Method and apparatus for intelligent control of devices in a shared office based on user status

Abstract

The application discloses a shared office equipment intelligent control method and device based on user state, a server method comprising: acquiring multi-modal sensing data of each shared office at each historical time in a preset period, the multi-modal sensing data comprising motion data, seat pressure data and personal article state data; performing key feature extraction and feature fusion on the motion data, the seat pressure data and the personal article state data to obtain multi-modal features at each historical time; determining a user state corresponding to each shared office according to the multi-modal features at each historical time, a predefined behavior mode or a pre-tuned user state recognition model; and sending the user state corresponding to each shared office to a preset resource scheduling platform to automatically control running equipment of each shared office. Therefore, by using the application, the operating cost can be reduced, the stability of the system can be improved, and the shared office can adapt to frequent personnel flow and diversified complex scenes.

Description

Technical Field

[0001] This application relates to the field of intelligent environmental sensing and control technology, and in particular to a method and apparatus for intelligent control of equipment in a shared office based on user status. Background Technology

[0002] In modern shared office environments, users move frequently, and their presence status (e.g., present, short-term absence, permanent absence) is highly dynamic. In this scenario, the usage of office equipment (such as air conditioning, lighting, and electrical outlets) needs to be flexibly adjusted based on the actual presence of personnel to achieve energy conservation and emission reduction.

[0003] In related technologies, shared office environments commonly use a single sensor to detect the presence of personnel. This is achieved by detecting the thermal signals of the human body. When the detected signal disappears, the administrator is notified to check video surveillance to confirm the personnel's actual status. The administrator can then manually turn off relevant equipment (such as air conditioning and lights) based on their judgment.

[0004] However, a single sensor can only detect the thermal signal of a person and cannot accurately distinguish whether a person is leaving temporarily or permanently, leading to delays and inaccuracies in equipment control. Secondly, administrators need to manually analyze and judge through video surveillance, which is inefficient and difficult to adapt to the complex scenarios of frequent personnel movement and diverse behavioral patterns in shared office environments. Summary of the Invention

[0005] This application provides a method and apparatus for intelligent control of equipment in a shared office based on user status. To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not intended as a general description, nor is it intended to identify key / important components or describe the scope of protection of these embodiments. Its sole purpose is to present some concepts in a simple form as a prelude to the detailed description that follows.

[0006] In a first aspect, embodiments of this application provide a method for intelligent control of equipment in a shared office based on user status, applied on a server side, the method comprising:

[0007] Acquire multimodal perception data for each shared office at various historical moments within a preset period. The multimodal perception data includes motion data, seat pressure data, and personal item status data.

[0008] Key features are extracted and fused from motion data, seat pressure data, and personal item status data to obtain multimodal features for each historical moment;

[0009] Based on the multimodal features of each historical moment, predefined behavioral patterns, or pre-tuned user state recognition models, the user state corresponding to each shared office is determined. The predefined behavioral patterns are used to characterize the correspondence between different combinations of user behaviors in the shared office and the user state. The pre-tuned user state recognition model is trained by combining unlabeled general model weight learning and labeled model fine-tuning learning.

[0010] Send the user status corresponding to each shared office to the preset resource scheduling platform to automatically control the operating equipment of each shared office.

[0011] Optionally, before acquiring the multimodal sensing data of each shared office at various historical moments within a preset period, the following steps are also included:

[0012] When a user activates a shared office, various sensing devices pre-deployed in the shared office are activated; these sensing devices include millimeter-wave radar, pressure sensors, and personal item detection components.

[0013] Millimeter-wave radar is used to collect users' movements, directions of movement, and speeds within the shared office space, which serve as motion data.

[0014] The pressure data of the chairs in the shared office is obtained through pressure sensors;

[0015] The personal items detection component monitors the status of the user's personal items and obtains personal item status data.

[0016] Preprocessing and aligning motion data, seat pressure data, and personal item status data with time windows yields multimodal perception data for each historical moment.

[0017] Store multimodal sensing data to a time-series data platform.

[0018] Optionally, key features can be extracted and fused from motion data, seat pressure data, and personal item status data to obtain multimodal features for each historical moment, including:

[0019] By analyzing the data, we can determine whether the user gets up, moves towards the door, and the speed of movement, thus obtaining movement trajectory characteristics.

[0020] By using seat pressure data, the occupancy status of the seat is determined, and the seat pressure characteristics are obtained;

[0021] By analyzing the status data of personal belongings, we can determine whether users have personal belongings in the shared office and obtain the characteristics of the items they carry.

[0022] The movement trajectory features, seat pressure features, and item carrying features are spliced ​​together to obtain multimodal features for each historical moment.

[0023] Optionally, the multimodal features at each historical moment include movement trajectory features, seat pressure features, and item carrying features; predefined behavioral patterns include behavioral patterns corresponding to the in-place state, behavioral patterns corresponding to the short-term departure state, and behavioral patterns corresponding to the permanent departure state.

[0024] Based on the multimodal characteristics and predefined behavioral patterns at each historical moment, the user status corresponding to each shared office is determined, including:

[0025] The results of statistically analyzing the most frequently occurring trajectory features at each historical moment were obtained.

[0026] By analyzing the seat pressure characteristics at various historical moments, the results of the most frequently occurring pressure characteristics were obtained.

[0027] The results of statistical analysis of the most frequently occurring item characteristics were obtained by analyzing the characteristics of items carried at various historical moments.

[0028] Based on the behavioral patterns corresponding to the in-place state, the behavioral patterns corresponding to the short-term departure state, the behavioral patterns corresponding to the permanent departure state, the results of the trajectory characteristics, the results of the stress characteristics, and the results of the item characteristics, the user status corresponding to each shared office is analyzed.

[0029] Optionally, based on the behavioral patterns corresponding to the in-place state, the behavioral patterns corresponding to the short-term departure state, the behavioral patterns corresponding to the permanent departure state, and the results of trajectory characteristics, stress characteristics, and item characteristics, the user status corresponding to each shared office is analyzed, including:

[0030] When the pressure characteristics indicate that the seat is occupied, the trajectory characteristics indicate that the user has not gotten up, has not moved towards the door, and is moving slowly, and the item characteristics indicate that personal belongings are indoors, the behavioral pattern corresponding to the in-place state is determined; the in-place state is taken as the user state for each shared office; or...

[0031] When the pressure characteristics indicate that the seat is unoccupied, the trajectory characteristics indicate that the user gets up, moves towards the door, and moves quickly, and the item characteristics indicate that personal belongings are indoors, the behavioral pattern corresponding to the short-term departure state is determined; the short-term departure state is taken as the user state for each shared office; or...

[0032] When the pressure characteristic indicates that the seat is not occupied, the trajectory characteristic indicates that the user gets up, moves towards the door, or moves quickly, and the item characteristic indicates that personal items are not indoors, the behavior pattern corresponding to the permanent departure state is determined; the permanent departure state is taken as the user state corresponding to each shared office.

[0033] Optionally, based on the multimodal features of each historical moment and a pre-tuned user status recognition model, the user status corresponding to each shared office is determined, including:

[0034] Initialize the pre-tuned user state determination model;

[0035] In chronological order, the multimodal features of each historical moment are sequentially input into the pre-tuned user state determination model;

[0036] Output the user status for each shared office.

