmission, to the electronic supervision device, of data representative of the prediction.

[0112] In certain embodiments, the quantity of predicted kinetic energy corresponds to a quantity of kinetic energy recoverable by this electronic device during a time interval, called the second interval, subsequent to the instant associated with a prediction and of duration for example equal to the prediction duration. .

[0113] During this same step, the second "timer" elapses during a time interval of duration is reset. Then the prediction process loops back to step S120.

[0114] In step S130, once the source of kinetic energy has been determined, the electronic device 10 determines, in a step S140, whether the source determined in step S130 is different from that determined in the previous determination. If this is not the case (e.g., if the source determined in step S130 is identical to that determined in the previous determination, choice "N"), then the prediction method implements step S160 previously described.

[0115] If, on the other hand, it is determined in step S140 that the source of kinetic energy has changed (choice "Y"), the electronic device 10 implements a step S150 during which at least the last calculated prediction (i.e., the prediction subsequent to the instant during which a change of source is detected) is deleted from the list of predictions. Furthermore, during this same step, the first and second "timers" are reset. Then the prediction method loops back to step S190 previously described.

[0116] Experimental results

[0117] The experimental results set out below are for example implemented by the electronic device 10 previously described, in one of its embodiments. It is important to note that these experimental results are only examples, and are not limiting of the invention. In particular, the devices used in the context of this experiment and the implementation choices do not in any way limit the scope of the present application.

[0118] These experimental results were obtained by considering different human activities as sources of kinetic energy. This field of study is sometimes called "human activity recognition" (HAR), and in the following, the "energy source classification model" is also called "human activity recognition model".

[0119] Determination of human activities

[0120] Input data

[0121] The raw data considered for this experiment come from the "PAMAP2" physical activity monitoring dataset. These data were processed and missing and / or erroneous data were removed. Table 1 summarizes the dataset considered as input to the model (corresponding to the "energy source classification model" previously mentioned).

[0122] ActivityDuration of accelerations (min)Number of samplesPercentage (%)Static position94.656829.37Ironing39.623812.34Vacuum cleaner29.11759.07Stairs3722211.5Walking38.623212.03Nordic walking311869.63Cycling27.51658.52Running16.16975Skipping rope8.1492.53TOTAL321.81931100

[0123] Feature extraction

[0124] The raw data were divided into labeled samples with a duration of 10 seconds. The final dataset considered contained 1931 samples.

[0125] The samples were then sorted by activity and the power spectral density (PSD) of each sample was calculated. The spectrum was then decomposed into frequency bands, (0-0.5), (0.5-1), (1-4), (4-7), (7-10), (10-13), (13-16), (16-19), (19-22), (22-25), (25-30) and (30-50).

[0126] Results of human activity recognition

[0127] The dataset was divided into 80% training data and 20% testing or inference data. Only low-resource-intensive techniques that could be embedded in microcontrollers were studied. In particular, two classifiers were preselected:

[0128] – support vector machines (SVM);

[0129] – a multi-layer perceptron (MLP) comprising 2 hidden layers of 24 and 19 nodes. This architecture was determined by a “grid search” type optimization.

[0130] Tables 2 and 3 illustrate the confusion matrices using the SVM and MLP models, respectively. It can be concluded that, overall, both models have almost similar results, with an average F1 score of 90.41% and 91.45%, respectively. These results are superior to those of other studies working on the same dataset, which obtain an F1 score varying between 82% for SVM and 95% for convolutional neural network (CNN) models.

