System and method for selecting an action strategy relating to at least one asset and associated control
Patent Information
- Application Number
- FR2023008563
- Authority / Receiving Office
- FR · FR
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-08-08
- Publication Date
- 2025-02-14
Smart Images
Figure 00000000_0000_ABST
Abstract
Description
Title of the invention: System and method for selecting an action strategy relating to at least one asset and associated control TECHNICAL FIELD OF THE INVENTION
[0001] The field of the invention is that of computing.
[0002] More specifically, the invention relates to a system and method for selecting an action strategy relating to at least one asset and associated control.
[0003] The invention finds applications in particular for determining an action relating to an asset such as a raw material, energy, a machine, an infrastructure. The asset can also be linked to an unalterable register, replicated on a multitude of IT infrastructures, such as a block chain, also known by the English term "blockchain". STATE OF THE ART
[0004] Methods for controlling actions relating to one or more assets are known from the prior art. They are used to carry out one or more actions, for example, to allocate resources in different assets in order to achieve specific objectives. Users of these methods therefore seek to control the action or set of actions optimized to achieve a pre-established objective, in a given context, in real time.
[0005] Among the existing methods, control methods based on a set of predefined rules can be used. However, the rules are general; they do not take into account the specificities of the real-time context. They are also not precisely adapted to each asset. In addition, they are not able to evaluate the effect of each action and their possible correlations. Their implementation is often very time-consuming. Thus, these methods are not suitable for controlling an optimized action in order to achieve a pre-established objective in real time.
[0006] There are also Markowitz models. These models, integrated into a control process, allow the optimization of asset allocation based on predefined objectives, for example, expected risk and return. Methods based on these models deal with the parallel management of several assets, allowing the selection and determination of actions around asset combinations. In other words, Markowitz models take into account the correlation between assets to make allocation decisions that meet a pre-established objective.
[0007] However, these models have the disadvantage of being too general, they are not unable to take into account the influence of the environment. In addition, they do not take into account all the actions that can be carried out. Thus, the optimization based on this model that could be carried out would be on a very limited field, and not very precise.
[0008] In recent years, predictive stock control methods have emerged. They use machine learning and data analysis techniques to predict the performance of stocks relating to an asset. However, these have a limited scope of application, they do not allow for comparing several stocks, determining the influence of the parameters constituting each of them, and being used in real time.
[0009] None of the current methods can meet all the required needs simultaneously. Statement of the invention
[0010] The present invention aims to remedy all or part of the drawbacks of the state of the art cited above.
[0011] The invention relates to a method for controlling an action relating to an asset, implemented by a computer system, comprising steps of: • Collection of data on the asset evolving over a PI period, • Determination of a plurality of action strategies to be carried out on the asset over the PI period, • Development of a database associating with each strategy, the asset data and at least one result of an evaluation over the PI period of said action strategy relating to the asset, • Generation of parameters of a machine learning algorithm trained on the previously developed database, • Prediction of a result over a period P2, distinct from PI, for each strategy using the machine learning algorithm, • Command of an action relative to the asset by automatically selecting the strategy with the best result.
[0012] The invention advantageously makes it possible to establish a command for an action relating to the asset linked to the most relevant strategy among a set of strategies.
[0013] Also, this invention makes it possible to take into account the influence of the environment of the asset, in particular of the period P2.
[0014] The fact that a set of strategies are compared with each other makes it possible to see the correlation between the effects of the different actions and the environment of the asset.
[0015] The scope of application of the invention is particularly broad insofar as a large field of strategies can be tested for a multitude of assets.
[0016] In a particularly advantageous embodiment, the step of determining a plurality of action strategies to be carried out over the period PI comprises the following sub-steps for each strategy: • development of at least one action associated with said strategy, and a time associated with said action, • determination of at least one performance indicator associated with said strategy, each strategy being distinct.
[0017] This mode of implementation is particularly advantageous because it makes it possible to generate numerous strategies very easily and quickly and to take into account the impact of several actions and to model it.
[0018] For example, the actions may be chosen from an action of a first type, and an action of a second type, opposite to the first type.
[0019] This advantageously makes it possible to control optimized actions as a function of the period P2 during which the method is implemented.
[0020] Alternatively or in addition, the step of ordering an action relating to the asset includes a sub-step of classifying the results obtained by each strategy. This particular mode of implementation makes it possible to better measure the performance of each of the strategies in relation to each other.
[0021] In particular embodiments of the invention, the step of developing a database comprises a sub-step of preprocessing the data. This implementation mode offers better precision of the results obtained compared to an alternative in which the database would not be preprocessed. This makes it possible to accelerate the convergence of the prediction step.
