CN113112021B - Reasoning method of artificial behavior decision model - Google Patents
Reasoning method of artificial behavior decision model Download PDFInfo
- Publication number
- CN113112021B CN113112021B CN202110441527.8A CN202110441527A CN113112021B CN 113112021 B CN113112021 B CN 113112021B CN 202110441527 A CN202110441527 A CN 202110441527A CN 113112021 B CN113112021 B CN 113112021B
- Authority
- CN
- China
- Prior art keywords
- reasoning
- rules
- knowledge
- generating
- rule
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 90
- 230000008569 process Effects 0.000 claims abstract description 34
- 238000011156 evaluation Methods 0.000 claims abstract description 33
- 230000008447 perception Effects 0.000 claims abstract description 28
- 230000003993 interaction Effects 0.000 claims abstract description 21
- 230000019771 cognition Effects 0.000 claims abstract description 14
- 230000001364 causal effect Effects 0.000 claims abstract description 4
- 238000004519 manufacturing process Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 7
- 239000013598 vector Substances 0.000 claims description 7
- 238000004458 analytical method Methods 0.000 claims description 4
- 230000006872 improvement Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 230000001960 triggered effect Effects 0.000 claims description 3
- 230000009471 action Effects 0.000 claims 3
- 238000012854 evaluation process Methods 0.000 claims 1
- 230000006399 behavior Effects 0.000 description 30
- 241000220324 Pyrus Species 0.000 description 22
- 235000014443 Pyrus communis Nutrition 0.000 description 16
- 241000220223 Fragaria Species 0.000 description 12
- 235000016623 Fragaria vesca Nutrition 0.000 description 9
- 235000011363 Fragaria x ananassa Nutrition 0.000 description 9
- 230000007613 environmental effect Effects 0.000 description 7
- 241000160720 Olinia ventosa Species 0.000 description 6
- 230000001149 cognitive effect Effects 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 235000021017 pears Nutrition 0.000 description 6
- 230000003542 behavioural effect Effects 0.000 description 4
- 238000009826 distribution Methods 0.000 description 4
- 235000013399 edible fruits Nutrition 0.000 description 4
- 230000007935 neutral effect Effects 0.000 description 4
- 210000000936 intestine Anatomy 0.000 description 3
- 210000002784 stomach Anatomy 0.000 description 3
- 235000021012 strawberries Nutrition 0.000 description 3
- 241000167854 Bourreria succulenta Species 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 235000019693 cherries Nutrition 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 241001513358 Billardiera scandens Species 0.000 description 1
- 235000004936 Bromus mango Nutrition 0.000 description 1
- 240000007228 Mangifera indica Species 0.000 description 1
- 235000014826 Mangifera indica Nutrition 0.000 description 1
- 206010034912 Phobia Diseases 0.000 description 1
- 235000009184 Spondias indica Nutrition 0.000 description 1
- 230000002378 acidificating effect Effects 0.000 description 1
- 125000002015 acyclic group Chemical group 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 208000030251 communication disease Diseases 0.000 description 1
- 238000001816 cooling Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 208000019899 phobic disease Diseases 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Probability & Statistics with Applications (AREA)
- Algebra (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Feedback Control In General (AREA)
Abstract
An inference algorithm of a human behavior decision model belongs to the technical field of human-computer interaction, and particularly relates to an inference algorithm of a human behavior decision model. The invention provides an inference algorithm of a human-like behavior decision model. The invention comprises a perception layer, an evaluation layer and a decision layer, and is characterized in that a method of generating a system structure is used for performing behavior cognition modeling in the perception layer, the generating formula is used as a basic knowledge structure unit in an artificial intelligent system, the generating formula system is used as a basic mode, the generating formula rule is used for representing reasoning process and behavior, the generating formula rule is used for representing causal relationship, rules among knowledge are represented, and a logic thinking reasoning process when a human solves a problem is simulated; in the evaluation layer, the service robot understands and judges the current situation according to the fuzzy state information output by the perception layer and by referring to experience and priori knowledge contained in a knowledge base, and adopts a behavior modeling method based on cognition.
Description
Technical Field
The invention belongs to the technical field of man-machine interaction, and particularly relates to an inference method of a human behavior decision-making model.
Background
In the service research of intelligent robots, in the process of completing a certain task, the robots mostly build a specific rule base, a knowledge base and an inference mechanism by programming by people to conduct task planning, analyze the single specific environment information and user behavior information, and not embody intelligent decision of the robots, which greatly restricts autonomous cognition of the service robots. The demand of the society for service robots is gradually increased, people have higher expectations for the service level of the server robots, when the service robots serve the people, the robots are required to simply complete tasks, the robots are required to have similar thinking reasoning and decision making capabilities as people, the service intelligence of the robots is reflected, the service process is based on people, the user is satisfied with the task execution process, and the service robots help the user to efficiently complete various service tasks.
