METHOD FOR PARAMETING A DATA PROCESSING LINE

Description

technical field

[0001] The present invention relates to a method for parameterizing a data processing chain.

[0002] The invention also relates to a computer program and a device implementing such a method.

[0003] The invention applies to the field of predictive maintenance. State of the art

[0004] As part of the implementation of an IoT project (from English "Internet of Things", or Internet of Things), it is known to have connected objects, in particular sensors, in a given environment to be monitored (for example, an installation), in order to make it "smart", that is to say capable of sending back data (such as measurements) representative of events which take place there.

[0005] Furthermore, predictive maintenance strategies are known to be implemented to anticipate potential anomalies within the environment, at the level of these objects (for example, malfunctions), in order to prevent failures and limit downtime. In this way, connected objects are part of a predictive maintenance system. Document FR 3111250 A1 discloses a method for detecting anomalies in a communication network, including the determination by a validation module of a reward value for a detected anomaly, positive if the anomaly corresponds to a failure, and negative otherwise. This reward value accompanies the measurement data in a validation message sent to the detection module, which evaluates a rate of positive rewards in order to adjust the frequency of the learning phases applied to the predictive model during the execution phase.In the event of a persistently low rate, the model is updated with a new training dataset or replaced.

[0006] However, such predictive maintenance systems do not provide complete satisfaction.

[0007] Indeed, known predictive maintenance systems are prone to false positives, meaning they can predict, based on data from one or more sensors, the occurrence of an anomaly within the monitored installation, even when the installation is actually behaving normally. This leads to unnecessary on-site interventions by operators responsible for maintaining the installation.

[0008] However, physical access to all or part of the installation is sometimes complex, for example due to the nature and profile of the terrain on which it is located (sewers, pipes, turbines, for example). Such interventions are therefore lengthy and costly, and are likely to result in direct losses for the operator. Furthermore, they cause unnecessary disruption for the operators.

[0009] Such false positives are also likely to trigger interventions requiring the plant to be shut down for maintenance, resulting in additional financial losses.

[0010] Therefore, the cost inherent in the management and maintenance of an IoT project is a real issue for communities and businesses, as this cost is likely to compromise the profitability of the project.

[0011] One object of the present invention is to remedy at least one of the drawbacks of the prior art.

[0012] Another objective of the invention is to offer a predictive maintenance solution that has a better ability to predict anomalies than existing solutions. Description of the invention

[0013] To this end, the invention relates to a method of the aforementioned type, the processing chain comprising a prediction stage implementing a current artificial intelligence model, previously trained on the basis of a training dataset, to predict an anomaly in a monitored environment equipped with at least one sensor, based on input data received from each sensor. The current artificial intelligence model is associated with a current accuracy score, representing a concordance between, on the one hand, a first predicted state of the monitored environment, determined by the current artificial intelligence model from test data dependent on the input data, and, on the other hand, an actual state of the monitored environment for said test data. The parameterization method is implemented by computer and includes the implementation of a reinforcement learning algorithm comprising the steps of: selection of a dataset from a set of datasets stored in memory, based on a reward associated with each dataset; training of an artificial intelligence model on the basis of the selected dataset to obtain an experimental artificial intelligence model; calculation of an experimental accuracy score of the experimental artificial intelligence model, representative of a concordance between, on the one hand, a second predicted state of the monitored environment, determined by the experimental artificial intelligence model from the test data, and, on the other hand, the actual state of the monitored environment for the test data; and based on a result of a comparison between the current accuracy score and the experimental accuracy score: replacement or not, in the prediction stage, of the current artificial intelligence model by the experimental artificial intelligence model;and update of the reward associated with the selected dataset.

[0014] Indeed, such a process results in a continuous optimization of the artificial intelligence model implemented for anomaly prediction, which gives it greater reliability than existing solutions.