[0037] Optionally, a pre-tuned user state determination model is generated by following these steps:

[0038] Collect and preprocess historical multimodal sensing data of each shared office within a preset historical time period to obtain historical modal data of each shared office at each historical moment;

[0039] The historical modal data of adjacent historical moments in each historical moment of each shared office are combined into positively correlated first perception data pairs;

[0040] From the historical modal data of each historical moment in each shared office, randomly select any two historical modal data from different shared offices and combine them into a negatively correlated second perception data pair;

[0041] By performing unlabeled general model weight learning on each first perception data pair and each second perception data pair, a basic model that can identify the correlation features between adjacent time points is obtained.

[0042] From the historical modal data at various historical moments, a portion of the historical modal data is selected and labeled to reflect the actual user state under the historical modal data, resulting in a small number of model fine-tuning samples;

[0043] Based on the model fine-tuning samples, the basic model is subjected to labeled model fine-tuning learning to obtain a pre-fine-tuned user state determination model.

[0044] Optionally, based on each first-sensory data pair and each second-sensory data pair, unlabeled general model weights are learned to obtain a basic model capable of identifying correlation features between adjacent time points, including:

[0045] Create an initial network architecture, which includes a feature embedding module, a similarity calculation module, and a loss function for unlabeled learning, connected in sequence.

[0046] Normalize the two historical modal data in each first perception data pair and each second perception data pair;

[0047] Each normalized first perception data pair and each normalized second perception data pair are input into the feature embedding module to encode the two historical modal data in the perception data pair, thereby obtaining the embedded representation of each feature.

[0048] The embedding representation of each feature is concatenated to obtain two overall embedding representations for each first perceptual data pair and two overall embedding representations for each second perceptual data pair.

[0049] The two wholes are embedded to represent the input similarity calculation module, and the cosine similarity of each first perceptual data pair and the cosine similarity of each second perceptual data pair are calculated.

[0050] The correlation loss value is calculated using a loss function and cosine similarity; when the correlation loss value reaches its minimum, a basic model capable of identifying correlation features between adjacent time points is generated.

[0051] Among them, the loss function value The calculation formula is:

[0052]

[0053] Wherein, the cosine similarity of each first perception data pair is , It is the first Cosine similarity of a second-sensory data pair It is the number of second-sensory data pairs. It is a temperature parameter used to control the distribution of similarity values.

[0054] Optionally, based on model fine-tuning samples, labeled model fine-tuning learning is performed on the base model to obtain a pre-fine-tuned user state determination model, including:

[0055] Extract the basic data features and label features that reflect the real user status for each model fine-tuning sample; the basic data features include movement trajectory features, seat pressure features, and item carrying features;

[0056] The basic data features and label features are mapped to fixed-dimensional embedding vectors to obtain the feature vector of each model fine-tuning sample;

[0057] The feature vector is input into the base model for labeled model fine-tuning, and the cross-entropy loss value between the user state predicted by the base model and the real user state of each model fine-tuning sample is output.

[0058] When the cross-entropy loss value reaches its minimum, a pre-fine-tuned user state determination model is generated.

[0059] Alternatively, if the cross-entropy loss value has not reached its minimum, continue performing the step of inputting the feature vector into the base model for labeled model fine-tuning until the cross-entropy loss value reaches its minimum.

[0060] Secondly, embodiments of this application provide an intelligent control device for a shared office based on user status, the device comprising:

[0061] The multimodal perception data acquisition module is used to acquire multimodal perception data of each shared office at various historical moments within a preset period. The multimodal perception data includes motion data, seat pressure data, and personal item status data.

[0062] The feature processing module is used to extract and fuse key features from motion data, seat pressure data, and personal item status data to obtain multimodal features for each historical moment.

[0063] The user status determination module is used to determine the user status corresponding to each shared office based on the multimodal features of each historical moment, predefined behavioral patterns, or pre-tuned user status recognition models. The predefined behavioral patterns are used to characterize the correspondence between different combinations of user behaviors in the shared office and the user status. The pre-tuned user status recognition model is trained by combining unlabeled general model weight learning and labeled model fine-tuning learning.

[0064] The automation control module is used to send the user status corresponding to each shared office to the preset resource scheduling platform in order to automatically control the operating equipment of each shared office.

[0065] The technical solutions provided in this application embodiment may include the following beneficial effects:

[0066] In this embodiment, on the one hand, by collecting motion data, seat pressure data, and personal item status data, combined with key feature extraction and feature fusion technologies, the system can comprehensively and accurately perceive user behavior and status. This allows the system to accurately distinguish whether a user is leaving temporarily or permanently, thereby achieving precise equipment control and improving energy efficiency. On the other hand, predefined behavioral patterns can characterize the correspondence between different combinations of user behaviors and user status within the shared office, providing the system with clear judgment criteria. Simultaneously, the model training method, combining unlabeled general model weight learning and labeled model fine-tuning learning, enables the model to be quickly fine-tuned to adapt to various complex scenarios. These two methods significantly improve management efficiency, reduce operating costs, and enhance system stability and reliability, making it adaptable to the complex scenarios of frequent personnel movement and diverse behavioral patterns in shared offices.

[0067] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0068] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0069] Figure 1 This is a schematic flowchart of a method for intelligent control of equipment in a shared office based on user status, provided in an embodiment of this application.

[0070] Figure 2 This is a schematic diagram of a perception interface for multimodal sensing data provided in an embodiment of this application;

[0071] Figure 3 This is a visual schematic diagram of a fused feature vector provided in an embodiment of this application;

[0072] Figure 4 This is a schematic block diagram illustrating a process for intelligent control of equipment in a shared office based on user status, as provided in an embodiment of this application.

[0073] Figure 5 This is a status diagram of the status of each shared office displayed in the background, provided in an embodiment of this application;

[0074] Figure 6 This is a flowchart illustrating a model training method for a user state determination model provided in an embodiment of this application.

[0075] Figure 7 This is an architecture diagram of an initial network architecture provided in an embodiment of this application;

[0076] Figure 8 This is a schematic diagram of the structure of a smart control device for a shared office based on user status, provided in an embodiment of this application.

[0077] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0078] The following description and accompanying drawings fully illustrate specific embodiments of this application to enable those skilled in the art to practice them.

[0079] It should be understood that the described embodiments are merely some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0080] In the following description, when referring to the accompanying drawings, the same numbers in different drawings denote the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0081] In the description of this application, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances. Furthermore, in the description of this application, unless otherwise stated, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship.

[0082] Currently, shared office environments commonly use a single sensor to detect the presence of personnel. This sensor determines occupancy by detecting the thermal signals of the human body. When the detected signal disappears, the administrator is notified to check video surveillance to confirm the personnel's actual status. The administrator can then manually turn off relevant equipment (such as air conditioning and lights) based on their judgment.

[0083] The inventors realized that a single sensor can only detect the thermal signal of a person and cannot accurately distinguish whether a person is leaving temporarily or permanently, leading to delays and inaccuracies in equipment control. Secondly, administrators need to manually analyze and judge through video surveillance, which is inefficient and difficult to adapt to the complex scenarios of frequent personnel movement and diverse behavioral patterns in shared office environments.

[0084] To address the existing technical problems, this application provides a method and apparatus for intelligent equipment control in a shared office based on user status, thereby resolving the issues mentioned above. In this application's embodiments, on one hand, by collecting motion data, seat pressure data, and personal item status data, combined with key feature extraction and feature fusion techniques, the system can comprehensively and accurately perceive user behavior and status. This allows the system to accurately distinguish between short-term and permanent user departures within a short time, achieving precise equipment control and improving energy efficiency. On the other hand, predefined behavioral patterns can characterize the correspondence between different combinations of user behaviors and user status within the shared office, providing the system with clear judgment criteria. Simultaneously, a model training method combining unlabeled general model weight learning and labeled model fine-tuning learning allows the model to be quickly fine-tuned to adapt to various complex scenarios. These two methods significantly improve management efficiency, reduce operating costs, and enhance system stability and reliability, adapting to the complex scenarios of frequent personnel movement and diverse behavioral patterns in shared offices. Exemplary embodiments are described in detail below.