Therefore, the real intention of the human is presumed by grasping the essence of human cognitive action decision and artificial intelligence, simulating the mode of human thinking reasoning and decision to make the ambiguous and uncertain task clear and specific process, and describing the real intention as corresponding language variable to be transmitted to the robot so as to construct the robot with similar human intelligent reasoning and decision.
Therefore, how to provide a method for simulating human thinking reasoning and decision making process is a technical problem to be solved by those skilled in the art.
Disclosure of Invention
The invention aims at the problems and provides an inference method of a human-like behavior decision model.
In order to achieve the purpose, the invention adopts the following technical scheme that the invention comprises a perception layer, an evaluation layer and a decision layer, and is characterized in that the perception layer carries out behavior cognition modeling by using a method of generating a system structure, the generating formula is used as a basic knowledge structure unit in an artificial intelligent system, the generating formula system is used as a basic mode, the generating formula rule is used for representing reasoning process and behavior, the generating formula rule is used for representing causal relationship, rules among the knowledge are represented, and the logic thinking reasoning process when a human solves a problem is simulated;
in the evaluation layer, the service robot understands and judges the current situation according to the fuzzy state information output by the perception layer and by referring to experience and priori knowledge contained in a knowledge base, and adopts a behavior modeling method based on cognition;
in the decision layer, the service robot inquires the opinion of the user through a man-machine interaction mode according to the evaluation result of the evaluation layer, finally decides a final target task, if the user cannot make a decision, makes a decision according to priori knowledge and common knowledge, and feeds back the final result to the comprehensive database and the knowledge base, so that conclusion information of each layer in reasoning and judging is updated.
As a preferred embodiment, the production System of the present invention includes a Global Database (Global Database), a production rule base (Set of Product Rules), and a Control System (Control System);
the comprehensive database is used for storing the data structures of various current task information in the solving process, wherein the data are the processing objects of the generating rules, when the premise of a certain generating rule in the rule base is matched with the facts in the comprehensive database, the rule base is activated, and the conclusion of reasoning is stored into the comprehensive database as new facts to become the known facts of later reasoning;
the generating rule base is common knowledge and heuristic knowledge for describing application field and is used for storing generating rules, and the generating rule base is a set of rule acts for knowledge in a certain field and comprises all transformation rules for converting a problem from an initial state to a target state; these rules describe general knowledge of the problem domain, and the rule base is the basis for the problem solving of the production system;
the control system is an inference engine of the production system; the method is an interpreter for generating rules, and the rules in a rule base are utilized to carry out logic operation on data in a comprehensive database; the control system is responsible for matching the preconditions or conditions of the generating type rules with the data in the comprehensive database, reasoning all the successfully matched rules according to a certain logic operation, ending the operation of the generating type system when the condition is proper, determining the priority reasoning sequence of the types and the attributes in the task objects, and inquiring the user to provide a further fact set through man-machine interaction to perform reasoning and judgment when the reasoning does not reach the conclusion.
As another preferable scheme, the processing object of the generating rule comprises initial state information of a task, input facts or evidences, intermediate reasoning conclusions, questions answered by a user, conclusions of rules successfully matched and final reasoning results.
As another preferable scheme, the invention stores priori knowledge used in reasoning and newly added specific task information, sets three attributes of task objects in the comprehensive database, and the states of the attributes are respectively two.
As another preferred embodiment, the invention uses a generating rule in the reasoning and behavior cognition process, which is in the form of: IF P THEN Q, where P is a precondition for regular use; q is the conclusion obtained when the rule condition is satisfied.
As another preferable scheme, a control system (inference engine) in the generating system adopts a forward direction reasoning mode, and a reasoning mode is performed according to the direction of reasoning out the task object conclusion by a given fact set in the comprehensive database; the reasoning process is:
step one: according to the task instruction given by the user, providing some initial known facts for the comprehensive database, matching the current data with the preconditions of the rules in the production rule base by the control system, judging whether the comprehensive database contains a solution of the problem, if yes, completing the task, otherwise, performing the step two;
step two: traversing rules in the generated rules, if the matching is successful, adding the back part of the triggered rules into the comprehensive database as new facts, matching the updated facts in the comprehensive database with rules in the rule base, updating the conclusion part of the updated facts into the comprehensive database, and discarding the rules if the matching is failed;
step three: and (3) repeating the process of the second step until no matchable generating rule and no new fact are added into the comprehensive database, forming a knowledge set after all the rules which are successfully matched are adopted by a set operation method, obtaining a matched rule set, wherein if the rule set is solved, the rule set indicates reasoning to obtain a task target, and if the rule set is not solved, inquiring a user to provide a further fact set through man-machine interaction, and carrying out reasoning and judgment.
Secondly, the modeling method in the assessment layer adopts a Bayesian network, the assessment layer adds a learner, the knowledge base is modified through feedback of a decision layer, and in the assessment process based on the Bayesian network method, the learner is connected, and the influence and improvement on the parameters in the knowledge base are exerted through supervised learning; and evaluating the reasoning by a Bayesian network reasoning method, and then sending the evaluated situation information to a decision layer for decision judgment, wherein the knowledge base comprises priori knowledge and Bayesian network form reasoning knowledge (namely the structure and the parameters of the Bayesian network).
