Process for the iterative construction of an explanatory model
A process for iterative construction of an explanatory model including at least one rule calculated from a plurality of experiments, each of which rules is associated with at least one indicator of the quality of a corresponding rule including determining a rule, of a set of new experiments in an application space, conducting the new experiments to obtain the corresponding results, and calculating and optionally updating indicators of quality of the rule as a function of the experiments and of corresponding results.
1. A method implemented by a computer programmed as a rule generation apparatus for iterative construction of an explanatory model of an industrial process, the method comprising:
determining, by a processor of the computer, based on experimental data obtained from the industrial process, a rule in the explanatory model that explains at least a part of the experimental data;
displaying the determined rule to a user and allowing the user to modify the rule;
determining a set of new experiments for evaluating the rule;
conducting the determined new experiments to obtain corresponding results;
calculating at least one quality indicator that indicates a quality of the determined rule based on the corresponding results of the experiments; and
determining whether the at least one quality indicator satisfies predetermined conditions for including the rule in the explanatory model of the industrial process.
2. The method according to claim 1 , further comprising:
determining new rules from initial experiments and from the new experiments when the at least one quality indicator is not satisfactory.
3. The method according to claim 1 , further comprising:
modifying at least one initial rule determined from initial experiments and from the new experiments when the at least one quality indicator is not satisfactory.
4. The method according to claim 1 , wherein an initial set of rules is empty.
5. The method according to claim 1 , wherein the steps of determining new experiments and conducting new experiments are repeated until the at least one quality indicator satisfies the predetermined conditions.
6. The method according to claim 1 , wherein the steps of determining new experiments and conducting new experiments are repeated as long as the at least one quality indicator improves between two cycles.
7. The method according to claim 1 , wherein the step of determining new experiments is realized manually for certain experiments of the new experiments.
8. The method according to claim 1 , wherein the step of determining new experiments is performed, at least for certain experiments, by maximizing a foreseeable effect of the new experiments on the rule.
9. The method according to claim 1 , wherein the step of determining new experiments is performed taking account of a parameter representative of an expense of the new experiments.
10. The method according to claim 1 , wherein the step of determining new experiments is performed taking account of a parameter representative of feasibility of the new experiments.
11. The method according to claim 1 , wherein the step of determining new experiments is performed at least for certain experiments by maximizing a compromise between a foreseeable effect of the new experiments on the rule and a number of new experiments.
12. The method according to claim 1 , wherein the step of determining new experiments is performed at least for certain experiments by maximizing a compromise between a foreseeable effect of the new experiments on the rule and a cumulative expense of the new experiments.
13. The method according to claim 1 , wherein the step of determining new experiments is performed at least for certain experiments by maximizing distances two-by-two between initial experiments and the new experiments in an application space of the rule.
14. The method of claim 1 , wherein the at least one quality indicator includes a quality indicator indicating one of an interest, purity, relative size, robustness, sensitivity, complexity, density, and volume of the rule.
15. The method of claim 1 , wherein the at least one quality indicator includes a purity of the rule.
16. The method of claim 1 , wherein the at least one quality indicator includes a robustness of the rule.
17. The method of claim 1 , wherein the at least one quality indicator includes a complexity of the rule.
18. The method of claim 1 , wherein the at least one quality indicator includes a volume of the rule.
19. The method of claim 1 , wherein the at least one quality indicator includes a density of the rule.