[0131] ActivitiesPlanned ActivityStatic PositionIroningVacuumingStairsWalkingNordic WalkingCyclingRunningSkipping RopeReminder%Actual ActivityStatic Position1090111001096.46Ironing339230010081.25Vacuuming223100000088.57Stairs400372100084.09Walking111143000091.49Nordic Walking0000037000100Cycling303010260078.79Running000000117189.47Skipping Rope jump00110000880Accuracy%89.3492.8679.4986.0591.4997.3792.8694.4488.89F1 = 90.41

[0132] ActivitiesPlanned ActivityStatic PositionIroningVacuum CleanerStairsWalkingNordic WalkingCyclingRunningSkipping RopeReminder%Actual ActivityStatic Position1100120000097.35Ironing342110010087.5Vacuum Cleaner313100000088.57Stairs411352100079.55Walking011144000093.62Nordic Walking0000037000100Cycling220011270081.82Running100000018094.74Skipping Rope skip00100000990Precision%89.4389.3686.1189.7493.6294.8796.43100100F1 = 91.45

[0133] Energy prediction model

[0134] To test the hypothesis that the same activity generates approximately the same amount of energy over a given period of time, a solution was developed to generate a data set based on the PSDs of the samples available for each activity.

[0135] More specifically, several PSDs of the same activity were randomly mixed with random phases. Thus, random time series were generated for each activity with different durations. The generated signals were then used to control a vibration source (in the experiment, the PM20 permanent magnet shaker from DynaLab ©) to approximately reproduce the vibrations generated by human activity. A piezoelectric cantilever "S452-J1FR-1808XB", with dimensions of 71 mm × 25.4 mm × 1.32 mm, was fixed on the vibration source. Since its resonant frequency was 246 Hz, a weight was added to reduce its resonant frequency to about 20 Hz, which is a suitable frequency to simulate the vibrations of human activities.

[0136] Twenty time signals were created with durations of 30, 60, 90, 120, and 300 seconds. The illustrates the time evolution of the relative standard deviation for several human activities, and the illustrates the time evolution of the relative standard error for these human activities.

[0137] It can be seen that the relative standard deviation (also called "coefficient of variation") decreases as the vibration duration increases. For a duration of 300 seconds, it can be seen that the relative standard deviation fell to 11.49, 8.78, 9.52 and then 6.76%. In a relatively similar manner, the standard error decreased significantly. These results therefore validate the hypothesis that an activity can generate approximately the same amount of energy during a certain time interval, here 300s. Therefore, it can be deduced that it may be possible to estimate the harvestable energy of each activity using a moving average and statistical models.

[0138] Prediction models

[0139] To be able to apply energy prediction models, signals of much longer duration than the samples were required for each activity.

[0140] As before, three 2-hour signals were generated for each activity. Then the results were divided into 300-second windows. Therefore, for each activity, 72 samples were available. The effective voltage of each sample was then calculated, and 72 observations were thus obtained.

[0141] The time series were divided into two datasets, one for training and one for validation. Five models were studied: the exponential moving average (EMA) model, the autoregressive (AR) model, the moving average (MA) model, the autoregressive integrated moving average (ARIMA) model, and the vector error correction (VEC) model.

[0142] The last 10 values ​​of the energy quantities actually obtained were used to predict the next recoverable energy quantity. The results in terms of relative error are summarized in Table 4.

[0143] ActivitiesEMAARMAARIMAVECCycling5.355.615.305.804.74Nordic Walking7.506.526.616.616.89Vacuum Cleaner8.048.266.627.257.45Running6.978.2510.897.068.05Average6.977.167.366.686.78

[0144] It can be seen that all predictions have a relative error of less than 10%. High-frequency activities, such as cycling and Nordic walking, perform better in this experiment than the other two activities. On average, the ARIMA model has the best accuracy. It can also be seen that the EMA model, which requires few resources to implement, performs well for all activities, even outperforming the ARIMA model for cycling and running.

[0145] Laillustrates the experimental results of actual human activity changes and as detected by an electronic device according to the present application and laillustrates the experimental results of actual energy quantities and as predicted by the electronic device of the present application.

[0146] To obtain these experimental results, an 8-hour signal representative of several human activities with random durations was generated. In addition, a human activity recognition model of the SVM type, an energy prediction model of the EMA type, a time interval of a duration of 120 s between two energy source determinations and a time interval of a duration of 300 s between two predictions of the quantity of recoverable energy were considered.

[0147] The human activity recognition model illustrates the determination of human activities by considering this 8-hour signal. It can be concluded that human activities are well recognized, with a slight delay caused by the 120-s time interval of the human activity recognition model. It can also be observed that, in this experiment, over the entire 8 hours, only one false determination lasting 2 false minutes was made.