In addition, the invention uses Bayesian network to make reasoning analysis with four steps:
step one: determining vectors and state representations of problems in task objects, defining nodes, constructing a Bayesian network, and constructing a directed acyclic graph according to the relation between the nodes;
step two: constructing a conditional probability table which is primarily generated according to experience and has a certain subjectivity;
step three: utilizing Bayesian network reasoning;
step four: updating the Bayesian network.
The invention has the beneficial effects that.
The invention provides an inference method for a service robot behavior decision model, which comprises the steps of simulating human behavior inference decision by dividing a human behavior decision-like model layered framework structure (figure 2) into a perception layer, an evaluation layer and a decision layer, providing a method for realizing functions of each layer according to the characteristics of each layer of the perception layer, the evaluation layer and the decision layer model, and respectively adopting a generating system, a behavior modeling method based on cognition and a method for making decisions in a human-computer interaction inquiry mode to simulate the ambiguity and the uncertainty of human cognition behavior. Firstly, receiving task information in a perception layer, taking a generating system as a main structure of artificial intelligence, establishing a category rule base and an attribute rule base of corresponding task objects based on different task information through generating rules, and simulating a thinking process when a human solves a problem; then in the evaluation layer, the service robot understands and judges the current situation information according to the historical state and the current state information and the events output by the perception layer and by referring to experience and priori knowledge contained in a knowledge base and the structure and parameters of the Bayesian network, and evaluates the possibility (probability magnitude) of various events and situation conditions appearing in the next step. Finally, in the decision layer, the service robot completes final decision according to the evaluation result of the evaluation layer, the priori knowledge in the knowledge base, and obtains more information with the least interaction times in a man-machine interaction mode, thereby determining the final task object type and attribute, and feeding back the final result to the perception layer and the evaluation layer to update the information stored in the comprehensive database and the structure and parameters of the Bayesian network in the knowledge base.
Drawings
The invention is further described below with reference to the drawings and the detailed description. The scope of the present invention is not limited to the following description.
FIG. 1 is a method of reasoning about a human-like behavioral decision model of the present invention.
FIG. 2 is a hierarchical framework of a user's behavioral decision-like model using the present invention.
Fig. 3-6 are directed acyclic graphs of the present invention.
Fig. 7 and 8 are partial enlarged views of fig. 1.
Detailed Description
As shown in the figure, the algorithm of the invention adopts a hierarchical framework structure (attached
FIG. 2), dividing into a perception layer, an evaluation layer and a decision layer to simulate human behavior reasoning decision, providing a method for realizing functions of each layer according to the characteristics of each layer of the perception layer, the evaluation layer and the decision layer model, and respectively adopting a method for making decisions by a generating system, a Bayesian network and a human-computer interaction inquiry mode to simulate the ambiguity and the uncertainty of human cognitive behaviors;
the method is characterized in that a method of generating a system structure is used for conducting behavior cognition modeling in a perception layer, a generating formula is used as a basic knowledge structural unit in an artificial intelligent system, the generating formula system is used as a basic mode, generating formula rules are used for representing reasoning processes and behaviors, the generating formula rules are used for representing causal relationship knowledge, rules among the knowledge can be clearly and definitely represented, the representation form is similar to logic thinking when a human solves a problem, and the method is reasonable and easy to understand, so that the logic thinking reasoning process when the human solves the problem is simulated. The production System consists of three parts, namely a Global Database (Global Database), a production rule base (Set of Product Rules) and a Control System (Control System).
In the human behavior decision model inference algorithm, a comprehensive database is a data structure for storing various current task information in the solving process, wherein data is a processing object of a production rule, such as initial state information of a task, input facts or evidence, intermediate inference conclusion, questions answered by a user, conclusion of a successfully matched rule, and final inference result, when the premise of a certain production rule in the rule base is matched with the facts in the comprehensive database, the rule base is activated, and the conclusion of the inference is stored in the comprehensive database as new facts to become known facts of later inference. And storing priori knowledge used in reasoning and newly added specific task information. The comprehensive database is provided with three attributes of the task objects, and the states of the attributes are represented as two.
A generative rule base is a collection of rule acts that describe common sense and heuristic knowledge of the application domain, and is used to store the generative rules, and is a collection of domain knowledge, including all transformation rules that transform a problem from an initial state to a target state. These rules describe general knowledge of the problem domain, and the rule base is the basis for the problem solving by the generative system. The use of generative rules in the reasoning and behavioral awareness process is generally of the form: IF P THEN Q, where P is a precondition for regular use; q is the conclusion obtained when the rule condition is satisfied.
