Big data life habit and behavior analysis method and system for smart home

Abstract

The present application relates to the technical field of smart home, especially to a big data living habit and behavior analysis method and system for smart home, the present application proposes the following scheme, collects user behavior data, environment data and device state data, processes multi-dimensional data to generate behavior preference label, outputs user's living behavior mode through behavior analysis model, and generates smart home device control strategy combined with real-time data, the present application realizes the linkage and personalized control of smart home device in multiple scenes, effectively improves the response efficiency and user experience of the system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart home, and in particular to a big data living habit and behavior analysis method and system for smart home. BACKGROUND

[0002] The existing smart home system mainly relies on fixed rules or single data sources (such as time trigger or environmental conditions) to control the operation of devices. This way has significant limitations in complex home scenarios:

[0003] Firstly, the existing method lacks in-depth analysis of user behavior habits and living rules, and can only provide simple device linkage, which cannot meet the user's demand for personalized scenes. Secondly, in the case of multiple scenes coexisting (such as "getting up mode" and "watching mode" existing at the same time), the traditional method is difficult to effectively distinguish and manage different scenes, which is easy to cause device conflict or strategy confusion. In addition, the existing technology lacks flexibility in responding to dynamic environmental changes (such as light and temperature fluctuations), which leads to the inability of device control strategy to adapt to user needs in real time, affecting the user experience.

[0004] Therefore, how to construct a smart home system with data-driven capability and scenario-based control logic based on user behavior patterns and dynamic environmental changes has become an important technical problem to be solved in the current smart home field.

[0005] For example, the Chinese patent with the authorization announcement number CN109542944B discloses a smart home user operation behavior recommendation method based on time sequence causal relationship analysis, which divides a large amount of user behavior habit operation data, then extracts the user frequent operation sequence from the data through a sequence pattern mining algorithm, and constructs a Bayesian network to mine the causal relationship between the sequences based on the frequent operation sequence, to construct a smart home operation behavior recommendation scheme. The invention combines the advantages of pattern mining algorithm and Bayesian model, reduces the entire data set to the selected frequent operation sequence set through pattern mining, greatly reduces the calculation complexity and noise in causal relationship calculation, is conducive to more efficiently forming a smart home operation behavior recommendation scheme that is more suitable for user experience, and enhances the time strong correlation between each device in the smart home operation behavior recommendation scheme, filling the shortcomings of traditional algorithms.

[0006] The above existing technologies all have the problems proposed in the background art: the smart home system has limitations in personalized scene recognition, multi-scene management and device linkage strategy. In order to solve the above problems, the present application designs a big data living habit and behavior analysis method and system for smart home. SUMMARY

[0007] The technical problem to be solved by the present application is to provide a big data living habit and behavior analysis method and system for smart home, which collects user behavior data, environment data and device state data, processes multi-dimensional data to generate behavior preference labels, outputs user living behavior patterns through a behavior analysis model, and generates smart home device control strategies in combination with real-time data.

[0008] To achieve the above object, the present application provides the following technical scheme.

[0009] The big data living habit and behavior analysis method for smart home comprises the following steps.

[0010] Collecting multi-dimensional data when a user uses smart home;

[0011] Processing the multi-dimensional data to output behavior preference labels;

[0012] Inputting the behavior preference labels into a preset behavior analysis model, and outputting user living behavior patterns through the behavior analysis model;

[0013] Generating smart home device control strategies in combination with real-time data according to the living behavior patterns.

[0014] The processing of the multi-dimensional data comprises the following steps.

[0015] Uniformly modeling the multi-dimensional data to generate a correlation matrix;

[0016] Performing pattern analysis on the correlation matrix to extract behavior features;

[0017] Generating behavior preference labels through a clustering algorithm according to the behavior features.

[0018] The uniformly modeling of the multi-dimensional data comprises the following steps.

[0019] Constructing a time-environment bidirectional index structure;

[0020] Analyzing the interaction relationship of the time-environment bidirectional index structure according to a predefined correlation rule to obtain a behavior correlation matrix;

[0021] Updating the behavior correlation matrix through an attention mechanism to generate a correlation matrix.

[0022] The pattern analysis of the correlation matrix comprises the following steps.

[0023] The data distribution in the correlation matrix is analyzed by a pattern mining algorithm to extract time regularity patterns, device linkage patterns and environment adaptation patterns of user behaviors.

