METHOD AND DEVICE FOR DATA PROCESSING FOR MONITORING THE FILL LEVEL OF A LIQUID

Description

Technical field

[0001] A method and a processing device for monitoring the level of a liquid are described. The method and the device can for example be implemented to monitor the level of a liquid in a container, such as the fuel level in a tank, and detect events related to variations in this level, for example liquid theft. Technical background

[0002] Many applications require liquid level monitoring. One such application is, for example, monitoring the fuel level in a tank powering a backup generator. Such generators must be activated as soon as a primary power source fails, for example, in the event of a power grid outage or, for stand-alone applications, a malfunction of the power source (solar panels, wind turbines) and / or the batteries that the power source charges. Another application concerns domestic tanks or tanks in the agricultural sector.

[0003] The level of a liquid can be monitored in a manner known per se by implementing suitable sensors or probes. The sensor or probe makes it possible to obtain data representative of the evolution of the level of the liquid monitored. However, the signal obtained may be noisy or contain ambiguous information. The signal will then be likely to be misinterpreted and will give rise to unjustified alarms.

[0004] It is desirable to have a method of monitoring the level of a liquid that reduces these disadvantages.

[0005] Documents CN 111 980 104 A, US 2022 / 028189 A1 and US 2015 / 310678 A1 concern the technical field in question. Summary

[0006] A first aspect relates to a method implemented by a processing device comprising a processor comprising obtaining a signal, obtained from measurements of a probe, representative of the evolution of the level of a liquid in a container; searching, in the signal, for at least one event of a first type; where a given event is associated with one or more signal characteristics allowing the identification of the given event in the representative signal and where an event of the first type is an event for which no alarm is generated; labeling, of time periods of the representative signal corresponding to an event of the first type, with the event(s) of the first type identified for these periods; following the labeling, searching, in the representative signal, for at least one given event of a second type; where an event of the second type is an event for which an alarm is generated;searching for a given event of the second type including disqualifying signal time periods labeled with certain events of the first type, and maintaining signal time periods labeled for other events of the first type; generating an alarm for the second type event(s) identified in the signal.

[0007] For example, an event of the first type for which a time period is disqualified when searching for an event of the given second type is an event likely to give rise to an erroneous identification of the event of the given second type, and therefore to a false alarm.

[0008] For example, an event of the first type for which a time period is maintained while searching for an event of the given second type is an event that can hide the event of the given second type.

[0009] According to one or more embodiments, the method comprises, before the search phases, a preprocessing of the representative signal, said preprocessing comprising at least one of noise reduction based on wavelet decomposition; generation of missing data in the representative signal.

[0010] According to one or more embodiments of the method, the probe is a differential pressure probe and searching for a first type event comprises searching for a clogged condition of the probe.

[0011] According to one or more embodiments of the method, a clogged state of the probe is identified during the search if a daily periodic pattern is identified in the representative signal; and for a day for which a periodic pattern is identified, a maximum of the representative signal over the day is found in a time interval around sunrise and a minimum of the representative signal for the day is found in a time interval including sunset, the time intervals being disjoint.

[0012] According to one or more embodiments, the method comprises, during the phase of searching for at least one event of the second type, parts of the signal corresponding to a period of time including at least the day for which a blocked probe event was identified.

[0013] According to one or more embodiments of the method, searching for a first type event comprises searching for a probe disconnection and reconnection event.

[0014] According to one or more embodiments of the method, the search for a disconnection and reconnection of a probe comprises: identifying in the signal a phase of variation of the signal comprising a variation in the negative direction then in the positive direction, respectively the opposite; determining that the liquid level indicated by the representative signal before and after a phase of variation is substantially the same, and in the event of a positive determination, identifying a disconnection and reconnection event without a parallel flight event, and in the event of a negative determination, identifying a disconnection and reconnection event with potentially a parallel flight event.

[0015] According to one or more embodiments of the method, the search for an event of the first type comprises the search for an event of filling the container with liquid.

[0016] According to one or more embodiments of the method, the search for a filling of the container with liquid comprises: identifying in the signal a positive variation in the liquid level, determining whether the positive variation is greater than a liquid quantity threshold over a given period of time and, if the determination is positive, identifying a refilling event and, if not, not identifying a refilling event.

[0017] According to one or more embodiments of the method, the search for a second type event comprises the search for an event of filling the container with liquid.

[0018] According to one or more embodiments of the method, searching for a second type event comprises searching for a liquid theft event.