The control system is an inference engine of the generative system. It is an interpreter of the production rules, which uses rules in the rule base to logically manipulate the data in the comprehensive database. The control system is responsible for matching the preconditions or conditions of the generating type rules with the data in the comprehensive database, reasoning all the successfully matched rules according to a certain logic operation, ending the operation of the generating type system when the condition is proper, determining the priority reasoning sequence of the types and the attributes in the task objects, and inquiring the user to provide a further fact set through man-machine interaction to perform reasoning and judgment when the reasoning does not reach the conclusion.
The control system (inference engine) in the generating system adopts a forward direction inference mode, which is an inference mode conducted according to the direction of inferring the task object conclusion from the given fact set in the comprehensive database. The reasoning process is:
step one: according to the task instruction given by the user, providing some initial known facts for the comprehensive database, matching the current data with the preconditions of the rules in the production rule base by the control system, judging whether the comprehensive database contains a solution of the problem, if yes, completing the task, otherwise, performing the step two;
step two: traversing rules in the generated rules, if the matching is successful, adding the back part of the triggered rules into the comprehensive database as new facts, matching the updated facts in the comprehensive database with rules in the rule base, updating the conclusion part of the updated facts into the comprehensive database, and discarding the rules if the matching is failed;
step three: and (3) repeating the process of the second step until no matchable generating rule and no new fact are added into the comprehensive database, forming a knowledge set after all the rules which are successfully matched are adopted by a set operation method, obtaining a matched rule set, wherein if the rule set is solved, the rule set indicates reasoning to obtain a task target, and if the rule set is not solved, inquiring a user to provide a further fact set through man-machine interaction, and carrying out reasoning and judgment.
In the evaluation layer, the service robot understands and judges the current situation by referring to experience and priori knowledge contained in a knowledge base according to fuzzy state information output by a perception layer, adopts a behavior modeling method based on cognition, and adopts a Bayesian network. Since there is a subconscious reinforcement of the experience knowledge and the active learning and updating of the knowledge structure by the person during the inference and decision process, the evaluation layer may incorporate a learner to enable the knowledge base to be adapted by feedback from the decision layer, and during the evaluation based on the bayesian network method, the influence and improvement of the parameters in the knowledge base may also be exerted by the supervised learning by connecting the learner. And evaluating the reasoning by a Bayesian network reasoning method, and then sending the evaluated situation information to a decision layer for decision judgment, wherein the knowledge base comprises priori knowledge and Bayesian network form reasoning knowledge (namely the structure and the parameters of the Bayesian network). The reasoning analysis by using the Bayesian network mainly comprises four steps:
step one: determining vectors and state representations of problems in task objects, defining nodes, constructing a Bayesian network, and constructing a directed acyclic graph according to the relation between the nodes;
step two: the conditional probability table is constructed mainly based on experience and has a certain subjectivity.
Step three: bayesian network reasoning is utilized.
Step four: updating the Bayesian network. The various factors in the bayesian network are also constantly changing, and there may be a certain result at a certain time.
In the decision layer, the service robot inquires the opinion of the user through a human-computer interaction mode according to the evaluation result of the evaluation layer, obtains more information quantity with the minimum interaction times to finish the decision, finally decides out a final target task, if the user cannot make the decision, makes the decision according to priori knowledge and common knowledge, and further accords with the human thinking reasoning and decision making process, and feeds the final result back to the comprehensive database and knowledge base, so that the conclusion information of each layer in reasoning and judgment is updated.
Taking a task of eating fruits by the old as an example, assuming that the teeth of the old are good and intestines and stomach are bad, reasoning according to the types and the priorities of the attributes in the task objects, wherein the three attributes in the task objects are set, the states of the attributes are respectively two, and under the condition that the deduced conclusions exist, the algorithm reasoning process of the behavior decision model of the person is as follows:
(1) firstly, obtaining unknown task object types according to known information in tasks, then judging that a solution of a target task is not contained in a database, and establishing a task object-based type rule base, wherein an initial fact set obtained by a service robot according to time period and environment information is as follows: spring |afternoon, table 1 is a category rule base based on task objects;
(2) sequentially taking rules in the category database, matching the rules with an initial fact set in the database by using the front piece of the rules, taking the rules R1 to successfully match, adding the conclusion of the rules as new facts into the initial fact set in the database, and changing the initial fact set into: neutral or cool in spring/afternoon; the matching is unsuccessful when taking rule R2; unsuccessful in taking rule R3; when taking rule R4, the method is unsuccessful, and the initial fact set is changed into spring/afternoon/neutrality or cooling/coolness; taking rule R5 successfully; taking the rule R6 and the rule R7 is unsuccessful, and taking the rule R8 is successful, so that the old people can deduce that the old people need to eat pears or strawberries;
table 1 class rule base based on task objects
| IF | THEN | |
| R1 | Spring and summer | Neutral or cool nature |
| R2 | Season in autumn and winter | Neutral or acidic |
| R3 | Morning and evening | Thermal property of warm |
| R4 | Afternoon | Cool nature |
| R5 | Neutral | Apple or strawberry |
| R6 | Acidity | Apple or cherry |
| R7 | Thermal property of warm | Cherry or mango |
| R8 | Cool nature | Pear or strawberry |
When the type of the task object is inferred, the vector and the state representation of the problem are represented by a Bayesian network method, nodes are defined, a Bayesian network is constructed, and a directed acyclic graph is constructed according to the relation between the nodes, as shown in fig. 3:
(3) the Bayesian network method is used for expressing the vector of the problem and the state thereof when the probability of pear and strawberry is inferred are expressed as follows: time (morning, evening, afternoon), environment (spring, summer, autumn, winter), fruit (pear, strawberry), constructing a bayesian network, and constructing a directed acyclic graph according to the relationship between nodes, as shown in fig. 4:
the probabilities of pear and strawberry, under conditions of known time and environment, are expressed as: p (f=p|t=a, e=ss), P (f=s|t=a, e=ss), where f=p denotes a pear, f=s denotes a strawberry, t=a denotes a afternoon, and e=ss denotes a spring and summer season.