[0015] Furthermore, thanks to the implementation of a reinforcement learning algorithm, the artificial intelligence model is trained on the dataset that yields the most accurate results with respect to the test data, which depends on the input data—that is, the data actually collected by the sensors. Therefore, the establishment and ongoing modification, by a human agent, of explicit rules relating to the datasets on which the artificial intelligence models are to be trained is unnecessary, which makes the method according to the invention very simple to implement.

[0016] Advantageously, the process according to the invention has one or more of the following characteristics, taken individually or in any technically feasible combination: If the experimental accuracy score is higher than the current accuracy score: the current artificial intelligence model is replaced by the experimental artificial intelligence model; and the update of the reward associated with the selected dataset is an increase in said reward; if the experimental accuracy score is lower than the current accuracy score: the current artificial intelligence model is not replaced by the experimental artificial intelligence model; the update of the reward associated with the selected dataset is a reduction in said reward; the selected dataset is the dataset associated with the maximum reward; the parameterization process further includes the steps of: comparing a current predicted state of the monitored environment, predicted by the current artificial intelligence model from the input data, with an actual state of the monitored environment;and updating the test data based on a comparison result; updating the test data includes adding to the test data all or part of the input data from which the current predicted state was determined, associated with a label representing whether the prediction is correct or incorrect. The parameterization process further includes the steps of: comparing a current predicted state of the monitored environment, predicted by the current artificial intelligence model from the input data, with an actual state of the monitored environment; in case of a discrepancy between the current predicted state and the actual state of the monitored environment, creating an additional dataset by modifying the dataset on which the current artificial intelligence model was trained, based on the predicted state and the actual state; and storing the created dataset in memory;The additional dataset created comprises the data from the dataset on which the current artificial intelligence model was trained, to which have been added the input data from which the current predicted state was determined, associated with a label representing whether the prediction is correct or incorrect; upon its creation, the created dataset is associated with a reward having a value greater than the value of the reward associated with each other dataset stored in memory; the parameterization process further includes the steps of: selecting at least a portion of the input data, preferably by implementing a feature selection process, to generate at least one additional dataset; and storing the generated dataset in memory; the reinforcement algorithm is a Q-learning algorithm, a Deep Q-learning algorithm, or a neural network.

[0017] According to another aspect of the invention, a computer program is proposed comprising executable instructions which, when executed by computer, implement the steps of the process as defined above.

[0018] The computer program can be in any computer language, such as for example machine language, C, C++, JAVA, Python, etc.

[0019] According to another aspect of the invention, a parameterization device for a data processing chain is proposed, the processing chain comprising a prediction stage implementing a current artificial intelligence model, previously trained on the basis of a training dataset, to predict an anomaly in a monitored environment equipped with at least one sensor, from input data received from each sensor, the current artificial intelligence model being associated with a current accuracy score, representative of a concordance between, on the one hand, a first predicted state of the monitored environment, determined by the current artificial intelligence model from test data dependent on the input data, and, on the other hand, a real state of the monitored environment for said test data, the parameterization device being configured to implement a reinforcement algorithm comprising the steps of: selection of a dataset from a set of datasets stored in a memory, according to a reward associated with each dataset; training of an artificial intelligence model on the basis of the selected dataset to obtain an experimental artificial intelligence model;calculation of an experimental accuracy score for the experimental artificial intelligence model, representative of the agreement between, on the one hand, a second predicted state of the monitored environment, determined by the experimental artificial intelligence model from the test data, and, on the other hand, the actual state of the monitored environment for the test data; and, depending on the result of a comparison between the current accuracy score and the experimental accuracy score: replacement or not, in the prediction stage, of the current artificial intelligence model by the experimental artificial intelligence model; and updating of the reward associated with the selected dataset.

[0020] The device according to the invention can be any type of device such as a server, a computer, a tablet, a calculator, a processor, a computer chip, programmed to implement the method according to the invention, for example by executing the computer program according to the invention. Brief description of the figures

[0021] The invention will be better understood upon reading the following description, given solely by way of non-limiting example and made with reference to the accompanying drawings in which: there figure 1 is a schematic representation of a processing chain associated with a parameterization device according to the invention; the figure 2 is a flowchart of a parameterization process implemented by the device of the figure 1 .

