Search

CN-122018353-A - Skill generating method, control device and control system for intelligent body

CN122018353ACN 122018353 ACN122018353 ACN 122018353ACN-122018353-A

Abstract

The application is suitable for the technical field of intelligent home control, and provides a skill generating method, control equipment and a control system for an intelligent body. The intelligent household equipment management method comprises the steps of obtaining a first behavior sequence formed by interaction events of a user and intelligent household equipment in a first time period, conducting multidimensional mining on the first behavior sequence to mine a plurality of behavior modes, generating corresponding skill packages according to any behavior mode and based on the behavior modes, and when a trigger source is detected, controlling the corresponding intelligent household equipment to execute corresponding actions by an intelligent body according to the generated skill packages. The application can excavate the behavior mode from multiple dimensions, understand complex behaviors and realize self-adaptive intelligent home personalized service.

Inventors

  • Request for anonymity
  • Request for anonymity
  • Request for anonymity
  • LI ZHICHEN
  • PAN YANG
  • LI JIANGANG

Assignees

  • 卧安科技(深圳)有限公司

Dates

Publication Date
20260512
Application Date
20260413

Claims (17)

  1. 1. A skill generating method for a body-building agent, comprising: acquiring a first behavior sequence formed by each interaction event of a user and intelligent home equipment in a first time period; Multi-dimensional mining is carried out on the first behavior sequence, a plurality of behavior modes are mined, the plurality of behavior modes comprise at least two of a periodic mode, a first sequence mode, a first association mode and a scene mode, the periodic mode represents a user behavior mode with a time rule, the first sequence mode represents a time sequence relation among different device actions, the device actions refer to actions executed by the intelligent home device, the first association mode represents an association relation between action execution conditions and the device actions, and the scene mode represents a behavior scene of the user; Generating a corresponding skill package based on the behavior patterns for any one of the behavior patterns; when the triggering source is detected, the intelligent body controls the corresponding intelligent home equipment to execute corresponding actions according to the generated skill packages.
  2. 2. The skill generating method according to claim 1, wherein the periodic pattern comprises a periodic sequence pattern and/or a periodic association pattern, and the mining method of the periodic sequence pattern comprises: Grouping the second behavior sequences according to different second time periods and date types aiming at the second behavior sequences of each day in the first behavior sequences to obtain first groups; extracting frequently-occurring first action subsequences from each first packet, and determining the first action subsequences as a second sequence mode; If the second sequence pattern appears stably in N continuous observation periods, determining the second sequence pattern as the periodic sequence pattern, wherein N is an integer greater than 1; The mining mode of the periodic association mode comprises the following steps: organizing the first behavior sequence into transactions according to a fixed time window, and attaching a time tag to each transaction, wherein the time tag comprises a time period corresponding to the transaction and the date type; grouping the transactions with the same time labels into the same group to obtain second groups; mining an association relationship between the action execution condition and the device action from each of the second packets, and determining the association relationship between the action execution condition mined from each of the second packets and the device action as a second association pattern; And if the support degree of the second association mode in the continuous M observation periods is stable and is concentrated in a certain time period, determining the second association mode as the periodic association mode, wherein M is an integer greater than 1.
  3. 3. The skill generation method according to claim 1, wherein the mining method of the first association mode includes: Under the condition that the action execution condition is another device action, aiming at a third time period in the first action sequence, mining linkage relations among different device actions to obtain an inter-device association mode; Under the condition that the action execution condition is an environmental context condition, aiming at a fourth time period in the first action sequence, mining an association relation between the environmental context condition and the equipment action to obtain a context and equipment association mode; When the action execution condition is that another equipment state and the environmental context condition are simultaneously included, aiming at a fifth time period in the first action sequence, the association relationship between the other equipment state and the environmental context condition and the equipment action is mined to obtain a first mixed mode, wherein the other equipment state refers to the running state of the other intelligent household equipment; when the action execution condition simultaneously comprises the other equipment action and the environmental context condition, aiming at a sixth time period in the first behavior sequence, mining an association relationship between the other equipment action and the environmental context condition and the equipment action to obtain a second mixed mode; at least one of the inter-device association mode, the context and device association mode, the first mixed mode and the second mixed mode is the first association mode.
  4. 4. The skill generating method according to claim 1, wherein the mining method of the first sequence pattern includes: And extracting a frequently-occurring second action sub-sequence for a seventh time period in the first action sequence, recording an average time interval between the equipment actions in the second action sub-sequence, and determining the second action sub-sequence as the first sequence mode.
  5. 5. The skill generation method according to claim 1, wherein the mining method of the scene mode includes: dividing a second behavior sequence of each day in the first behavior sequence into a plurality of behavior segments; vectorizing the behavior segments to obtain a plurality of vectors; and clustering the vectors to obtain the behavior scene, and determining the behavior scene as the scene mode.
  6. 6. A skill generation method according to any of claims 1-5, wherein the generating a corresponding skill package based on the behavioral patterns for any of the behavioral patterns comprises: Inputting the behavior pattern into a first large model to identify the intention of the behavior pattern through the first large model, and generating a section of natural language description, wherein the natural language description is used for explaining the user intention of the behavior pattern representation; the skill pack is generated based on the behavioral patterns and the natural language descriptions.
  7. 7. A skill generation method according to claim 6, further comprising, after said generating said skill pack: generating a semantic description for the skill package based on the natural language description; Pushing the skill package to the user and attaching the semantic description; in response to a confirmation instruction to the skill pack, marking the skill pack as active, Or marking the skill package as pending or discarded in response to a denial of the skill package.