According to the Bayes formulaIt can be derived that:
assuming that only the elderly eat pears and strawberries within 100 days of the spring and summer season, D (f=p) =60, D (f=s) =40, D (t=a|f=p) =15, D (t=a|f=s) =12
The method can obtain the following steps:
P(T＝A,E＝SS|F＝P)＝1/4，P(T＝A,E＝SS|F＝S)＝3/10
according to the actual situation, the time is divided into two time periods of morning, evening and afternoon, the environment is divided into spring and summer seasons and autumn and winter seasons, and P (T) = (2/3, 1/3) and P (E) = (1/2 ) can be obtained.
Since time and environmental information (seasons) are relatively independent, it is possible to:
P(T＝A,E＝SS)＝P(T＝A)P(E＝SS)＝1/6
from the comprehensive database, the environmental information is known as spring and summer season, and the distribution of days of pear and strawberry in each time period within 100 days of spring and summer season is shown in table 2, and the probability of establishing time and environmental information condition under the condition that pear and strawberry are eaten by the old is known as shown in table 3.
TABLE 2 distribution of days of pear and strawberry for each time period within 100 days of spring and summer
TABLE 3 conditional probability based on time and environmental information under fruit category
It can be derived that:
P(T＝A,E＝SS,F＝P)＝P(F＝P)×P(T＝A,E＝SS|F＝P)＝3/5×1/4＝3/20
P(T＝A,E＝SS,F＝S)＝P(F＝S)×P(T＝A,E＝SS|F＝S)＝2/5×3/10＝3/25
because 9/10>18/25, the pear eating probability of the old is high.
(4) The probability of obtaining pears is high in an evaluation layer through Bayesian network calculation, then the attributes of task objects are judged to be less than three according to task information in a comprehensive database, a pear-based attribute rule base (table 4) is established, and the pear-based attribute rule base is matched with known facts in the comprehensive database to obtain soft or hard pears which are small and normal-temperature and eaten by old people;
table 4 pear-based attribute rule base
| IF | THEN | |
| R9 | Good pear AND tooth mouth | Soft or hard |
| R10 | Pear AND tooth stump | Soft and soft |
| R11 | Good stomach AND intestines | Big or small AND normal temperature or cold storage |
| R12 | Poor stomach AND intestine function of pear | Small or ambient temperature |
When the task object attribute is inferred, the problem and the state representation thereof are represented by a Bayesian network method based on the task object type, nodes are defined, a Bayesian network is constructed, and a directed acyclic graph is constructed according to the relation between the nodes, as shown in fig. 5:
(5) the Bayesian network method is used for representing the vector and state of the problem when the probability of soft and hard pears is inferred: time (morning, evening, afternoon), environment (spring, summer, autumn, winter), taste (soft, hard), a bayesian network is constructed, and a directed acyclic graph is constructed according to the relationship between nodes, as shown in fig. 6:
according to the Bayes formulaIt can be derived that:
assuming that only the elderly eat pears and strawberries in 100 days of the spring and summer, D (m=s) =60, D (m=h) =40, D (m=s|t=a, f=p) =12, D (m=h|t=a, f=p) =10
The method can obtain the following steps:
P(M＝S)＝3/5，P(M＝H)＝2/5，P(T＝A,E＝SS,F＝P|M＝S)＝1/5，P(T＝A,E＝SS,F＝P|M＝H)＝1/4
under the conditions of time, environmental information and known pear eating condition of the old, the probability of soft and hard pears is expressed as follows: p (m=s|t=a, e=ss, f=p), P (m=h|t=a, e=ss, f=p), where m=s represents soft, m=p x Indicating hard.
According to the comprehensive database, the environment information is known as spring and summer season, the distribution of the days of eating soft and hard pears by the old people in each time period within 100 days of spring and summer season is shown in table 5, and the probability table of the time and environment information condition is known to be shown in table 6 under the condition that the old people eat soft and hard pears.