[0019] According to one or more embodiments of the method, searching for a liquid theft event comprises: searching for at least one drop in the liquid level in the representative signal; determining for each drop identified whether this drop is greater than a liquid quantity threshold during a given period of time and, if the determination is positive, identifying a theft event and otherwise, not identifying a theft event.

[0020] According to one or more embodiments of the method, when an identified drop follows or overlaps with an event of the first type, the liquid quantity threshold during a given time period is lower compared to the threshold used when an identified drop does not follow or overlap with an event of the first type.

[0021] A processing device comprising means for implementing one of the methods described. These means may comprise a processor and a memory comprising software code which, when executed by the processor, causes the device to implement one of the methods.

[0022] A recording medium readable by a processing device having a processor, said medium comprising software code which, when the software is executed by a processor of a processing device, causes the processing device to implement one of the methods. Brief description of the figures

[0023] Other characteristics and advantages will appear during the reading of the detailed description which follows for the understanding of which one will refer to the attached drawings among which: there Figure 1is a block diagram of a liquid level monitoring system according to a non-limiting exemplary embodiment; Figure 2 is a block diagram of a processing device according to a non-limiting exemplary embodiment; the Figure 3 is a flowchart of a method according to an exemplary embodiment; the Figure 4 is a flowchart giving details of a pre-processing phase of the above method according to a non-limiting exemplary embodiment; Figure 5 is a flowchart of a noise reduction method according to a non-limiting exemplary embodiment; Figure 6 is a set of graphs representing an initial signal and signals following noise reduction; the Figure 7a is a graph representing a signal with missing data; the Figure 7b is a graph of the signal of the Figure 6 in which the missing data have been interpolated; figure 8is a flowchart of a method for detecting a blocked probe by implementing autocorrelation according to a non-limiting exemplary embodiment; Figure 9 is a graph of a liquid level signal in the context of a non-limiting example application of the method of the figure 8 ; there Figure 10 is a graph showing the result of the autocorrelation of the signal of the Figure 9 according to a non-limiting example of embodiment; Figure 11 is an enlargement of one period of the signal of the figure 8 ; there Figure 12 represents a liquid level signal indicating a clogged probe; the Figure 13 is a flowchart of a method for detecting a probe disconnection event, according to a non-limiting exemplary embodiment; Figure 14 includes graphs of a liquid level signal and its derivative in the context of detecting a disconnection event; Figure 15is a flowchart of a method for detecting filling according to a non-limiting exemplary embodiment; figure 16 includes the graphs of a liquid level signal and its derivative in the context of detecting a filling event; Figure 17 is a graph of a liquid level signal illustrating the detection of flight events in conjunction with probe disconnection events; figure 18 is a graph of a liquid level signal illustrating the detection of liquid theft events after a tank filling event; figure 19 is a graph of a liquid level signal represents a liquid level signal in centimeters over a period of more than one month, before pretreatment; the figure 20 is a flowchart of a method for detecting an event of the second type according to a particular exemplary embodiment. Detailed description

[0024] In the following description, identical, similar or analogous elements will be designated by the same reference numbers.

[0025] The block diagrams, flowcharts, and message sequence diagrams in the figures illustrate the architecture, functionality, and operation of systems, devices, processes, and computer program products according to one or more exemplary embodiments. Each block of a block diagram or each phase of a flowchart may represent a module or a portion of software code comprising instructions for implementing one or more functions. In some implementations, the order of the blocks or phases may be changed, or the corresponding functions may be implemented in parallel. The process blocks or phases may be implemented using circuitry, software, or a combination of circuitry and software, in a centralized manner, or in a distributed manner, for all or some of the blocks or phases.The systems, devices, methods and methods described may be modified, added to and / or deleted while remaining within the scope of this description. For example, the components of a device or system may be integrated or separated. Also, the described functions may be implemented using more or fewer components or phases, or with other components or through other phases. Any suitable data processing system may be used for the implementation. A suitable data processing system or device, for example, comprises a combination of software code and circuitry, such as a processor, controller or other circuitry suitable for executing the software code. When the software code is executed, the processor or controller causes the system or device to implement some or all of the functionalities of the blocks and / or phases of the methods or methods according to the exemplary embodiments.For example, the processor or controller may implement all or part of a method, and in this context also communicate with other components to control them and / or transmit data to them and / or receive data, for the implementation of the method. The processor or controller therefore leads, alone or in conjunction with other components of a device in which it is integrated, to implement the steps of a method. The software code may be stored in a memory or a readable medium accessible directly or through another module by the processor or controller.