[0085] The following will be combined with the appendix Figure 1 -Appendix Figure 7 This application provides a detailed description of the intelligent device control method for shared offices based on user status, as provided in its embodiments. This method can be implemented using a computer program and can run on an intelligent device control system for shared offices based on the von Neumann architecture. The computer program can be integrated into an application or run as a standalone utility application.

[0086] Please see Figure 1 This document provides a flowchart illustrating a method for intelligent device control in a shared office based on user status, applicable to the server side. Figure 1 As shown, the method in this application embodiment includes the following steps:

[0087] S101, acquire multimodal perception data of each shared office at each historical moment within a preset period. The multimodal perception data includes motion data, seat pressure data, and personal item status data.

[0088] A shared office is a workplace that provides shared workspaces equipped with basic facilities such as desks, chairs, and internet access. A preset period is a pre-defined time range used for periodic data collection and analysis; for example, a period of 1 minute or 2 minutes. Historical moments are each consecutive point in time within the preset period. Multimodal sensing data refers to different types of data collected through various sensors. Motion data is information collected by sensors regarding the user's actions, direction of movement, and speed within the shared office. Millimeter-wave radar can detect whether a user is walking around the office, moving towards the door, and at what speed. Seat pressure data refers to information collected by pressure sensors regarding whether a user is sitting in a chair. Pressure sensors are installed on the chair to detect whether the user is seated. Personal belongings status data refers to the status of personal belongings carried by the user, collected by sensors (such as RFID tags, cameras, etc.), such as whether the items are left in the office.

[0089] In some embodiments of this application, when the duration between the historical moment and the current moment of the previous execution of the shared office device intelligent control method based on user status is equal to a preset period, a time series data platform can be connected. Using the historical moment and the current moment as query parameters, motion data, seat pressure data, and personal item status data of each shared office at each historical moment within the preset period can be obtained from the time series data platform, thereby obtaining multimodal perception data of each shared office at each historical moment within the preset period.

[0090] In one possible implementation, with a preset period of 1 minute, the historical time when the last execution of the shared office device intelligent control method based on user status was 16:00 on November 24, 2025, and the current time is 16:01 on November 24, 2025. At this time, the one-minute interval is met, so it is possible to connect to the time series data platform and query 60 seconds of multimodal sensing data within the period from 16:00 on November 24, 2025 to 16:01 on November 24, 2025.

[0091] It should be noted that by acquiring multimodal perception data within one minute, the user's status can be analyzed in a timely manner, thereby improving the system's response performance.

[0092] In some embodiments of this application, the specific generation process of real-time storage of multimodal sensing data by the time-series data platform includes: when the shared office is activated by a user, activating multiple sensing devices pre-deployed in the shared office; the multiple sensing devices include millimeter-wave radar, pressure sensors, and personal item detection components; collecting the user's actions, movement direction, and speed in the shared office through millimeter-wave radar as motion data; acquiring seat pressure data of the seats in the shared office through pressure sensors; monitoring the status of the personal items carried by the user through the personal item detection components to obtain personal item status data; preprocessing and aligning the motion data, seat pressure data, and personal item status data with time windows to obtain multimodal sensing data for each historical moment; and storing the multimodal sensing data to the time-series data platform.

[0093] In one possible implementation, millimeter-wave radar, pressure sensors, and personal item detection components are installed in a shared office space. The millimeter-wave radar is mounted on the ceiling or walls to detect user movement. The pressure sensors are mounted on the chairs to detect whether a user is sitting. The personal item detection components are installed in the shared office to detect the status of personal items. Data can be collected every minute. The millimeter-wave radar collects user movement data, including motion, direction of movement, and speed. The pressure sensors collect pressure data from the chairs to determine whether a user is sitting. The personal item detection components collect status data of personal items to determine whether the items are left in the office. Noise and outliers are removed from the collected data, data from different sensors are normalized to the same units and time-aligned, and the pre-processed data is stored in a time-series data platform.

[0094] For example, the shared office is equipped with millimeter-wave radar, pressure sensors, and personal belongings detection components (such as RFID readers), which can collect data once per second. At 10:00:00, the millimeter-wave radar detects a user moving around in the office, heading towards the door at a relatively fast speed; the pressure sensor detects no pressure on the chair, indicating that the user is not in the chair; the RFID reader detects that the user's personal belongings are not in the office. The real-time acquired sensing data is preprocessed and aligned with a time window, and the aligned data can be stored in a time-series data platform. Through the backend, the multimodal sensing data of office A-101 within the last 30 seconds can be viewed, such as... Figure 2 As shown.

[0095] It should be noted that the personal belongings detection component can be a camera or an RFID reader. The camera analyzes captured images to determine if a user's personal belongings are in the office. The RFID reader is installed near the door, work area, or locker. Passive RFID tags are attached to the user's personal belongings (such as laptops, backpacks, etc.). When a user enters or leaves the office, the RFID reader emits a radio frequency signal. If the RFID tag on the user's personal belongings enters the reader's signal range, the tag is activated and sends the data stored in its chip (such as the item's unique identifier) ​​back to the reader. Upon receiving the data, the reader decodes and records the item's status (such as whether the item is indoors). The shared office in this application is a single-person room.

[0096] S102, extract and fuse key features from motion data, seat pressure data, and personal item status data to obtain multimodal features for each historical moment;

[0097] In some embodiments of this application, the specific process of extracting and fusing key features from motion data, seat pressure data, and personal item status data to obtain multimodal features at each historical moment includes: analyzing whether the user got up, moved towards the door, and the speed of movement based on the motion data to obtain motion trajectory features; determining the occupancy status of the seat based on the seat pressure data to obtain seat pressure features; determining whether the user has personal items in the shared office based on the personal item status data to obtain item carrying features; and concatenating the motion trajectory features, seat pressure features, and item carrying features to obtain multimodal features at each historical moment.

[0098] The movement trajectory features are user movement-related characteristics obtained by analyzing motion data, including whether the user got up, moved towards the door, and the speed of movement. For example, if a user gets up and moves quickly towards the door, the movement trajectory feature can be represented as "got up, moved towards the door, quickly". The seat pressure feature is a feature obtained by analyzing seat pressure data, indicating whether the seat is occupied. Pressure on the seat indicates that the user is sitting in the chair; no pressure indicates that the user is not in the chair. The item carrying feature is a feature obtained by analyzing personal item status data, indicating whether the user's personal items are in the shared office. If the user's personal items are indoors, the item carrying feature is "indoors"; if personal items are not indoors, the item carrying feature is "not indoors".

[0099] In one possible implementation, motion data collected by millimeter-wave radar is analyzed to detect if the user has changed from a stationary state to a moving state. The user's direction of movement is analyzed to determine if they are heading towards a door. The user's speed is calculated to determine whether the movement is fast or slow. The results of these determinations are used as motion trajectory features. Seat pressure data collected by a pressure sensor is analyzed to determine if there is pressure on the seat, thus determining whether the user is sitting in the chair. The results of this determination are used as seat pressure features. Personal item status data collected by an RFID reader is analyzed to determine if the user's personal items are in the office. The results of this determination are used as item carrying features. The extracted motion trajectory features, seat pressure features, and item carrying features are concatenated to form a comprehensive feature vector. For example, these features can be combined into a vector form, such as: [motion trajectory features, seat pressure features, item carrying features].