TABLE 5 distribution of days for Soft and hard pears over the 100 days of the spring and summer season
TABLE 6 conditional probability based on time, environmental information and mouthfeel under pears
It can be derived that:
because of 4/5>2/3, the probability of eating soft pear by the old is high.
(6) And (3) the situation information estimated by the estimation layer is sent to the decision layer, and decision planning is carried out in a man-machine interaction mode according to priori knowledge in the knowledge base. Firstly judging whether specific task information is newly added or not, if no new task information is added, judging the number of the task objects which are deduced, if the number of the task objects which are deduced is one, then judging whether the old can make a decision or not, if the old can make a decision, then inquiring whether the old agrees with the deduced conclusion or not in a man-machine interaction mode, and if the old agrees with the conclusion, finally obtaining soft, small and normal-temperature pears which the old needs to eat.
T in each formula represents a time variable, E represents an environmental variable, F represents a fruit variable, S represents a shape variable, M represents a mouthfeel variable, tem represents a temperature variable, D represents a number of days, and B represents an event in a Bayesian formula.
As shown in fig. 1, the invention provides an inference algorithm for a service robot behavior decision model, wherein the algorithm is divided into a perception layer, an evaluation layer and a decision layer to simulate human behavior inference decision through a layered framework structure (fig. 2) of the human behavior decision-making model, and the method for realizing functions of each layer is provided according to the characteristics of each layer of the perception layer, the evaluation layer and the decision layer model, and the ambiguity and uncertainty of human cognitive behaviors are simulated by adopting a decision making method of a generating system, a bayesian network and a human-computer interaction query mode respectively. In the process of simulating human logic thinking reasoning, the personality characteristics (such as communication disorder, weak language expression capability, selection phobia and the like) of the user are considered, certain specific information (such as physical condition and preference of the user) of the user is concerned, and newly added target related specific information is considered. And under the condition that the inferred conclusions exist, reasoning is carried out according to the types and the priorities of the attributes in the task objects, wherein the attributes in the task objects are set to be three, and the states of the attributes are represented to be two.
In the perception layer, a method of generating a System structure is used for performing behavior cognition modeling, and a generating formula is used as a basic knowledge structural unit in an artificial intelligent System, so that the generating formula System is used as a basic mode, and is composed of three parts, namely a Global Database (Global Database), a generating formula rule base (Set of Product Rules) and a Control System (Control System). The control system (inference engine) in the generating system adopts a forward direction inference mode, which is an inference mode conducted according to the direction of inferring the task object conclusion from the given fact set in the comprehensive database.
In the evaluation layer, the service robot understands and judges the current situation by referring to experience and priori knowledge contained in a knowledge base according to fuzzy state information output by a perception layer, adopts a behavior modeling method based on cognition, and adopts a Bayesian network. Firstly, determining vectors and state representations of problems in task objects, defining nodes, constructing a Bayesian network, and constructing a directed acyclic graph according to the relation between the nodes; and then constructing a conditional probability table, carrying out Bayesian network reasoning, and finally determining the result of evaluation reasoning.
In the decision layer, the service robot acquires more information to complete final decision in a man-machine interaction mode according to the evaluation result of the evaluation layer and the priori knowledge in the knowledge base, so as to determine the final task object type and attribute, and feeds the final result back to the perception layer and the evaluation layer to update the information stored in the comprehensive database and the structure and parameters of the Bayesian network in the knowledge base.
As shown in fig. 2, another aspect of the present invention further provides a hierarchical framework structure of a behavioral decision model, where the method includes:
the method comprises the steps of simulating human behavior logic reasoning and decision by three layers of perception processing, situation assessment, decision planning and the like, wherein the three layers are respectively called a perception layer, an assessment layer and a decision layer, the algorithm is added with feedback connection between the decision layer and each layer while highlighting the human perception process, the active cognitive behavior of a human under a decision target task is enhanced, a real-time online learning mechanism is added, a method for realizing functions of each layer is provided by taking a generating system as a main line, the method for realizing functions of each layer is provided according to the characteristics of each layer of a perception layer, the assessment layer and a decision layer model, ambiguity and uncertainty of human cognitive behaviors are simulated by adopting the method for carrying out decision making by adopting a generating system, a Bayesian network and a human-computer interaction query mode, the modeling process of autonomous learning of a simulator, analysis processing of uncertain factors and experiences and the like is realized, and the cognitive behavior process of the human can be well described.
It should be understood that the foregoing detailed description of the present invention is provided for illustration only and is not limited to the technical solutions described in the embodiments of the present invention, and those skilled in the art should understand that the present invention may be modified or substituted for the same technical effects; as long as the use requirement is met, the invention is within the protection scope of the invention.