[0026] There Figure 1 is a block diagram of a liquid level monitoring system according to a non-limiting exemplary embodiment.

[0027] The system of the Figure 1comprises a tank 101 containing a liquid 102. The liquid is, for example, fuel intended to power a device 103, for example, a generator set. A probe 104 is placed in the tank so as to measure a quantity representative of the height of the liquid 102. According to the present exemplary embodiment, the probe is a differential pressure probe which measures the pressure difference between atmospheric pressure and the pressure in the tank - the height (level) of liquid in the tank can be deduced therefrom in a manner known per se. The probe is, for example, placed at the bottom of the tank. The volume of liquid can, if necessary, be deduced from this information depending on the characteristics of the tank. The probe has a cable 105 and a capillary 106 connected to a connection box 107. The capillary makes it possible to apply atmospheric pressure to the rear of the sensor of the probe. The pressure of the liquid column above the probe is thus measured.The measurement data from the probe 104 are collected by the connection box 107, converted into liquid height and transmitted to a controller device 108, one role of which is to collect the data over time and transmit them to a server 111 of a supervision center 112 via a communication network 110 (for example a TCP / IP network). The server 111 stores the data and makes them available to a device 114 allowing the data to be displayed - for example to an operator. The device 114 is also adapted to transmit configuration data to the controller device 108 so that the latter can implement them at the production site 113. This configuration data may, for example, include the triggering and stopping parameters of the generator set 103. A single supervision center 112 may interact with several production sites.The data representative of the liquid height are also transmitted to a processing device 109. This processing device can be an integral part of the supervision center 112, or even be connected temporarily.

[0028] This is a typical implementation of a system where the tank is remote from the processing device or the control center that hosts this device. However, according to other exemplary embodiments, the storage and processing of the data from the probe 104 can be done closer to the tank - in this case, the functionalities of the processing device 109 can for example be integrated into the controller device 108. The advantage of the illustrated system is that the history of the data from the probe is available in the server 111 for other applications and can thus be easily retrieved to be subjected to the method described later.

[0029] Other types of probes or sensors than a differential probe can be implemented to obtain data representative of the level or volume of liquid.

[0030] It should be noted that in what follows, the data obtained from the probe refers to both direct measurements and data derived from direct measurements. For example, for a differential probe, 'data' can include pressure measurements, but also data obtained from the measurements, such as the height or volume of liquid. In what follows, the height or level of the liquid will generally be taken as an example, knowing that the principles explained can be applied to other representations of the measurements. Furthermore, we will also speak of 'signal'; the data represents a signal obtained directly or indirectly from the probe measurements and representing a physical quantity.

[0031] It should be noted that in the case of a probe using a capillary, the latter can become blocked by condensation and the condensation water bubble can move in the capillary.

[0032] There Figure 2 is a block diagram of a processing device 109 according to an exemplary embodiment. The processing device 109 comprises a processor 201, a memory 202, a user interface 203, a communication interface 204 configured to communicate for example with the server 111, a display 206 adapted for displaying data to a user, as well as a working memory 207.

[0033] The various components of the processing device 109 are connected through a communication bus 208. The memory 202 comprises software code 205. The memory 207 is used to store and manage the data to be transmitted. When the processor executes the software code 205, it causes the processing device 109 to implement a method according to one or more exemplary embodiments described. The processing device 109 may be a computer or another suitable processing device.

[0034] There Figure 3 is a flowchart of a method according to an exemplary embodiment as such not covered by the claims. The method is implemented by the processing device 109 and comprises three main steps: Pre-processing the signal (S301). Pre-processing aims to prepare the signal with a view to improving its quality in order to limit possible erroneous decisions during the detection steps. Pre-processing is in itself optional - its implementation may for example depend on the quality of the incoming data. The detection, based on the pre-processed signal, of one or more events of a first type other than those sought (S302). The detection, in the pre-processed signal, of one or more events of a second type corresponding to events sought (S303). The second detection step uses the results of the first step in the sense that in the event of detection of events of the first type, the corresponding part of the signal will be marked accordingly and will not be able to give rise to the detection of events of the second type.