[0100] For example Figure 3 As shown, Figure 3 This is a visualization of the fused feature vector of a shared office A-101 provided in this application. In shared office A-101, the system collected the following data in the last 10 seconds: movement trajectory features:

[0101] Getting up: No (The user did not get up);

[0102] Movement direction: Towards the door (the user is moving towards the door);

[0103] Movement speed: 1.0 m / s (user movement speed is 1.0 meter per second);

[0104] Dwell time: 2.3 seconds (the user stayed at a certain location for 2.3 seconds);

[0105] Seat pressure characteristics:

[0106] Pressure value: 48kg (48 kg of pressure on the seat, indicating that the user is sitting in the chair);

[0107] Pressure change: Rapid release (the seat pressure decreases rapidly, indicating that the user stands up quickly);

[0108] Occupancy status: Occupied (the seat is occupied, and the user is sitting in the chair);

[0109] Pressure distribution: Uneven (The uneven pressure distribution on the seat may be caused by the user's sitting posture);

[0110] Characteristics of the items carried:

[0111] Backpack: Not present (user is not carrying a backpack);

[0112] Laptop: Existing (the user is carrying a laptop).

[0113] Water cup: Not present (the user did not bring a water cup);

[0114] Mobile phone: Exists (the user is carrying a mobile phone);

[0115] The system fuses the above features to form a 12-dimensional feature vector, with each dimension representing a specific feature, as shown below: [0, 1, 1, 2.3, 48, 1, 1, 0, 0, 1, 0, 1];

[0116] Dimension 1: Getting up (0 indicates no, 1 indicates yes);

[0117] 2nd dimension: direction of movement (1 indicates towards the door);

[0118] The third dimension: movement speed (1.0 m / s, 0.5 after MinMaxScaler normalization);

[0119] 4th dimension: Dwell time (2.3 seconds, 0.115 after MinMaxScaler normalization);

[0120] 5th dimension: Pressure value (48kg, 0.8 after MinMaxScaler standardization);

[0121] 6th dimension: Pressure change (1 indicates rapid release);

[0122] 7th dimension: Occupancy status (1 indicates occupancy);

[0123] Dimensions 8-11: Item carrying status (0 indicates non-existence, 1 indicates existence);

[0124] 12th dimension: The mobile phone exists (1 indicates existence);

[0125] Feature weights:

[0126] Movement trajectory: 30% (accounting for 30% of the weight in the feature vector);

[0127] Seat pressure: 30% (accounting for 30% of the weight in the feature vector);

[0128] Item carrying: 40% (accounting for 40% weight in the feature vector).

[0129] S103. Based on the multimodal features of each historical moment, predefined behavioral patterns, or pre-tuned user state recognition models, determine the user state corresponding to each shared office. The predefined behavioral patterns are used to characterize the correspondence between different combinations of user behaviors in the shared office and the user state. The pre-tuned user state recognition model is trained by combining unlabeled general model weight learning and labeled model fine-tuning learning.

[0130] The multimodal features at each historical moment include movement trajectory features, seat pressure features, and item carrying features; the predefined behavioral patterns include behavioral patterns corresponding to the in-place state, behavioral patterns corresponding to the short-term departure state, and behavioral patterns corresponding to the permanent departure state.

[0131] In this context, "In-position" means the user is in the shared office and is working or active. "Short-term absence" means the user has temporarily left the shared office but is expected to return shortly. "Permanent absence" means the user has left the shared office and does not intend to return shortly.

[0132] In some embodiments of this application, the specific process of determining the user status corresponding to each shared office based on the multimodal characteristics and predefined behavioral patterns at each historical moment includes: statistically analyzing the trajectory features that appear most frequently based on the movement trajectory features at each historical moment; statistically analyzing the pressure features that appear most frequently based on the seat pressure features at each historical moment; statistically analyzing the item features that appear most frequently based on the item carrying features at each historical moment; and analyzing the user status corresponding to each shared office based on the behavioral patterns corresponding to the in-place state, the behavioral patterns corresponding to the short-term departure state, the behavioral patterns corresponding to the permanent departure state, the results of the trajectory features, the results of the pressure features, and the results of the item features.

[0133] Specifically, the process of analyzing the user status for each shared office based on the behavioral patterns corresponding to the in-place state, the behavioral patterns corresponding to the short-term departure state, the behavioral patterns corresponding to the permanent departure state, the results of trajectory characteristics, the results of stress characteristics, and the results of item characteristics includes: when the stress characteristics indicate that the seat is occupied, the trajectory characteristics indicate that the user has not gotten up, has not moved towards the door, and the movement speed is slow, and the item characteristics indicate that personal items are indoors, the behavioral pattern corresponding to the in-place state is determined; the in-place state is taken as the user status for each shared office. Alternatively, when the stress characteristics indicate that the seat is not occupied, the trajectory characteristics indicate that the user has gotten up, moved towards the door, and the movement speed is fast, and the item characteristics indicate that personal items are indoors, the behavioral pattern corresponding to the short-term departure state is determined; the short-term departure state is taken as the user status for each shared office. Alternatively, when the stress characteristics indicate that the seat is not occupied, the trajectory characteristics indicate that the user has gotten up, moved towards the door, and the movement speed is fast, and the item characteristics indicate that personal items are not indoors, the behavioral pattern corresponding to the permanent departure state is determined; the permanent departure state is taken as the user status for each shared office.

[0134] In one possible implementation, the state of user A within the last minute is determined. Within 60 seconds, user A's characteristic is moving towards the door for 35 seconds; the most frequent trajectory feature during this time is moving towards the door. Within 60 seconds, the seat is not occupied for 40 seconds (user A is not in the seat); the pressure feature during this time is not in the seat. Within 60 seconds, personal belongings are not in the room for 50 seconds; the most frequent item feature during this time is personal belongings not in the room. At this point, user A's user state can be determined to be permanently absent.

[0135] It's worth noting that predefined behavioral patterns identify user states by setting a series of experience-based rules. Their advantages lie in their simplicity and high computational efficiency. Because these rules are intuitive, the system can directly apply them to make quick decisions, significantly saving computing resources. Furthermore, this pattern offers fast response times, reacting to user behavior in real time, which is particularly important for applications requiring immediate feedback. Predefined behavioral patterns are also easy to maintain and extend, allowing the system to be adjusted according to new needs while maintaining low development and operational costs.

[0136] In other embodiments of this application, the specific process of determining the user status corresponding to each shared office based on the multimodal features of each historical moment and the pre-tuned user status recognition model includes: initializing the pre-tuned user status determination model; sequentially inputting the multimodal features of each historical moment into the pre-tuned user status determination model in chronological order; and outputting the user status corresponding to each shared office.

[0137] It's worth noting that the pre-tuned user state recognition model utilizes machine learning techniques to learn user behavior patterns by analyzing historical data, thereby providing higher accuracy in state recognition. This model can handle complex datasets and extract deep information from multimodal features to improve recognition accuracy. It can flexibly respond to different user behaviors and environmental changes, making it suitable for diverse application scenarios.