Claims (7)
1. The reasoning method of the artificial action decision model comprises a perception layer, an evaluation layer and a decision layer, and is characterized in that the perception layer carries out action cognition modeling by using a method of a generating system structure, the generating system is used as a basic knowledge structure unit in an artificial intelligent system, the generating system is used as a basic mode, the generating rule is used for representing a reasoning process and actions, the generating rule is used for representing causal relations, rules among knowledge are represented, and a logic thinking reasoning process when a human solves a problem is simulated;
in the evaluation layer, the service robot understands and judges the current situation information according to the historical state and the current state information and events output by the perception layer and by referring to experience and priori knowledge contained in a knowledge base and the structure and parameters of a Bayesian network, and evaluates the probability of various events and situation conditions appearing in the next step;
in the decision layer, the service robot acquires more information to complete final decision in a man-machine interaction mode according to the evaluation result of the evaluation layer and the priori knowledge in the knowledge base, so as to determine the final task object type and attribute, and feeds the final result back to the perception layer and the evaluation layer to update the information stored in the comprehensive database and the structure and parameters of the Bayesian network in the knowledge base;
the generating system comprises a comprehensive database, a generating rule base and a control system;
the comprehensive database is used for storing the data structures of various current task information in the solving process, wherein the data are the processing objects of the generating rules, when the premise of a certain generating rule in the rule base is matched with the facts in the comprehensive database, the rule base is activated, and the conclusion of reasoning is stored into the comprehensive database as new facts to become the known facts of later reasoning;
the generating rule base is common knowledge and heuristic knowledge for describing application field and is used for storing generating rules, and the generating rule base is a set of rule acts for knowledge in a certain field and comprises all transformation rules for converting a problem from an initial state to a target state; these rules describe general knowledge of the problem domain, and the rule base is the basis for the problem solving of the production system;
the control system is an inference engine of the production system; the method is an interpreter for generating rules, and the rules in a rule base are utilized to carry out logic operation on data in a comprehensive database; the control system is responsible for matching the preconditions or conditions of the generating type rules with the data in the comprehensive database, reasoning all the successfully matched rules according to a certain logic operation, ending the operation of the generating type system when the condition is proper, determining the priority reasoning sequence of the types and the attributes in the task objects, and inquiring the user to provide a further fact set through man-machine interaction to perform reasoning and judgment when the reasoning does not reach the conclusion.
2. The method of reasoning about a human-like behavior decision model according to claim 1, wherein the processing object of the production-like rule includes initial state information of a task, input facts or evidence, intermediate reasoning conclusions, questions answered by a user, conclusions of rules that match successfully, final reasoning results.
3. The reasoning method of the human-like behavior decision model according to claim 1, wherein the three attributes of the task object are set in the comprehensive database by prior knowledge used in the process of storing reasoning and newly added specific task information, and the states of the attributes are represented by two.
4. The method of reasoning about a human-like behavior decision model according to claim 1, characterized by using a generating rule in the course of reasoning and behavior cognition in the form of: IF P THEN Q, where P is a precondition for regular use; q is the conclusion obtained when the rule condition is satisfied.
5. The reasoning method of the human-like behavior decision model according to claim 1, wherein the control system in the production system adopts a forward reasoning mode, and a reasoning mode is performed according to the direction of reasoning out the task object conclusion by a given fact set in the comprehensive database; the reasoning process is:
step one: according to the task instruction given by the user, providing some initial known facts for the comprehensive database, matching the current data with the preconditions of the rules in the production rule base by the control system, judging whether the comprehensive database contains a solution of the problem, if yes, completing the task, otherwise, performing the step two;
step two: traversing rules in the generated rules, if the matching is successful, adding the back part of the triggered rules into the comprehensive database as new facts, matching the updated facts in the comprehensive database with rules in the rule base, updating the conclusion part of the updated facts into the comprehensive database, and discarding the rules if the matching is failed;
step three: and (3) repeating the process of the second step until no matchable generating rule and no new fact are added into the comprehensive database, forming a knowledge set after all the rules which are successfully matched are adopted by a set operation method, obtaining a matched rule set, wherein if the rule set is solved, the rule set indicates reasoning to obtain a task target, and if the rule set is not solved, inquiring a user to provide a further fact set through man-machine interaction, and carrying out reasoning and judgment.
6. The reasoning method of the human-like behavior decision model according to claim 1, characterized in that the evaluation layer is added with a study Xi Qi, the knowledge base is modified through feedback of the decision layer, and in the evaluation process based on the Bayesian network method, the study is connected, and the influence and improvement are exerted on the parameters in the knowledge base through supervised study; and evaluating the reasoning by a Bayesian network reasoning method, and then sending the evaluated situation information to a decision layer for decision judgment, wherein a knowledge base contains priori knowledge and Bayesian network form reasoning knowledge.
