Computer-implemented method for analysing water quality
Patent Information
- Application Number
- EP2024701448
- Authority / Receiving Office
- EP · EP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-01-27
- Filing Date
- 2024-01-26
- Publication Date
- 2025-12-03
AI Technical Summary
Current methods for monitoring fecal contamination indicator bacteria in water, such as Escherichia coli and intestinal enterococci, are too slow for effective management of bathing water quality, as they do not distinguish between viable culturable and non-culturable bacteria, which are crucial for regulatory decisions.
A computer-implemented method that acquires environmental parameters and uses a statistical regression module, trained with machine learning algorithms, to estimate the quantity of viable culturable bacteria in water samples, allowing for rapid and accurate detection.
Enables quick and robust estimation of viable culturable bacteria, facilitating timely decisions on water quality management by distinguishing between viable culturable and non-culturable populations.
Smart Images
Figure EP2024051980_02082024_PF_FP
Abstract
Description
Description Title of the invention: Computer-implemented method for analyzing water quality
[0001] The invention relates to a computer-implemented method for analyzing water quality. More particularly, the invention relates to a computer-implemented method for estimating a parameter representative of a quantity of viable culturable fecal contamination indicator bacteria in a sample of collected water.
[0002] Monitoring of faecal contamination indicator bacteria in recreational waters is an important public health measure to minimize waterborne diseases. Thus, the European Bathing Water Directive 2006 / 7 / EC requires monitoring of Escherichia coli and intestinal enterococci, whose presence in the water indicates faecal contamination. Although these microbial germs do not in themselves pose a danger to bathers at the usually high thresholds, their presence can indicate the simultaneous presence of pathogenic germs. It is therefore relatively important to measure the concentration of these bacteria in the water.
[0003] In order to quantify the presence of these bacteria in water, several measurement methods are known.
[0004] In particular, a miniaturized method using microplates for seeding in liquid media is known (respectively ISO 9308-3 and 7899-1 for Escherichia coli and enterococci). This standardized method is based on the ability of these bacteria to multiply in a suitable environment. However, it has the disadvantage of a long response time, generally between 36 and 72 hours, which is relatively unsuitable for active management of bathing waters. Indeed, deteriorations in water quality are generally punctual and dissipate quickly depending on hydroclimatic conditions. Water quality analysis measurements must therefore be obtained as quickly as possible to avoid exposing the public to pathogens, particularly in swimming pools and beaches.
[0005] Over the past few decades, new methods have been developed to reduce this response time and enable decisions to be made within hours of sample collection.
[0006] Enzymatic methods developed in recent years are based on the same principle as the previous miniaturized microplate method. Each of the two bacteria has an enzyme that is quasi-specific to it: PD-glucuronidase for E. coli and PD-glucosidase for enterococci. These enzymes have the ability to hydrolyze artificial compounds by releasing a colored or fluorescent product that allows visual quantification of the bacteria initially present in the sample. Many protocols and detection devices have been developed in recent years based on this principle. The response times of these different devices are then shorter than those observed for the miniaturized microplate technique but remain too long to allow active management of bathing water. These times are in fact between 12h and 24h depending on the method used to obtain a result <100 bacteria / 100ml.
[0007] Various methods have been developed to further reduce analysis time. In particular, we know of so-called polymerase chain reaction techniques, generally abbreviated as PCR, which allow for rapid and relevant results.
[0008] The quantitative PCR method, known as qPCR, is based on the detection and quantification of a gene section of the target bacteria. Bacteria are quantified using fluorescent molecules or probes that are released in the same proportion as the targeted genes after amplification cycles. It is a sensitive and specific method that also detects bacteria with repressed or unexpressed glucuronidase or glucosidase activity, and requires less than 3 hours to produce a result.
[0009] However, the problem with this type of method is that it is not discriminatory with respect to the different physiological states of bacteria that can be encountered in a bacterial population: culturable, viable, senescent and dead bacteria.
[0010] The RT-qPCR (Reverse Transcriptase - quantitative Polymerase Chain Reaction) method is based on the extraction and amplification of ribonucleic acid (or RNA), RNA being more representative of cell viability than DNA. The method consists of extracting the target RNA, then retranscribing it into a complementary DNA strand. This strand complementary DNA, often abbreviated as cDNA, then serves as a template for a conventional quantitative PCR reaction. This is a sensitive and specific method that detects only viable bacteria and can provide results on both regulatory bacteria E. coli and fecal enterococci in less than 3 hours and for up to forty samples simultaneously.