[0035] Events of the first type detected beforehand will not be reclassified as events of the second type. Events of the first type will then not give rise to an erroneous detection of an event of the second type. By giving priority in the order of identification of events to events of the first type, the number of false alarms for events of the second type is reduced by having previously detected events corresponding, for example, to known normal situations or not requiring a particular action such as a site visit. An alarm is raised for an event of the second type but no alarm is raised for events of the first type. The expression 'alarm' designates, in a non-limiting manner, any indication (visual, audible or specific labeling of data identifying an event of the second type, such as a data label...) or any information or signal suitable for informing a user of the presence of an event of the second type. An alarm is such that it distinguishes an event of the second type from an event of the first type. No alarm is generated for events of the first type.

[0036] The detection of certain events of the first type disqualifies a period of time during which the event occurs for subsequent detection of an event of the second type.

[0037] Specific processing is performed in the case where the data indicate the possibility of an event of the second type being covered by certain events of the first type. In other words, an event of the second type may be hidden by an event of the first type. This is the case, for example, for certain events of the first type comprising an increase in the liquid level followed by a decrease - or vice versa - and that an event of the second type comprises only a decrease in the level (for example following a theft), or only an increase in the level (for example a filling, in the variant where this is an event of the second type). Pretreatment

[0038] There Figure 4 is a flowchart that details the preprocessing process of the Figure 3 and gives examples of the events within the context of the Figure 1 .

[0039] In a first step (S400), the historical data representative of the liquid level are obtained by the processing device.

[0040] The history can include data over several days, for example, a week. The sampling frequency used to obtain the data varies depending on the specific application and the events to be detected. For example, it is less than one hour, typically 15 minutes.

[0041] Preprocessing (S401) is then applied to the data. According to one exemplary embodiment, the preprocessing comprises noise reduction. According to another exemplary embodiment, the preprocessing further comprises the generation of missing data. According to one exemplary embodiment, the missing data is generated by interpolation, for example by linear interpolation between the two data values ​​surrounding a data-free interval.

[0042] The detection of known but unsearched events with known characteristics is then carried out (S402). In the example of the Figure 4 , these events include: Clogged probe conditions Probe disconnections Tank fillings

[0043] The data is labeled with the detected events (S403).

[0044] The detection of sought events is then carried out (S403). In the example of the Figure 4 , one such event is the theft of liquid from the tank.

[0045] According to another exemplary embodiment, the assignment of an event to a given type can be reviewed. For example, an event of the first type can become an event of the second type or vice versa, depending on the needs of a particular application. For example, the filling of the tank can be an event for which an alarm is sought to be raised, and therefore an event of the second type.

[0046] There Figure 5 is a flowchart of a noise reduction method according to an exemplary embodiment is as follows. Noise reduction aims to prepare the data - or the signal represented by the data - for the detection steps, in particular by reducing short but high variations in the signal represented by the data to limit the possibility that these variations erroneously give rise to the detection of an event. It has been found that wavelet denoising provides good results - such filtering allows in fact to keep the local properties of the signal and not to filter the characteristics by which the events will then be detected. For example and without limitation, a Daubechies wavelet decomposition can be used. Order 4 can be used.

[0047] According to the process of the Figure 5, the processing device first obtains the data representative of the signal to be processed (S500). A wavelet decomposition is then carried out (S501) - corresponding coefficients are thus obtained. The decomposition is followed by an estimation of the noise level (S502), the determination of a threshold (S503), and the reduction (S504) of the wavelet coefficients on the basis of this threshold, by deleting (or setting to zero) the coefficients lower than this threshold. A wavelet reconstruction (S505) of the signal is then carried out to obtain (S506) the denoised signal (and consequently the data which represents it). The threshold can for example be determined by statistical analysis of a sample of known data. The threshold is, by way of non-limiting example, equal to the standard variation of the noise of the signal multiplied by the square root of 2*In(N), where N is the size of the signal.The signal size is defined by the number of signal points per period considered. For a signal sample every 15 minutes, therefore for a period of one day, the size will therefore be 96 points.

[0048] There Figure 6 consists of three graphs (a), (b) and (c). Graph (a) is a graph representing the initial liquid level signal in centimeters over two days, from August 18 to August 20. Graph (b) represents the same signal, wavelet denoised as described above, while graph (c) shows the signal from graph (a), but denoised with a Savitzy-Golay filter. Comparing the three graphs, we see that wavelet denoising preserves local properties without smoothing them as much as the Savitzy-Golay filter.