[0138] For example Figure 4 As shown, Figure 4 This application provides a schematic block diagram of a shared office equipment intelligent control process based on user status, activating various sensing devices deployed within the shared office, such as millimeter-wave radar, pressure sensors, and personal item detection components. The activated sensing devices collect different types of data, including motion data, seat pressure data, and personal item status data. The collected raw data undergoes preprocessing operations such as cleaning and formatting, ensuring temporal consistency. The preprocessed data is stored in a time-series database for subsequent analysis and processing. Key features of multimodal sensing data from a preset historical period are extracted from the stored data and fused to form multimodal features. Based on the extracted multimodal features, a predefined behavioral pattern or a pre-trained user status recognition model is selected for analysis. For example, predefined behavioral patterns, such as "present," "short-term absence," and "permanent absence," are used to determine the user's current status. For instance, a pre-tuned user status determination model is initialized, trained jointly through unlabeled general model weight learning and labeled model fine-tuning learning. The fused multimodal features are input into the user status determination model. The model analyzes the input features and outputs the user's current status. The determined user status is sent to the resource scheduling platform for equipment control. Based on the user status, the automated control system controls the shared office equipment, such as lighting and air conditioning. The status of each shared office is displayed in the backend, for example... Figure 5 As shown.

[0139] S104, send the user status corresponding to each shared office to the preset resource scheduling platform to automatically control the operating equipment of each shared office.

[0140] The operating equipment refers to various devices within the shared office, such as lights, air conditioning, and printers. These devices can be automatically controlled based on user status. For example, the office's intelligent lighting system can automatically adjust brightness based on whether a user is seated.

[0141] In one possible implementation, the system identifies user A as "present." The system sends user A's "present status" to a pre-defined resource scheduling platform. Upon receiving user A's "present status," the resource scheduling platform instructs the lighting system in office A-101 to maintain brightness and the air conditioning system to continue operating to ensure user A's comfortable working environment. When user B temporarily leaves to use the restroom (short-term departure), the lighting brightness is reduced and the air conditioning fan speed is decreased. When user C finishes work and leaves the office (permanent departure), the lights are turned off, the air conditioning is shut off, and the computer and other electronic devices are completely turned off.

[0142] In this embodiment, on the one hand, by collecting motion data, seat pressure data, and personal item status data, combined with key feature extraction and feature fusion technologies, the system can comprehensively and accurately perceive user behavior and status. This allows the system to accurately distinguish whether a user is leaving temporarily or permanently, thereby achieving precise equipment control and improving energy efficiency. On the other hand, predefined behavioral patterns can characterize the correspondence between different combinations of user behaviors and user status within the shared office, providing the system with clear judgment criteria. Simultaneously, the model training method, combining unlabeled general model weight learning and labeled model fine-tuning learning, enables the model to be quickly fine-tuned to adapt to various complex scenarios. These two methods significantly improve management efficiency, reduce operating costs, and enhance system stability and reliability, making it adaptable to the complex scenarios of frequent personnel movement and diverse behavioral patterns in shared offices.

[0143] Please see Figure 6 This is a flowchart illustrating a method for generating a user state determination model, as provided in an embodiment of this application. Figure 6 As shown, the method in this application embodiment may include the following steps:

[0144] S201, Collect and preprocess historical multimodal perception data of each shared office within a preset historical time period to obtain historical modal data of each shared office at each historical moment;

[0145] S202, combine the historical modal data of adjacent historical moments in the historical modal data of each historical moment in each shared office into positively correlated first perception data pairs;

[0146] In this embodiment, positive correlation refers to a certain relationship or similarity between data from two adjacent time points. By combining historical modal data from adjacent historical moments, the model is trained to identify the continuity and consistency of user states. This data helps the model learn patterns and regularities in user behavior.

[0147] S203, randomly select any two historical modal data from different shared offices from the historical modal data of each historical moment and combine them into a negatively correlated second perception data pair;

[0148] In this embodiment, negative correlation refers to a significant difference or dissimilarity between data from two different time points. This data, obtained by combining historical modal data from any two historical moments in different shared offices, is used to train the model to identify differences and changes in user states. This data helps the model learn its ability to distinguish between different user states.

[0149] S204. Based on each first perception data pair and each second perception data pair, perform unlabeled general model weight learning to obtain a basic model that can identify the correlation features between adjacent time points.

[0150] In this embodiment of the application, by combining these two types of data for unlabeled machine learning, the model is able to distinguish and identify correlation features between adjacent time points, as well as perceive differences and changes in user status.

[0151] In some embodiments of this application, the specific process of obtaining a basic model capable of recognizing correlation features between adjacent time points by performing unlabeled general model weight learning on each first perception data pair and each second perception data pair includes: creating an initial network architecture, which includes a feature embedding module, a similarity calculation module, and a loss function for unlabeled learning connected in sequence; normalizing the two historical modal data in each first perception data pair and each second perception data pair; inputting the normalized first perception data pair and each normalized second perception data pair into the feature embedding module to encode the two historical modal data in the perception data pair to obtain the embedding representation of each feature; concatenating the embedding representation of each feature to obtain two overall embedding representations for each first perception data pair and two overall embedding representations for each second perception data pair; inputting the two overall embedding representations into the similarity calculation module to calculate the cosine similarity of each first perception data pair and each second perception data pair; calculating the correlation loss value using the loss function and cosine similarity; and generating a basic model capable of recognizing correlation features between adjacent time points when the correlation loss value reaches its minimum.

[0152] For example Figure 7As shown, the initial network architecture includes a feature embedding module, a similarity calculation module, and a loss function for unlabeled learning, which are connected in sequence.

[0153] Among them, the loss function value The calculation formula is:

[0154]

[0155] Wherein, the cosine similarity of each first perception data pair is , It is the first Cosine similarity of a second-sensory data pair It is the number of second-sensory data pairs. It is a temperature parameter used to control the distribution of similarity values.

[0156] S205: Select a portion of historical modal data from historical modal data at various historical moments to label and reflect the actual user state under the historical modal data, and obtain a small number of model fine-tuning samples;

[0157] S206, based on model fine-tuning samples, performs labeled model fine-tuning learning on the basic model to obtain a pre-fine-tuned user state determination model.

[0158] In some embodiments of this application, the specific process of performing labeled model fine-tuning learning on the base model based on model fine-tuning samples to obtain a pre-fine-tuned user state determination model includes: extracting basic data features and label features reflecting the real user state for each model fine-tuning sample; the basic data features include movement trajectory features, seat pressure features, and item carrying features; mapping the basic data features and label features to fixed-dimensional embedding vectors to obtain the feature vector of each model fine-tuning sample; inputting the feature vector into the base model for labeled model fine-tuning, and outputting the cross-entropy loss value between the user state predicted by the base model and the real user state of each model fine-tuning sample; generating the pre-fine-tuned user state determination model when the cross-entropy loss value reaches its minimum; or, if the cross-entropy loss value has not reached its minimum, continuing to execute the step of inputting the feature vector into the base model for labeled model fine-tuning until the cross-entropy loss value reaches its minimum.

[0159] In some embodiments of this application, fine-tuning learning with a small number of data samples enables the model to learn more accurate user state recognition capabilities on limited labeled data.

[0160] In this embodiment, on the one hand, by collecting motion data, seat pressure data, and personal item status data, combined with key feature extraction and feature fusion technologies, the system can comprehensively and accurately perceive user behavior and status. This allows the system to accurately distinguish whether a user is leaving temporarily or permanently, thereby achieving precise equipment control and improving energy efficiency. On the other hand, predefined behavioral patterns can characterize the correspondence between different combinations of user behaviors and user status within the shared office, providing the system with clear judgment criteria. Simultaneously, the model training method, combining unlabeled general model weight learning and labeled model fine-tuning learning, enables the model to be quickly fine-tuned to adapt to various complex scenarios. These two methods significantly improve management efficiency, reduce operating costs, and enhance system stability and reliability, making it adaptable to the complex scenarios of frequent personnel movement and diverse behavioral patterns in shared offices.