[0024] The time regularity patterns, device linkage patterns and environment adaptation patterns are reduced in dimension by a feature embedding technology to generate behavior features.

[0025] The behavior preference label is generated by a clustering algorithm, comprising:

[0026] The behavior features are processed according to the DBSCAN and K-means clustering algorithms, and a clustering result is generated according to the similarity measurement of user behavior features.

[0027] The behavior preference label is generated for each clustering result.

[0028] The behavior analysis model comprises:

[0029] A feature input layer is configured to receive a user behavior feature vector generated by the behavior preference label;

[0030] A time series modeling layer is configured to capture the time dependence and periodicity of behaviors according to the time series data in the user behavior feature vector, and output a user behavior trend;

[0031] A behavior pattern analysis layer is configured to obtain a scenario-based behavior pattern of the user according to the output of the time series modeling layer and the behavior preference label;

[0032] A behavior pattern output layer is configured to enhance and classify the scenario-based behavior pattern by an adjacency matrix and a mask, and output a life behavior pattern through semantic conversion.

[0033] The time series modeling layer comprises:

[0034] A time window division sub-layer is configured to divide the time series data in the user behavior feature vector according to a dynamic time window to extract key behaviors in a time period, wherein the dynamic time window is calculated according to the volatility of the user behavior feature vector;

[0035] A time dependence analysis sub-layer is configured to generate a time dependence relationship graph by a recurrent neural network;

[0036] A regularity identification sub-layer is configured to extract the periodicity of behaviors by frequency spectrum analysis.

[0037] The intelligent home device control strategy is generated in combination with real-time data, comprising:

[0038] The real-time data is matched with the life behavior pattern of the user to determine whether the current scene conforms to the existing behavior pattern;

[0039] Generate a scene operation suggestion according to the matching result, determine the intelligent home devices and their operation sequence of linkage;

[0040] Generate a control instruction, and send the control instruction to the corresponding intelligent home device.

[0041] The big data life habit and behavior analysis system for smart home comprises a data acquisition module, a data modeling module, a behavior pattern management module and a control strategy generation module;

[0042] The data acquisition module is used for collecting user behavior data, environment data and device state data in real time and performing preprocessing.

[0043] The data modeling module is used for uniformly modeling the collected multi-dimensional data, generating a correlation matrix and extracting behavior features, and identifying the life behavior pattern and periodicity of the user.

[0044] The behavior pattern management module is used for outputting the life behavior pattern of the user.

[0045] The control strategy generation module is used for generating a scene operation suggestion and a device control instruction according to the matching result generated by real-time data and the life behavior pattern.

[0046] The data modeling module comprises:

[0047] The correlation modeling unit is used for uniformly modeling the preprocessed multi-dimensional data, generating a correlation matrix, and capturing the interaction relationship between user behavior and time, environment and device state.

[0048] The feature extraction unit is used for extracting the behavior features of the user from the correlation matrix, including time regularity features, device linkage features and environment adaptation features.

[0049] Compared with the prior art, the present application has the following advantages:

[0050] The present application accurately extracts user behavior preferences and life rules through multi-dimensional data modeling and behavior pattern analysis, and generates personalized device control strategies in combination with real-time data, thereby solving the problems of inaccurate scene recognition, multi-scene management conflict and inflexible device response in the prior art, and significantly improving the adaptability, linkage efficiency and user experience of the smart home system. BRIEF DESCRIPTION OF DRAWINGS

[0051] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the accompanying drawings:

[0052] Figure 1A flowchart of a big data life habit and behavior analysis method for smart home in Embodiment 1 of the present application is shown in FIG. 1.

[0053] Figure 2 A behavior analysis model structure diagram in Embodiment 1 of the present application is shown in FIG. 3.

[0054] Figure 3 A time sequence modeling layer structure diagram in Embodiment 1 of the present application is shown in FIG. 4.

[0055] Figure 4 A module diagram of a big data life habit and behavior analysis system for smart home in Embodiment 2 of the present application is shown in FIG. 5. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments.

