FR3051943A1 - METHOD FOR ASSISTING THE DETERMINATION OF A PLAN THAT MEETS PREFERENCES - Google Patents

Abstract

The invention relates to a method for assisting the determination of the plan that best meets the preferences, the method being implemented by a computer and comprising the steps of: providing input data comprising the initial state, the state final search, possible actions, and constraints, - conversion of input data into an automatic planning language, - provision of a set of preferences, - for each preference, determination of a preference score function by analysis the evolution of the satisfaction according to the value of the corresponding preference, - determination of a global score function from the preference score functions, - conversion of the global score function into the same automatic planning language , and - generating at least one plane from the converted data and the converted function.

Description

Method of assisting the determination of a plan respecting preferences

The present invention relates to a method of assisting in the determination of the plan best respecting preferences. The invention also relates to a support system for determining the plan that best meets preferences. The invention also relates to an associated computer program product and an information carrier on which is encoded such a computer program product.

In many areas, it is desirable to have decision support tools because real-time decision-making, often with the consideration of a plurality of criteria, is difficult or impossible for a human being.

For example, document US 2006/0224426 A1 discloses a computer-implemented method for manipulating transport plans, the method comprising a step of accessing a series of transport orders and a step of accessing load plan actions that can be implemented. The method also includes a plane plane exception identification step relating to the transport plane and a step of automatically identifying a candidate plane action to resolve the plane exception. The method then selectively updates the load plan actions based on data from the identified candidate plan action.

However, such a method only makes it possible to manage an exception and is not adapted to handle a plurality of exceptions, let alone exceptions from different decision makers that can be interpreted as preferences.

There is therefore a need for a method of assisting the determination of plans, allowing the consideration of both constraints but also preferences, especially in the presence of at least two different decision makers. For this purpose, there is provided a method of assisting in the determination of the plan that best respects preferences, a plan being a series of actions leading a system from an initial state to a final state that is sought by respecting predefined constraints, the method being implemented by a computer and comprising the following steps: providing input data, the input data comprising: the initial state of the system, the desired final state of the system, the set of actions possible, and constraints, the conversion of input data into an automatic planning language to obtain converted data, the provision of a set of preferences, for each preference, determining a preference score function by analysis of the evolution of the satisfaction according to the value of the corresponding preference, the determination of a global score function from the functions s of preference score, the conversion of the global score function into the same automatic scheduling language to obtain a converted function, and the generation of at least one plan from the converted data and the converted function.

According to a particular embodiment, the method comprises one or more of the following characteristics, taken in isolation or in all the technically possible combinations: the set of possible actions are carried out by several distinct entities, the step of providing the preferences being conducted by at least two separate entities. the set of possible actions are carried out by several distinct entities, at least one entity providing a plurality of distinct sets of preferences at the supply stage, a game being a subset of the set of preferences. the global score function is obtained by implementing a multi-attribute utility model using an aggregation function, the aggregation function being an integral of Choquet or an integral of Choquet whose capacity is 2- additive. the automatic scheduling language includes preference management functions. a plurality of planes is generated at the generating step, the method comprising a step of displaying the plurality of planes and the value of each preference score function for each of the plurality of planes. the method being reiterated at least once, at least one of the group consisting of the provided preferences and the determination of the preference score functions being distinct from one iteration to another.

There is also provided a system for assisting the determination of the plan that best respects preferences, a plan being a series of actions leading a system from an initial state to a desired end state while respecting predefined constraints, the system comprising at least one computer and being adapted to receive input data, the input data comprising: the initial state of the system, the desired final state of the system, the set of possible actions, and the constraints, converting input data into an automatic scheduling language for obtaining converted data, receiving a set of preferences, for each preference, determining a preference score function by analyzing the evolution of the satisfaction as a function of the value of the corresponding preference, determine a global score function from the score preference functions, convert the scor function e giobal in the same automatic scheduling language to obtain a converted function, and generate at least one plan from the converted data and the converted function.

There is also provided a computer program product comprising software instructions which, when executed by a computer, implement a method as previously described.

It is also proposed an information carrier on which is stored a computer program product as previously described. The features and advantages of the invention will appear on reading the description which follows, given solely by way of nonlimiting example, and with reference to the appended drawings, in which: FIG. 1 is a diagrammatic view of an example of a system for implementing a method of assisting the determination of the plan best respecting preferences, and - Figure 2 is a flowchart of an example of implementation of a method of assisting determining the plan that best meets the preferences.