[0011] However, the RT-qPCR method has the disadvantage of not discriminating between viable culturable and viable non-cultivable bacterial populations, whereas it is useful to know only the viable culturable population in certain regulatory contexts.
[0012] A viable, non-culturable bacterium is a living bacterium that has lost its ability to multiply on culture media. Bacterial cells enter a viable, non-culturable state when they are subjected to stress related to environmental factors in the medium in which they are found.
[0013] However, in many regulatory frameworks, these non-culturable viable bacteria are excluded from the concentrations taken into account when deciding whether a "recreational" body of water, such as a swimming pool, or the sea near a beach, should be closed to the public due to bacterial pollution.
[0014] There is therefore a need for a method for quantifying bacteria, particularly Escherichia coli and intestinal enterococci, which is rapid, simple to use, and allows the quantification of viable culturable bacterial populations.
[0015] To this end, a computer-implemented method is proposed for estimating a parameter representative of a quantity of viable culturable faecal contamination indicator bacteria in a sample of water taken, comprising:
[0016] - A step of acquiring values representative of a plurality of environmental parameters associated with the collection of said water sample;
[0017] - A step of quantifying the viable bacteria present in the said water sample taken;
[0018] - A step of implementing a statistical regression module adapted to estimate a representative value of the quantity of viable cultivable bacteria based on the representative values of the acquired environmental parameters and said quantification of viable bacteria in said water sample.
[0019] This allows us to quickly and robustly estimate the quantity of viable bacteria that can be cultivated in a sample of water taken.
[0020] In particular, the method then implements a display step or a transmission step of the representative value of the estimated quantity of viable cultivable bacteria.
[0021] In particular, the display can be achieved, for example, by a human-machine interface such as a screen, on any type of digital medium or on a printed medium.
[0022] The transmission of the representative value of the estimated quantity of viable cultivable bacteria may be a digital transmission, for example via a communication network, wired or wireless, to a remote server or to a database.
[0023] Advantageously and in a non-limiting manner, the step of quantifying the viable bacteria present in said sample of water taken comprises the implementation of a polymerase chain reaction method, in particular a real-time quantitative polymerase chain reaction. Thus, a quantification result can be obtained particularly quickly.
[0024] In particular, said bacteria indicative of fecal contamination comprise a fecal enterococcus bacterium and / or an Escherichia coli bacterium.
[0025] Advantageously, the environmental parameters are chosen from: a parameter representative of the salinity of the water, a parameter representative of the turbidity of the water, a parameter representative of the water height, a parameter representative of the wind, a parameter representative of the solar radiation, a parameter representative of the rainfall and a parameter representative of the swell. These parameters are particularly relevant to allow a reliable estimation by the regression module.
[0026] In particular, for each environmental parameter, an instantaneous value is acquired, or an average value over a measurement period of said environmental parameter or a cumulative value of the measurements taken over a measurement period of said environmental parameter or a time series representative of a measurement period of said environmental parameter.
[0027] Advantageously, said statistical regression module is a machine learning module trained from a set of training data, in which each training data associates:
[0028] - Representative values of environmental parameters associated with a sample of water taken;
[0029] - A value representative of the quantity of viable bacteria, including viable culturable and non-culturable bacteria in said sample; and
[0030] - A value representative of a quantity of viable bacteria cultivable in the said water sample taken;
[0031] Said machine learning module being trained to infer, for each data set, said representative value of a quantity of viable cultivable bacteria from the associated environmental parameters and the representative value of the quantity of viable bacteria, including viable cultivable and non-culturable bacteria.
[0032] In particular, said machine learning module is chosen from a gradient boosting machine learning module, a support vector machine, an artificial neural network, a k-nearest neighbor method, a decision tree module, a decision tree forest machine learning module.
[0033] The invention also relates to a computer-implemented training method for a machine learning module for implementing a method as described above, comprising:
[0034] A step of providing a training data set, in which each training data associates:
[0035] - Representative values of environmental parameters associated with a water sample;
[0036] - A value representative of a quantity of viable bacteria, including viable culturable and non-culturable bacteria; and
[0037] - A value representative of a quantity of viable non-cultivable bacteria in said water sample;
[0038] - A step of training said machine learning module to infer, for each data set, said representative value of a quantity of viable cultivable bacteria from the associated environmental parameters and the representative value of the quantity of viable bacteria, including viable cultivable bacteria and non- cultivable.