[0057] Embodiment 1

[0058] Referring to FIG. 1, Figure 1 An embodiment provided by the present application is a big data life habit and behavior analysis method for smart home, and the specific steps of the method are as follows:

[0059] S1: Collecting multi-dimensional data when a user uses smart home;

[0060] In this embodiment, the multi-dimensional data includes user behavior data (such as device operation records), environmental data (such as indoor temperature, humidity, and light intensity), and device state data (such as the current running state and parameter setting of the device). The sensor network deployed in the smart home collects these data in real time and uploads the data to the central processing module for summarization and preliminary cleaning. Through comprehensive data collection covering multi-dimensional information of user behavior and environment, the basic data for subsequent modeling and behavior analysis has breadth and depth.

[0061] S2: Processing the multi-dimensional data to output behavior preference labels;

[0062] In this embodiment, the processing of the multi-dimensional data includes two parts: first, generating a correlation matrix through a data modeling module to capture the correlation between user behavior and environment, time, and device state; and then analyzing the data distribution in the correlation matrix through a clustering algorithm (such as DBSCAN and K-means) to extract the behavior preference features of the user and generate behavior preference labels.

[0063] Specifically, the behavior preference label can concisely and efficiently express the core behavior characteristics of the user, facilitate subsequent analysis model calculation, and dynamically reflect the time regularity and scenario demand of the user for the device use, thereby overcoming the problem of insufficient detailed understanding of user behavior in the traditional method.

[0064] S3: inputting the behavior preference label into a preset behavior analysis model, and outputting a life behavior mode of the user through the behavior analysis model;

[0065] In the embodiment, the behavior analysis model includes a feature input layer, a time sequence modeling layer, a behavior mode analysis layer, and a behavior mode output layer. The behavior preference label is input, the time dependence and periodicity of the user behavior are analyzed through the time sequence modeling layer, and the scenario life behavior mode (such as “morning getting up mode” or “evening watching mode”) is output through the behavior mode analysis layer in combination with the relationship mining of the correlation matrix.

[0066] Specifically, the complex behavior characteristics of the user are abstracted as the life behavior mode, which helps the smart home system to understand the demand of the user in a specific scenario, and solves the problem of lack of scenario logic in the device response in the existing smart home. At the same time, through semantic conversion and enhanced classification, the generated life behavior mode has stronger generalization and adaptability.

[0067] S4: generating a smart home device control strategy according to the life behavior mode in combination with real-time data;

[0068] In the embodiment, by matching the life behavior mode and the real-time collected data, a scenario operation suggestion is generated, and is converted into a device control instruction. For example, when the user enters the “watching mode” and the real-time light intensity is low, a control strategy of automatically dimming the light, closing the curtain, and starting the television is generated.

[0069] Specifically, the introduction of the real-time data enables the control strategy to dynamically adapt to the changing needs of the user, and the scenario operation of the behavior mode reduces the complexity of the device operation and improves the user experience. The problem of single control strategy and inflexible response in the traditional smart home is solved.

[0070] Specifically, in the prior art, smart home systems can perform device linkage control through user's basic operation data (such as turning on / off lights, adjusting air conditioner temperature, etc.) and sensor collected data (such as indoor temperature, humidity, etc.). However, in actual application, these methods face the following problems: first, the dimension of data is relatively single, lacking deep mining of user behavior patterns, and unable to accurately capture user's personalized needs; second, existing methods usually trigger device operation with fixed rules, and this static response mechanism has obvious limitations in dynamic and complex home environment, and is difficult to adapt to rapid changes in user behavior and environmental state. In the case of multiple scenes coexisting in the home, control conflicts or strategy confusion problems are prone to occur, for example, after recognizing the user's "getting up mode" and "watching mode" at the same time, the linkage strategy of the device is difficult to distinguish the priority, resulting in poor user experience.

[0071] In this embodiment, through deep modeling and behavior pattern analysis of user's multi-dimensional data, a smart home device control strategy dynamically adapting to user's needs is generated.

[0072] For example, this embodiment takes "smart bedroom getting up scene" as an example for illustration. When the user enters the bedroom in the early morning, the traditional method may only automatically open the curtains based on a simple time trigger rule, but cannot consider the user's multi-dimensional behavior and environmental needs, while this embodiment realizes precise scene control through the following steps:

[0073] Firstly, the system collects user's multi-dimensional data through the data collection module, including time characteristics (such as 6:30 to 7:00 in the morning), device state (such as closed curtains, running air conditioner) and environmental parameters (such as insufficient indoor light intensity of less than 300 lux). Through the data modeling module, these multi-dimensional data are associated and modeled to generate user's behavior preference label, for example, "likes natural light in the morning but is sensitive to temperature".

[0074] Then, the system inputs the behavior preference label into the behavior analysis model, combines time dependence and periodicity rules, and outputs user's life behavior pattern, for example, identifies that the user is in "getting up mode". On this basis, the system further generates personalized control strategy in combination with real-time environmental data, for example, in the case of detecting insufficient light, the curtains are preferentially opened, and the air conditioner temperature is adjusted to a comfortable range (such as 24℃). This strategy not only considers the user's behavior pattern, but also dynamically adapts to environmental changes, avoiding the simplification problem of traditional rule triggering.

[0075] Further, when the user gets up in the bedroom and prepares to go to the kitchen, the system recognizes that the user's behavior pattern has changed and dynamically adjusts the device control strategy according to real-time data and historical behavior patterns. For example, in the kitchen scene, the system can start the coffee machine in advance and brighten the lights in the kitchen. This dynamic switching mechanism not only ensures the continuity of the scene and the accuracy of the device linkage, but also improves the user's whole-house smart experience.

[0076] The specific steps of S2 are as follows:

[0077] S2.1: Uniformly model the multi-dimensional data to generate a correlation matrix;

[0078] In this embodiment, the multi-dimensional data includes user behavior data (such as device operation records, user movement trajectories, etc.), environmental data (such as indoor temperature, humidity, and light intensity), and device state data (such as the running mode of the air conditioner and the opening and closing state of the curtain). After being collected, these data are normalized and processed using a principal component analysis dimension reduction algorithm to remove redundant information, thereby ensuring the efficiency and consistency of the data. Subsequently, by constructing a time-environment bidirectional index structure, the time dimension of user behavior (such as morning, noon, and night) is cross-correlated with environmental parameters and device states to form a mapping relationship between the multi-dimensional data. On this basis, an association rule mining algorithm (such as the Apriori algorithm) is used to calculate the correlation strength between each data feature and generate a correlation matrix, with the elements in the matrix representing the correlation weights between different data features.

[0079] Specifically, the correlation matrix can not only effectively depict the complex correlations between user behavior, environmental state, and devices, but also dynamically adapt to changes in user behavior characteristics, solving the problems of data isolation and weak correlation.

[0080] S2.2: Perform pattern analysis on the correlation matrix to extract behavior characteristics;

[0081] In this embodiment, first, the system uses a frequent pattern mining technique to identify high-frequency operation patterns in user behavior, such as the opening of curtains, adjustment of air conditioners, and turning on of lights at a fixed time period every morning. Subsequently, a time series-based pattern analysis method (such as sliding window detection) is used to capture the periodicity of user behavior in the time dimension, such as the difference between weekend and weekday operation patterns. In addition, through the mining of device linkage patterns, the system can discover the collaborative operation relationship between multiple devices, such as dimming the lights and closing the curtains when the user watches TV.

[0082] Specifically, the core features of user behavior are extracted from the correlation matrix, including time regularity, device linkage, and environmental adaptability, thereby providing comprehensive data support for the generation of subsequent behavior preference labels.

[0083] S2.3: generating behavior preference labels by a clustering algorithm according to the behavior features;

[0084] In this embodiment, the DBSCAN algorithm is first used to perform density clustering on the behavior features, and operation habit groups of the user in different time periods are divided. For example, the morning behavior mode can include a "getting up mode" (opening the curtains, adjusting the air conditioner) and a "breakfast mode" (starting the coffee maker, brightening the kitchen light), and the evening mode can include a "movie watching mode" (closing the curtains, dimming the light). Subsequently, the K-means algorithm is used to further refine the clustering results, and the behavior features are further classified according to the operation frequency and scene dependency, thereby generating refined behavior preference labels such as "high-frequency movie watching mode" and "low-light adjustment priority mode".

[0085] Specifically, the clustering algorithm can convert complex user behavior features into simple and clear behavior preference labels, which directly reflect the user's operation tendency and scene demand. It can dynamically adapt to changes in user behavior, ensuring that the behavior preference labels always remain consistent with the user's current needs.

[0086] Further, first, the behavior features are clustered according to the DBSCAN clustering algorithm, and the operation habit core points are obtained. The number of operation habit core points is used to determine the number of clustering centers, and the operation habit core points after DBSCAN clustering are used as the initial center clusters of the K-means clustering algorithm. The Euclidean distance of the behavior features to each initial center cluster is calculated, and clustering division is performed according to the Euclidean distance. The clustering average error of the initial center cluster is calculated, the center cluster is updated according to the clustering average error, a convergence condition is defined, and it is judged whether the updated center cluster satisfies the convergence condition. If the convergence condition is not satisfied, the center cluster is continuously updated, and if the convergence condition is satisfied, the clustering result is output.

[0087] The specific steps of S2.1 are as follows:

[0088] S2.1.1: Construct a time-environment bidirectional index structure;

[0089] In this embodiment, the time-environment bidirectional index structure is used to establish a bidirectional association between the user's operation behavior data and the time features and environment states in which it occurs. The time features include the occurrence time point, duration, and periodicity of the behavior, and the environment states include temperature, humidity, and light intensity and other related physical parameters.

[0090] Specifically, first, the multi-dimensional data is time-sliced, i.e., the user behavior data is divided according to time points, and the behavior record of each time point is attached with its corresponding environmental state parameter. Then, the environmental state parameter is used as a conditional index to associate the user behaviors under similar environmental conditions. For example, under the condition that the indoor temperature is higher than 28℃ and the light intensity is lower than 300 lux, the user tends to start the air conditioner and increase the light brightness, and this behavior and its associated conditions are stored as an index entry.

[0091] Further, the index structure supports bidirectional query, i.e., the user's behavior and environmental conditions in a time period can be queried through the time period, and the user's typical behavior under similar environment can also be deduced through a specific environmental state. The complex relationship between user behavior and environmental change can be captured, thereby providing more comprehensive context information for subsequent behavior pattern analysis.

[0092] S2.1.2: Analyzing the interaction relationship of the time-environment bidirectional index structure according to the predefined association rule to obtain a behavior association matrix;

[0093] In this embodiment, the predefined association rule is used to mine the deep interaction relationship between user behavior and environmental conditions from the time-environment bidirectional index structure. The content of the association rule includes time regularity rule (e.g., high-frequency behavior at fixed time point), environmental adaptability rule (e.g., frequent use of air conditioner under high temperature condition), and device linkage rule (e.g., turning off the curtain when starting the TV).

[0094] Specifically, by frequency statistics on the time-environment bidirectional index structure, the confidence of the user behavior and its corresponding condition is extracted, and a behavior association matrix is generated according to the confidence. Each row of the behavior association matrix represents a behavior feature of the user, each column represents a time, environmental or device linkage feature related thereto, and the weight value in the matrix represents the strength of the association. For example, the user opens the curtain in the living room at 7:00 every morning, and the association weight between this behavior and the indoor light intensity lower than 200 lux is 0.8, which indicates that the probability of this behavior under this condition is relatively high. Through this mining process, the behavior association matrix not only accurately reflects the user's behavior habits, but also reveals the internal relationship between multi-dimensional features.

[0095] S2.1.3: Updating the behavior association matrix through an attention mechanism to generate an association matrix.

[0096] In this embodiment, a query vector (Query), a key vector (Key), and a value vector (Value) are constructed for each behavior feature (such as turning on the air conditioner, adjusting the light) and its corresponding condition feature (such as time, environment, and linked device) in the behavior association matrix. The query vector is generated from action data and represents the demand weight of the current user behavior. The key vector and the value vector are generated from historical data of the behavior association matrix and describe the static association between behavior features and condition features. Taking the morning mode as an example, the query vector may be composed of real-time indoor light intensity, time point, and user behavior of entering the bedroom. The key vector and the value vector represent the behavior preferences in the morning condition in history (such as the record of the user frequently opening the curtains).

[0097] In this embodiment, the dot product between the query vector and the key vector is calculated to obtain the matching degree of the behavior feature and the current real-time condition, and an attention score is generated. After Softmax normalization, the attention score represents the priority of each behavior feature. For example, under the condition of insufficient morning light, the priority score of opening the curtains may be significantly higher than that of other behaviors.

[0098] In this embodiment, the value vector is weighted and summed by the attention score to dynamically adjust the weight value in the behavior association matrix. The weighted matrix is represented as an association matrix, in which the weight value not only reflects the behavior rule in the historical data, but also adaptively combines the current scene demand. For example, the weight value of the "turn on the air conditioner" behavior in the association matrix will be dynamically amplified under high temperature conditions, and may be weakened in the morning low temperature environment.