[0049] There Figure 7a is a graph representing an initial liquid level signal in centimeters over two days and including missing data. The Figure 7bis a graph of the same signal, where the missing part has been replaced by interpolated data. Detection of unsought events with known characteristics

[0050] The second step seeks to identify unsought events, such as a blocked probe or a probe disconnection, but detectable by their characteristics in order to avoid confusing them with sought events, such as a theft. Detection of a blocked probe

[0051] When the probe used is a differential pressure probe, it determines the difference between the pressure in the tank and the atmospheric pressure to measure the liquid level in the tank. When the probe is plugged, it will therefore measure the atmospheric pressure of the air.

[0052] This feature is used in this exemplary embodiment to determine whether the probe is clogged or not. Several approaches can be used: A first approach is to use a history of data labeled for clogged probe events in conjunction with a supervised machine learning algorithm (e.g., a support vector machine). The algorithm can then detect events corresponding to a clogged probe in the liquid level data to be analyzed.

[0053] A second approach is based on meteorological data (pressure, or temperature (pressure depending on temperature) ...) or solar panel production data, over the period considered. A probe is considered clogged if a correlation exists between meteorological data or production data on the one hand and liquid level data on the other.

[0054] A third approach implements autocorrelation of liquid level data. The advantage of this method is that it does not require data history or data as described with the second approach. In addition, a transition to the frequency domain is not necessary.

[0055] Autocorrelation is used to detect a periodic repetition of a pattern in signal level data.

[0056] Autocorrelation is performed on data from two consecutive days. If the autocorrelation has a periodic form, then a periodicity is detected in the autocorrelated data. The first peak (after the central peak) and the first valley following this peak are then identified. The existence of a peak indicates a periodicity in the data. The duration of the period can be determined if necessary by the time interval between the first peak and the first valley.

[0057] The processing device further checks the liquid level data for the existence of the maximum level and the minimum level in specific time intervals during the day. These specific time intervals are chosen to correspond respectively to intervals in which a maximum and a minimum are expected with respect to the usual evolution of atmospheric pressure during a day. Typically, a maximum is found in a time interval including sunrise and a minimum is found in a time interval including sunset, the variations in atmospheric pressure being correlated with these two events.

[0058] For example, the time interval in which a maximum is to be located is 6 a.m. and noon, and the time interval in which a minimum is to be located is 4 p.m. and 8 p.m. There will be at least 4 hours but no more than 14 hours between the peak and the valley, which will distinguish the periodic pattern due to atmospheric pressure from patterns due to other phenomena. Note that the precise intervals are given as illustrations of the principles discussed, but the limits of these intervals may be different in other implementations. For example, one can also verify that one has the minimum of the day between 2 p.m. and 6 p.m. and the maximum between 6 a.m. and 9 a.m.

[0059] There figure 8 is a flowchart of a method for detecting a blocked probe by implementing autocorrelation according to an exemplary embodiment. Figure 9is a graph of a liquid level signal in centimeters over seven days which will be used to give a detailed example in relation to the method of figure 8 The method is applied successively to data corresponding to the interval Px of two consecutive periods of one day each, given that we are seeking to detect a daily periodicity of a phenomenon. For example, we take into account the data on the interval P1 comprising the first and second days, then those on the interval P2 comprising the second and third days, etc., until all the data have been processed.

[0060] A selection of data corresponding to a period covering two days is therefore carried out initially (S801) in the preprocessed data. An autocorrelation is carried out on the signal of these two periods (S802). The Figure 10is a graph showing the autocorrelated signal for period P1. It is then determined (S803, S804) whether at least one peak exists in the autocorrelation signal outside the central peak - in the Figure 10 , this peak is noted 'A', and the valley that follows is noted 'B'. If this is not the case, we conclude that no periodic pattern is present in the signal (S806). If on the other hand at least one peak is detected, we conclude that there is indeed a periodic pattern (S805). In this case, an additional check is carried out in S807 and S808 with respect to the liquid level signal (illustrated by the Figure 11 , which is an enlargement of the period P1 of the signal of the Figure 9) to determine whether the liquid level signal has a maximum in a first time slot in which a maximum is expected depending on the evolution of atmospheric pressure in a day. We do the same for a minimum, in relation to a second time slot. The minimum and the maximum are noted 'C' and 'D' in the Figure 11 . Here we consider the minimum and maximum of the signal over a day. If this check is negative (S809), and although a periodic pattern has been detected, it is concluded that the probe is not blocked, because this periodicity does not appear to be linked to a variation in atmospheric pressure. Otherwise (S810), the probe is considered to be blocked. In the case of Figure 10 , the probe is considered clogged.