[0039] The invention also relates to a device for estimating a parameter representative of a quantity of bacteria indicative of viable culturable fecal contamination in a sample of water taken, comprising:
[0040] - Means for acquiring a plurality of values representative of environmental parameters associated with the collection of said water sample;
[0041] - Means of acquiring a quantification of the viable bacteria present in the said water sample taken;
[0042] - A unit for implementing a statistical regression module adapted to estimate a representative value of the quantity of viable cultivable bacteria based on the representative values of the acquired environmental parameters and said quantification of viable bacteria in said water sample.
[0043] The invention also relates to a device for training a machine learning module for implementing an estimation method as described previously, comprising:
[0044] - Means for providing a set of training data, in which each training data associates:
[0045] — Representative values of environmental parameters associated with a water sample;
[0046] — A value representative of a quantity of viable bacteria, including both culturable and non-culturable viable bacteria; and
[0047] — A value representative of a quantity of non-culturable viable bacteria in said water sample;
[0048] - A training unit of said machine learning module for inferring, for each data set, said representative value of a quantity of viable cultivable bacteria from the associated environmental parameters and a representative value of a quantity of viable bacteria, including viable cultivable and non-culturable bacteria.
[0049] Other features and advantages of the invention will emerge from reading the description given below of a particular embodiment of the invention, given for information purposes, but not as a limitation, with reference to the appended drawings in which:
[0050] Fig. 1 is a schematic representation of the devices for implementing the estimation method according to the invention and the training method according to the invention;
[0051] Fig. 2 is a flowchart of the steps of the estimation method according to the main embodiment of the invention; and
[0052] Fig. 3 is a flowchart of the steps of the training method according to the main embodiment of the invention.
[0053] Figures 1 to 3 relate to the same embodiment and will be commented on simultaneously.
[0054] The invention relates to a computer-implemented method 1 for estimating the quantity of viable culturable bacteria in a sample of collected water.
[0055] In the main embodiment of the invention, method 1 aims to detect E. coli bacteria or intestinal enterococci, but it could be adapted to the detection of any other bacteria indicative of fecal contamination.
[0056] Thus, within the meaning of the invention, the general term viable bacteria means bacteria indicating viable fecal contamination.
[0057] Generally speaking, method 1 according to the invention is suitable in its implementation for a type of bacteria and a type of water sampled.
[0058] Thus, by way of example, and in a non-limiting manner, several methods 1 can be implemented: a method 1 according to the invention for quantifying viable cultivable E. coli bacteria and a method 1 according to the invention for quantifying viable cultivable intestinal enterococci bacteria. The two quantifications obtained make it possible to verify whether the regulatory health conditions are met.
[0059] Similarly, different processes can be implemented depending on the water sampling environment. For example, processes can be provided for seawater environments and processes for freshwater environments.
[0060] Thus, in an exemplary embodiment of the invention, four different methods 1 can be provided: a method 1 for quantifying E. Coli for seawater, a method 1 for quantifying E. Coli for freshwater, a method 1 for quantifying intestinal enterococci for seawater and a method 1 for quantifying intestinal enterococci for freshwater.
[0061] However, method 1 in its main embodiment, whatever the type of bacteria and type of water sampled is implemented as follows.
[0062] First of all, a step 10 is implemented for acquiring representative values of environmental parameters associated with the collection of said water sample.
[0063] The set of environmental parameters selected is intended to represent the environmental conditions in which the water was taken.
[0064] The representative values of these environmental parameters are, in the main embodiment of the invention, average values of the measurements carried out over determined measurement periods or cumulative values over a defined measurement period, depending on the environmental parameters, or instantaneous values.
[0065] Instantaneous values are preferably acquired at the time of sampling, for example for salinity or turbidity.
[0066] Depending on the parameters, it will also be possible to obtain an average value, for example for swell, wind, or cumulative values, for example for rain or solar radiation.
[0067] However, the values could be, in an alternative embodiment, time series, or a combination of time series, instantaneous values, average values and cumulative values.