[0022] It is understood that the embodiments described below are by no means exhaustive. In particular, variants of the invention may be conceived comprising only a selection of the features described below, isolated from the other features described, if this selection of features is sufficient to confer a technical advantage or to differentiate the invention from the prior art. This selection includes at least one feature, preferably a functional one without structural details, or with only a portion of the structural details if this portion alone is sufficient to confer a technical advantage or to differentiate the invention from the prior art.

[0023] In particular, all the variants and embodiments described can be combined with each other if there are no technical obstacles to this combination.

[0024] In the figures and in the rest of the description, elements common to several figures retain the same reference. Detailed description

[0025] A parameterization device 2 according to the invention, for parameterizing a data processing chain 4 (hereinafter referred to as the "processing chain"), is illustrated by the figure 1 .

[0026] The processing chain 4 is configured to process a stream 6 of input data (also called "input data"), received from at least one sensor 8, in particular from at least one connected object.

[0027] Each sensor 8 is intended for monitoring a predetermined environment 9 (referred to as the "monitored environment"), such as a facility, machine, or device. More specifically, the input data 6 comprise measurements and observations representative of a state of the monitored environment 9.

[0028] The processing chain 4 includes a prediction stage 10 adapted to, in operation, implement a current artificial intelligence model 12 in order to predict, from the input data 6, the occurrence of an anomaly at the level of the monitored environment 9. More generally, the prediction stage 10 is preferably configured to implement the current artificial intelligence model 12 in order to predict, from the input data 6, a future state of the monitored environment 9, called the "current predicted state".

[0029] The current artificial intelligence model 12 was previously trained on the basis of a corresponding training dataset.

[0030] In addition, the current artificial intelligence model is associated with a current accuracy score PC, representative of a concordance between, on the one hand, a first predicted state of the monitored environment, determined by the current artificial intelligence model 12 from test data 14 depending on the input data 6, and, on the other hand, a real state of the monitored environment 9 for said test data (i.e. a state of the monitored environment 9 for which the input data 6 are identical to the test data).

[0031] The evolution of such test data over time will be described later.

[0032] The parameterization device 2 is intended to modify a parameterization of the processing chain 4, and more particularly of its prediction stage 10. More specifically, the parameterization device 2 is intended to modify the parameterization of the processing chain 4 according to predictions made by the processing chain 4 from the input data 6.

[0033] The parameter setting device 2 may be in a physical form, such as a computer, a server, a processor, an electronic chip, etc. Alternatively, or additionally, the parameter setting device 2 may be in a software form, such as a computer program, or an application, for example an application for a user device such as a tablet or smartphone.

[0034] The parameter setting device 2 is associated with a memory 16 configured to store the test data 14.

[0035] Memory 16 is also configured to store at least one dataset 18. Each dataset 18 is associated with a respective reward, also stored in memory 16. Obtaining such datasets 18 will be described later.

[0036] In addition, memory 16 is configured to store at least one artificial intelligence model 20, referred to as "experimental". Each experimental artificial intelligence model 20 has been previously trained on the basis of a corresponding dataset 18 from among the set of datasets 18 stored in memory 16.

[0037] To configure the processing chain 4, the parameterization device 2 is configured to implement a parameterization process 21, schematically illustrated by the figure 2 .

[0038] The parameterization procedure 21 is likely to be implemented several times, successively over time.

[0039] As shown in this figure, the parameterization process 21 involves the implementation of a reinforcement algorithm comprising a step 22 of selecting a dataset (called "selection step"), a training step 24, a step 26 of calculating the accuracy score (called "calculation step") and a successive configuration step 28.

[0040] Optionally, the parameterization process 21 also includes a step 30 of enriching the test data (called the "enrichment step") and / or a step 32 of generating datasets (called the "generation step").