A system 10 and a computer program product 12 are shown in FIG. 1. The interaction of the computer program product 12 with the system 10 makes it possible to implement a method of assisting in the determination of the plan that best respects preferences.

The system 10 is a computer.

More generally, the system 10 is an electronic calculator adapted to manipulate and / or transform data represented as electronic or physical quantities in the registers of the system 10 and / or memories in other similar data corresponding to physical data in the data. memories, registers or other types of display, transmission or storage devices.

The system 10 includes a data processing unit 14 comprising a processor 16, memories 18 and an information carrier reader 20. The system 10 also comprises a keyboard 22 and a display unit 24.

The computer program product 12 comprises a readable information medium 20.

A readable information medium 20 is a medium readable by the system 10, usually by the information processing unit 14. The readable information medium 20 is a medium suitable for storing electronic instructions and capable of being coupled. to a bus of a computer system. By way of example, the readable information medium 20 is an optical disk, a CD-ROM, a magneto-optical disk, a ROM memory, a RAM memory, an EPROM memory, an EEPROM memory, a magnetic card or a memory card. optical card.

Alternatively, the readable information medium 20 is part of a cloud (also referred to as "cloud").

On the readable information medium 20 is stored the program comprising program instructions.

The computer program is loadable on the data processing unit 14 and is adapted to cause the implementation of a method of assisting the determination of the plan best respecting preferences when the computer program is put implemented on the processor 16. By the expression "respecting the best", it is understood as "optimizing".

The operation of the system 10 interacting with the computer program product 12 is now described with reference to FIG. 2 which illustrates an exemplary implementation of a method of assisting the determination of the plan best respecting preferences.

The illustrated method is a method of assisting the determination of the plan best respecting preferences.

A plan is a sequence of actions leading a system from an initial state to a desired end state under pre-defined constraints.

In the context illustrated, the plan involves a plurality of entities or stakeholders. Each entity with the power to influence (or even determine) the adapted plan is called a decision maker. To show the interest of the process, it is assumed that there are two decision makers.

The notions of 'plan' and 'preference' will be specified in the following description.

The method comprises a first supply step 100, a first conversion step 102, a second supply step 104, a first determination step 106, a second determination step 108, a second conversion step 110, a generation step 112 and a display step 114. At the first delivery step 100, input data is provided.

The input data include the initial state So of the system, the desired final state Sf of the system, the set of possible actions A and the constraints C. The initial state Sq represents the situation of the system before the implementation. plan.

In the illustrated example, initially, there is no manufactured ship.

On the other hand, some of the stakeholders already have some of the equipment to make the ship. The desired end state Sf represents the state of the system at the end of the implementation of the plan.

In the example, in the desired final state Sf, the ship is manufactured. The set of possible actions A collects the actions to switch from a system state to another state of the system.

For example, a stakeholder manufactures the rudder from given materials. By implementing this action, the system thus passes from a state where the material for making the rudder is available to a state in which the rudder is manufactured.

The constraints C are goals to be achieved absolutely.

For example, the ship must be delivered on a predefined date.

In another example, its tonnage can not exceed a predefined value.

The constraints C are distinguished from the preferences P by their absolute nature.

The first supply step 100 is, for example, implemented by means of a keyboard and a mouse. By way of illustration, the first supply step 100 is implemented by the decision-makers.

According to an alternative embodiment, the set of possible actions A is established on the basis of a description of the (operational) capacities of the various stakeholders, the implementation of which makes it possible to change the system from one state to another. other state.

In the first conversion step 102, the input data is converted to an automatic scheduling language.

A planning language is a formal language dedicated to the field of planning.

More specifically, a planning language is a set of vocabulary and grammar. The vocabulary is a set of words and symbols whereas the grammar is a set of rules that specifies how to construct valid formulas from the basic elements of this vocabulary. For example, the automatic planning language is a PLM language, which stands for Planning Domain Definition Language, which literally means "planning domain definition language".

PDDL proposes to represent a planning problem using two sets of data. The first set of data relates to the description of the field of study, a description of the system and possible actions. The second set of data relates to the description of the problem to be solved, a description of the initial state of the system and the desired end state of the system.

The two sets of data are, for example, files.

In one example, the automatic scheduling language includes preference management functions.