[0068] For the remainder of the description, the "measurement period" is defined as a time period beginning before the water sample is taken and ending at the time of sampling. Thus, a 12-hour measurement period is understood as the acquisition of measurement data beginning twelve hours before sampling and ending at the time of sampling.
[0069] However, the invention is not limited to this sole interpretation of the measurement period, the measurement periods being able in particular to be shifted to end before or after the sample.
[0070] Indeed, measurements offset in time, for example ending one hour before or one hour after the sample, for example, could prove to be just as reliable within the framework of the invention.
[0071] Each environmental parameter measured can be measured during its own measurement period, possibly different from the other environmental parameters measured.
[0072] The invention is not limited to a fixed list of environmental parameters, but, in the main embodiment of the invention, the following environmental parameters are measured:
[0073] - the salinity of the water, measured at the time of sampling. Typically, salinity is measured using a conductivity meter or salinity probe. These devices measure the ability of water to conduct electricity, which is directly related to the concentration of dissolved salts in the water. The higher the salt concentration, the more conductive the water is. Salinity is typically expressed in electrical conductivity (EC) units, and in conductivity units in microsiemens per centimeter (pS / cm). Note that water temperature affects conductivity, and it is usually necessary to correct conductivity measurements for temperature to obtain an accurate salinity measurement.
[0074] - the turbidity of the water, measured at the time of sampling. Turbidity is generally measured using a turbidimeter, which uses a beam of light through the water sample and measures the amount of light that is reflected or absorbed by the particles suspended in the water. The more particles there are, the more light is reflected and absorbed, and the higher the turbidity. This turbidity is generally quantified by quantities dimensioned in NTU (Nephelometric Turbidity Units) or FTU (Formazine Turbidity Units) or by the amount of suspended matter generally quantified in mg / l or ppm.
[0075] - the wind, measured or modeled, and averaged over a measurement period ranging from 1h to 24h, in this embodiment the measurement period being 12h and 24h before sampling. To quantify the wind associated with the sample, a wind speed is acquired, in meters / second. However, it is possible, as an alternative, to combine a wind speed and a general wind direction value, in degrees. Generally speaking, these representative wind data can be measured directly in the vicinity of the water sample collection location or acquired via external data on remote databases, or are extracted from weather modeling and prediction systems.
[0076] - rainfall, in particular values representing cumulative rainfall and maximum hourly intensities over measurement periods ranging from 1 to 72 hours before each sample.
[0077] - solar radiation, calculated as the sum of the radiation solar (in Watts / m2), over a measurement period generally starting at midnight the day before the sampling day and ending when the water sample is taken.
[0078] - waves, or swell, when the sample is taken in an area with significant waves, such as an ocean area. In this embodiment of the invention, waves, or swell, are measured in the case of seawater measurement. However, the invention is not limited to this particular choice. The average wave height measured or modeled at the point closest to the sampling point, over a measurement period ranging from 1 to 24 hours before the sample, here 6 hours, 12 hours and 24 hours. In general, we calculate the average wave height by modeling it at the sampling point, based on data obtained from external sources. In an alternative embodiment, it could be provided that the data representative of the waves combines both an average height and a wave direction, obtained for example from numerical models.
[0079] - the water height, linked to the tides, acquired at the time of sampling, is here taken from a water height forecast model, for example a model produced specifically for the measurement site, generally measured in meters.
[0080] In addition, according to a particular embodiment, it is possible to add to the environmental parameters location values of the sample, such as geolocation coordinates, or a name, such as a city name or a beach name.
[0081] According to an alternative embodiment, the water temperature can also be chosen from environmental parameters.
[0082] A quantification step 11 of viable faecal contamination indicator bacteria present in the said water sample taken is also implemented.
[0083] In this embodiment, the bacteria quantified are Escherichia coli, also called E. coli, and intestinal enterococci. However, the invention is not limited to these two bacteria alone, which nevertheless represent two particularly advantageous indicators of bacterial pollution of water.
[0084] This quantification step 11 is not necessarily carried out after the acquisition step 10. It can be implemented before or in parallel with acquisition step 10.
[0085] This quantification step 11 is in our embodiment of the invention implemented according to a method described in document EP3599283A1 “Method for evaluating fecal pollution in water”, which discloses an RT-qPCR method for quantifying the bacteria indicating fecal contamination sought.