[0061] There Figure 12is a graph representing a liquid level signal in centimeters over 31 days, with the probe becoming blocked on the 23rd day (period marked 'X' in the figure). This graph shows both the daily periodicity of the signal from the 23rd day onwards and the location of a maximum in the morning and a minimum in the afternoon for each day when the probe is blocked. Observation of this graph validates the choice of characteristics retained for the detection of a blocked probe.

[0062] A clogged probe is an event that particularly results in erroneous detection of other events, such as refueling or theft. Combining a clogged probe detection with a subsequent detection of a theft event is a particularly useful embodiment.

[0063] According to one embodiment, the two days during which a clogged probe is identified are excluded from the detection of liquid theft. The signal (i.e. the corresponding data) is labeled with the nature of the event found for this time interval. Detection of a probe disconnection

[0064] A probe disconnection, for example, involves removing and then replacing a probe. This can happen during a maintenance operation. The probe can also be disconnected by a malicious person. Liquid theft can occur during a disconnection phase.

[0065] The detection of a probe disconnection is based on the detection of positive and negative, or negative and positive variations in the liquid level over a limited period of time, and on a comparison of the liquid level before and after the variation.

[0066] There Figure 13is a flowchart of a method for detecting the disconnection (followed by reconnection) of a probe according to one embodiment. The preprocessed signal is used as a basis for detection (S1301). First, a derivative of the signal is obtained (S1302). The positive / negative or negative / positive variation phases are then identified in the derivative signal (S1303) - such a phase corresponds in the liquid level signal to either an increase in level, then to a decrease in level - or vice versa - during a period of time having a maximum duration. According to the present example, the phases with variations greater than a threshold 'XL' during a time T are retained. As a non-limiting example, in the context of certain applications, XL is for example equal to the accuracy of the probe (20 liters in the context of a real example) and T is two hours.

[0067] The detected variation phases are then checked individually - the liquid levels before and after a variation phase are compared. It is determined whether the liquid returns to the same level or not (S1304). According to the present example, this check is done to the nearest XL. If the liquid level returns to approximately the same level, it is considered that there has been a disconnection event, but a priori without parallel liquid theft (S1305). If the liquid level is considered not to return to the same level and is lower than before the variation phase, it is considered that there has been a disconnection event, with potentially an associated theft (S1306). According to a non-limiting example, the levels before and after a variation phase are taken respectively 30 minutes before and 30 minutes after the variation phase, for example in the case where the data sampling period is 15 minutes.

[0068] The signal (i.e. the corresponding data) is labeled with the nature of the event found for the time interval corresponding to the variation phase.

[0069] There Figure 14 includes two graphs showing a liquid level signal in centimeters over two days (reference (a)) and its derivative (reference (b)). The rectangles identify the positive / negative or negative / positive variation phases. In the liquid level signal graph, the variation phase at 0h between the two days corresponds to a disconnection event with possible theft given the level difference before and after the variation phase. The other variation phases correspond to a simple disconnection / reconnection. Detection of a tank filling event

[0070] A refill is defined as a positive change in liquid level above a threshold liquid quantity over a given period of time.

[0071] The method for detecting a tank filling according to the present exemplary embodiment is similar to that used for detecting a disconnection, but only considering positive variations in the liquid level.

[0072] There Figure 15 is a flowchart of a filling detection method according to an exemplary embodiment. The preprocessed signal is used as a basis for detection (S1501). First, a derivative of the signal is obtained (S1502). The positive variation phases are then identified in the derived signal (S1503). According to the present example, the phases with variations greater than a threshold 'XL' for a time T are retained. As a non-limiting example, in the context of certain applications, XL is for example equal to the accuracy of the probe (20 liters in the context of a real example) and T is two hours.

[0073] The detected variation phases are then checked individually - the liquid levels before and after a variation phase are compared. It is determined whether the liquid level after a variation phase is higher than the liquid level before the variation phase (S1504). According to the present example, this check is done to the nearest XL. If the check gives a positive result, then a filling event is identified (S1505). Otherwise, the observed variation does not correspond to a filling event (S1506). According to a non-limiting example, the levels before and after a variation phase are taken respectively 30 minutes before and 30 minutes after the variation phase, for example in the case where the data sampling period is 15 minutes.