[0041] Preferably, the reinforcement algorithm implemented by the parameterization device 2 during the execution of parameterization process 21 is a Q-learning algorithm, a Deep Q-Learning algorithm, or a neural network. Selecting a dataset

[0042] The parameterization device 2 is configured to select, during the selection step 22, a dataset 18 from the set of datasets 18 stored in memory 16. Such a selection is made on the basis of the reward associated with each dataset 18.

[0043] For example, in the case where an administrator has configured setting device 2 to adopt a so-called "optimal choice" strategy, setting device 2 is configured to select dataset 18 associated with the maximum reward.

[0044] However, other strategies may be adopted. For example, parameter setting 2 may be configured to adopt an "exploratory" strategy. In this case, parameter setting 2 is configured to select a dataset 18 that is not necessarily associated with the maximum reward. The choice between an optimal choice strategy and an exploratory strategy depends, for example, on factors such as a variable following a probability distribution (such as a Bernoulli distribution) or the occurrence of an event, depending on the specific task in question. Training

[0045] Furthermore, the parameterization device 2 is configured to train, during training step 24, an artificial intelligence model based on the selected dataset 18. In this way, an experimental artificial intelligence model 20 is obtained.

[0046] The parameterization device 2 is also configured to save the experimental artificial intelligence model 20 obtained in memory 16.

[0047] The artificial intelligence model used in training step 24 is likely to be any artificial intelligence model known to a person skilled in the art, for example a neural network. Calculating the accuracy score

[0048] In addition, the parameterization device 2 is configured to calculate, during the calculation step 26, an experimental accuracy score PE of the experimental artificial intelligence model 20 obtained at the end of the training step 24.

[0049] The experimental accuracy score PE is representative of a concordance between, on the one hand, a second predicted state of the monitored environment, determined by the experimental artificial intelligence model 20 from the test data 14, and, on the other hand, the actual state of the monitored environment 9 for the test data (i.e. a state of the monitored environment 9 for which the input data 6 are identical to the test data). Configuration

[0050] In addition, the parameterization device 2 is adapted to configure, during the configuration step 28, the processing chain 4 according to the experimental accuracy score PE determined at the end of the calculation step 26.

[0051] More specifically, parameterization device 2 is configured to compare the current accuracy score PC and the experimental accuracy score PE.

[0052] Furthermore, parameter setting 2 is configured to, depending on the result of the comparison: order the replacement or not, in the prediction stage 10, of the current artificial intelligence model 12 by the experimental artificial intelligence model 20; and update the reward associated with the selected dataset 18.

[0053] Advantageously, if the experimental accuracy score is higher than the current accuracy score, parameter setting 2 is preferably configured to: order the replacement of the current artificial intelligence model 12 with the experimental artificial intelligence model 20; and increase the reward associated with the dataset 18 selected during the selection step 22 on the basis of which the experimental artificial intelligence model 20 was obtained.

[0054] This is advantageous, since the artificial intelligence model implemented by the processing chain is the most efficient model with regard to the test data.

[0055] Advantageously, if the experimental accuracy score PE is strictly less than the current accuracy score PC, parameter setting 2 is configured to: command the maintenance of the current artificial intelligence model 12 as the artificial intelligence model implemented by the prediction stage 10 of the processing chain 4; and reduce the reward associated with the dataset 18 selected during the selection stage 22 on the basis of which the experimental artificial intelligence model 20 was obtained.

[0056] This is advantageous, since the dataset 18 that led to a poorly performing trained model will be less likely to be selected during a subsequent iteration of the parameterization process 21. Enrichment of test data

[0057] As previously stated, the parameterization process 21 advantageously includes the enrichment step 30.

[0058] Preferably, the parameterization device 2 is configured to implement the enrichment step 30 following an intervention triggered by a prediction, by the current artificial intelligence model 12 of the processing chain 4, of the occurrence of an anomaly.

[0059] More specifically, when an operator is sent to the site in connection with an anomaly predicted by the processing chain 4, the operator is required to provide feedback on this intervention: the operator indicates whether this intervention was justified or not, that is to say whether the anomaly prediction was incorrect or not.