[0086] The invention is not limited to this particular method and any rapid quantification method, making it possible to detect viable faecal contamination indicator bacteria, such as E. coli or faecal enterococci, can be implemented.
[0087] Quantification step 11 provides the cycle threshold value, abbreviated Ct for "cycle threshold", which corresponds to the number of amplification cycles required to reach a fluorescence threshold value. This intersection point confirms the presence of the target in the sample and determines its concentration. The Ct is inversely proportional to the amount of genetic material present in the sample.
[0088] This threshold cycle value can then be transformed, by regression, into a bacteria count.
[0089] However, whatever the technique used, it is well known that in rapid or semi-rapid approaches, for example over periods of less than 12 hours, the quantifications obtained do not allow the discrimination of viable cultivable bacteria from viable non-cultivable bacteria.
[0090] However, for regulatory purposes, it is necessary to know only the viable cultivable bacteria that are taken into account in the regulations for recreational bathing waters.
[0091] To this end, a statistical regression module is then implemented 12 adapted to estimate a representative value of the quantity of viable cultivable bacteria as a function of the representative values of the acquired environmental parameters and of said quantity of viable bacteria in said water sample.
[0092] In the main embodiment of the invention, the regression module is adapted to estimate a count of viable cultivable bacteria, here sized in most probable number per 100ml, abbreviated MPN per 100ml. However, any other unit could be used.
[0093] In an alternative embodiment, however, one could employ a regression module adapted to provide another quantification or count value.
[0094] This statistical regression module 50 is in this embodiment of the invention a machine learning module 50 by gradient amplification, better known in English as gradient boosting.
[0095] Gradient boosting is a machine learning algorithm used in particular to solve regression problems. It uses an optimization technique that iteratively adds new models to a set of existing models.
[0096] Each new model is generated to correct the errors made by previous models. The models are usually decision trees, but can be other types of models.
[0097] Gradient boosting is particularly powerful for regression tasks because it allows for the generation of highly accurate models by combining many weaker base models.
[0098] However, the invention is not limited to this particular machine learning module 50. Other regression techniques can be used, whether it is a support vector machine, called SVM, an artificial neural network, a k-nearest-neighbors method, a decision tree module, a machine learning module using a forest of decision trees, called Random Forest.
[0099] Advantageously, a machine learning module 50 trained specifically for the type of bacteria considered is generally implemented, this making it possible to obtain more precise results for each type of bacteria.
[0100] In particular, since European regulations require the monitoring of these two types of bacteria, in a particular implementation of the invention we implement two distinct detection methods 1 each comprising a distinct machine learning module trained separately.
[0101] Thus, a machine learning module 50 trained for E. coli bacteria and / or a machine learning module 50 trained for intestinal enterococcus bacteria is implemented, depending on the bacteria to be monitored.
[0102] Furthermore, it is also possible to train a 50-unit machine learning module per type of bacteria and per type of sampling medium, such as than a marine environment or a freshwater environment.
[0103] Indeed, in a non-salty water environment, the training parameters may be different, for example the swell, or the waves, may not be taken into account.
[0104] For the purposes of the invention, the term representative value of the quantity of viable cultivable bacteria may be an indicator of the quantity of viable cultivable bacteria, such as a count of the quantity of viable cultivable bacteria.
[0105] Since the quantity of viable cultivable bacteria can be directly calculated from the quantity of viable non-culturable bacteria, by subtracting from the total count that can be calculated from the threshold cycle value Ct obtained during quantification step 11, the representative value of the quantity of viable cultivable bacteria can also be a count of the quantity of viable non-culturable bacteria.
[0106] The method 1 optionally implements a step of displaying and / or transmitting 40 the value representative of the estimated quantity of viable cultivable bacteria, the displaying and transmission being able to be carried out alone or in combination.
[0107] In particular, the display 40 can be produced for example as a human-machine interface such as a screen, on any type of digital medium or on a printed medium.
[0108] The transmission 40 of the representative value of the estimated quantity of viable cultivable bacteria may be a digital transmission, for example via a communication network, wired or wireless, to a remote server or to a database. For example via an intranet network, or via the Internet.
[0109] The invention according to a particular embodiment of the invention, comprises a step of comparison between the quantity of viable cultivable bacteria determined and a predetermined threshold value, such as a pollution threshold value.