  8. 8. A skill generation method according to any of claims 1 to 5, wherein upon detection of a trigger source, the body-building agent controls the corresponding smart home device to perform a corresponding action according to each of the generated skill packs, comprising: Upon detecting a trigger source, the body-building agent determines a target skill package matching the trigger source from the generated skill packages; the intelligent agent triggers the target skill package to control the corresponding intelligent home equipment to execute corresponding actions.
  9. 9. A skill generating method according to claim 8, further comprising, prior to detecting the trigger source: In the case of the activation of the body-building agent, the body-building agent scans a skill library to read metadata in each of the skill packages and constructs a skill index table based on the metadata, the skill library including at least one of the skill packages; The intelligent agent detects whether an trigger source exists or not based on the skill index table to obtain a detection result; the step of determining a target skill package matched with the trigger source by the body intelligent agent from the generated skill packages when the trigger source is detected comprises the following steps: If the detection result indicates that a trigger source is present, the body-building agent retrieves the skill library by vector and determines the target skill package that matches the trigger source in response to the trigger source.
  10. 10. The skill generation method according to claim 9, wherein the skill index table includes at least one of a time trigger condition, a sequence trigger condition, a state trigger condition, and a natural language description of each of the skill packages, the natural language description being used to explain a user intention of a corresponding behavior pattern characterization, wherein the body-building agent detects whether a trigger source exists based on the skill index table, and obtaining a detection result includes: The intelligent agent with body periodically detects whether the first current time is matched with any time triggering condition in the skill index table, and if the first current time is matched with any time triggering condition, the intelligent agent with body determines that a detection result of a triggering source exists; And/or under the condition of subscribing to a device state change event, the self-contained agent detects whether the device state change event is matched with any sequence triggering condition or any state triggering condition in the skill index table, and if the device state change event is matched with any sequence triggering condition or any state triggering condition, a detection result of a triggering source is determined; and/or, responding to a user instruction, the intelligent body detects whether the user instruction is matched with any natural language description in the skill index table, and if the user instruction is matched with the natural language description, determining that a detection result of an trigger source exists.
  11. 11. The skill generating method according to claim 8, wherein in case that the number of the target skill packages is plural, the body-building agent triggers the target skill packages to control the corresponding smart home devices to perform the corresponding actions, comprising: The self-contained intelligent agent detects whether a skill conflict exists between at least two target skill packages in a plurality of target skill packages, and determines a conflict relation between the at least two target skill packages under the condition that the skill conflict exists between the at least two target skill packages, wherein the skill conflict comprises at least one of resource conflict, logic conflict and intention conflict, the resource conflict means that the at least two target skill packages use the same intelligent household equipment and the equipment actions are mutually exclusive, the logic conflict means that the execution results of the at least two target skill packages are semantically contradictory, and the intention conflict means that the user intention reflected by the at least two target skill packages is contradictory; Generating a prompt word by the body-building agent based on a task target, a candidate skill list, the conflict relation, a current environmental context and a user recent feedback, wherein the candidate skill list comprises skill names, skill descriptions, trigger conditions, execution plans and trigger priorities of a plurality of target skill packages, and the user recent feedback characterizes the acceptance or rejection condition of the user to the plurality of target skill packages in an eighth time period, and the end time of the eighth time period is earlier than a second current time; The body-building intelligent agent inputs the prompt word into a second large model so as to understand the skill intention of a plurality of target skill packages through the second large model and obtain a judging result, wherein the judging result comprises a skill package set for selecting execution and the execution sequence of each skill package in the skill package set; The intelligent agent analyzes the execution plan of each skill pack in the skill pack set into each skill execution instruction, and controls the corresponding intelligent home equipment to execute the corresponding skill execution instruction according to the execution sequence.
  12. 12. The skill generating method according to claim 11, wherein the arbitration result further comprises a decision reason including a reason for selecting the skill package set and a reason for executing each of the skill packages in the skill package set in the execution order, and further comprising, after the obtaining the arbitration result: the body-building agent presents the decision reason to the user.
  13. 13. The skill generating method according to claim 8, further comprising, after the body-building agent triggers the target skill package: Recording the user behavior of the target skill package after being triggered; If the user behavior is a correction behavior and the correction behavior aiming at the target skill package frequently occurs, deleting the target skill package, wherein the correction behavior is used for correcting the equipment action corresponding to the target skill package.
  14. 14. A skill generation method according to any of claims 1 to 5, further comprising, after the generating the corresponding skill package: And continuously monitoring the recent behavior of the user, if the behavior mode corresponding to a certain skill pack is detected to be no longer obvious based on the recent behavior of the user, marking the certain skill pack as expired, and reducing the triggering priority of the certain skill pack, or confirming whether to deactivate the certain skill pack to the user, wherein the recent behavior of the user refers to an interaction event of the user with the intelligent household equipment in a ninth time period, and the ending time of the ninth time period is earlier than a third current time.
  15. 15. The skill generating method according to any of claims 1 to 5, wherein obtaining a first sequence of actions formed by interaction events of the user with the smart home device during a first time period comprises: Cleaning and aligning each interaction event to obtain each processed interaction event; and ordering each processed interaction event according to time to obtain the structured first behavior sequence.
  16. 16. A control device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, causes the control device to implement the skill generation method of any of claims 1 to 15.
  17. 17. A smart home control system comprising a control device and at least one smart home device, the control device performing the skill generating method of any of claims 1 to 15.