[0148] Illustrates the predicted energy quantity and the energy actually obtained considering this 8-hour signal. The vertical lines represent the instants during which the activity changed. It can be observed that the curve representing the predicted energy quantity follows that of the energy actually obtained, and that variations in energy quantity are anticipated when the activity changes.

[0149] It can also be observed that the energy quantity prediction model achieved a relative error of 11.25%. The increase in the relative error compared to the root mean square deviation when the activities are analyzed separately can be explained by the fact that the mixing of activities during the generation of the 8-hour signal biased the model previously trained on distinct human activities. Nevertheless, these results prove that the electronic device of the present application effectively predicts the short-term recoverable kinetic energy, even with random and / or variable accelerations.

Claims

A method for predicting an amount of energy recoverable by an electronic device (10, 10A, 10B), the method comprising a prediction (S190), by an energy prediction model, of an amount (PRED) of kinetic energy recoverable by the electronic device as a function of a type of a first source (SRC) of kinetic energy and a history of amounts of kinetic energy actually obtained with said first source of kinetic energy. The method of claim 1, further comprising adapting the functionality and / or performance of the electronic device, based on the predicted amount of recoverable kinetic energy. The method of claim 1 or claim 2, further comprising recording (S180) a value representative of an amount of energy actually obtained by the electronic device during a first time interval, in the history of kinetic energy amounts. The method of claim 3, wherein the predicted amount of kinetic energy corresponds to an amount of kinetic energy recoverable during a second time interval. Method according to claim 3 or claim 4, the value representative of a quantity of energy actually obtained by the electronic device during a first time interval corresponding to an average of values of quantity of energy obtained by the electronic device and with said first source of kinetic energy during the first time interval. Method according to one of claims 1 to 5, comprising a determination (S130) of the type of the first source (SRC) of kinetic energy by an energy source classification model from a signal representative of an acceleration of said electronic device. The method of claim 6, wherein the determination (S130) of the type of the first kinetic energy source is carried out regularly at a frequency , and the prediction (S190) of a quantity of recoverable kinetic energy being implemented regularly at a frequency , with . Method according to claim 7, further comprising:detecting (S140) a change of kinetic energy source, the acceleration of said electronic device then being generated by a second kinetic energy source; and,correcting (S150) at least one prediction of the amount of recoverable kinetic energy subsequent to the detection time, by considering a history of amounts of kinetic energy actually obtained by said electronic device with the second kinetic energy source. The method of claim 8, further comprising, prior to correcting a prediction of recoverable kinetic energy quantity, synchronizing the lower bound of a time interval between two predictions of kinetic energy quantity with the detection time (S140) of a change of kinetic energy source. Method according to one of claims 6 to 9, further comprising,– obtaining the signal representative of an acceleration of said electronic device;– determining a power spectral density (or "Power Spectral Density", PSD, according to English terminology) of the signal obtained;– decomposing the power spectral density of the signal into frequency bands; and– determining, for each frequency band, a quadratic mean, so as to obtain a plurality of characteristics representative of the acceleration signal. A method according to one of claims 1 to 10, wherein the first and / or second source of kinetic energy corresponds to a motor vehicle selected from a plurality of motor vehicles, a road of a certain type selected from a plurality of road types, or a human activity selected from a plurality of human activities. Electronic device (10, 10A, 10B) configured to predict a quantity of recoverable energy comprising a module (MOD_PRED) for predicting, by an energy prediction model, a quantity (PRED) of kinetic energy recoverable by the electronic device as a function of a type of a source (SRC) of kinetic energy and a history of quantities of kinetic energy actually obtained with said source of kinetic energy. Electronic device (10, 10A, 10B) according to claim 12, comprising a module (MOD_SRC) for determining at least one type of source (SRC) of kinetic energy by an energy source classification model from a signal representative of an acceleration of said electronic device. Computer program comprising instructions for implementing the method according to one of claims 1 to 11, when said program is executed by a processor. A computer-readable recording medium on which the computer program according to claim 14 is recorded.

Images