[0074] The signal (i.e. the corresponding data) is labeled with the nature of the event found for the time interval corresponding to the variation phase.

[0075] There figure 16 includes two graphs showing a liquid level signal in centimeters over seven days (reference (a)) and its derivative (reference (b)). The rectangles identify the phases of positive variation. In the example in this figure, the two positive variations at the end of day 4 and at the end of day 5 are identified as refilling events.

[0076] According to a particular embodiment, the tank filling event is an event of the second type, i.e. a sought-after event. In this case, the detection of this event is only undertaken after the detection of the events of the first type. Detection of wanted events

[0077] Once the steps of preprocessing and identifying unsearched events have been completed, the third step for identifying the searched events whose characteristics are known is addressed. According to the present exemplary embodiment, these searched events include the theft of liquid and / or the filling of the tank.

[0078] The identification of a filling event has already been described. In the following, the detection of the theft event according to an exemplary embodiment will be described. Detection of a liquid theft event

[0079] A theft of liquid can occur with or without a connection to another event. Three examples of a theft event are as follows: During a disconnection event as previously described, a drop in the liquid level observed after reconnection compared to the liquid level before disconnection is potentially representative of liquid theft. According to an exemplary embodiment, the level of this drop during the disconnection event is compared to a threshold. This threshold is, according to the present example, a minimum rate of drop in the quantity of liquid per unit of time. This rate can be defined in relation to theft statistics, in particular in terms of liquid volume. According to an exemplary embodiment, this rate is 12 liters / 15 minutes. In the context of the present example, the duration chosen for calculating the rate is taken equal to the periodicity (granularity) of the liquid level data. When the rate is reached or exceeded, a theft event is identified. Figure 17illustrates such a situation. This figure is a graph of the liquid level signal per centimeter over a nine-day period. Two disconnection events are identified by rectangular boxes corresponding to previously labeled events in the liquid level signal. A theft can also occur after a tank fill. The figure 18is a graph of the liquid level signal in centimeters over five days. A fill is identified by the boxed positive change in liquid level. This fill is followed by a drop in liquid level potentially indicative of liquid theft. As with the example of theft after disconnection, a rate of drop in liquid quantity per unit of time can be used to decide whether or not a drop corresponds to a theft event. In an exemplary embodiment, this rate is identical to the rate in the case above related to the disconnection event. A theft can also occur when the liquid level signal has missing data over a period of time T and a drop in the liquid level indicated by the probe is observed when the data is resumed compared to when the data was stopped.As with the previous two cases, a rate of decline in the amount of liquid per unit time can be used to decide whether or not a decline corresponds to a theft event. Such a situation is illustrated by . figure 19 , which represents the liquid level signal in centimeters over a period of more than one month, before preprocessing. Missing data, identified by boxes in the figure, will have been linearly interpolated during the preprocessing phase.

[0080] Apart from the cases mentioned above, according to the present exemplary embodiment, a drop in the liquid level beyond a certain amount per unit of time is indicative of a theft event. In some applications, a liquid theft may have a duration exceeding two hours.

[0081] There figure 20is a flowchart of a method for detecting an event of the second type according to a particular exemplary embodiment. In this exemplary embodiment, the event of the second type is a theft. In other examples, it is a filling. In other examples, both the theft and filling events are events of the second type.

[0082] According to the flowchart of the figure 20 : We first identify (S2001) the events of the first type and we label the signal with the detected events.

[0083] In the context of this example, we seek to identify the events of blocked probe, tank filling, disconnection.

[0084] The periods of the liquid level signal corresponding to a clogged probe are excluded from further signal analysis (S2002).

[0085] The liquid level signal is then analyzed to detect the second type event (or events of the second type).

[0086] For the purposes of this example, a liquid theft event is identified in the following cases: S2003 - after a filling event, variations of more than 12I / 15 minutes; S2004 - after a probe disconnection event, downward variations of more than 12I / 15 minutes; S2005 - in case of downward variation of more than 20I / 15 minutes.

[0087] An alarm is generated for detected theft events (S2006).