[0060] In particular, the parameterization device 2 is configured to, during the enrichment step 30, compare a current predicted state of the monitored environment 9, predicted by the current artificial intelligence model 12 from the input data 6, with a real state of the monitored environment.

[0061] In addition, parameter setting device 2 is configured to update test data 14 from a comparison result.

[0062] In this case, the parameterization device 2 is configured to enrich the test data by including the input data 6 on the basis of which the maintenance operation was triggered, associated with a label representing whether the prediction was correct or incorrect (i.e., representative of whether the predicted anomaly was true or not).

[0063] More generally, the parameterization device 2 is configured to implement the enrichment step 30 at each observation of the actual state of the monitored environment 9. In this case, the parameterization device 2 is configured to enrich the test data by including the input data 6 on the basis of which a state of the monitored environment 9, at the time of said observation, was predicted by the current artificial intelligence model 12. In this case, the input data 6 is associated with a label representing whether the prediction is correct or incorrect, the label being determined by said observation of the actual state of the monitored environment 9. Dataset generation

[0064] As mentioned previously, parameterization process 21 is also likely to include generation step 32.

[0065] Preferably, the parameterization device 2 is configured to implement the generation step 32 after an observation of the real state of the monitored environment 9. Such an observation is, for example, the result of an intervention, at the level of the monitored environment 9, triggered by a prediction of an anomaly by the current artificial intelligence model 12 of the processing chain 4.

[0066] The parameterization device 2 is specifically configured to generate, during the generation step 32, at least one new dataset 18 from the datasets 18 stored in memory 16.

[0067] In particular, the parameterization device 2 is configured to, during generation step 32, compare a current predicted state of the monitored environment, predicted by the current artificial intelligence model 12 from the input data 6, with the actual state of the monitored environment 9.

[0068] In addition, in the event of a discrepancy between the current predicted state and the actual state of the monitored environment 9, the parameterization device 2 is configured to create an additional dataset 18 by modifying the dataset on the basis of which the current artificial intelligence model 12 was trained (called the "current training dataset").

[0069] In particular, the parameterization device 2 is configured to modify the current training dataset to include the input data 6 on the basis of which the current predicted state was determined, associated with a label representing whether the prediction was correct or incorrect (i.e., representative of a concordance or not between the current predicted state and the actual observed state).

[0070] The parameterization device 2 is also configured to store the additional dataset 18 thus created in memory 16.

[0071] Advantageously, in this case, the parameterization device 2 is configured to associate the dataset 18 created, at its creation, with a reward having a value greater than the value of the reward associated with each other dataset 18 stored in memory 16.

[0072] Optionally, the parameterization device 2 is configured to generate, during the generation step 32, at least one new dataset 18 from the input data 6.

[0073] In this case, parameter setting 2 is configured to apply a selection process to the input data 6 in order to extract so-called "useful" data. By "useful data," we mean data providing relevant information for the predictions to be made, that is, enabling the artificial intelligence model to make reliable predictions.

[0074] To extract the useful data, parameter setting 2 is, in particular, configured to implement a feature selection process (in English, "feature selection"). Such a process is, for example, based on the use of a statistical analysis such as the X 2< (Chi-2) test, or on mutual information (“ mutual information " in English).

[0075] The parameterization device 2 is also configured to store the additional dataset 18 thus created in memory 16. Functioning

[0076] The operation of parameter setting device 2 will now be described with reference to the figure 2 .

[0077] During the selection step 22, the parameterization device 2 selects, based on the reward associated with each dataset 18, a dataset 18 from the set of datasets 18 stored in memory 16.

[0078] Then, during training step 24, the parameterization device 2 trains an artificial intelligence model based on the selected dataset 18. In this way, an experimental artificial intelligence model 20 is obtained.

[0079] Then, during calculation step 26, the parameterization device 2 calculates the experimental accuracy score PE of the experimental artificial intelligence model 20.

