Family life event atlas construction method and system based on cooperation of multiple intelligent calculation engines

Abstract

The invention discloses a family life event atlas construction method and system based on cooperation of multiple intelligent computing engines, and relates to the technical field of knowledge atlas construction.The method comprises the steps that a family environment continuous audio stream is collected through an end side, effective fragments are screened through voice activity detection, and original audio is destroyed after acoustic feature vectors are extracted; and then a plurality of parallel intelligent calculation engines respectively recognize non-language environment sounds, emotion and voiceprint identities, semantics and potential intentions, a standardized family event object is generated through multi-modal fusion, the standardized family event object is stored in an independent research and development data warehouse of a time sequence + graph mixed framework, a family life knowledge graph is constructed, and finally graph changes are monitored through an inference engine, so that the family life knowledge graph is obtained. And actively triggering home equipment adjustment or pushing suggestions. According to the invention, active perception and semantic understanding of family events are realized, the risk of privacy disclosure is avoided, and the problems of passive response, data islands and privacy safety hidden troubles of the existing smart home are solved.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph construction technology, and more specifically to a method and system for constructing a family life event graph based on the collaboration of multiple intelligent computing engines. Background Technology

[0002] Currently, existing smart home and home hub technologies (such as smart speakers and home control screens) mainly operate using a "command interaction" mode: after the user issues a specific wake word and a clear command, the system uses a microphone to pick up the sound, cloud-based speech-to-text (ASR) and keyword matching to finally issue device control commands. Regarding data storage, existing data backends only record device physical status logs or user-initiated operation records, lacking semantic understanding of "people" and "events."

[0003] However, existing technologies have the following significant drawbacks: 1. Passive response, lack of proactive perception: It must rely on users to actively initiate interaction. If the user is unconscious (such as an elderly person who has fallen and cannot call for help), unable to speak (such as a baby crying), or in a non-interactive state (such as a couple arguing or a family gathering), the system cannot perceive what is happening in the family, resulting in a huge "blind spot" in intelligent services.

[0004] Data silos and semantic gaps: Existing data warehouses only store discrete "device logs," lacking semantic understanding of "people" and "events." The system knows "the light is off," but doesn't know if it's because "the person is sleeping" or "they've left home," failing to form a family knowledge graph with reasoning capabilities.

[0005] Privacy and security risks: Traditional solutions often require uploading the original recordings to the cloud for identification, which poses a risk of recording leakage and makes users reluctant to keep their microphones on all the time.

[0006] Therefore, how to provide a family life event map that can achieve proactive perception, semantic understanding, and privacy protection is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides a method and system for constructing a family life event graph based on the collaboration of multiple intelligent computing engines, so as to solve the problems existing in the background technology.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: A method for constructing a family life event graph based on multi-intelligent computing engine collaboration includes: S1: Acquire a continuous audio stream from the home environment, perform voice activity detection, and filter valid audio segments; extract acoustic feature vectors from the valid audio segments and destroy the original PCM audio waveform data; S2: The acoustic feature vector is distributed to multiple parallel intelligent computing engines, and each intelligent computing engine outputs the recognition result synchronously: Intelligent Computing Engine A: Identifies non-verbal environmental sounds through an audio spectrum classification model; Intelligent Computing Engine B: Analyzes audio intonation, energy, and speech rate, and outputs emotion tags and voiceprint identity ID; Intelligent Computing Engine C: Based on a local lightweight large model, it combines contextual analysis of fuzzy semantics to identify potential intentions; S3: Perform multimodal fusion on the output results of intelligent computing engines A, B, and C to generate standardized family event objects; S4: Store the standardized family event objects in a self-developed data warehouse using a time-series + graph hybrid architecture, and construct a family life knowledge graph; S5: The inference engine monitors changes in the family life knowledge graph in real time. When the graph matches a preset pattern, it automatically triggers adjustments to home appliances or pushes suggestions.

[0009] Optionally, the logic for determining valid audio segments is as follows: The edge chip calculates the audio energy distribution and zero-crossing rate through voice activity detection. When the signal envelope exceeds the preset dynamic threshold and the duration meets the set range, it is determined to be a valid segment.

[0010] Optionally, the voice activity detection specifically includes: A pre-trained speech activity detection model is built upon a large-scale acoustic dataset of home environments containing various background noise levels and a human speech / life event dataset to ensure high recall and low false alarm rates in complex home environments. Optionally, the intelligent computing engine B specifically comprises: Intonation: Extract the fundamental frequency variation curve and determine emotional fluctuations through standard deviation; Energy: Short-term energy threshold is used as one of the criteria for judging emotional arousal; Speech rate: Calculates the syllable density per unit time, with the threshold dynamically adjusted based on the historical average of family members; Voiceprint ID determination: Combining the family_members relationship table in the self-developed data warehouse, the stored member voiceprint features are matched using vector distance metric, supporting multi-tenant / multi-member isolation recognition.

[0011] Optionally, S2 further includes an anti-interference method, specifically: The system utilizes intelligent computing engine A to initially identify background ambient sounds, which serve as a noise mask to assist intelligent computing engine B in emotion filtering; alternatively, it employs dual verification using "voiceprint + location context," combining the access point ID in the devices table to pinpoint the sound source location, reducing confusion when multiple people are speaking. Optionally, the data structure of the standardized family event object is {Time: [specific moment], Person: [identity corresponding to voiceprint ID], Behavior: [acoustic event / semantic intent], Emotion: [emotion tag], Ambient Sound: [background sound type]}.

[0012] Optionally, the self-developed data warehouse with the time-series + graph hybrid architecture specifically includes: Temporal layer: records the long-term changing trends of sound characteristics; Graph layer: Based on a preset ontology, it automatically establishes the spatiotemporal relationships between various family event objects and constructs a family life knowledge graph.

[0013] Optionally, after the family life knowledge graph is constructed, the following steps are also included: S6: For the event nodes and memory nodes stored in the self-developed data warehouse, obtain the first search result set and the second search result set based on the full-text search index and the vector semantic search index, respectively, and use the inverse ranking fusion algorithm to merge and sort the candidate nodes in the first search result set and the second search result set to obtain the fusion score; S7: Sort the candidate nodes according to the fusion score, filter low-relevance candidate nodes based on a preset threshold, and use the remaining nodes as the target retrieval results for user queries.

[0014] Optionally, the fusion score of the inverse ranking fusion algorithm is calculated according to the following formula:

[0015] in, As a candidate node, This is the set of rankings of the candidate nodes in the full-text search channel and the vector semantic search channel. For candidate nodes in the channel The ranking order in the middle, It is a preset positive constant.

[0016] A system for constructing a family life event graph based on multi-intelligent computing engine collaboration includes: The audio preprocessing module acquires a continuous audio stream from the home environment, performs voice activity detection, and filters valid audio segments; it extracts acoustic feature vectors from the valid audio segments and destroys the original PCM audio waveform data. The intelligent computing routing and distribution module distributes the acoustic feature vector to multiple parallel intelligent computing engines; Multiple intelligent computing engines work together, with each engine simultaneously outputting recognition results. Intelligent Computing Engine A: Identifies non-verbal environmental sounds through an audio spectrum classification model; Intelligent Computing Engine B: Analyzes audio intonation, energy, and speech rate, and outputs emotion tags and voiceprint identity ID; Intelligent Computing Engine C: Based on a local lightweight large model, it combines contextual analysis of fuzzy semantics to identify potential intentions; The graph construction module performs multimodal fusion on the output results of the intelligent computing engines A, B, and C to generate standardized family event objects; the standardized family event objects are stored in a self-developed data warehouse with a time-series + graph hybrid architecture, and a family life knowledge graph is constructed. We have independently developed a data warehouse that uniformly stores acoustic feature vectors, time-series events, and relational data, and supports the stratification of hot and cold data. The proactive service control interface monitors changes in the family life knowledge graph in real time through the inference engine. When the graph matches a preset pattern, it automatically triggers adjustments to home appliances or pushes suggestions.

[0017] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a method for constructing a family life event graph based on multi-intelligent computing engine collaboration, which has the following beneficial effects: The leap from "passive instructions" to "active perception" and from "isolated logs" to "semantic graphs".

[0018] 1. Existing technologies rely on a passive response mode of "wake word + explicit command". This application realizes non-command perception, which can automatically identify non-interactive states such as elderly people falling, babies crying or emotional conflicts, filling the "perception blind spot" of intelligent services.

[0019] 2. Enhanced Decision-Making Depth: Existing technologies only record discrete device status logs, lacking semantic understanding. This application constructs a life graph through a spatiotemporal correlated data warehouse, enabling causal reasoning (such as identifying "nighttime childcare" scenarios), achieving an evolution from simple automation to "proactive service decision-making."

[0020] 3. Thorough Privacy Protection: Existing solutions pose a risk of cloud-based audio recording leakage. This application employs edge-side privacy protection technology, which immediately destroys the original audio after feature extraction, physically eliminating the possibility of leakage. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the method flow provided by the present invention; Figure 2 This is a schematic diagram illustrating the principle and structure of the method provided by the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] This invention discloses a method for constructing a family life event graph based on multi-intelligent computing engine collaboration, such as... Figure 1 and Figure 2 As shown, it includes: S1: Non-invasive audio stream acquisition and end-to-end privacy desensitization By acquiring a continuous audio stream of the home environment through a microphone array, voice activity detection (VAD) is immediately performed on the edge chip to filter valid audio segments. Acoustic feature vectors are extracted from the valid audio segments, and the original PCM audio waveform data is destroyed to ensure that no recoverable recording files are stored or uploaded, thus solving the privacy problem at its source.

[0025] S2: Multi-engine collaborative computing based on expert hybrid model (MoE)

[0026] The acoustic feature vectors are distributed to multiple parallel intelligent computing engines, and each engine outputs the recognition results synchronously. Intelligent Computing Engine A (Acoustic Event Expert): Identifies "what happened". Utilizing an audio spectrum classification model, it identifies non-verbal environmental sounds (e.g., breaking glass, baby crying, hurried footsteps, snoring). Intelligent Computing Engine B (Sub-Language / Emotion Expert): Identifies "emotional states." It analyzes the tone, energy, and speed of audio, and outputs emotion tags (anxiety, anger, joy) and voiceprint identity ID.

[0027] Intelligent Computing Engine C (Semantic Reasoning Expert): Identifies "latent intents". It utilizes a local lightweight large model (LLM) and combines contextual analysis to understand fuzzy semantics (e.g., a user saying "It's so cold" is inferred as "thermal comfort needs" rather than simply transcribing text).

[0028] S3: Data Fusion and Event Instantiation

[0029] The outputs of the intelligent computing engines A, B, and C are fused in a multimodal manner to generate standardized family event objects. The data structure of the standardized family event objects is {Time: [specific moment], Person: [identity corresponding to voiceprint ID], Behavior: [acoustic event / semantic intent], Emotion: [emotion tag], Ambient sound: [background sound type]}.

[0030] Example data structure: {Time:22:15, Person: Father, Behavior: Continuous coughing, Emotion: Fatigue, Ambient sound: Bedroom background sound}.

[0031] S4: Autonomous Data Warehouse and Life Graph Construction

[0032] Standardized family event objects are stored in a self-developed data warehouse employing a hybrid time-series and graph architecture, and a family life knowledge graph is constructed; specifically: Temporal layer: records the long-term trend of changes in sound characteristics (e.g., the curve of cough frequency over the past month).

[0033] Knowledge Graph: Automatically establishes connections based on a predefined ontology. For example, it automatically connects "late night time node" + "baby crying node" + "mother's hurried footsteps node" to construct a knowledge graph structure for "nighttime childcare scene".

[0034] S5: Graph-based proactive service decision-making

[0035] The inference engine monitors changes in the family life knowledge graph in real time. When the graph matches a preset pattern (such as "argument mode" or "health abnormality mode"), it automatically triggers adjustments to home appliances (such as adjusting the lighting atmosphere or playing music) or pushes suggestions.

[0036] Specifically, S1 also includes: Judgment logic: The audio energy distribution and zero-crossing rate are calculated by the Local Edge chip through Voice Activity Detection (VAD). When the signal envelope exceeds the preset dynamic threshold and the duration meets the set range, it is judged as a "valid segment".

[0037] Preprocessing includes noise reduction, gain control, and feature extraction (such as extracting embedding vectors).

[0038] Model Training: This speech activity detection model needs to be pre-trained by comparing it with a large-scale home environment acoustic dataset (containing various background noises such as wind and appliance noise) and a human speech / life event dataset to ensure high recall and low false alarm rate in complex home environments.