[0110] In this particular embodiment, it is also possible to implement a step of sending an alert to a user, such as a message on a human-machine interface, sending a written message, such as an email, communicating an alert frame to a remote server, when the value representing the quantity of viable cultivable bacteria is greater than a determined threshold, such as an alert threshold, so as to indicate contamination with a bacterium indicating fecal contamination.
[0111] However, to implement the method 1 according to the invention, it is necessary, when it is a machine learning module 50, to train T before its implementation.
[0112] For this purpose, the invention also relates to a training method 2 of the automatic learning module 50.
[0113] This training method firstly comprises a step 20 of providing a set of training data, in which each training data associates:
[0114] - Representative values of environmental parameters associated with a water sample;
[0115] - A value representative of a quantity of viable bacteria, including viable culturable and non-culturable bacteria; and
[0116] - A value representative of a quantity of viable bacteria that can be cultivated in said water sample;
[0117] The representative values of the environmental parameters associated with the water sampling are the same as those of the estimation method 1 described previously. It is indeed necessary to train the learning module on the same parameters as those which will then be provided in the implementation steps of the machine learning module.
[0118] In the main embodiment of the invention, the training data is provided so as to train the machine learning module 50 specifically for a type of bacteria indicative of fecal contamination, in particular E. coli, or intestinal enterococci bacteria. In this embodiment, in addition the module is trained specifically for a type of water sampled: seawater or freshwater.
[0119] However, the invention is not limited to such training data, and the model could be trained for several types of bacteria simultaneously and / or for several types of water simultaneously.
[0120] The method then comprises a training step 21 of said machine learning module 50 to infer, for each data set, said representative value of a quantity of viable cultivable bacteria from the associated environmental parameters and the representative value of the quantity of viable bacteria, including viable cultivable and non-culturable bacteria.
[0121] This training step is carried out, within the framework of gradient boosting, in the following manner:
[0122] First of all, an initialization step is implemented, during which a basic model, here a decision tree, is trained on the training data to produce predictions, here the prediction is the representative value of a quantity of viable bacteria that can be cultivated in said water sample.
[0123] An error calculation step. The errors made by the basic model are calculated by comparing the predictions obtained with the actual values.
[0124] We then proceed to a step of training the following models: a new model is trained to correct the errors made by the previous model using an optimization technique, such as a gradient descent algorithm.
[0125] The models are then combined. The trained models are combined to produce a final prediction. This combination can be done in different ways, such as using a weighted average or using a majority vote.
[0126] These steps are then repeated iteratively, for example over a predetermined number of iterations, or until a defined error rate is obtained.
[0127] The characteristics of the embodiments described above can be taken individually or in combination according to all technically possible combinations.
[0128] The estimation method according to the invention is implemented by a device 3 which comprises means 30 for acquiring the representative values of the environmental parameters, such as input-output communication means 30 which can be of any known type, Ethernet port, COM port, USB port, wired or wireless communication device, and storage means adapted to store in memory, either in RAM or on a storage medium, the representative values acquired.
[0129] The device s, here a computer, further comprises means 31 for acquiring a quantification of the viable bacteria present in said sample of water taken.
[0130] This quantification 31 is generally carried out by implementing RT-qPCR type quantification means.
[0131] The results of this RT-qPCR quantification are then acquired by the device 3 by the acquisition means 31 which may be identical or distinct from the acquisition means 30 of the values representative of the environmental parameters.
[0132] The device 3 also comprises a unit for implementing a statistical regression module 50 as described previously.
[0133] Here the implementation unit 32 is a processor 32, or any digital computer suitable for implementing a statistical regression module 50 according to the invention.
[0134] The training method 2 is implemented by a training device 4, such as a computer, comprising means 33 for supplying a set of training data, such as a storage medium, for example a hard disk or a flash memory or any input-output communication means suitable for transmitting training data.
[0135] The training device 4, here a computer, comprises a training unit 34 of the automatic learning module 50.
[0136] The training unit 34 is a processor, a computer, or any digital assembly suitable for training a machine learning module 50.
[0137] Furthermore, the training unit 34 can implement the training steps of the training method 2 with parallel computing processors, such as GPUs implemented in a general computing context, known in English by the acronym GPGPU, for General-Purpose computing on Graphics Processing Units.