7. The method for reasoning about a human-like behavior decision model as recited in claim 6, wherein the reasoning analysis using a bayesian network has four steps:
step one: determining vectors and state representations of problems in task objects, defining nodes, constructing a Bayesian network, and constructing a directed acyclic graph according to the relation between the nodes;
step two: constructing a conditional probability table, and primarily generating according to experience;
step three: utilizing Bayesian network reasoning;
step four: updating the Bayesian network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110441527.8A CN113112021B (en) | 2021-04-23 | 2021-04-23 | Reasoning method of artificial behavior decision model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110441527.8A CN113112021B (en) | 2021-04-23 | 2021-04-23 | Reasoning method of artificial behavior decision model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113112021A CN113112021A (en) | 2021-07-13 |
| CN113112021B true CN113112021B (en) | 2024-03-08 |
Family
ID=76719569
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110441527.8A Active CN113112021B (en) | 2021-04-23 | 2021-04-23 | Reasoning method of artificial behavior decision model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113112021B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114416792A (en) * | 2022-01-11 | 2022-04-29 | 中国人民解放军国防科技大学 | Probability-based data stream processing method and system |
| CN116227361B (en) * | 2023-03-06 | 2023-08-15 | 中国人民解放军32370部队 | Intelligent body decision method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104506338A (en) * | 2014-11-21 | 2015-04-08 | 河南中烟工业有限责任公司 | Fault diagnosis expert system based on decision tree for industrial Ethernet network |
| CN112200319A (en) * | 2020-10-20 | 2021-01-08 | 北京计算机技术及应用研究所 | Rule reasoning method and system for achieving unmanned vehicle navigation obstacle avoidance |
| CN112419126A (en) * | 2020-12-07 | 2021-02-26 | 安徽四创电子股份有限公司 | Urban public safety video monitoring and evaluating system based on expert system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8346683B2 (en) * | 2009-04-30 | 2013-01-01 | Exprentis, Inc. | System, program, and method for representation, utilization, and maintenance of regulatory knowledge |
-
2021
- 2021-04-23 CN CN202110441527.8A patent/CN113112021B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104506338A (en) * | 2014-11-21 | 2015-04-08 | 河南中烟工业有限责任公司 | Fault diagnosis expert system based on decision tree for industrial Ethernet network |
| CN112200319A (en) * | 2020-10-20 | 2021-01-08 | 北京计算机技术及应用研究所 | Rule reasoning method and system for achieving unmanned vehicle navigation obstacle avoidance |
| CN112419126A (en) * | 2020-12-07 | 2021-02-26 | 安徽四创电子股份有限公司 | Urban public safety video monitoring and evaluating system based on expert system |
Non-Patent Citations (1)
| Title |
|---|
| 类人答题系统中立体几何问题自动求解的研究及实现;张文迪;中国优秀硕士学位论文全文数据库信息科技辑;20181015(第10期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113112021A (en) | 2021-07-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Qiang et al. | Reinforcement learning model, algorithms and its application | |
| Watthayu et al. | A Bayesian network based framework for multi-criteria decision making | |
| CN102402712B (en) | Robot reinforced learning initialization method based on neural network | |
| Konar et al. | Reasoning and unsupervised learning in a fuzzy cognitive map | |
| Papageorgiou | Review study on fuzzy cognitive maps and their applications during the last decade | |
| Zhang et al. | CORPP: Commonsense reasoning and probabilistic planning, as applied to dialog with a mobile robot | |
| CN113112021B (en) | Reasoning method of artificial behavior decision model | |
| Yesil et al. | Fuzzy cognitive maps learning using artificial bee colony optimization | |
| CN109496305A (en) | Nash equilibrium strategy on continuous action space and social network public opinion evolution model | |
| Peters et al. | Reinforcement learning with approximation spaces | |
| Hu et al. | Adaptive exploration strategy with multi-attribute decision-making for reinforcement learning | |
| Nguyen et al. | A framework for learning to request rich and contextually useful information from humans | |
| Chen et al. | Policy gradient from demonstration and curiosity | |
| Farias et al. | BDI Agents with Fuzzy Perception for Simulating Decision Making in Environments with Imperfect Information. | |
| Khan et al. | Fuzzy cognitive maps and intelligent decision support–a review | |
| CN116167415A (en) | Policy decision method in multi-agent cooperation and antagonism | |
| Hasegawa et al. | Inferring place-object relationships by integrating probabilistic logic and multimodal spatial concepts | |
| Naim et al. | A general type-2 fuzzy logic based approach for multi-criteria group decision making | |
| Raz et al. | Trust-based Cognitive Decision Making by Social Things-A Case Study of Cancer Treatment | |
| Jiya et al. | A Review of Fuzzy Cognitive Maps Extensions and Learning | |
| Luengo et al. | Framework for simulating the human behavior for intelligent virtual agents. part ii: Behavioral system | |
| Oliehoek et al. | A decision-theoretic approach to collaboration: Principal description methods and efficient heuristic approximations | |
| Kallel et al. | Towards a fuzzy evaluation of the adaptivity degree of an evolving agent | |
| Luo et al. | The synchronization of K-valued Fuzzy cognitive maps | |
| Sandoval et al. | Predicting Agent's Behavior: A Systematic Mapping Review. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |