Intelligent application management method and system for a hotel

Abstract

The application discloses an intelligent application management method and system for a hotel, and relates to the technical field of hotel management.The intelligent application management system for the hotel comprises a hotel user behavior reasoning module and an intelligent application management module.The application collects equipment application data of a user in a guest room and behavior trajectory data of the user in a public space of the hotel, and fuses historical preference information of the user, so that global modeling and state sensing of user behavior are realized, and the coverage range and recognition accuracy of the system for real behavior of the user are improved.The different behavior data of the user in and out of the hotel space are aligned in time and space and graph construction is performed, complex behavior states of the user at a specific moment are subjected to deep semantic understanding, accurate behavior intention labels are inferred, and the system has intelligentization and generalization capabilities.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of hotel management, and in particular to an intelligent application management method and system for a hotel. BACKGROUND

[0002] With the rapid development of artificial intelligence, Internet of Things and big data analysis technology, intelligent hotel management systems have been widely applied in the hotel industry. The existing intelligent hotel management systems mainly realize centralized control and automatic adjustment of devices such as room environment, light, air conditioner and television by deploying sensing devices, intelligent terminals and background control platforms, thereby improving the experience of guests and the operation efficiency. However, most of the existing technologies rely on fixed rules or preset scene modes for device scheduling, lack dynamic perception and intelligent response to user personalized behaviors, and are difficult to meet the diversified and real-time service needs of guests. For example, the existing systems usually control the air conditioner switch and lighting mode according to the time period, preset instructions or room card state, and fail to fully combine the activity behavior data of users inside and outside the hotel to make targeted service adjustments. In addition, the current systems are limited to device usage data in the room in terms of behavior perception, lack effective tracking and correlation analysis of user behavior activities in public areas of the hotel, and cannot realize comprehensive perception of the overall behavior path of the user.

[0003] Therefore, an intelligent hotel application management method based on user multi-source behavior data fusion and capable of realizing user behavior reasoning and intelligent device scheduling coordination is needed to improve the adaptability of hotel device applications and the refinement degree of user services. SUMMARY

[0004] The application aims to provide an intelligent application management method and system for a hotel, which can improve the adaptability of hotel device applications and the refinement degree of user services.

[0005] An intelligent application management method for a hotel comprises the following steps:

[0006] For at least one in-stay user in a target hotel, user device operation data of the in-stay user in a hotel room are collected in real time; the user device operation data include environment perception data and device application switch state data; corresponding public user behavior trajectory data and hotel preference data are obtained; the user behavior trajectory data include nth behavior data, n = 1, 2, …, N, and the nth behavior data is the nth behavior event of the user recorded in a public place of the hotel;

[0007] Activity state reasoning is performed on the current corresponding nth behavior data to obtain user external activity prediction data; a check-out timestamp T n of the nth behavior data is obtained; and a check-in timestamp T n-1 of the nth behavior data is obtained.n The user device operation data is combined with the inference network to analyze user behavior and obtain user internal activity prediction data.

[0008] The predicted data of users' internal activities and external activities are spatiotemporally aligned, and the time stamp of leaving the room T is identified. n The corresponding behavioral intent tags are used to match the real-time device application update index in the guest rooms. Based on the real-time device application update index, the hotel device applications in and outside the guest rooms are intelligently adjusted to achieve intelligent management of device applications in the target hotel.

[0009] As a preferred embodiment of the present invention, the specific steps for performing activity state reasoning on the current corresponding nth row of data include:

[0010] Based on the data of the nth behavior, continuous spatiotemporal features are extracted to obtain the user spatiotemporal feature set; the user spatiotemporal feature set includes the behavior path, duration and corresponding location coordinates of the hotel in public places;

[0011] The behavioral sequence context features are constructed by acquiring behavioral data from the 1st to the (n-1st)th behavior. The user's spatiotemporal feature set, behavioral sequence context features, and hotel preference data are used as inputs to the pre-trained behavioral sequence learning model for analysis, thereby obtaining the user's external activity prediction data.

[0012] As a preferred technical solution of the present invention, the departure timestamp T is obtained. n-1 Time stamp until check-out T n The specific steps for analyzing user behavior by combining user device operational data with inference networks include:

[0013] Based on environmental perception data, identify the user's intention to change the environment; based on device application on / off status data, identify the user's intention to perform human actions; based on the user's intention to change the environment and the user's intention to perform human actions, construct contextual features to obtain an internal behavior feature vector sequence.

[0014] Acquire historical check-in data of users currently staying in guest rooms at the target hotel; establish comprehensive check-in context characteristics based on historical check-in data and hotel preference data;

[0015] The vector sequence of integrated occupancy context features and internal behavioral features is input into the inference network to analyze the departure timestamp T. n-1 Time stamp until check-out T n The system analyzes the device usage patterns, environmental change trends, and human activity patterns within the user's premises; by combining these patterns, the system obtains predictive data on user activities within the premises.

[0016] As a preferred technical solution of the present application, the identification of the time stamp T n The specific steps of the corresponding behavior intention label include:

[0017] The user internal activity prediction data and the user external activity prediction data are spatio-temporally aligned to obtain a user comprehensive context behavior vector; and behavior intention matching is performed based on the user comprehensive context behavior vector to obtain a behavior intention label;

[0018] The specific steps of spatio-temporal alignment include: extracting all behavior events in the user internal activity prediction data and the user external activity prediction data; mapping the user internal activity prediction data and the user external activity prediction data to a user behavior prediction coordinate system, and labeling spatial position coordinates and event absolute time stamps according to the behavior event granularity of the behavior events;

[0019] A dynamic time window is defined according to the behavior event granularity; a plurality of event corresponding relationships are established based on the dynamic time window for the user internal activity prediction data and the user external activity prediction data; and a behavior response tolerance interval is set for each event corresponding relationship;

[0020] Taking each behavior event as a node and the event corresponding relationship as an edge, a graph structure after spatio-temporal alignment is constructed, and the graph structure after alignment is vectorized by a behavior semantic embedding model to obtain a user comprehensive context behavior vector.

[0021] As a preferred technical solution of the present application, based on the identified behavior intention label, the specific steps of matching the real-time device application update index of the in-house guest room include:

[0022] A device application control factor library containing a plurality of hotel device applications is constructed; the device application control factor library contains a plurality of device application control factors;

[0023] Based on the behavior intention label, a target device application set associated therewith is matched, a corresponding device application control factor in the device application control factor library is matched for each target device application in the target device application set as a behavior association weight coefficient; a device application running parameter of the current in-house guest room is obtained; and a real-time device application update index is calculated by combining the device application running parameter and the behavior association weight coefficient.

[0024] As a preferred technical solution of the present application, the inference network takes a Bayesian inference network as a basic network structure for user behavior analysis.

[0025] An intelligent application management system for a hotel includes:

[0026] The hotel user behavior inference module comprises a data perception unit and an activity prediction unit; the data perception unit is used for collecting user equipment running data of a user in a guest room in a hotel room of a target hotel in real time; the user equipment running data comprises environment perception data and device application switch state data; user behavior trajectory data and hotel preference data are obtained; the user behavior trajectory data comprises nth behavior data, n=1, 2, …, N, and the nth behavior data is the nth behavior event of the user recorded in a public place of the hotel; the activity prediction unit is used for performing activity state inference on the corresponding nth behavior data to obtain user external activity prediction data; the off-room timestamp T n is obtained; the user equipment running data between the off-room timestamp T n-1 and the off-room timestamp T n is obtained, and user behavior analysis is performed on the user equipment running data by using an inference network to obtain user internal activity prediction data.

[0027] The intelligent application management module comprises a chip control unit; the chip control unit is used for performing space-time alignment on the user internal activity prediction data and the user external activity prediction data, simultaneously identifying a behavior intention label corresponding to the off-room timestamp T n ; based on the identified behavior intention label, a real-time device application update index in the guest room is matched; based on the real-time device application update index, the hotel device applications in the guest room and the hotel device applications outside the guest room are intelligently adjusted, and intelligent management of the device applications in the target hotel is realized.

[0028] The application has the following advantages:

[0029] 1. The application realizes user behavior global modeling and state perception by collecting device application data of the user in the guest room and behavior trajectory data of the user in the public space of the hotel, and fusing historical preference information of the user, breaks through the limitation of traditional hotel intelligent systems which are limited to single rooms or preset scene rules, and improves the coverage range and recognition accuracy of the system for real user behaviors; by performing space-time alignment and graph construction on different behavior data of the user in and outside the hotel space, and generating a comprehensive context behavior vector of the user by means of a behavior semantic embedding model, the system can perform deep semantic understanding on the complex behavior state of the user at a specific moment, and infer an accurate behavior intention label, which has more intelligent and generalization ability than the mode of judging only according to a single action or static rule.

[0030] 2、The application matches the target device set related to the intention through the recognized behavior intention label, combines the device application control factor library and the current device running state of the user, calculates the real-time update index of each device, realizes the quantitative management of the device response priority, and ensures that high-energy-consumption devices, user preferred devices or environment sensitive devices can be preferentially regulated, thereby improving the resource utilization efficiency and the service personalization level. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 A structural schematic diagram of an intelligent application management system for a hotel used in an embodiment of the application. DETAILED DESCRIPTION

[0032] In order to enable personnel in the art to better understand the technical solutions in the application, the technical solutions in the embodiments of the application will be clearly and completely described below in combination with the drawings in the embodiments of the application.

[0033] Embodiment 1, an intelligent application management method for a hotel, comprising the following steps:

[0034] Customer flow prediction; context-aware service; device data and running state data; lost article management; air monitoring (micro scene regulation); obstacle area and barrier-free area; judging whether cleaning is needed (for at least one day of customers), multiple sensing (without involving privacy) considering how to perform device self-cleaning cycle;

[0035] For at least one in-stay guest room user in the target hotel, real-time collection of user device running data of the in-stay guest room user in the hotel room; the user device running data includes environment sensing data and device application switch state data; obtaining corresponding public user behavior flow data and hotel preference data; the user behavior flow data includes nth behavior data, n=1, 2, …, N, the nth behavior data is the nth behavior event recorded in the public place of the hotel;

[0036] It should be noted that the user equipment operation data refers to the recordable and analyzable data set generated by the user in the process of interacting with various intelligent devices in the hotel room; the data mainly includes two types of content: one is environmental perception data, that is, physical environment information automatically collected by sensors deployed in the room, including temperature, humidity, light intensity, noise decibel value, air quality index (such as PM2.5, carbon dioxide concentration) and presence detection information (for example, whether there is a person in the room); the second is device application switch state data, that is, the control operation information of the user to the devices in the room through the remote control, panel, voice or mobile application, such as the on-off state of the air conditioner, the set temperature, the running mode, the on-off of the lighting lamp, the brightness level, the color temperature selection, the opening degree of the curtain, the on-off of the television, the channel, the volume, and the current use state of the air purifier, the incense, the music player and other devices; after these data are collected in real time through standardized communication protocols, they form a data record stream with device identification, time stamp, operation type and parameter value as basic fields, reflecting the specific activities and use preferences of the user in the room, which is an important basis for behavior recognition and scene intelligent control;

[0037] At the same time in the target hotel, all data are transmitted by integrated chips to realize device interconnection.

[0038] The disclosed user behavior trajectory data refers to the activity track and behavior record of the user in the public area of the hotel, which is mainly derived from the perception system deployed in the hotel, such as video analysis system, face recognition camera, infrared mobile detector, Bluetooth positioning device, wireless network probe or card swiping access control system, etc. These systems can identify the appearance time, stay position, movement path and behavior event of the user in different open areas of the hotel (such as lobby, elevator hall, gym, restaurant, conference room, etc.), and record them as continuous behavior sequences; each behavior data includes the occurrence time of the behavior event, the event type (for example, "get into the elevator", "enter the restaurant", "leave the gym"), the position code (floor, area number), the duration and possible additional information (such as user identity confirmation method, behavior frequency, etc.), which can be combined into a complete behavior trajectory according to the user identity and time sequence, used to judge the current state of the user and predict the behavior purpose or the room scene to be entered.

[0039] The nth behavior data is used to represent the nth time-ordered behavior event of the user being identified in the public area of the hotel, wherein n is used to establish a clear time sequence index relationship for the user behavior event, for behavior path analysis, context modeling and behavior intention reasoning; the behavior event refers to an independent behavior unit generated by the user in the public area of the hotel, having observable attributes, identifiable boundaries and being closely related to space position and time; one behavior event can be entering or leaving an area, performing a certain functional activity or interacting with a certain system, common behavior event types include but are not limited to: entering the elevator room, reaching the restaurant and staying, entering the gym through the access control, staying in the public sofa area, queuing in the lobby to check in, checking out from the room door with a card, etc.

[0040] The hotel preference data refers to a set of information related to the behavior mode, usage habit and personalized setting of a specific user in the hotel, which is derived from the historical records formed in the user's previous stays, the service options actively set, the preference templates or the habitual behavior data summarized by the system through learning; the preference data content includes but is not limited to: the type of guest room equipment preferred by the user (such as habitually using intelligent sound or intelligent incense), the commonly used temperature and humidity range in the room, the lighting brightness and color temperature setting, whether the user prefers a quiet environment, whether the user is used to opening the curtains or playing background music at specific time periods in the morning and evening, etc. These preference data are usually stored with the user as the primary key, and each preference item contains fields such as behavior label, device attribute, priority level and trigger condition, which are used as context input in the subsequent system decision process to participate in device intelligent control and service personalized configuration, thereby improving the consistency and comfort of the guest experience.

[0041] For the current corresponding nth behavior data, activity state reasoning is performed to obtain user external activity prediction data; the check-out timestamp T n is obtained; the user device running data between the check-out timestamp T n-1 and the check-out timestamp T n are obtained to perform user behavior analysis by using a reasoning network to obtain user internal activity prediction data;

[0042] The specific steps of activity state reasoning for the current corresponding nth behavior data include:

[0043] Based on the nth behavior data, continuous space-time features are extracted to obtain a user space-time feature set; wherein the user space-time feature set contains a behavior path, a duration and corresponding hotel public place position coordinates;

[0044] The user spatiotemporal feature set is extracted by structurally analyzing the current nth behavior data and combining the continuous positioning behavior events in the adjacent time period. When extracting, first, the activity path segment of the user in the public area of the hotel is determined, the spatial nodes, stay time, moving speed and route mode of the user moving from the previous event location to the current event location are identified, and semantic labels are given in combination with the behavior event type (such as dining, fitness, meeting, etc.); then the spatial coordinates at the time of the event, the function label of the corresponding hotel area (such as public rest area or service passage), the event duration, the event occurrence time and the event triggering mode (such as whether accompanied by access control, positioning or code scanning operation) are extracted; the above information is composed into the spatiotemporal feature set at the current time in the form of a vector, and the data structure usually includes a timestamp sequence, a spatial position index, a behavior category code, a time length value and path topology information, which reflects the behavior state and spatial positioning relationship of the user in the hotel space at this moment, and provides rich spatial context and short-time dynamic features for subsequent model analysis, especially helps to judge whether the current behavior belongs to continuous path, habitual behavior or abnormal stay behavior.

[0045] The first to nth-1 behavior data is acquired to construct behavior sequence context features; the user spatiotemporal feature set, the behavior sequence context features and the hotel preference data are taken as inputs and input to a pre-trained behavior sequence learning model for analysis to obtain user external activity prediction data;

[0046] The behavior sequence context features of the first to nth-1 behavior data are obtained by sequence encoding, time alignment and semantic abstraction of historical behavior events; the first n-1 behavior events of the user in the hotel are extracted in chronological order, ensuring that each event contains complete timestamp, position coordinates, behavior category label, stay time and other fields; these events are sequentially embedded into a time sequence, and unequal interval events are aligned by time normalization method; the system converts each event into a fixed-length behavior vector through a behavior embedding algorithm (such as position embedding, event type encoding, behavior semantic graph embedding, etc.), and then uses a sliding window mechanism or a recursive structure (such as a recurrent neural network) to establish the evolution association and context dependence relationship between events, and finally forms a behavior sequence context feature reflecting the short-term and medium-term behavior tendency of the user.

[0047] The user external activity prediction data is the behavior state, behavior purpose or activity intention of the user currently or to be performed inferred by the system based on the behavior event data, spatiotemporal features and historical behavior context of the user in the public area of the hotel. It is a dynamic and high semantic level explanation and prediction of the user's behavior outside the guest room, expressed in the form of behavior labels (such as going out to eat, short-time leaving the room, attending a meeting, exercising) and corresponding confidence or probability distribution;

[0048] The pre-trained behavior sequence learning model is a deep learning-based time series modeling network, which is usually trained using long short-term memory networks, gated recurrent units, transformer structures or graph neural networks; the training process mainly relies on a large amount of historical user behavior trajectory data of the hotel, and through supervised learning, the historical behavior event sequence and user preference data are input, and the label is the actual external activity intention or behavior state of the user (such as short-time check-out, going to the gym, going out to eat, accompanying visitors, etc.); During the construction of the training data set, the final behavior intention of each behavior sequence needs to be manually annotated or semi-automatically annotated, and the corresponding time period, spatial path and preference label are constructed; the model learns how to predict the current behavior state according to the previous behavior characteristics and generalizes different behavior patterns through the loss function optimization in the training; the trained model can capture the semantic evolution of behavior, the migration of spatial preference and the influence of preference habits, so as to have the ability of prediction and induction when facing new users or new sequences.

[0049] Obtain the check-out timestamp T n-1 to the check-out timestamp T n The specific steps of user behavior analysis by combining device running data between the check-in timestamp T and the check-out timestamp T with the inference network include:

[0050] Identify the user's environmental change intention based on environmental perception data; identify the user's human action intention based on device application switch state data; construct the scene feature based on the user's environmental change intention and the user's human action intention, and obtain the internal behavior feature vector sequence;

[0051] The user's environmental change intention refers to the implicit demand or potential intention of the user to adjust the room environment inferred by the system through analyzing the trend of the environmental perception data in the room; for example, when the system detects that the room temperature continues to rise and the humidity decreases, and the user does not directly operate the air conditioner, the system may infer that the user has an environmental change intention to adjust the temperature; this intention is not expressed by the user through explicit control, but is reflected by the user's implicit demand through passive fluctuation of environmental variables, which is usually identified by analyzing the deviation between the change rate, fluctuation amplitude, duration and the established comfort interval, to identify the possible target, such as the need to cool down, the desire to dim the light, the desire to ventilate, etc. This intention reflects the user's perception and expected adjustment of the environmental state, and is an important basis for understanding the behavior context.

[0052] User action intention refers to the explicit behavior purpose expressed by the system through analyzing the user's active control behavior on the room equipment; for example, when the user successively turns off the TV, air conditioner and closes the curtains in a short time, and at the same time triggers the door lock switch, the system can identify the human action intention of leaving the room or preparing to go out; unlike environmental change intention, human action intention reflects the direct intention expressed by the user through actual operation of the equipment, including on-off control, mode switching, parameter adjustment, etc., which is often based on equipment switch state data, combined with the timing of behavior operation, the sequence of linkage between devices and consistency with past behavior to determine the intention; identifying such intention helps to clarify the actual goal of user operation behavior and improve the system's ability to respond to user state changes in real time.

[0053] Scenario feature construction refers to the fusion of user device usage behavior and environmental change data in the time period before leaving the room to generate a set of high-level behavior features that can reflect the actual behavior context of the user; in specific operation, the system will extract key feature variables from environmental change intention and human action intention, such as temperature change rate, light change amplitude, device operation frequency, device combination usage mode, operation sequence stability, etc., and organize these variables into a time sequence feature sequence through time window sliding, event density analysis and semantic label coding, etc. The system uses a time sequence modeling structure (such as a recurrent network or a time sequence encoder) to integrate it into a structured internal behavior feature vector sequence, each vector in the sequence represents the behavior state in a certain time slice, and its content includes user operation intensity, response frequency, environmental deviation, etc. information, used to represent the continuous behavior trajectory in the entire leaving window, and the feature sequence is the basic data for modeling the user's internal behavior in the room, and is an important input for the inference model to identify patterns and infer intentions.

[0054] Obtain historical stay data of the user staying in the room in the target hotel; based on the historical stay data and hotel preference data, establish comprehensive in-stay scenario features;

[0055] The comprehensive in-stay context feature refers to a global semantic representation of personalized behavior background and environmental adaptation characteristics constructed by the system based on historical stay behavior data and preference configuration of the user in the current hotel; the system will call the user's past stay records in the hotel or similar hotels, including stay duration, stay frequency, commonly used equipment combination, typical operation mode, equipment adjustment habit and other information, and combine the user's preference template set before staying (such as the most comfortable air conditioning temperature, whether to automatically pull the curtain, whether to use the silent mode, etc.), extract the high-frequency behavior mode in the room use scene; through multi-dimensional feature fusion (such as frequency vectorization, behavior heat map coding, etc.) and label embedding processing, a unified user in-stay portrait is formed, which is the comprehensive in-stay context feature, which is the understanding carrier of the system for the user's long-term behavior preference and environmental expectation, and helps to improve the consistency judgment ability of the reasoning network for the current behavior and historical habits.

[0056] The comprehensive in-stay context feature and the internal behavior feature vector sequence are input into the reasoning network to analyze the check-out timestamp T n-1 to the check-out timestamp T n The device usage mode, environmental change trend and human activity mode in the internal behavior feature vector sequence are analyzed to obtain the user internal activity prediction data; the reasoning network takes a Bayesian reasoning network as a basic network structure to analyze the user behavior;

[0057] The user internal activity prediction data is an intelligent recognition result of the typical activity state of the user in the guest room output by the system after analyzing the environmental changes, human operation behaviors, historical check-in habits and preference factors in the off-occupancy time period; the device usage mode is the first type of behavior feature identified by the inference network, mainly reflecting the usage combination, frequency, change trend and operation rhythm of the user on the intelligent devices in the off-occupancy time period; for example, whether the air conditioner is turned on, whether the temperature setting is adjusted, whether the television is repeatedly turned on and off, whether the lighting is gradually dimmed or turned off, etc., these operation combinations can reflect whether the user is preparing to leave, whether the user is in a resting state or whether the user is performing office activities; through the collaborative control features and operation sequence rhythm of the devices, the system can infer whether the user is in an active use state or in an automatic energy-saving response; the environmental change trend is the analysis result of the system on the change process of the environmental perception data in the room, reflecting the indirect expression of the user's environmental adjustment intention without directly controlling the devices; for example, whether the light intensity gradually decreases, whether the indoor temperature is maintained within a constant range, whether the noise shows a downward trend, etc.; the environmental change trend is usually compared and analyzed in combination with the device response result and the user's historical preference, to judge whether the current environment has tended to the user's desired state, and whether the environment has stabilized or fallen due to the user's imminent departure; the human activity mode refers to the purposeful behavior combination and behavior logic expressed by the user through device switch operation, movement triggering, access control behavior, etc. The human activity mode not only looks at whether the operation occurs, but also focuses on the combination order, time interval, repeatability and typicality of the operation; for example, the user continuously turns off the air conditioner, turns off the lights, opens the door lock and triggers the door magnet within 10 minutes, which may constitute a typical human activity mode of preparing to leave; if the user repeatedly adjusts the air conditioner and the curtains during the noon period, it may indicate an office or stay state. These patterns are modeled through probabilistic reasoning, constituting the main clues for identifying the user's internal activity state.

[0058] The process of the inference network for user behavior analysis is based on the structure of the Bayesian inference network, which integrates the user's comprehensive in-occupancy situation features and its T n-1 to T nThe internal behavior feature vector sequence formed in the time period establishes the conditional probability dependency relationship between the features, thereby systematically inferring the behavior state of the user in this time period. The reasoning network is constructed in the following manner: taking historical user behavior data as training samples, combining artificial expert rules and data statistical analysis results to establish an initial network topology structure, and then using parameter learning and structure learning algorithms to optimize the dependency relationship between nodes; in the construction process, first determine the core variable nodes in the network, including user behavior labels, device operation features, environmental change indicators, and preference constraint factors, etc., then identify the statistical correlation and causal path between variables through data analysis to form an initial directed acyclic graph; on this basis, the maximum expectation algorithm or Bayesian scoring function is used to iteratively optimize the network structure to ensure the establishment of the conditional independence assumption and the balance of the model generalization ability, and the conditional probability distribution under each node is parameterized through maximum likelihood estimation or Bayesian estimation to ensure that the model can effectively infer the target behavior label according to the partial variable state in actual use.

[0059] The user internal activity prediction data and the user external activity prediction data are spatio-temporally aligned, and the check-out timestamp T n corresponding behavior intention label; based on the identified behavior intention label, matching a real-time device application update index in the guest room; based on the real-time device application update index, intelligently adjusting the hotel device application in the guest room and the hotel device application outside the guest room to realize intelligent management of the target hotel device application;

[0060] The check-out timestamp T n The specific steps of identifying the corresponding behavior intention label include:

[0061] The user internal activity prediction data and the user external activity prediction data are spatio-temporally aligned to obtain a user comprehensive context behavior vector; based on the user comprehensive context behavior vector, behavior intention matching is performed to obtain a behavior intention label; first, the user comprehensive context behavior vector generated in the previous stage is used as input, which integrates the internal activity features and external behavior features of the user before and after checking out and has completed semantic integration through a spatio-temporal alignment graph structure. This vector has high-dimensional semantic expression capability and can reflect the complete behavior state of the user in the current context; then, the system extracts a group of modeled behavior intention label vector templates from the built-in behavior intention label prototype library, each label vector is trained or expert semantically modeled according to a large amount of historical user data, represents a specific behavior intention, and each behavior intention label includes standardized behavior feature distribution, typical device operation mode, preference consistency weight, and behavior background feature range.

[0062] By the behavior semantic similarity calculation method, the semantic similarity measurement method (such as cosine similarity or Euclidean distance) is used to compare the current behavior vector of the user with each label vector one by one, calculate the similarity value in the semantic space, and form a similarity score list; according to the similarity sorting result, the system selects the label with the highest score and exceeding the confidence threshold as the main behavior intention label at the current time, and the secondary label with higher score can be used as an auxiliary behavior reference;

[0063] The specific steps of spatio-temporal alignment are: extracting all behavior events in the user internal activity prediction data and the user external activity prediction data; mapping the user internal activity prediction data and the user external activity prediction data into a user behavior prediction coordinate system, and labeling the spatial position coordinates and event absolute time stamps according to the behavior event granularity of the behavior events;

[0064] The extraction of behavior events is to analyze the structured behavior state information contained in the prediction data, and identify the smallest event unit with independent behavior semantics; the system first deconstructs the prediction data, parses the fields such as labels representing user activities, corresponding time intervals, spatial positions, device operation characteristics, and environment response states into standardized behavior description units; each activity with independent behavior intention expression, such as turning off the air conditioner, entering the elevator, dimming the light, and leaving the gym, is regarded as a behavior event. In the extraction process, the system arranges all events in time sequence with event trigger time as the main anchor point, and eliminates time overlap or behavior conflict to ensure that each event has clear boundaries in time and semantics.

[0065] The method of mapping the user internal activity prediction data and the user external activity prediction data into the user behavior prediction coordinate system is to establish a two-dimensional or three-dimensional coordinate system structure with time and space as the main axis to uniformly express and align the two types of prediction behavior events; the user behavior prediction coordinate system is a structured data mapping space, the time dimension is usually in units of standard time stamps, and the space dimension is represented by pre-defined spatial position codes in the hotel scene, such as each floor, each functional area, and each guest room having a unique space number; at a finer granularity, it can also be labeled as a two-dimensional coordinate; in the mapping process, the system will position each behavior event on the time axis according to the time information carried by the behavior event, and drop it on the spatial dimension according to the position information of the behavior occurrence, while labeling the specific content of the event in the coordinate system, including behavior type, event confidence, trigger mode, etc., to complete the projection of the behavior event in the coordinate system; the spatial position can be converted into position coordinates through the spatial label mapping table, and the time stamp can be directly normalized to form a unified alignment basis.

[0066] define a dynamic time window according to a behavior event granularity; establish a plurality of event corresponding relationships based on the dynamic time window and user internal activity prediction data and user external activity prediction data; and set a behavior response tolerance interval for each event corresponding relationship;

[0067] The dynamic time window defined according to the behavior event granularity is that the system automatically adjusts the matching time range allowed for each type of event according to the time length, continuity and behavior rhythm of the behavior event itself, so as to establish a more adaptive behavior corresponding relationship. The dynamic time window represents the allowable alignment interval of each behavior event on the time axis, and is a time period with adjustable length, which is used to determine whether the two events can establish a time association. The system defines an initial window range for each type of behavior according to the semantic attributes and historical behavior characteristics of the event, and automatically adjusts the window width through model learning by using a sliding window mechanism or a rhythm estimation algorithm.

[0068] In the dynamic window, the system selects events with high time overlap and close behavior semantics from internal and external events to establish event corresponding relationships, that is, to identify whether two events express the same continuous behavior or causally related behavior. Each event corresponding relationship further defines a behavior response tolerance interval, which is the maximum offset range that two behavior events can tolerate in time. It can not only handle the delay existing in the perception system, but also adapt to the asynchronization in user behavior. For example, turning off all lamps within 5 minutes after leaving the house is still considered as part of the leaving behavior. The role of the response tolerance interval is to improve the tolerance of the system to the uncertainty of the time sequence of the real behavior, prevent misjudgment of unimportant time differences, and improve the robustness and fault tolerance of behavior understanding.

[0069] Taking each behavior event as a node and the event corresponding relationship as an edge, a graph structure after time and space alignment is constructed, and the graph structure after alignment is represented by a behavior semantic embedding model to obtain a user comprehensive context behavior vector;

[0070] In the graph construction, the system takes each internal or external behavior event that has been time and space aligned as a node, and connects the events that are determined to be associated through edges to form a graph structure. Each edge records the time and space distance, semantic similarity, response strength and other attributes of the two events, forming a multi-attribute heterogeneous graph. Subsequently, the graph is sent to a behavior semantic embedding model for learning and vectorization processing. The model can be a graph convolutional neural network or a structured transformation model, which is used to encode the nodes and their context structure into a fixed-dimensional vector representation. The model compresses the entire graph into a comprehensive context behavior vector by capturing the causal relationship, spatial adjacency, time sequence logic and semantic coupling characteristics between events. This vector is the overall abstract expression of the user behavior in the current time period, which is used for behavior intention recognition and control strategy matching.

[0071] It should be noted that in all data processing and data analysis processes, when the input data structure is inconsistent, the system aligns the features and fills in the missing values through the data standardization module to ensure that all behavior event data can be uniformly entered into the analysis process, and the data format generated from different sources (such as devices, sensors, and behavior recognition modules) is mapped to an internal uniform behavior event data structure. If the data granularity is different, for example, some external behavior events are composite behaviors, the system will disassemble the events and re-model and label them according to the standard behavior granularity. Through the above standardization and repair mechanism, the system can ensure that all behavior data can be accurately located in the coordinate system and participate in the unified graph modeling and semantic reasoning process, ensuring that the behavior intent recognition still has consistency and high accuracy under the condition of multi-source heterogeneous data.

[0072] Based on the recognized behavior intent label, the specific steps of matching the real-time device application update index in the guest room include:

[0073] A device application control factor library containing multiple hotel device applications is constructed. The device application control factor library contains several device application control factors. The device application control factors include but are not limited to device type identification, user usage frequency, device current running state, environmental response priority, energy consumption level, and device belonging scenario category label.

[0074] Based on the behavior intent label, the associated target device application set is matched. For each target device application in the target device application set, a corresponding device application control factor in the device application control factor library is matched as a behavior association weight coefficient. The device application running parameters of the current guest room are obtained. The real-time device application update index is calculated by combining the device application running parameters and the behavior association weight coefficient.

[0075] The construction mode of the device application control factor library is based on the classification modeling of all intelligent controllable device resources in the hotel. According to the multi-dimensional factors such as functional attributes, user behavior correlation, use frequency and energy efficiency characteristics, the application characteristics of each type of device are structured defined. The system imports the list of all devices from the hotel management system in the initialization stage, and extracts the key attributes that can affect the device control decision according to the functional categories, service scenarios, historical user interaction data, energy consumption data and other information of the devices, and establishes a standardized control factor set. Each device application object forms a device control factor record in the library, which is composed of multiple fields, covering device type identification (such as lighting, air conditioning, television), user use frequency (usage times and duration statistics during the current or historical stay period), device current running state (on, off, adjusting, etc.), environmental response priority (indicating the contribution weight of the device to environmental comfort or user satisfaction), energy consumption level (high, medium, low corresponding to unit time consumption), device belonging to scene category label (such as rest scene, off scene, welcome scene); the control factor library adjusts the weight distribution of each factor in real time according to user feedback and use behavior through regular updating mechanism, to ensure that the device response and user demand maintain high consistency.

[0076] The device type identification in the device application control factor is used to identify the functional attribution of the device in the control logic, which is convenient for matching with the target behavior scene in the behavior intention label; the user use frequency represents the frequency of the device being enabled in the current user historical behavior, which is used to judge the degree of dependence and habitual use tendency of the user to the device; the device current running state reflects the necessity of the current control, and if the device is already in an ideal state, the response priority can be adjusted downward; the environmental response priority evaluates whether the device directly affects the comfort or the behavior scene, such as the contribution of air conditioning to temperature regulation is higher than that of background music; the energy consumption level measures the influence of device response on energy consumption, which can be used for intelligent balancing in energy saving scenarios; the device belonging to scene category label is used to bind the device to a specific behavior intention type, for example, the lighting and curtains are preferentially controlled under the preparation for rest label, rather than the television and incense. These factors constitute the core input of the decision model and provide quantitative basis in the update index calculation.

[0077] The matching of the associated target device application set is to take the behavior intention label as a query condition to search for a corresponding device list from a preset behavior-device mapping relationship model; the mapping relationship model is built based on historical behavior data mining, user operation path analysis and expert experience knowledge, and common behavior intentions are correspondingly mapped with typical device groups; for example, preparing to leave the room is usually associated with air conditioning, lighting, curtains and door locks, and returning to the room is preferentially matched with lighting, background music and air conditioning temperature setting; after the user behavior intention label is identified, the system extracts the corresponding device set as a response candidate target in the relationship model, and further dynamically filters according to user individual preferences, room configuration differences and current occupancy states to obtain the target device set that can actually respond in the guest room.

[0078] The calculation method of the real-time device application update index is that after the target device application set associated with the current behavior intention label is matched, the system extracts the corresponding control factor of each device from the device application control factor library, and combines the current operating parameters of the device in the in-stay guest room to comprehensively calculate the response priority score of each device in the current behavior context; specifically, the system first converts the control factor of each device into a standardized score vector, mainly including user usage frequency score, running state deviation value, environmental response priority weight, energy consumption level adjustment coefficient and scene category matching degree; at the same time, the current parameter values such as switch state, adjustment intensity, running mode are obtained from the device running state, and the deviation degree between the current parameter values and the required state of the target behavior intention is weighted calculated; after the calculation is completed, all the update indexes of the devices are sorted, and the higher the value, the more the device needs to be adjusted or responded in the current behavior scene; this calculation method takes into account the user behavior intention, the actual state of the device and the energy efficiency strategy, and is the core mechanism for realizing the intelligent linkage of guest room devices under the behavior driving.

[0079] The real-time device application update index is used to intelligently adjust the hotel device applications in the in-stay guest room and the hotel device applications outside the in-stay guest room, to realize intelligent management of the target hotel device applications;

[0080] The specific adjustment method is: for example, in an actual application scenario, a user checks into a smart hotel and is identified by the system at 12:30 noon with a behavior intention label of short-time leaving; the system first extracts the user internal activity prediction data and user external activity prediction data of the user in the current time period, and performs spatio-temporal alignment based on the granularity of the behavior event to construct a spatio-temporal alignment graph structure, and generates a user comprehensive context behavior vector through a behavior semantic embedding model; then, the system matches and analyzes the vector with multiple labels in the behavior intention label library to determine that the current behavior intention is short-time leaving;

[0081] According to the identified behavior intention label, a target device application set associated with the short-time off-label is obtained in a preset behavior-device mapping rule, covering device types such as air conditioners, lighting, curtains, televisions, air purifiers, and door locks; the system extracts corresponding device application control factors for each target device from a device application control factor library, including device type identification, user usage frequency, current running state, environmental response priority, energy consumption level, and device belonging scenario category label information; combined with the current running parameters of each device, the system calculates the update index of each device, and obtains a higher real-time device application update index value of the air conditioner, television, and curtain, indicating that it needs to be adjusted first under the current behavior intention.

[0082] According to the sorting result of the update index, the system sequentially performs intelligent adjustment operation on the target device application; specifically including: adjusting the air conditioner to energy-saving mode and increasing the set temperature, turning off the television and all main lighting lamps, only keeping the entrance induction lamp, automatically closing the curtain to the full-closed state, and keeping the air purifier in the original running state unchanged, because the user has air quality preference record; after the user leaves the room, the system automatically locks the door lock controller after detecting that the door magnetic state becomes closed for 3 seconds, and pushes the guest room state feedback information to the user; at the same time, the system judges that the user is going to the gym area, automatically wakes up the corridor lighting and the gym welcome interface, and provides the user with a coherent and low-perception intelligent experience. Through the above operation, according to the user behavior intention label, combined with the device application control factor library and the device running state data, the real-time device application update index is calculated, and the hotel device application in the in-room and out-of-room is intelligently adjusted, realizing the whole-process intelligent management of the target hotel device application, and improving the energy use efficiency and user service experience.

[0083] Embodiment 2, an intelligent application management system for a hotel, as shown in Figure 1 , comprising:

[0084] The hotel user behavior reasoning module comprises a data perception unit and an activity prediction unit; the data perception unit is used for real-time collection of user device running data of a user in a target hotel in-room guest room in the hotel room; the user device running data comprises environmental perception data and device application switch state data; the corresponding public user behavior trajectory data and hotel preference data are obtained; the user behavior trajectory data comprises nth behavior data, n = 1, 2, …, N, and the nth behavior data is the nth behavior event recorded in the public place of the user in the hotel; the activity prediction unit is used for activity state reasoning for the current corresponding nth behavior data to obtain user external activity prediction data; the off-time stamp T n is obtained; the off-time stamp T n-1 is obtained; the off-time stamp T nThe user equipment operation data between the two time points is combined with an inference network to analyze user behavior and obtain user internal activity prediction data;

[0085] The intelligent application management module includes a chip control unit; the chip control unit is used for spatiotemporal alignment of the user internal activity prediction data and the user external activity prediction data, and simultaneously identifies a time stamp T n corresponding behavior intention labels; based on the identified behavior intention labels, a real-time device application update index in the guest room is matched; based on the real-time device application update index, the hotel device applications in the guest room and outside the guest room are intelligently adjusted, thereby achieving intelligent management of the target hotel device applications.

[0086] It should be understood that, for those skilled in the art, improvements or changes can be made according to the above description, and all these improvements and changes shall belong to the protection scope of the appended claims of the present application. The parts not described in detail in the specification belong to the prior art known to those skilled in the art.

Claims

1. A method for intelligent application management for a hotel, the method comprising: The method comprises the following steps: For at least one in-stay room user in the target hotel, real-time collection of in-stay room user's user equipment running data in the hotel room; the user equipment running data includes environmental perception data and device application switch state data; obtaining corresponding public user behavior path data and hotel preference data; the user behavior path data includes the nth behavior data, n = 1, 2, …, N, and the nth behavior data is the nth behavior event recorded in the hotel public place; Carrying out activity state reasoning on the current corresponding nth behavior data to obtain user external activity prediction data; acquiring a check-out timestamp T of the nth behavior data n ; acquiring the check-out timestamp T n-1 between the check-out timestamp T n and the check-out timestamp T n , user equipment running data is combined with an inference network to analyze user behavior and obtain user internal activity prediction data; The user internal activity prediction data and the user external activity prediction data are spatio-temporally aligned, and a time stamp T of leaving the room is identified n The corresponding behavior intention label; based on the identified behavior intention label, a real-time device application update index in the guest room is matched; based on the real-time device application update index, the hotel device application in the guest room and the hotel device application outside the guest room are intelligently adjusted, and the intelligent management of the target hotel device application is realized.

2. The intelligent application management method for a hotel according to claim 1, wherein, The specific steps of carrying out activity state reasoning on the current corresponding nth behavior data include: Based on the nth behavior data, continuous spatio-temporal features are extracted to obtain a user spatio-temporal feature set; wherein the user spatio-temporal feature set contains behavior path, duration and corresponding hotel public place location coordinates; The first to nth-1 behavior data are obtained to construct behavior sequence context features; the user spatio-temporal feature set, the behavior sequence context features and the hotel preference data are taken as inputs and input into a pre-trained behavior sequence learning model for analysis to obtain user external activity prediction data.

3. The intelligent application management method for a hotel according to claim 2, wherein, acquiring a check-out timestamp T n-1 between the check-in timestamp T n The specific steps of user behavior analysis by combining user equipment running data between the check-in timestamp T and the check-out timestamp T with an inference network include: Based on the environmental perception data, the user environmental change intention is identified; based on the device application switch state data, the user artificial action intention is identified; based on the user environmental change intention and the user artificial action intention, the scene feature is constructed to obtain an internal behavior feature vector sequence; The historical stay data of the in-stay room user in the target hotel is obtained; based on the historical stay data and the hotel preference data, comprehensive in-stay scene features are established; The integrated in-home context feature and internal behavior feature vector sequence is input to the inference network to analyze the departure time stamp T n-1 to the departure time stamp T n Device usage patterns, environmental change trends, and human activity patterns within the home; combining device usage patterns, environmental change trends, and human activity patterns to obtain user internal activity prediction data.

4. The intelligent application management method for a hotel according to claim 3, wherein, Identifying a time stamp T of leaving the room n The specific steps of the corresponding behavior intention label include: The user internal activity prediction data and the user external activity prediction data are spatio-temporally aligned to obtain a user comprehensive context behavior vector; based on the user comprehensive context behavior vector, behavior intention matching is carried out to obtain a behavior intention label; The specific steps of spatio-temporal alignment include: extracting all behavior events in the user internal activity prediction data and the user external activity prediction data; mapping the user internal activity prediction data and the user external activity prediction data to a user behavior prediction coordinate system, and labeling spatial position coordinates and event absolute time stamps according to the behavior event granularity of the behavior events; Defining a dynamic time window according to the behavior event granularity; based on the dynamic time window, a plurality of event corresponding relationships are established for the user internal activity prediction data and the user external activity prediction data; setting a behavior response tolerance interval for each event corresponding relationship; Taking each behavior event as a node and the event corresponding relationship as an edge, a spatio-temporally aligned graph structure is constructed, and the aligned graph structure is vectorized by a behavior semantic embedding model to obtain a user comprehensive context behavior vector.

5. The intelligent application management method for a hotel according to claim 4, wherein, Based on the identified behavior intention label, the specific steps of matching the real-time device application update index of the in-stay room include: A device application control factor library containing a plurality of hotel device applications is constructed; the device application control factor library contains a plurality of device application control factors; Based on the behavior intention label, a target device application set associated with the behavior intention label is matched, and a device application control factor corresponding to each target device application in the target device application set is matched in a device application control factor library as a behavior correlation weight coefficient; a device application running parameter of a current in-house guest room is acquired; and a real-time device application update index is calculated by combining the device application running parameter and the behavior correlation weight coefficient.

6. The intelligent application management method for a hotel according to claim 5, wherein, The reasoning network takes a Bayesian reasoning network as a basic network structure to analyze user behaviors.

7. An intelligent application management system for a hotel, characterized by, The system applies the intelligent application management method for a hotel according to any one of claims 1-6, comprising: The hotel user behavior inference module comprises a data perception unit and an activity prediction unit. The data perception unit is configured to collect, in real time, user equipment running data of a user staying in a room in a target hotel, the user equipment running data comprising environment perception data and device application switch state data; obtain corresponding public user behavior trajectory data and hotel preference data; the user behavior trajectory data comprises nth behavior data, n=1, 2, …, N, the nth behavior data being an nth behavior event of the user recorded in a public place of the hotel; the activity prediction unit is configured to perform activity state inference on the current corresponding nth behavior data to obtain user external activity prediction data; obtain a check-out timestamp T n ; obtain user equipment running data between the check-out timestamp T n-1 and a check-in timestamp T n ; and perform user behavior analysis on the user equipment running data by using a reasoning network to obtain user internal activity prediction data. The intelligent application management module comprises a chip control unit; the chip control unit is used for spatiotemporal alignment of user internal activity prediction data and user external activity prediction data, and simultaneously recognizing a time stamp T of leaving a room n The corresponding behavior intention label; based on the recognized behavior intention label, a real-time device application update index in the guest room is matched; based on the real-time device application update index, the hotel device application in the guest room and the hotel device application outside the guest room are intelligently adjusted, so as to realize intelligent management of the target hotel device application.

Images