Description

Skill generating method, control device and control system for intelligent body Technical Field The application belongs to the technical field of intelligent home control, and particularly relates to a skill generating method, control equipment and control system for an intelligent body. Background Currently, user personalization services of smart home systems mainly rely on two technical schemes of "manual rule configuration" and "automation based on simple statistics". Manual rule configuration, wherein a user manually sets a timing task or a linkage scene (for example, a living room lamp is turned on 19:00 every day) through an APP or a voice assistant, and the system performs fixed operation according to a preset rule. Automation based on simple statistics the system collects device usage frequency, recommends possible automation rules to the user (e.g. "you often turn on the living room lights in the evening, is automation to be created. This approach typically only counts frequencies in a single dimension and does not understand complex behavior. Disclosure of Invention The embodiment of the application provides a skill generating method, control equipment and a control system for an intelligent body, which can excavate a behavior mode from multiple dimensions, understand complex behaviors and realize self-adaptive intelligent home personalized service. In a first aspect, an embodiment of the present application provides a skill generating method for an agent with body, including: acquiring a first behavior sequence formed by each interaction event of a user and intelligent home equipment in a first time period; Multi-dimensional mining is carried out on the first behavior sequence, a plurality of behavior modes are mined, the plurality of behavior modes comprise at least two of a periodic mode, a first sequence mode, a first association mode and a scene mode, the periodic mode represents a user behavior mode with a time rule, the first sequence mode represents a time sequence relation among different device actions, the device actions refer to actions executed by the intelligent home device, the first association mode represents an association relation between action execution conditions and the device actions, and the scene mode represents a behavior scene of the user; Generating a corresponding skill package based on the behavior patterns for any one of the behavior patterns; When the triggering source is detected, the intelligent body controls the corresponding intelligent home equipment to execute corresponding actions according to the generated skill packages. In the embodiment of the application, the first behavior sequence formed by each interaction event of the user and the intelligent home equipment in the first time period is obtained, and multidimensional mining is carried out on the first behavior sequence, so that a high-order behavior mode comprising a plurality of equipment actions, action execution conditions and user behavior scenes with time rules and time sequence relations can be automatically mined, the complex behaviors of the user are understood, the skill package adapting to the life habits of the user can be automatically generated based on the high-order behavior modes, and when a trigger source is detected, the intelligent home equipment corresponding to each generated skill package is controlled by the intelligent body to execute corresponding actions, so that the intelligent body can realize self-adaptive intelligent home personalized service. In some embodiments of the first aspect, the periodic pattern comprises a periodic sequence pattern and/or a periodic association pattern, and the mining mode of the periodic sequence pattern comprises: Grouping the second behavior sequences according to different second time periods and date types aiming at the second behavior sequences of each day in the first behavior sequences to obtain first groups; extracting frequently-occurring first action subsequences from each first packet, and determining the first action subsequences as a second sequence mode; If the second sequence pattern appears stably in N continuous observation periods, determining the second sequence pattern as the periodic sequence pattern, wherein N is an integer greater than 1; The mining mode of the periodic association mode comprises the following steps: organizing the first behavior sequence into transactions according to a fixed time window, and attaching a time tag to each transaction, wherein the time tag comprises a time period corresponding to the transaction and the date type; grouping the transactions with the same time labels into the same group to obtain second groups; mining an association relationship between the action execution condition and the device action from each of the second packets, and determining the association relationship between the action execution condition mined from each of the second packets and the device action as a second