These preferences management functions take into account the nature of the preferences (binary or fuzzy), which represent the semantics of the preferences and their aggregation.

An example of a preference management function is a function associating with the preference an integer of the set {0; 1} if the preference corresponds to a binary satisfaction or an integer of the interval [0; 1] if the preference corresponds to to a representable satisfaction for example with the formalism of fuzzy logic. As an illustration, the version 3.0 PPDL is an automatic scheduling language including some of these preferences management functions. The conversion step is carried out by the data processing unit 14. At the end of the first conversion step 102, converted data is obtained.

In the second supply step 104, a set of preferences is provided.

For example, decision makers enter preferences via keyboard or mouse.

A preference is a wish, which can be seen as an optional "pseudo constraint" (unlike true constraints). It may be a goal that it is desired to achieve without obligation.

The number of preferences is greater than two, advantageously greater than five, advantageously greater than ten.

For the purposes of the example, it is assumed that the preferences are three in number and relate to the cost of construction, the amount of energy consumed and the period of occupancy of the dry dock.

In another example, a decision maker provides a plurality of distinct sets of preferences at the provisioning stage, a game being a subset of the set of preferences.

It is possible to consider, at least formally, that this case can be reduced to the illustrated case, namely a plurality of decision-makers, each with possibly overlapping preferences.

During the first determination step 106, a preference score function is determined by analyzing the satisfaction evolution as a function of the value of the corresponding preference, for each preference provided at the second delivery step 104. .

For example, a satisfaction worth 0 corresponds to a state of total dissatisfaction while a satisfaction worth 1 corresponds to a state of total satisfaction.

According to the proposed example, it is determined for each of the three preferences a preference score function.

In the following, the score preference functions are denoted u ,, i being an integer varying from 1 to 3.

A first score function preferably Ui associated with the construction cost is thus determined, the second score function preferably U2 associated with the quantity of energy consumed and a third preferably score function Ua associated with the occupancy time of the hold.

Each function Uj is a function whose arrival set is an interval of E.

For example, the arrival set is the interval [0, 1].

According to the example described, for the determination of the different score functions Ui, the analysis is implemented by interaction with at least one decision maker, at the end of which a score function of preference for each preference is obtained. The interaction includes obtaining answers on the graduated appreciation of the decision maker's satisfaction for a finite plurality of preference values.

In the proposed example, this means that after the first determination step 106, the three score preference functions Ui, Ua and U3 are known.

In the second determination step 108, an overall score function is determined from the preference score functions.

The global score function makes it possible to obtain for a given plan a unique score value.

The global score function is obtained by implementing a mathematical technique.

For example, it uses a multi-attribute utility model also referred to as MAUT (for "Multi-attribute utility theory" in English).

Such a model uses an aggregation function.

According to one example, the aggregation function is an integral of Choquet.

In a variant, the aggregation function is a Choquet integral whose capacity is 2-additive.

In another example, a generative additive independence model is used to obtain the giobai score function. At the end of the second determination step 108, an overall score function is obtained.

In the second conversion step 110, the global score function is converted into the same automatic scheduling language. At the end of the second conversion step 110, a converted function is obtained.

In the generation step 112, a plurality of plans are generated from the converted data and the converted function.

According to one particular example, in the generation step 112, a single plan is generated. The step of generating plans is based on three modules - referred to here as lexeur, parseur and solver - which are invoked one after the other.

The first (the lexer) processes the text files obtained at the end of the conversion steps (and thus expressed according to the PDDL automatic planning language) in order to convert them into sequences of lexical entities.

These syntactic elements are then analyzed by the second module (the parser) which associates them with semantics. This is determined by the method of resolution employed; in our case, it is a graph whose nodes are characterized by logical formulas.

The last module implemented (the solver) relies on computer objects previously built to implement the problem solving algorithm.

The solution returned by the algorithm corresponds to the plane obtained at the end of the generation step 112. In our case, it is a search algorithm called "forward" guided by a heuristic.

The heuristics we propose are based on several estimates of the potential of a node (relative to objectives to be respected and preferences to be optimized) that are aggregated with each other using a Ghoquet integral ( which is none unrelated to the Choquet integral used to aggregate preferences in the determination step 108). At display step 114, the plurality of generated planes are displayed.

In addition, it is also displayed scores for each of the plans.

In a particular example, the value of the global score function and the value of each of the preference score functions are displayed.

According to another particular example, in the case where a decision maker has several sets of distinct preferences, it is displayed the value of the global score function and the value of each of the preference score functions for the set of preference games. considered. The display step 114 is implemented for each decision maker who can thus participate in the decision-making. The method thus makes it possible to determine plans that better meet the preferences of the decision-makers. This facilitates the work of decision makers.

In particular, the method makes it possible to provide decision-makers with all the information enabling them to choose with accuracy, relevance and with full knowledge of the facts, the solution which seems to them the most adapted to the present case.

In addition, the method for assisting the determination of plans makes it possible to take into account a multitude of preferences, either in a global manner (the search for a plan simultaneously takes into account all the preferences), or alternatively (preferences are grouped into different models and the search determines as many plans as models), especially in the presence of at least two different decision-makers.

The generalization to any number of preferences is, in fact, immediate, the proposed mathematical techniques for the second determination step 108 being able to operate for a large number of preferences.

According to a more elaborate version, the method is reiterated at least once, the preferences provided and / or the determination of the preference score functions being distinct from one iteration to another.

According to one variant, the system 10 is a support system for determining the plan that best meets the preferences that can be used on several sites.

For example, the system 10 comprises a plurality of terminals, each terminal being accessible for the entities.

This facilitates exchanges and increases the speed of determination of the best plan.

In one example, the method is implemented successively as many times as desired, and this dynamically by changing the input data.

According to another example, the method is implemented "cold" (that is to say outside dynamic context - on case studies for example) during the preparation of mission or exercise, or for purposes training to produce plans in accordance with a doctrine for example or in anticipation of possible future situations.

According to an example of use, the method is applied to hypothetical input data, so as to be able to anticipate choices in the context of possible futures, for example, or to "play" other roles than its own role ( in the case of multi-entity situations).

Claims (10)

A method of assisting the determination of the plan that best respects preferences, a plan being a sequence of actions leading a system from an initial state to a desired final state while respecting predefined constraints, the method being implemented implemented by a computer and comprising the following steps: - supply of input data, the input data comprising: • the initial state of the system, • the desired final state of the system, • the set of possible actions, and constraints, - converting input data into an automatic planning language to obtain converted data, - providing a set of preferences, - for each preference, determining a preference score function by analyzing the data. the evolution of the satisfaction as a function of the value of the corresponding preference, - determination of a global score function from the score functions of preferably, - converting the global score function into the same automatic scheduling language to obtain a converted function, and - generating at least one plan from the converted data and the converted function. 2. - The method of claim 1, wherein the set of possible actions are conducted by several separate entities, the preference providing step being conducted by at least two separate entities. 3. - Method according to claim 1 or 2, wherein the set of possible actions are carried out by several different entities, at least one entity providing a plurality of distinct sets of preferences at the supply stage, a game being a sub -all set of preferences. 4. - Method according to any one of claims 1 to 3, wherein the global score function is obtained by implementing a multi-attribute utility model using an aggregation function, the aggregation function. being an integral of Choquet or an integral of Choquet whose capacity is 2-additive. The method of any one of claims 1 to 4, wherein the automatic scheduling language includes preference management functions. 6. - Method according to any one of claims 1 to 5, a plurality of plans is generated in the generating step, the method comprising a step of displaying the plurality of plans and the value of each score function preferably for each of the plurality of planes. 7. The method according to any one of claims 1 to 6, the method being reiterated at least once, at least one of the group consisting of the provided preferences and the determination of the preference score functions being distinct from each other. one iteration to another. 8. - Support system for determining the plan that best respects preferences, a plan being a series of actions leading a system from an initial state to a desired end state under predetermined constraints, the system comprising at least a computer and being able to: - receive input data, the input data including: • the initial state of the system, • the desired end state of the system, • the set of possible actions, and • the constraints, - converting input data into an automatic planning language to obtain converted data, - receiving set of preferences, - for each preference, determining a preference score function by analyzing the evolution of satisfaction according to the value of the corresponding preference, - determining a global score function from the preference score functions, - converting the function of global score in the same automatic scheduling language to obtain a converted function, and - generate at least one plan from the converted data and the converted function. 9, - Computer program product comprising software instructions which, when executed by a computer, implement a method according to any one of claims 1 to 7. 10. - Computer readable recording medium on which a computer program is recorded for performing the steps of the method according to any one of claims 1 to 7.