[0161] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0162] Please see Figure 8 This illustration shows a schematic diagram of a user-state-based intelligent control device for a shared office, provided in an exemplary embodiment of this application. This user-state-based intelligent control device for a shared office can be implemented as all or part of an electronic device through software, hardware, or a combination of both. The device 1 includes a multimodal sensing data acquisition module 10, a feature processing module 20, a user state determination module 30, and an automation control module 40.

[0163] The multimodal perception data acquisition module 10 is used to acquire multimodal perception data of each shared office at each historical moment within a preset period. The multimodal perception data includes motion data, seat pressure data, and personal item status data.

[0164] Feature processing module 20 is used to extract and fuse key features from motion data, seat pressure data, and personal item status data to obtain multimodal features for each historical moment.

[0165] The user state determination module 30 is used to determine the user state corresponding to each shared office based on the multimodal features of each historical moment, predefined behavior patterns, or pre-tuned user state recognition models. The predefined behavior patterns are used to characterize the correspondence between different combinations of user behaviors in the shared office and the user state. The pre-tuned user state recognition model is trained by combining unlabeled general model weight learning and labeled model fine-tuning learning.

[0166] The automation control module 40 is used to send the user status corresponding to each shared office to the preset resource scheduling platform in order to automatically control the operating equipment of each shared office.

[0167] It should be noted that the above-described intelligent control device for shared offices based on user status, when executing the intelligent control method for shared offices based on user status, is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the intelligent control device for shared offices based on user status provided in the above embodiments and the intelligent control method embodiments for shared offices based on user status belong to the same concept, and their implementation process is detailed in the method embodiments, which will not be repeated here.

[0168] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0169] In this embodiment, on the one hand, by collecting motion data, seat pressure data, and personal item status data, combined with key feature extraction and feature fusion technologies, the system can comprehensively and accurately perceive user behavior and status. This allows the system to accurately distinguish whether a user is leaving temporarily or permanently, thereby achieving precise equipment control and improving energy efficiency. On the other hand, predefined behavioral patterns can characterize the correspondence between different combinations of user behaviors and user status within the shared office, providing the system with clear judgment criteria. Simultaneously, the model training method, combining unlabeled general model weight learning and labeled model fine-tuning learning, enables the model to be quickly fine-tuned to adapt to various complex scenarios. These two methods significantly improve management efficiency, reduce operating costs, and enhance system stability and reliability, making it adaptable to the complex scenarios of frequent personnel movement and diverse behavioral patterns in shared offices.

[0170] This application also provides a computer-readable medium having program instructions stored thereon, which, when executed by a processor, implement the user-state-based intelligent control method for shared office devices provided in the above-described method embodiments.

[0171] This application also provides a computer program product containing instructions that, when run on a computer, causes the computer to execute the user-state-based intelligent control method for shared office devices in the various method embodiments described above.

[0172] Please see Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 9As shown, the electronic device 1000 may include: at least one processor 1001, at least one network interface 1004, a user interface 1003, a memory 1005, and at least one communication bus 1002.

[0173] The communication bus 1002 is used to realize the connection and communication between these components.

[0174] The user interface 1003 may include a display screen and a camera. Optionally, the user interface 1003 may also include a standard wired interface and a wireless interface.

[0175] The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0176] The processor 1001 may include one or more processing cores. The processor 1001 connects to various parts within the electronic device 1000 using various interfaces and lines. It executes various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 1005, and by calling data stored in the memory 1005. Optionally, the processor 1001 may be implemented using at least one hardware form selected from Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). The processor 1001 may integrate one or more of the following: a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), and a modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip, without being integrated into the processor 1001.

[0177] The memory 1005 may include random access memory (RAM) or read-only memory. Optionally, the memory 1005 may include a non-transitory computer-readable storage medium. The memory 1005 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 1005 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 1005 may also be at least one storage system located remotely from the aforementioned processor 1001. Figure 9 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a smart control application for shared office devices based on user status.

[0178] exist Figure 9 In the illustrated electronic device 1000, the user interface 1003 is mainly used to provide an input interface for the user and to acquire user input data; while the processor 1001 can be used to call the shared office device intelligent control application based on user status stored in the memory 1005, and specifically perform the following operations:

[0179] Acquire multimodal perception data for each shared office at various historical moments within a preset period. The multimodal perception data includes motion data, seat pressure data, and personal item status data.

[0180] Key features are extracted and fused from motion data, seat pressure data, and personal item status data to obtain multimodal features for each historical moment;

[0181] Based on the multimodal features of each historical moment, predefined behavioral patterns, or pre-tuned user state recognition models, the user state corresponding to each shared office is determined. The predefined behavioral patterns are used to characterize the correspondence between different combinations of user behaviors in the shared office and the user state. The pre-tuned user state recognition model is trained by combining unlabeled general model weight learning and labeled model fine-tuning learning.

[0182] Send the user status corresponding to each shared office to the preset resource scheduling platform to automatically control the operating equipment of each shared office.

[0183] In one embodiment, before acquiring multimodal sensing data for each shared office at various historical moments within a preset period, the processor 1001 also performs the following operations:

[0184] When a user activates a shared office, various sensing devices pre-deployed in the shared office are activated; these sensing devices include millimeter-wave radar, pressure sensors, and personal item detection components.

[0185] Millimeter-wave radar is used to collect users' movements, directions of movement, and speeds within the shared office space, which serve as motion data.

[0186] The pressure data of the chairs in the shared office is obtained through pressure sensors;

[0187] The personal items detection component monitors the status of the user's personal items and obtains personal item status data.

[0188] Preprocessing and aligning motion data, seat pressure data, and personal item status data with time windows yields multimodal perception data for each historical moment.

[0189] Store multimodal sensing data to a time-series data platform.

[0190] In one embodiment, when the processor 1001 performs key feature extraction and feature fusion on motion data, seat pressure data, and personal item status data to obtain multimodal features for each historical moment, it specifically performs the following operations:

[0191] By analyzing the data, we can determine whether the user gets up, moves towards the door, and the speed of movement, thus obtaining movement trajectory characteristics.

[0192] By using seat pressure data, the occupancy status of the seat is determined, and the seat pressure characteristics are obtained;

[0193] By analyzing the status data of personal belongings, we can determine whether users have personal belongings in the shared office and obtain the characteristics of the items they carry.

[0194] The movement trajectory features, seat pressure features, and item carrying features are spliced ​​together to obtain multimodal features for each historical moment.

[0195] In one embodiment, when the processor 1001 determines the user status corresponding to each shared office based on the multimodal characteristics and predefined behavioral patterns at each historical moment, it specifically performs the following operations:

[0196] The results of statistically analyzing the most frequently occurring trajectory features at each historical moment were obtained.

[0197] By analyzing the seat pressure characteristics at various historical moments, the results of the most frequently occurring pressure characteristics were obtained.

[0198] The results of statistical analysis of the most frequently occurring item characteristics were obtained by analyzing the characteristics of items carried at various historical moments.

[0199] Based on the behavioral patterns corresponding to the in-place state, the behavioral patterns corresponding to the short-term departure state, the behavioral patterns corresponding to the permanent departure state, the results of the trajectory characteristics, the results of the stress characteristics, and the results of the item characteristics, the user status corresponding to each shared office is analyzed.

[0200] In one embodiment, when the processor 1001 analyzes the user status corresponding to each shared office based on the results of behavioral patterns corresponding to the in-place status, behavioral patterns corresponding to the short-term departure status, behavioral patterns corresponding to the permanent departure status, trajectory characteristics, stress characteristics, and item characteristics, it specifically performs the following operations:

[0201] When the pressure characteristics indicate that the seat is occupied, the trajectory characteristics indicate that the user has not gotten up, has not moved towards the door, and is moving slowly, and the item characteristics indicate that personal belongings are indoors, the behavioral pattern corresponding to the in-place state is determined; the in-place state is taken as the user state for each shared office; or...

[0202] When the pressure characteristics indicate that the seat is unoccupied, the trajectory characteristics indicate that the user gets up, moves towards the door, and moves quickly, and the item characteristics indicate that personal belongings are indoors, the behavioral pattern corresponding to the short-term departure state is determined; the short-term departure state is taken as the user state for each shared office; or...

[0203] When the pressure characteristic indicates that the seat is not occupied, the trajectory characteristic indicates that the user gets up, moves towards the door, or moves quickly, and the item characteristic indicates that personal items are not indoors, the behavior pattern corresponding to the permanent departure state is determined; the permanent departure state is taken as the user state corresponding to each shared office.

[0204] In one embodiment, when the processor 1001 executes a user state recognition model that is pre-tuned based on multimodal features from various historical moments to determine the user state corresponding to each shared office, it specifically performs the following operations:

[0205] Initialize the pre-tuned user state determination model;

[0206] In chronological order, the multimodal features of each historical moment are sequentially input into the pre-tuned user state determination model;

[0207] Output the user status for each shared office.

[0208] In one embodiment, when the processor 1001 executes the generation of a pre-tuned user state determination model, it specifically performs the following operations:

[0209] Collect and preprocess historical multimodal sensing data of each shared office within a preset historical time period to obtain historical modal data of each shared office at each historical moment;

[0210] The historical modal data of adjacent historical moments in each historical moment of each shared office are combined into positively correlated first perception data pairs;

[0211] From the historical modal data of each historical moment in each shared office, randomly select any two historical modal data from different shared offices and combine them into a negatively correlated second perception data pair;

[0212] By performing unlabeled general model weight learning on each first perception data pair and each second perception data pair, a basic model that can identify the correlation features between adjacent time points is obtained.

[0213] From the historical modal data at various historical moments, a portion of the historical modal data is selected and labeled to reflect the actual user state under the historical modal data, resulting in a small number of model fine-tuning samples;

[0214] Based on the model fine-tuning samples, the basic model is subjected to labeled model fine-tuning learning to obtain a pre-fine-tuned user state determination model.

[0215] In one embodiment, when processor 1001 performs unlabeled general model weight learning based on each first sensing data pair and each second sensing data pair to obtain a basic model capable of recognizing correlation features between adjacent time points, it specifically performs the following operations:

[0216] Create an initial network architecture, which includes a feature embedding module, a similarity calculation module, and a loss function for unlabeled learning, connected in sequence.

[0217] Normalize the two historical modal data in each first perception data pair and each second perception data pair;

[0218] Each normalized first perception data pair and each normalized second perception data pair are input into the feature embedding module to encode the two historical modal data in the perception data pair, thereby obtaining the embedded representation of each feature.

[0219] The embedding representation of each feature is concatenated to obtain two overall embedding representations for each first perceptual data pair and two overall embedding representations for each second perceptual data pair.

[0220] The two wholes are embedded to represent the input similarity calculation module, and the cosine similarity of each first perceptual data pair and the cosine similarity of each second perceptual data pair are calculated.

[0221] The correlation loss value is calculated using a loss function and cosine similarity. When the correlation loss value reaches its minimum, a basic model that can identify the correlation characteristics between adjacent time points is generated.

[0222] In one embodiment, when the processor 1001 performs labeled model fine-tuning learning on the base model based on model fine-tuning samples to obtain a pre-fine-tuned user state determination model, it specifically performs the following operations:

[0223] Extract the basic data features and label features that reflect the real user status for each model fine-tuning sample; the basic data features include movement trajectory features, seat pressure features, and item carrying features;

[0224] The basic data features and label features are mapped to fixed-dimensional embedding vectors to obtain the feature vector of each model fine-tuning sample;

[0225] The feature vector is input into the base model for labeled model fine-tuning, and the cross-entropy loss value between the user state predicted by the base model and the real user state of each model fine-tuning sample is output.

[0226] When the cross-entropy loss value reaches its minimum, a pre-fine-tuned user state determination model is generated.

[0227] Alternatively, if the cross-entropy loss value has not reached its minimum, continue performing the step of inputting the feature vector into the base model for labeled model fine-tuning until the cross-entropy loss value reaches its minimum.

[0228] In this embodiment, on the one hand, by collecting motion data, seat pressure data, and personal item status data, combined with key feature extraction and feature fusion technologies, the system can comprehensively and accurately perceive user behavior and status. This allows the system to accurately distinguish whether a user is leaving temporarily or permanently, thereby achieving precise equipment control and improving energy efficiency. On the other hand, predefined behavioral patterns can characterize the correspondence between different combinations of user behaviors and user status within the shared office, providing the system with clear judgment criteria. Simultaneously, the model training method, combining unlabeled general model weight learning and labeled model fine-tuning learning, enables the model to be quickly fine-tuned to adapt to various complex scenarios. These two methods significantly improve management efficiency, reduce operating costs, and enhance system stability and reliability, making it adaptable to the complex scenarios of frequent personnel movement and diverse behavioral patterns in shared offices.

[0229] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program for intelligent control of shared office equipment based on user status can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. The storage medium for the program for intelligent control of shared office equipment based on user status can be a magnetic disk, optical disk, read-only memory, or random access memory, etc.

[0230] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A method for intelligent control of equipment in a shared office based on user status, characterized in that, Applied to the server side, the method includes: Acquire multimodal perception data for each shared office at various historical moments within a preset period. The multimodal perception data includes motion data, seat pressure data, and personal item status data. The motion data, seat pressure data, and personal item status data are subjected to key feature extraction and feature fusion to obtain the multimodal features of each historical moment; Based on the multimodal features of each historical moment, predefined behavioral patterns, or pre-tuned user state recognition models, the user state corresponding to each shared office is determined; the predefined behavioral patterns are used to characterize the correspondence between different combinations of user behaviors in the shared office and the user state; the pre-tuned user state recognition model is obtained by jointly training unlabeled general model weight learning and labeled model fine-tuning learning. The multimodal features at each historical moment include movement trajectory features, seat pressure features, and item carrying features; the predefined behavioral patterns include behavioral patterns corresponding to the in-place state, the behavioral patterns corresponding to the short-term departure state, and the behavioral patterns corresponding to the permanent departure state; based on the multimodal features at each historical moment and the predefined behavioral patterns, the user state corresponding to each shared office is determined, including: using the movement trajectory features at each historical moment to count the most frequently occurring trajectory features; using the seat pressure features at each historical moment to count the most frequently occurring pressure features; using the item carrying features at each historical moment to count the most frequently occurring item features; When the pressure characteristic indicates that the seat is occupied, the trajectory characteristic indicates that the user has not gotten up, has not moved towards the door, and is moving slowly, and the item characteristic indicates that personal belongings are indoors, the behavior pattern corresponding to the "in-place" state is determined; this "in-place" state is taken as the user state for each shared office. When the pressure characteristic indicates that the seat is not occupied, the trajectory characteristic indicates that the user has gotten up, moved towards the door, and is moving quickly, and the item characteristic indicates that personal belongings are indoors, the behavior pattern corresponding to the "short-term departure" state is determined; this "short-term departure" state is taken as the user state for each shared office. When the pressure characteristic indicates that the seat is not occupied, the trajectory characteristic indicates that the user has gotten up, moved towards the door, and is moving quickly, and the item characteristic indicates that personal belongings are not indoors, the behavior pattern corresponding to the "permanent departure" state is determined; this "permanent departure" state is taken as the user state for each shared office. The user status corresponding to each shared office is sent to a preset resource scheduling platform to automatically control the operating equipment of each shared office.

2. The method according to claim 1, characterized in that, Before acquiring the multimodal sensing data of each shared office at various historical moments within a preset period, the process also includes: When the shared office is activated by a user, a variety of sensing devices pre-deployed in the shared office are activated; the various sensing devices include millimeter-wave radar, pressure sensors, and personal item detection components; The millimeter-wave radar collects the user's movements, direction of movement, and speed within the shared office as motion data. The pressure sensor is used to acquire seat pressure data of the chairs in the shared office; The personal item detection component monitors the status of the user's personal items and obtains personal item status data. The motion data, seat pressure data, and personal item status data are preprocessed and aligned with time windows to obtain multimodal perception data for each historical moment; The multimodal sensing data is stored in a time-series data platform.

3. The method according to claim 1, characterized in that, The process involves extracting and fusing key features from the motion data, seat pressure data, and personal item status data to obtain multimodal features for each historical moment, including: By analyzing the motion data, we can determine whether the user got up, moved towards the door, and the speed of movement, thus obtaining the movement trajectory characteristics. The occupancy status of the seat is determined by the seat pressure data, and the seat pressure characteristics are obtained. By using the personal item status data, it can be determined whether there are any personal items belonging to the user in the shared office, thus obtaining the item carrying characteristics; The movement trajectory features, seat pressure features, and item carrying features are spliced ​​together to obtain the multimodal features of each historical moment.

4. The method according to claim 1, characterized in that, Based on the multimodal features of each historical moment and the pre-tuned user state recognition model, the user state corresponding to each shared office is determined, including: Initialize the pre-tuned user state determination model; The multimodal features of each historical moment are sequentially input into the pre-tuned user state determination model in chronological order. Output the user status corresponding to each shared office.

5. The method according to claim 4, characterized in that, Generate a pre-tuned user state determination model according to the following steps: Collect and preprocess historical multimodal sensing data of each shared office within a preset historical time period to obtain historical modal data of each shared office at each historical moment; The historical modal data of adjacent historical moments in each historical moment of each shared office are combined into positively correlated first perception data pairs; From the historical modal data of each historical moment in each shared office, randomly select any two historical modal data from different shared offices and combine them into a negatively correlated second perception data pair; By performing unlabeled general model weight learning on each first perception data pair and each second perception data pair, a basic model that can identify the correlation features between adjacent time points is obtained. From the historical modal data at each historical moment, a portion of the historical modal data is selected and labeled to reflect the actual user state under the historical modal data, resulting in a small number of model fine-tuning samples; Based on the model fine-tuning samples, the base model is subjected to labeled model fine-tuning learning to obtain a pre-fine-tuned user state determination model.

6. The method according to claim 5, characterized in that, The step of learning unlabeled general model weights for each first sensing data pair and each second sensing data pair to obtain a basic model capable of identifying correlation features between adjacent time points includes: An initial network architecture is created, comprising a feature embedding module, a similarity calculation module, and a loss function for unlabeled learning, which are connected in sequence. Normalize the two historical modal data in each first sensing data pair and each second sensing data pair; Each normalized first perception data pair and each normalized second perception data pair are input into the feature embedding module to encode the two historical modal data in the perception data pair, thereby obtaining the embedded representation of each feature. The embedding representation of each feature is concatenated to obtain two overall embedding representations for each first perceptual data pair and two overall embedding representations for each second perceptual data pair. The two whole embedded representations are input into the similarity calculation module to calculate the cosine similarity of each first perceptual data pair and the cosine similarity of each second perceptual data pair; The correlation loss value is calculated using the loss function and the cosine similarity; when the correlation loss value reaches its minimum, a basic model capable of identifying correlation features between adjacent time points is generated.

7. The method according to claim 5, characterized in that, The step of performing labeled model fine-tuning learning on the base model based on the model fine-tuning samples to obtain a pre-tuned user state determination model includes: Extract the basic data features and label features reflecting the real user state for each model fine-tuning sample; the basic data features include movement trajectory features, seat pressure features, and item carrying features; The basic data features and the label features are mapped to fixed-dimensional embedding vectors to obtain the feature vector of each model fine-tuning sample; The feature vector is input into the base model for labeled model fine-tuning, and the cross-entropy loss value between the user state predicted by the base model and the real user state of each model fine-tuning sample is output. When the cross-entropy loss value reaches its minimum, a pre-tuned user state determination model is generated; Alternatively, if the cross-entropy loss value has not reached its minimum, continue executing the step of inputting the feature vector into the base model for labeled model fine-tuning until the cross-entropy loss value reaches its minimum.

8. A smart control device for shared office equipment based on user status, characterized in that, The device includes: The multimodal perception data acquisition module is used to acquire multimodal perception data of each shared office at each historical moment within a preset period. The multimodal perception data includes motion data, seat pressure data, and personal item status data. The feature processing module is used to extract and fuse key features from the motion data, seat pressure data, and personal item status data to obtain the multimodal features of each historical moment. The user state determination module is used to determine the user state corresponding to each shared office based on the multimodal features of each historical moment, predefined behavior patterns, or pre-tuned user state recognition models. The predefined behavior patterns are used to characterize the correspondence between different combinations of user behaviors in the shared office and the user state. The pre-tuned user state recognition model is trained by combining unlabeled general model weight learning and labeled model fine-tuning learning. The multimodal features at each historical moment include movement trajectory features, seat pressure features, and item carrying features; the predefined behavioral patterns include behavioral patterns corresponding to the in-place state, the behavioral patterns corresponding to the short-term departure state, and the behavioral patterns corresponding to the permanent departure state; based on the multimodal features at each historical moment and the predefined behavioral patterns, the user state corresponding to each shared office is determined, including: using the movement trajectory features at each historical moment to count the most frequently occurring trajectory features; using the seat pressure features at each historical moment to count the most frequently occurring pressure features; using the item carrying features at each historical moment to count the most frequently occurring item features; When the pressure characteristic indicates that the seat is occupied, the trajectory characteristic indicates that the user has not gotten up, has not moved towards the door, and is moving slowly, and the item characteristic indicates that personal belongings are indoors, the behavior pattern corresponding to the "in-place" state is determined; this "in-place" state is taken as the user state for each shared office. When the pressure characteristic indicates that the seat is not occupied, the trajectory characteristic indicates that the user has gotten up, moved towards the door, and is moving quickly, and the item characteristic indicates that personal belongings are indoors, the behavior pattern corresponding to the "short-term departure" state is determined; this "short-term departure" state is taken as the user state for each shared office. When the pressure characteristic indicates that the seat is not occupied, the trajectory characteristic indicates that the user has gotten up, moved towards the door, and is moving quickly, and the item characteristic indicates that personal belongings are not indoors, the behavior pattern corresponding to the "permanent departure" state is determined; this "permanent departure" state is taken as the user state for each shared office. An automated control module is used to send the user status corresponding to each shared office to a preset resource scheduling platform in order to automatically control the operating equipment of each shared office.

Images