[0088] Regarding the variation rates for S2003 and S2005, it was statistically found that for a theft, the variation could be between 12 and 15 L / 15min in the case where the theft is associated with other events. Such other events include refilling as above, or the fact that the generator is not running. In such cases, the variation threshold in S2005 is not always sufficient to detect this type of theft. Detection is thus refined. REFERENCE SIGNS

[0089] 101 - Tank 102 - Liquid 103 - Liquid-consuming device 104 - Probe 105 - Cable 106 - Capillary 107 - Connection box 108 - Controller device 109 - Processing device 110 - Communication network 111 - Server 201 - Processor 202 - Memory 203 - User interface 204 - Communication interface 205 - Software code 206 - Display 207 - Working memory 208 - Communication bus

Claims

1. Method carried out by a processing device (109) comprising a processor, the method comprising - obtaining (S400) a signal, obtained from probe measurements, representative of the changes in the level of a liquid in a container; - searching (S402), in the signal, for at least one event of a first type; a given event being associated with one or more signal characteristics enabling identification of the given event in the representative signal and an event of the first type being an event for which no alarm is generated; characterized by - labeling (S403) time periods of the representative signal corresponding to an event of the first type with the event or events of the first type identified for these periods; - following labeling, searching, in the representative signal, for at least one given event of a second type; an event of the second type being an event for which an alarm is generated; - searching for a given event of the second type having the disqualification of signal time periods labeled with certain events of the first type, and retaining signal time periods labeled with other events of the first type; - generating an alarm for the event or events of the second type identified in the signal.

2. Method according to claim 1, comprising, prior to the search phases, preprocessing (S401) of the representative signal, said preprocessing comprising at least one of the following - noise reduction (S401_2, S500-S506) based on wavelet decomposition; - generation (S401_1) of missing data in the representative signal.

3. Method according to either of claims 1 or 2, wherein the probe is a differential pressure probe and the search for an event of the first type comprises searching (S402_1) for a blocked state of the probe.

4. Method according to claim 3, characterized in that a blocked condition of the probe is identified during the search if - a daily-periodic pattern is identified in the representative signal; and - for a day for which a periodic pattern is identified, a maximum of the representative signal for the day is found in a time interval around sunrise and a minimum of the representative signal for the day is found in a time interval including sunset, the time intervals being disjointed.

5. Method according to either of claims 3 or 4, comprising the exclusion, during the phase of searching for at least one event of the second type, of the parts of the signal corresponding to a time period including at least the day for which a blocked probe event has been identified.

6. Method according to any of claims 1 to 5, wherein the search for an event of the first type comprises searching (S402_2) for a probe disconnection and reconnection event.

7. Method according to claim 6, wherein searching for disconnection and reconnection of a probe comprises: - identifying (S1303) in the signal a variation phase of the signal comprising a variation in the negative direction then in the positive direction, respectively the reverse; - determining (S1304) that the liquid level indicated by the representative signal before and after a variation phase is substantially the same, and in the event of a positive determination, identifying (S1305) a disconnection and reconnection event with no parallel theft event, and in the event of a negative determination, identifying (S1306) a disconnection and reconnection event with potentially a parallel theft event.

8. Method according to any of claims 1 to 7, wherein the search for an event of the first type comprises searching (S402_3) for an event of filling the container with liquid.

9. Method according to claim 8, wherein the search (S402_3) for a filling of the container with liquid comprises: - identifying (S1503) in the signal a positive liquid level variation, - determining (S1504) whether the positive variation is greater than a liquid quantity threshold over a given time period and, if the determination is positive (S1505), identifying a filling event and, if not, not identifying (S1506) a filling event.

10. Method according to any of claims 1 to 7, wherein the search for an event of the second type comprises searching for an event of filling the container with liquid.

11. Method according to any of claims 1 to 10, wherein the search for an event of the second type comprises searching for a liquid theft event.

12. Method according to claim 11 wherein the search for a liquid theft event comprises: - searching for at least one decrease in the liquid level in the representative signal; - determining (S2103, S2104, S2105) for each identified decrease whether this decrease is greater than a liquid quantity threshold over a given time period and, if the determination is positive, identifying a theft event and, if not, not identifying a theft event.

13. Method according to claim 12, wherein, when an identified decrease follows or overlaps with an event of the first type, the liquid quantity threshold over a given time period is lower than the threshold used when an identified decrease does not follow or overlap with an event of the first type.

14. Processing device (109), comprising means for carrying out a method according to any of claims 1 to 13.

15. Recording medium readable by a processing device provided with a processor, said medium comprising software program instructions which, when the software program is executed by a processing device processor, cause the processing device to carry out a method according to any of claims 1 to 13.