[0080] Then, during the configuration step 28, the parameterization device 2 configures the processing chain 4 according to the determined experimental accuracy score PE.

[0081] More specifically, depending on the result of the comparison, parameter setting 2: commands the replacement or not, in the prediction stage 10, of the current artificial intelligence model 12 by the experimental artificial intelligence model 20; and updates the reward associated with the selected dataset 18.

[0082] In addition, the input data 6, representative of the state of the monitored environment 9, are routed to the processing chain 4 (for determining the current predicted state of the monitored environment 9) and to the parameterization device 2.

[0083] Advantageously, the parameterization device 2 enriches the test data 14 over time, by implementing the enrichment step 30.

[0084] More specifically, after at least one observation of the actual state of the monitored environment 9, the parameterization device 2 compares the current predicted state of the monitored environment 9 with the actual state of the monitored environment.

[0085] Furthermore, parameterization device 2 updates the test data 14 based on a comparison result. In particular, parameterization device 2 enriches the test data by including the input data 6 on the basis of which the current predicted state of the monitored environment 9 was determined, associated with a label representing whether the prediction is correct or incorrect.

[0086] In this way, test data 14 is enriched, in order to optimize the choice of the artificial intelligence model implemented by the processing chain 4.

[0087] Advantageously, the parameterization device 2 generates datasets 18 over time, by implementing the generation step 32, in order to arrive at increasingly efficient trained artificial intelligence models.

[0088] More specifically, after at least one observation of the actual state of the monitored environment 9, the parameterization device 2 compares the current predicted state of the monitored environment 9 with the actual state of the monitored environment 9.

[0089] In addition, in the event of a discrepancy between the current predicted state and the actual state of the monitored environment 9, the parameterization device 2 modifies the current training dataset to include the input data 6 on the basis of which the current predicted state was determined, associated with a label representing whether the prediction was correct or incorrect (i.e., representative of a concordance or not between the current predicted state and the observed actual state).

[0090] In this way, a new dataset 18 is obtained, on the basis of which an artificial intelligence model can be trained during the training step 24.

[0091] Of course, the invention is not limited to the examples that have just been described.

Claims

1. Method (21) for configuring a data processing chain (4), the processing chain (4) comprising a prediction stage (10) implementing a current artificial intelligence model (12) trained beforehand on a training dataset, in order to predict an anomaly in a monitored environment (9) provided with at least one sensor, on the basis of input data (6) received from each sensor, the current artificial intelligence model (12) being associated with a current accuracy score representative of a concordance between, on the one hand, a first predicted state of the monitored environment (9), which first predicted state is determined by the current artificial intelligence model (12) on the basis of test data (14) which are dependent on the input data (6), and, on the other hand, an actual state of the monitored environment (9) for said test data (14), the configuring method being computer-implemented and comprising implementing a reinforcement algorithm comprising the steps of: - selecting a dataset (18) from a group of datasets (18) stored in a memory, on the basis of a reward associated with each dataset (18); - training an artificial intelligence model on the selected dataset (18) in order to obtain an experimental artificial intelligence model (20); - computing an experimental accuracy score for the experimental artificial intelligence model (20), which experimental accuracy score is representative of a concordance between, on the one hand, a second predicted state of the monitored environment (9), which second predicted state is determined by the experimental artificial intelligence model (20) on the basis of the test data (14), and, on the other hand, the actual state of the monitored environment (9) for the test data (14); and - on the basis of a result of a comparison between the current accuracy score and the experimental accuracy score: • replacing or not the current artificial intelligence model (12) with the experimental artificial intelligence model (20) in the prediction stage (10); and • updating the reward associated with the selected dataset (18).

2. Configuring method (21) according to claim 1, wherein, if the experimental accuracy score is higher than the current accuracy score: - the current artificial intelligence model (12) is replaced with the experimental artificial intelligence model (20); and - the update of the reward associated with the selected dataset (18) is an increase of said reward.