Claims
Claims 1. Computer-implemented method (1) for estimating a parameter representative of a quantity of viable culturable faecal contamination indicator bacteria in a sample of water taken, comprising: - A step of acquiring (10) values representative of a plurality of environmental parameters associated with the collection of said water sample; - A step of quantifying (11) the bacteria indicating viable fecal contamination present in said sample of water taken; - A step of implementing (12) a statistical regression module (50) adapted to estimate a representative value of the quantity of viable cultivable faecal contamination indicator bacteria as a function of the representative values of the acquired environmental parameters and of said quantification of viable bacteria in said water sample.
2. Method (1) according to claim 1, characterized in that the step of quantifying (11) the bacteria indicating viable fecal contamination present in said sample of water taken comprises the implementation of a polymerase chain reaction method, in particular a quantitative polymerase chain reaction in real time.
3. Method (1) according to claim 1 or 2, characterized in that said bacteria indicating fecal contamination comprise a fecal enterococcus bacterium and / or an Escherichia coli bacterium.
4. Method (1) according to any one of claims 1 to 3, characterized in that the environmental parameters are chosen from: a parameter representative of the salinity of the water, a parameter representative of the turbidity of the water, a parameter representative of the water height, a parameter representative of the wind, a parameter representative of the solar radiation, a parameter representative of the rainfall and a parameter representative of the swell.
5. Method (1) according to claim 1 to 4, characterized in that, for each environmental parameter, an instantaneous value, an average value over a measurement period of said environmental parameter or a cumulative value of the measurements carried out over a measurement period of said environmental parameter is acquired (10). environmental parameter or a time series representative of a measurement period of said environmental parameter.
6. Method (1) according to any one of claims 1 to 5, characterized in that said statistical regression module (50) is a machine learning module (50) trained from a set of training data, in which each training data associates: o Representative values of environmental parameters associated with a sample of water taken; o A representative value of a quantity of viable bacteria, comprising viable cultivable and non-culturable bacteria in said sample; and o A representative value of a quantity of viable cultivable bacteria in said sample of water taken; Said machine learning module (50) being trained to infer, for each data set, said representative value of the quantity of viable cultivable bacteria from the associated environmental parameters and the representative value of the quantity of viable bacteria, comprising the viable cultivable and non-culturable bacteria.
7. Method (1) according to claim 6, characterized in that said machine learning module (50) is chosen from a machine learning module by gradient amplification, a support vector machine, an artificial neural network, a k nearest neighbors method, a decision tree module, a machine learning module by decision tree forest.
8. Computer-implemented training method (2) of a machine learning module for implementing a method according to any one of claims 6 or 7 comprising: - A step of providing (20) a set of training data, in which each training data associates: o Representative values of environmental parameters associated with a water sample; o A value representative of a quantity of viable bacteria, including viable culturable and non-culturable bacteria; and o A value representative of a quantity of viable non-culturable bacteria in said water sample; A training step (21) of said machine learning module (50) to infer, for each data set, said representative value of a quantity of viable cultivable bacteria from the associated environmental parameters and the representative value of the quantity of viable bacteria, including viable cultivable and non-culturable bacteria.
9. Device for estimating (3) a parameter representative of a quantity of bacteria indicating viable culturable faecal contamination in a sample of water taken, comprising: - Means for acquiring (30) a plurality of values representative of environmental parameters associated with the taking of said water sample; - Means for acquiring a quantification (31) of the viable bacteria present in said sample of water taken; - An implementation unit (32) of a statistical regression module (50) adapted to estimate a representative value of the quantity of viable cultivable bacteria as a function of the representative values of the acquired environmental parameters and of said quantification of viable bacteria in said water sample.
10. Training device (4) of a machine learning module (50) for implementing a method according to any one of claims 6 or 7, comprising: - Means for providing (33) a set of training data, in which each training data item associates: o Representative values of environmental parameters associated with a water sample; o A value representative of a quantity of viable bacteria, including viable culturable and non-culturable bacteria; and o A value representative of a quantity of viable non-culturable bacteria in said water sample; A training unit (34) of said machine learning module (50) for inferring, for each data set, said representative value of a quantity of viable cultivable bacteria from the associated environmental parameters and a representative value of a quantity of viable bacteria, including viable cultivable and non-culturable bacteria.