[0039] Intelligent computing engine B specifically includes: 1. Tone: Extract the variation curve of the fundamental frequency (F0) and determine emotional fluctuations through the standard deviation; 2. Energy: Short-term energy threshold (e.g., >60dB) is used as one of the criteria for judging emotional arousal (anger / joy); 3. Speech rate: Calculates the syllable density per unit time, with the threshold dynamically adjusted based on the historical average of family members; 4. Voiceprint ID determination: Combining the family_members relationship table in the self-developed data warehouse, the stored member voiceprint features are matched using vector distance metrics (such as cosine similarity), supporting multi-tenant / multi-member isolation recognition.

[0040] S2 also includes anti-interference methods, specifically: Intelligent computing engine collaboration: Intelligent computing engine A (acoustic expert) is used to identify background ambient sounds (such as TV sound, pet barking) in advance, and it is used as a noise mask to assist engine B in emotion filtering.

[0041] Anti-obfuscation strategy: Employs dual verification of "voiceprint + location context" and combines the access point ID in the devices table to lock the location of the sound source, reducing the obfuscation level when multiple people are speaking.

[0042] Based on the above embodiments, the present invention preferably further includes, after constructing the family life knowledge graph, the following: 1) Hybrid Retrieval and Result Fusion: A self-developed data warehouse establishes full-text search indexes and vector semantic search indexes for family event objects and long-term preference memories, respectively. When a user initiates a natural language query (e.g., "Which days has Dad coughed most severely at night recently?" or "What's Mom's favorite tea?"), the system first obtains a first candidate set based on keywords in the full-text search channel and a second candidate set based on semantic similarity in the vector semantic channel, calculating the ranking order of candidate nodes in both sets. Then, a reciprocal ranking fusion algorithm is used to weightedly sum the rankings of each candidate node in different channels, for example, according to... The fusion score is obtained through a process that sorts the nodes from highest to lowest fusion score. Candidate nodes below a preset threshold are filtered out to obtain the final target retrieval results. As a candidate node, This is the set of rankings of the candidate nodes in the full-text search channel and the vector semantic search channel. For candidate nodes in the channel The ranking order in the middle, It is a preset positive constant.

[0043] 2) Voiceprint-based member identification and rule binding: The system extracts acoustic embedding vectors from multiple speech samples of the same family member and performs normalized averaging to construct a template voiceprint vector for that member, which is then stored in the family_members relation table. After extracting the embedding vector to be identified from the real-time acquired audio segments, the system calculates the cosine similarity with each template voiceprint vector. If the maximum similarity exceeds a threshold, the segment is bound to the corresponding family member node, thereby forming a multi-dimensional association of "time-person-behavior-emotion" in the life knowledge graph.

[0044] 3) Parsing of reminder rules for Chinese time and frequency expressions: When the system detects that the user's voice contains Chinese time and frequency expressions such as "every day", "every week", "every N minutes / hours / day", and "arrive home / arrive at the company", it first parses out the specific numerical parameters and time points based on the preset Chinese number mapping table and regular expressions (for example, parsing "every two hours" as an interval of 120 minutes, and parsing "9 am every day" as a timed rule of 09:00 every day). At the same time, it identifies conditional trigger events containing semantics such as "go home / arrive home". Subsequently, the parsing results are encapsulated into structured reminder rule objects and written into the rule table in the self-developed data warehouse. It is then associated with the corresponding family member nodes and scene nodes in the family life knowledge graph, so that the inference engine can automatically generate corresponding reminder events or device control commands when it recognizes the corresponding pattern or reaches the corresponding time point.

[0045] A system for constructing a family life event graph based on multi-intelligent computing engine collaboration includes: The audio preprocessing module acquires a continuous audio stream from the home environment, performs voice activity detection, and filters valid audio segments; it extracts acoustic feature vectors from the valid audio segments and destroys the original PCM audio waveform data. The intelligent computing routing and distribution module distributes acoustic feature vectors to multiple parallel intelligent computing engines; Multiple intelligent computing engines work together, with each engine simultaneously outputting recognition results. Intelligent Computing Engine A: Identifies non-verbal environmental sounds through an audio spectrum classification model; Intelligent Computing Engine B: Analyzes audio intonation, energy, and speech rate, and outputs emotion tags and voiceprint identity ID; Intelligent Computing Engine C: Based on a local lightweight large model, it combines contextual analysis of fuzzy semantics to identify potential intentions; The graph construction module performs multimodal fusion on the output results of the intelligent computing engines A, B, and C to generate standardized family event objects; the standardized family event objects are stored in a self-developed data warehouse that adopts a time-series + graph hybrid architecture, and a family life knowledge graph is constructed. We have independently developed a data warehouse that uniformly stores acoustic feature vectors, time-series events, and relational data, and supports the stratification of hot and cold data. The proactive service control interface monitors changes in the family life knowledge graph in real time through the inference engine. When the graph matches a preset pattern, it automatically triggers adjustments to home appliances or pushes suggestions.