[0099] It should be noted that in this embodiment, the attention mechanism adopts a multi-head attention mechanism, which independently calculates the weight for the time feature, the environment feature, and the device linkage feature. For example, one attention head focuses on the time dimension to calculate the correlation between the current time point and the historical behavior mode; another attention head focuses on the environment dimension to identify the influence of the current environment state on the user behavior. Finally, the output results of each attention head are integrated into a complete association matrix.

[0100] Specifically, based on the historical data of the behavior mode, the correlation score of each behavior feature and the current scene is calculated. For example, in the morning mode, the correlation of the curtain opening behavior may be higher than that of the air conditioner adjustment behavior, and in the high temperature environment, the correlation of the air conditioner adjustment behavior will be significantly improved. The attention mechanism assigns different weights to the behavior features, so that the association matrix can adaptively adjust the priority of the behavior features.

[0101] Further, in order to further optimize the calculation efficiency and accuracy of the matrix, firstly, the weight is adjusted in the time dimension, giving higher priority to the behavior of a specific time period; secondly, the weight is dynamically allocated in the environment dimension, for example, the linkage relationship between curtains and lights is given priority in insufficient light. The correlation matrix can reflect the user's current behavior demand and environmental conditions in real time, avoiding the behavior response lag problem caused by fixed weights in traditional static models.

[0102] The specific steps of S2.2 are as follows:

[0103] S2.2.1: Analyze the data distribution in the correlation matrix by a pattern mining algorithm, extract the time regularity pattern, device linkage pattern and environment adaptation pattern of user behavior;

[0104] In this embodiment, the Apriori algorithm is used to mine the frequently occurring specific behavior combinations and their correlation weights from the correlation matrix. These combinations can reflect the user's behavior regularity under specific time and environmental conditions. For example, in the morning period, through analysis of the correlation matrix, the time regularity pattern of the user's high-frequency operation "6:30 open curtains - 6:35 start coffee maker - 6:40 adjust air conditioner to 24°C" can be identified. In the case of high humidity, the environment adaptation pattern of the user's preference "prefer to start dehumidifying equipment" can be extracted. In addition, the device linkage pattern is identified by analyzing the linkage probability of device operation, such as "70% probability of dimming lights when turning on TV". These patterns are screened by weight threshold to ensure the accuracy of the results.

[0105] Specifically, through pattern mining, the behavior characteristics of users are accurately extracted from large-scale behavior data, so that the multi-dimensional behavior patterns of users are revealed from the data, not only improving the depth of behavior analysis, but also solving the defect that user behavior understanding is limited to a single time point or device.

[0106] S2.2.2: Reduce the dimensionality of the regularity pattern, device linkage pattern and environment adaptation pattern through feature embedding technology to generate behavior features.

[0107] In this embodiment, the results of pattern mining often contain multi-dimensional and multi-level feature combinations, such as time dimension (e.g. daily, weekly regularity), environmental dimension (e.g. humidity, light intensity) and device operation linkage dimension (e.g. linkage relationship between TV and lights). Direct use of these features will increase the computational complexity of subsequent analysis.

[0108] Specifically, by vectorization representation and dimensionality reduction, high-dimensional features are mapped to a low-dimensional feature space, the correlation between different pattern features is analyzed by principal component analysis, and the core behavior features of users are expressed in the form of principal components.

[0109] For example, for the user's high-frequency operation "6:30 open the curtain - 6:35 start the coffee machine - 6:40 adjust the air conditioner to 24°C", it can be converted into a compact vector representation through feature embedding, and the elements in the vector correspond to features such as time dependence, device linkage strength, and environment adaptation priority, respectively.

[0110] Referring to Figure 2 The behavior analysis model structure diagram of the embodiment of the present application comprises:

[0111] The feature input layer is used to receive the user behavior feature vector generated by the behavior preference label;

[0112] In this embodiment, the multi-dimensional data such as user operation behavior, time dimension, device linkage and environmental parameters are feature-encoded to generate a structured feature vector.

[0113] Specifically, these features are uniformly expressed through embedded coding technology, and the user behavior operation (such as the device switch operation sequence) and the time feature (such as the high-frequency behavior time period) are jointly embedded to express the weight and association of different dimensional features in the vector space.