[0022] The invention also relates to a computer system for controlling an action relating to an asset, comprising means of: • Collection of data on the asset evolving over a PI period, • Determination of a plurality of action strategies to be carried out on the asset over the PI period, • Development of a database associating with each strategy, the asset data and at least one result of an evaluation over the PI period of said action strategy relating to the asset, • Generation of parameters of a machine learning algorithm trained on the previously developed database, • Prediction of a result over a period P2, distinct from PI, for each strategy using the machine learning algorithm, • Command of an action relative to the asset by automatically selecting the strategy with the best result.
[0023] In a particular embodiment, the means for determining a plurality of action strategies is configured to develop at least one action associated with said strategy, and a time associated with said action and determine at least one performance indicator associated with said strategy.
[0024] In a particularly advantageous embodiment, the control means is configured to automatically select the strategy having the best result.
[0025] In addition, the means for developing a database is configured to preprocess the data. BRIEF DESCRIPTION OF THE FIGURES
[0026] Other advantages, aims and particular characteristics of the present invention will emerge from the following non-limiting description of at least one particular mode of implementation of the methods which are the subject of the present invention, with reference to the appended drawings, in which: • [Fig.l] is a synoptic view of a mode of implementation of a method for controlling an action relating to an asset according to the invention; • [Fig.2] is a synoptic view of a step of determining a plurality of action strategies of the method for controlling an action relating to an asset of an embodiment of the invention; • [Fig.3] is a schematic view of an embodiment of the system in computer according to the invention. DETAILED DESCRIPTION OF THE INVENTION
[0027] The present description is given without limitation, each characteristic of an embodiment being able to be combined with any other characteristic of any other embodiment in an advantageous manner.
[0028] [Fig.l] illustrates a method 100 for controlling an action relating to an asset such as a resource (raw material, energy, etc.), a machine, or even an infrastructure. The value of the asset can be recorded in an unalterable register, replicated on a multitude of IT infrastructures, such as a blockchain.
[0029] The objective of the method 100 according to the invention is to determine the best action strategy to be carried out with regard to the asset with regard to a known history linked to the asset over a period PL. The history may in particular comprise data on the state of the asset during the period PI, data linked to the environment of the asset during the period PI, a valuation of the asset fluctuating during the period PI, etc.
[0030] Different action strategies are evaluated over this period PI in order to determine the best strategy to carry out over a period P2 distinct from PI, generally subsequent to PI, taking into account all or part of the history.
[0031] It should be emphasized that this strategy evaluation is particularly suited to stochastic assets whose evolution over time is weakly or even difficult to predict.
[0032] The control method 100 comprises a first step 110 of collecting data on the asset evolving over a period PI having a predetermined duration. The period PI is generally a continuous time interval but may be composed of a plurality of time intervals which may or may not overlap.
[0033] Several strategies are developed during a second step 120 of the method 100, which can be carried out concurrently with step 110.
[0034] A strategy can be defined as a set comprising at least one action to be carried out on the asset, or even at least one performance indicator. A time can also be associated with each action. This time, which can be distinct for each action, makes it possible to program the action(s) of the strategy. The time can be of an absolute type relative to a reference or of a relative type, such as a step between two actions.
[0035] The plurality of strategies can be determined in different ways: • either automatically, for example randomly according to at least one predetermined pattern; • either manually by a user; • either an automatic and manual combination.
[0036] For example, step 120 comprises the following substeps: • development 121 of at least one action associated with said strategy, and of a time associated with said action; • determination 122 of at least one performance indicator associated with said strategy.
[0037] It should be emphasized that the strategies are generally distinct from each other.
[0038] Once the action strategies have been determined, each strategy is evaluated over the period PI in order to obtain a result making it possible to evaluate at least one performance indicator of said strategy. A database is thus developed during this third step 130 of the method 100. The database comprises a plurality of records, each record comprising a strategy, an evaluation period (for example PI), the data of the asset over said evaluation period and at least one result of the evaluation of the strategy.
[0039] It should be noted that the evaluation period may be different on each sheet and that the same strategy may be evaluated over various periods.
[0040] Parameters of a machine learning algorithm are generated during a fourth step 140 of the control method 100, by training on the database linked to the strategies evaluated during step 130.
[0041] The machine learning algorithm can be a so-called supervised algorithm.
[0042] It may be an artificial neural network, for example of the convolutional type, the nodes of which each comprise at least one parameter which is generated during learning.
[0043] Learning can be done in several stages, by partitioning the database into several parts: at least one for learning and one for testing the parameters generated during learning.