[0046] See Table 1 for details.

[0047] Table 1 System Module Function Introduction

[0048] The following specific examples further illustrate the method of the present invention.

[0049] Example 1: Health monitoring of elderly people living alone

[0050] Perception: The intelligent computing engine detected a "heavy object falling sound" (acoustic event) in the restroom area, followed by a "painful groan" (emotional event), but did not detect "standing footsteps".

[0051] Processing: The self-developed data warehouse correlated the three discrete events in the family life knowledge graph and matched them with the "suspected fall model".

[0052] Response: If the danger level is determined to be "high", an emergency alarm will be automatically sent to the child's mobile phone without the user needing to call for help, and a two-way call will be established for confirmation.

[0053] Example 2: Intelligent Adjustment of Home Atmosphere

[0054] Perception: The intelligent computing engine B recognized that the conversation between the two contained high-frequency negative words, and the emotion engine detected that the tone was high and the speaking speed was extremely fast (judged as "anger / argument").

[0055] Processing: The self-developed data warehouse records the current time period as "family conflict atmosphere".

[0056] Response: Automatically adjusts all room lighting to a soft, warm tone, turns off the TV background noise, and plays soothing white noise to help calm emotions.

[0057] Example 3: Privacy Protection Circuit Breaker

[0058] Perception: The intelligent computing engine C recognizes sensitive intent features such as "password", "bank card", and "ID card" in the dialogue.

[0059] Action taken: Trigger the underlying privacy circuit breaker mechanism to immediately stop subsequent feature extraction and data entry operations, and insert a "privacy shielding tag" containing only a timestamp into the self-developed data warehouse to ensure that sensitive information is not recorded.

[0060] The method of this invention automatically adjusts the color temperature of lights, background music, and indoor temperature and humidity according to the sound atmosphere of the home environment (such as laughter, arguments, and quiet reading), realizing the evolution from "smart single product" to "smart scene".

[0061] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0062] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for constructing a family life event graph based on multi-intelligent computing engine collaboration, characterized in that, include: S1: Acquire a continuous audio stream from the home environment, perform voice activity detection, and filter valid audio segments; extract acoustic feature vectors from the valid audio segments and destroy the original PCM audio waveform data; S2: The acoustic feature vector is distributed to multiple parallel intelligent computing engines, and each intelligent computing engine outputs the recognition result synchronously: Intelligent Computing Engine A: Identifies non-verbal environmental sounds through an audio spectrum classification model; Intelligent Computing Engine B: Analyzes audio intonation, energy, and speech rate, and outputs emotion tags and voiceprint identity ID; Intelligent Computing Engine C: Based on a local lightweight large model, it combines contextual analysis of fuzzy semantics to identify potential intentions; S3: Perform multimodal fusion on the output results of intelligent computing engines A, B, and C to generate standardized family event objects; S4: Store the standardized family event objects in a self-developed data warehouse using a time-series + graph hybrid architecture, and construct a family life knowledge graph; S5: The inference engine monitors changes in the family life knowledge graph in real time. When the graph matches a preset pattern, it automatically triggers adjustments to home appliances or pushes suggestions.

2. The method for constructing a family life event graph based on multi-intelligent computing engine collaboration as described in claim 1, characterized in that, The logic for determining valid audio segments is as follows: The edge chip calculates the audio energy distribution and zero-crossing rate through voice activity detection. When the signal envelope exceeds the preset dynamic threshold and the duration meets the set range, it is determined to be a valid segment.

3. The method for constructing a family life event graph based on multi-intelligent computing engine collaboration according to claim 1, characterized in that, The voice activity detection specifically includes: A pre-trained speech activity detection model is conducted based on a large-scale home environment acoustic dataset containing various background noises and a human speech / life event dataset to ensure high recall and low false alarm rate in complex home environments.

4. The method for constructing a family life event graph based on multi-intelligent computing engine collaboration according to claim 1, characterized in that, The intelligent computing engine B is specifically: Intonation: Extract the fundamental frequency variation curve and determine emotional fluctuations through standard deviation; Energy: Short-term energy threshold is used as one of the criteria for judging emotional arousal; Speech rate: Calculates the syllable density per unit time, with the threshold dynamically adjusted based on the historical average of family members; Voiceprint ID determination: Combining the family_members relationship table in the self-developed data warehouse, the stored member voiceprint features are matched using vector distance metric, supporting multi-tenant / multi-member isolation recognition.

5. The method for constructing a family life event graph based on multi-intelligent computing engine collaboration according to claim 1, characterized in that, The S2 also includes an anti-interference method, specifically: The system utilizes intelligent computing engine A to identify background ambient sounds in advance, which serve as a noise mask to assist intelligent computing engine B in emotion filtering; or it employs dual verification of "voiceprint + location context" and combines the access point ID in the devices table to lock the location of the sound source, thereby reducing the confusion when multiple people are speaking.

6. The method for constructing a family life event graph based on multi-intelligent computing engine collaboration according to claim 1, characterized in that, The data structure of the standardized family event object is {Time: [Specific moment], Person: [Identity corresponding to voiceprint ID], Behavior: [Acoustic event / semantic intent], Emotion: [Emotion tag], Ambient sound: [Background sound type]}.

7. The method for constructing a family life event graph based on multi-intelligent computing engine collaboration according to claim 1, characterized in that, The self-developed data warehouse with the time-series + graph hybrid architecture specifically includes: Temporal layer: records the long-term changing trends of sound characteristics; Graph layer: Based on a preset ontology, it automatically establishes the spatiotemporal relationships between various family event objects and constructs a family life knowledge graph.

8. The method for constructing a family life event graph based on multi-intelligent computing engine collaboration according to claim 1, characterized in that, After the family life knowledge graph is constructed, the following steps are also included: S6: For the event nodes and memory nodes stored in the self-developed data warehouse, obtain the first search result set and the second search result set based on the full-text search index and the vector semantic search index, respectively, and use the inverse ranking fusion algorithm to merge and sort the candidate nodes in the first search result set and the second search result set to obtain the fusion score; S7: Sort the candidate nodes according to the fusion score, filter low-relevance candidate nodes based on a preset threshold, and use the remaining nodes as the target retrieval results for user queries.

9. A method for constructing a family life event graph based on multi-intelligent computing engine collaboration as described in claim 8, characterized in that, The fusion score of the inverse ranking fusion algorithm is calculated according to the following formula: in, As a candidate node, This is the set of rankings of the candidate nodes in the full-text search channel and the vector semantic search channel. For candidate nodes in the channel The ranking order in the middle, It is a preset positive constant.

10. A system for constructing a family life event graph based on multi-intelligent computing engine collaboration, characterized in that, The method for constructing a family life event graph based on multi-intelligent computing engine collaboration, as described in any one of claims 1-9, includes: The audio preprocessing module acquires a continuous audio stream from the home environment, performs voice activity detection, and filters valid audio segments; it extracts acoustic feature vectors from the valid audio segments and destroys the original PCM audio waveform data. The intelligent computing routing and distribution module distributes the acoustic feature vector to multiple parallel intelligent computing engines; Multiple intelligent computing engines work together, with each engine simultaneously outputting recognition results. Intelligent Computing Engine A: Identifies non-verbal environmental sounds through an audio spectrum classification model; Intelligent Computing Engine B: Analyzes audio intonation, energy, and speech rate, and outputs emotion tags and voiceprint identity ID; Intelligent Computing Engine C: Based on a local lightweight large model, it combines contextual analysis of fuzzy semantics to identify potential intentions; The graph construction module performs multimodal fusion on the output results of the intelligent computing engines A, B, and C to generate standardized family event objects; the standardized family event objects are stored in a self-developed data warehouse with a time-series + graph hybrid architecture, and a family life knowledge graph is constructed. We have independently developed a data warehouse that uniformly stores acoustic feature vectors, time-series events, and relational data, and supports the stratification of hot and cold data. The proactive service control interface monitors changes in the family life knowledge graph in real time through the inference engine. When the graph matches a preset pattern, it automatically triggers adjustments to home appliances or pushes suggestions.

Images