3. Configuring method (21) according to claim 1 or 2, wherein, if the experimental accuracy score is lower than the current accuracy score: - the current artificial intelligence model (12) is not replaced with the experimental artificial intelligence model (20); - the update of the reward associated with the selected dataset (18) is a decrease of said reward.

4. Configuring method (21) according to any of claims 1 to 3, wherein the selected dataset (18) is the dataset (18) associated with the maximum reward.

5. Configuring method (21) according to any of claims 1 to 4, further comprising the steps of: - comparing a current predicted state of the monitored environment (9), which current predicted state is predicted by the current artificial intelligence model (12) on the basis of the input data (6), with an actual state of the monitored environment (9); and - updating the test data (14) on the basis of a result of the comparison.

6. Configuring method (21) according to claim 5, wherein updating the test data (14) comprises adding to the test data (14) some or all of the input data (6) on the basis of which the current predicted state was determined and which are associated with a label representative of whether the prediction is correct or incorrect.

7. Configuring method (21) according to any of claims 1 to 6, further comprising the steps of: - comparing a current predicted state of the monitored environment (9), which current predicted state is predicted by the current artificial intelligence model (12) on the basis of the input data (6), with an actual state of the monitored environment (9); - in the event of a discrepancy between the current predicted state and the actual state of the monitored environment (9), creating an additional dataset (18) by modifying, on the basis of the predicted state and the actual state, the dataset (18) on which the current artificial intelligence model (12) was trained; and - storing the created dataset (18) in the memory.

8. Configuring method (21) according to claim 7, wherein the created additional dataset (18) comprises the data from the dataset (18) on which the current artificial intelligence model (12) was trained and to which were added the input data (6) on the basis of which the current predicted state was determined, said input data being associated with a label representative of whether the prediction is correct or incorrect.

9. Configuring method (21) according to claim 7 or 8, wherein, on creation, the created dataset (18) is associated with a reward having a value greater than the value of the reward associated with each of the other datasets (18) stored in the memory.

10. Configuring method (21) according to any of claims 1 to 9, further comprising the steps of: - selecting at least some of the input data (6), preferably by implementing a feature selection process, in order to generate at least one additional dataset (18); and - storing the generated dataset (18) in the memory.

11. Configuring method (21) according to any of claims 1 to 10, wherein the reinforcement algorithm is a Q-learning algorithm, a Deep Q-Learning algorithm or a neural network.

12. Computer program comprising executable instructions which, when they are executed by a computer, implement the steps of the method (21) according to any of claims 1 to 11.

13. Device (2) for configuring a data processing chain (4), the processing chain (4) comprising a prediction stage (10) implementing a current artificial intelligence model (12) trained beforehand on a training dataset, in order to predict an anomaly in a monitored environment (9) provided with at least one sensor, on the basis of input data (6) received from each sensor, the current artificial intelligence model (12) being associated with a current accuracy score representative of a concordance between, on the one hand, a first predicted state of the monitored environment (9), which first predicted state is determined by the current artificial intelligence model (12) on the basis of test data (14) which are dependent on the input data (6), and, on the other hand, an actual state of the monitored environment (9) for said test data (14), the configuring device (2) being set up to implement a reinforcement algorithm comprising the steps of: - selecting a dataset (18) from a group of datasets (18) stored in a memory, on the basis of a reward associated with each dataset (18); - training an artificial intelligence model on the selected dataset (18) in order to obtain an experimental artificial intelligence model (20); - computing an experimental accuracy score for the experimental artificial intelligence model (20), which experimental accuracy score is representative of a concordance between, on the one hand, a second predicted state of the monitored environment (9), which second predicted state is determined by the experimental artificial intelligence model (20) on the basis of the test data (14), and, on the other hand, the actual state of the monitored environment (9) for the test data (14); and - on the basis of a result of a comparison between the current accuracy score and the experimental accuracy score: • replacing or not the current artificial intelligence model (12) with the experimental artificial intelligence model (20) in the prediction stage (10); and • updating the reward associated with the selected dataset (18).