[0044] Once the parameters of the machine learning algorithm have been refined on the basis of data during step 140, a prediction of a result over a period P2 distinct from PI can be made for each strategy during a fifth step 150 of the method 100.
[0045] A command for an action relating to the active can then be carried out during a sixth step 160 by automatically selecting the strategy having the best result.
[0046] It should be emphasized that this best result can be selected with regard to a previously chosen performance indicator. Depending on the performance indicator(s) chosen as a criterion, the selection of the strategy to be activated can thus be different. A performance indicator can for example be linked to a duration, an average, a peak value, etc.
[0047] This sixth step may comprise a sub-step 165 of ranking the results obtained for each strategy. Each result may be formalized in the form of a weight, i.e. a real number between 0 and 1. Thus, ranking the results obtained for each strategy makes it possible to better measure the performance of each of the strategies in relation to each other.
[0048] In the case where several performance indicators are associated with the same strategy, a result can be calculated for each performance indicator.
[0049] An overall result can be calculated by, for example, averaging all the results for each performance indicator for the same strategy.
[0050] It should be noted that the different performance indicator results for the same strategy can be weighted. This advantageously allows performance indicators to be given different importance. A weighting profile can then be created and used for all strategies.
[0051] If a performance indicator cannot be measured for a strategy, then the result associated with this performance indicator may be equal to 0.
[0052] Several strategies can thus be compared with each other, even if they do not have the same performance indicators.
[0053] The data on the asset collected during step 110 can be preprocessed during a sub-step 135 of step 130 of the development of the database. Said data can be formalized in the form of a table consisting of i rows and j columns. The table can include the values wij, corresponding to the result of each of the strategies over each test period, i corresponding to the tested strategy, j corresponding to the test period.
[0054] Pre-processing helps improve the training efficiency of the machine learning algorithm
[0055] The method may include an optional step 170 of formatting the results. These may be integrated in the form of graphs. Indicators measuring performance may be calculated. The method may also provide a graphical representation of the performance of the trading strategy.
[0056] The method can be implemented several times in a row to control a set of actions autonomously. Thus, a complementary step can calculate a performance value relative to a combination of strategies. More precisely, the method can then calculate the weights relative to the combination of actions. The combination of actions having the highest weight is considered to be the combination of actions having the best result.
[0057] As illustrated by [Fig. 3], the invention also relates to a computer system 200 for controlling an action relating to an asset. It comprises means of: • Collection of 210 data on the asset evolving over a PI period, • Determination 220 of a plurality of action strategies to be carried out on the asset over the PI period, • Development 230 of a database associating with each strategy, the asset data and at least one result of an evaluation over the PI period of said action strategy relating to the asset, • Generation of 240 parameters of a machine learning algorithm trained on the previously developed database, • Prediction 250 of a result over a period P2, distinct from PI, for each strategy using the machine learning algorithm, • Order 260 of an action relating to the asset by automatically selecting the strategy with the best result.
[0058] The means for implementing steps 210 to 250 may comprise a computer comprising a CPU, a RAM and a storage memory such as a cloud or a hard disk. It may also have a system for transmitting information to an actuator.
[0059] The computer system may also include an application with a Human Machine Interface (HMI), allowing to collect data from a user such as an asset to be considered, a period, and a performance indicator, and to format results.
[0060] The means for implementing a command 260 of an action relating to the asset can be carried out by the actuator according to the information transmitted by the computer.
[0061] The actuator is then defined as the element carrying out the action relating to the asset.
[0062] The means for determining a plurality of action strategies may be configured to develop at least one action associated with said strategy, and a time associated with said action and determine at least one performance indicator associated with said strategy.
[0063] In a particularly advantageous embodiment, the control means is configured to automatically select the strategy having the best result.
[0064] In addition, the means for developing a database is configured to preprocess the data.
[0065] Example implementation for an asset being a resource
[0066] In one embodiment, the invention relates to a method 100 for controlling an action relating to an asset such as a resource. For example, this resource may be a raw material or energy.
[0067] Many actions can be ordered. Among them, we can cite extraction, exploitation, storage or even distribution. It can also be production. The action relating to the resource can also be a purchase or sale of a given quantity of said resource.
[0068] It should be noted that the action may itself comprise two alternative actions. The choice between one or the other of the alternative actions may be made depending on a predefined condition which is met or not. For example, if the action concerns the addition of raw material from a tank, the predefined condition may be a sufficient empty volume in said tank to accommodate the raw material. Thus, if the condition is met, i.e. the tank has a sufficient empty volume to be filled by said raw material, the filling action may be commanded. If the condition is not met, an alternative action such as the routing of the raw materials to another tank may be carried out.
[0069] It should also be noted that the actions may be chosen from an action of a first type, and an action of a second type, opposite to the first type. An action opposite to a first action is an action which has an effect opposite to the first action.
[0070] An action of a first type may be a purchase of said asset, and an opposite action may be a sale of the same asset.
[0071] Similarly, an action of a first type may be a filling of a reservoir with a quantity of said active ingredient, and an action of a second type may be an evacuation from the reservoir of a quantity of the same active ingredient.
[0072] The actuator for performing these actions may, for example, be configured to extract raw materials at a given rate.
[0073] In this mode of implementation, the data collected during step 110 may concern a need for the resource, the price of this resource during the PI period, the environmental impact linked to the transport of the resource, or even data linked to security.
[0074] In this embodiment, various following performance indicators may be considered. A purchase price, a sale price, a moving average of a valuation of an asset as a function of time may in particular be defined as performance indicators.
[0075] Example of implementation for an asset being a machine or an infrastructure
[0076] In one embodiment, the invention relates to a method 100 for controlling an action relating to an asset such as a machine or infrastructure. For example, this machine may be a mechanical device requiring to be powered.
[0077] The action relating to the machine or infrastructure can also be its setting in motion, or its purchase, sale, distribution, or exchange.
[0078] The action relating to this asset can be its energy supply. The energy supply can be carried out from a choice between several energies.
[0079] The actuator can then be configured to distribute energy to this machine or this infrastructure, according to proportions established by strategies, for example depending on the energy sources.
[0080] The actuator may also be configured to enable energy storage in this infrastructure.
[0081] The actuator can also be configured to operate the machine or infrastructure, power them according to a particular pattern, or even set them in motion.
[0082] The data collected during step 110 may include position data, temperature data, pressure data, or other physical quantity of the machine or infrastructure. It may also be data on the operating mode of the machine or infrastructure, on its state, on its maintenance, or on the actions that it can perform.
[0083] Data on assets over past periods and / or in real time may be collected by monitoring devices. These may be sensors, regularly updated databases. Said sensors and databases may also provide data to determine possible actions on the asset.
[0084] The performance indicators, in this mode of implementation, may be, in a non-limiting manner, electricity consumption, environmental impact, the capacity of the machine or infrastructure to be powered by this energy, etc.
Claims
Claims
1. Method (100) for controlling an action relating to an asset, implemented by a computer system, comprising steps of: • Collection (110) of data on the asset evolving over a period PI, • Determination (120) of a plurality of action strategies to be carried out on the asset over the period PI, • Development (130) of a database associating with each strategy, the data of the asset and at least one result of an evaluation over the period PI of said action strategy relating to the asset, • Generation (140) of parameters of a machine learning algorithm trained on the previously developed database, • Prediction (150) of a result over a period P2, distinct from PI, for each strategy using the machine learning algorithm, • Control (160) of an action relating to the asset by automatically selecting the strategy having the best result.
2. Control method (100) according to claim 1, in which the step of determining a plurality of action strategies to be carried out over the period PI, comprises for each strategy the following sub-steps: • development (121) of at least one action associated with said strategy, and a time associated with said action, • determination (122) of at least one performance indicator associated with said strategy, each strategy being distinct.
3. A control method (100) according to claim 2 wherein the actions can be selected from an action of a first type, and an action of a second type, opposite to the first type.
4. Control method (100) according to one of the preceding claims in which the step (160) of controlling an action relating to the asset includes a sub-step of ranking the results obtained by each strategy.
5. Control method (100) according to one of the preceding claims, in which the step (130) of developing a database comprises a sub-step (135) of preprocessing the data.
6. Computer system (200) for controlling an action relating to an asset, comprising means for: • Collecting (210) data on the asset evolving over a period PI, • Determining (220) a plurality of action strategies to be carried out on the asset over the period PI, • Developing (230) a database associating with each strategy, the data of the asset and at least one result of an evaluation over the period PI of said action strategy relating to the asset, • Generating (240) parameters of a machine learning algorithm trained on the previously developed database, • Predicting (250) a result over a period P2, distinct from PI, for each strategy using the machine learning algorithm, • Controlling (260) an action relating to the asset by automatically selecting the strategy having the best result.
7. Computer system (200) according to claim 7 wherein the means for determining (220) a plurality of action strategies is configured to develop at least one action associated with said strategy, and a time associated with said action and determine at least one performance indicator associated with said strategy.
8. A computer system (200) according to one of claims 7 to 8, wherein the control means (260) is configured to automatically select the strategy having the best result.
9. Computer system (200) according to one of claims 7 to 9, in which the means (230) for developing a database is configured to preprocess the data.