[0114] The time sequence modeling layer is used to capture the time dependence and periodicity of the behavior according to the time sequence data in the user behavior feature vector, and output the user behavior trend;

[0115] In this embodiment, the long short-term memory network (LSTM) in the recurrent neural network (RNN) is used as the core technology of time sequence modeling to identify the time sequence and potential association between behaviors. The gating mechanism of LSTM can effectively solve the long-distance dependence problem in time sequence, such as the behavior rule of the user in a certain time period (such as early morning or evening).

[0116] Specifically, this layer generates a user behavior trend vector according to the time distribution pattern of the user behavior operation (such as the behavior habit of opening the curtain at 7:00 every day), and outputs a feature representation containing operation frequency, time rule and behavior change trend.

[0117] The behavior pattern analysis layer is used to obtain the user's scenario-based behavior pattern according to the output of the time sequence modeling layer and the behavior preference label;

[0118] In this embodiment, by introducing a graph neural network (GNN), a correlation graph between user behavior and device, environment state is constructed, and deep analysis is performed thereon. The nodes in the user behavior graph represent specific behaviors of the user (such as opening the curtain, starting the air conditioner), and the edges represent the time sequence dependency relationship between the behaviors and their adaptability to the environment. Through graph convolution operation, the behavior pattern analysis layer can extract high-order correlation information in the scene, for example, identify the device linkage habit of the user in the “movie watching mode” (such as closing the curtain, dimming the light).

[0119] In this embodiment, the weights of the nodes and edges are assigned by predefined rules or learning methods. For example, the node weight can be allocated based on the frequency or importance of the behavior, and the edge weight can be calculated by the correlation between the behavior and the environment parameter. The adjacency matrix A is used to represent the topological structure of the behavior graph.

[0120] In this embodiment, feature extraction is performed on each node, and the initial feature vector of each node is generated by jointly encoding the attribute values (such as behavior type, device state, and environment parameter) and external data (such as behavior preference label). The node feature vector can be represented as a high-dimensional vector by embedding method.

[0121] In this embodiment, graph convolution operation is performed on the nodes according to the initial feature vector, and the nodes are iteratively updated, wherein the calculation formula of the graph convolution operation is:

[0122]

[0123] wherein, represents the feature vector of node v in the k+1 layer, represents the feature vector of node u in the k layer, u represents the neighbor node of node v, N(v) represents the neighbor node set of node v, δ(·) represents an activation function, and W k represents the weight matrix in the k layer, A uv represents the adjacency matrix of node u and node v.

[0124] In this embodiment, through multi-layer graph convolution operation until the nodes of the graph neural network meet the pre-set convergence requirement, high-order correlation information in the scene is extracted according to the node features, including:

[0125] Device linkage relationship: analyze the cooperative operation mode between user behaviors, for example, the linkage relationship between “turning on the TV” and “closing the curtain”.

[0126] Environment adaptability analysis: capture the dynamic coupling between behavior and environment, for example, the adaptive operation of “dimming the light” under low light conditions.

[0127] Temporal sequence correlation: combined with the output of the time sequence modeling layer, capture the dependence of behavior in the time dimension, such as the device linkage strategy after "opening the curtain in the morning".

[0128] In this embodiment, the behavior nodes of the user are classified according to the high-order correlation information, and the labels (such as "movie watching mode" and "getting up mode") are generated, and the behavior is bound with the sceneized label according to the classification result, for example, the core operation sequence in the "movie watching mode" is extracted through the high-weight node and edge.

[0129] Specifically, in a complex scene, the user's behavior often has the characteristics of multi-device linkage and environment adaptation, and the traditional method is difficult to effectively model. Through the graph neural network, not only the operation of a single device can be recognized, but also the linkage relationship between devices and the sceneized behavior mode can be deeply mined, thereby enhancing the generalization ability and practical application value of the analysis model.

[0130] The behavior mode output layer is used for enhancing and classifying the sceneized behavior mode through the adjacency matrix and the mask, and outputting the life behavior mode through semantic conversion.

[0131] Specifically, according to the adjacency matrix fixed in the behavior mode analysis layer, the low-correlation behavior nodes are shielded through the mask, for example, in the "getting up mode", the system will preferentially retain the high-frequency operation of the user (such as opening the curtain and starting the coffee machine), and shield the low-frequency device operation (such as turning on the sound).

[0132] Further, the enhanced behavior mode is classified through semantic conversion, and is converted into a life behavior mode label such as "getting up mode" or "movie watching mode". Semantic conversion can abstract complex behavior relationships into easily understandable life scenes, which facilitates the intelligent home system to generate accurate device control strategies according to these scene modes.

[0133] Please refer to Figure 3 , the time sequence modeling layer structure diagram of the embodiment of the present application, the time sequence modeling layer comprises:

[0134] The time window division sub-layer is used for dividing the time sequence data in the user behavior feature vector according to a dynamic time window, and extracting key behaviors in the time period, wherein the dynamic time window is calculated according to the volatility of the user behavior feature vector.

[0135] In this embodiment, the division of the dynamic time window is based on the volatility of the user behavior feature vector, which is determined by analyzing the change rate of the behavior data, such as the frequency of behavior operation, the mean and variance of the time interval, etc. Taking the time interval of behavior operation as an example, if the behavior interval is short in a certain time period (such as high-frequency use of equipment in the morning), the time window is automatically reduced, thereby capturing the behavior details in detail; on the contrary, if the interval is long (such as sparse device operation at night), the time window is appropriately enlarged to improve processing efficiency. Through this dynamic adjustment mechanism, the time window can adapt to the characteristic changes of user behavior, solving the problem that the fixed time window may ignore the details of short-term high-frequency behavior or waste processing resources. In addition, the boundary calculation of the dynamic time window uses the moving average method, combined with the local trend of the behavior data, to ensure that the division of the time window is smoother and more stable, thereby avoiding the omission of important behaviors.

[0136] a time-dependent analysis sub-layer for generating a time-dependent relationship graph through a recurrent neural network;

[0137] In this embodiment, specific behavior sequences (such as operating curtains, starting air conditioners, adjusting lights) are input into the RNN model, and the model captures the temporal relationship between user behaviors through the state memory (hidden state) of multiple time steps. For example, if a user usually completes the behaviors of "opening the curtain → adjusting the air conditioner temperature → starting the coffee machine" in the morning in sequence, the RNN establishes the time-dependent relationship between these behaviors through the iteratively updated hidden state. In addition, in order to enhance the adaptability of the model to the suddenness and periodicity of behaviors, the sub-layer introduces an attention mechanism to dynamically allocate importance weights to different time steps, thereby more accurately identifying the temporal logic between key behaviors.

[0138] Specifically, the recurrent neural network can mine the time sequence characteristics of user behaviors and combine the attention mechanism to weight process multi-dimensional behavior features, effectively solving the problem that traditional rule methods are difficult to capture complex time-dependent relationships. For example, when the user has a behavior operation mutation (such as temporarily opening the curtain or adjusting the light) in the "evening movie mode", the system can determine whether these behaviors conform to the regular mode according to the time-dependent relationship graph, and give corresponding adjustment suggestions, thereby improving the intelligent degree of the system.

[0139] a rule identification sub-layer for extracting periodic characteristics of behaviors through frequency spectrum analysis.

[0140] In the embodiment, the behavior data of the user is mapped to a frequency domain, a high-frequency operation time period and a long-term behavior rule of each day are identified by decomposing the behavior frequency components through a fast Fourier transform (FFT), and a periodic rule is identified. For example, if a user has a high-frequency behavior mode (such as opening a curtain and starting a coffee machine) from 6:30 to 7:00 in the morning on weekdays, and the time is delayed to 7:30 to 8:00 on weekends, the sub-layer can accurately identify this periodic rule. The identification result is further quantized as a behavior feature label (such as “morning mode on weekdays” and “morning mode on weekends”), which provides a direct input for subsequent behavior mode analysis and scene control.

[0141] Specifically, the spectrum analysis can extract the implicit periodic characteristics from the long-term user behavior data, and especially distinguish the behavior mode changes in different situations such as weekends and weekdays; the generation of the behavior rule label helps the system to predict the operation demand of the user in a future time period, so as to realize early response. For example, after identifying the rule that the user “turns off the light at 22:00 every night”, the system can automatically optimize and adjust the related devices before this time, which significantly improves the user experience.

[0142] The intelligent home device control strategy generated in combination with the real-time data includes:

[0143] The real-time data is matched with the life behavior mode of the user to determine whether the current scene conforms to the existing behavior mode;

[0144] A scene operation suggestion is generated according to the matching result, and the intelligent home device and the operation sequence thereof to be linked are determined;

[0145] A control instruction is generated, and the control instruction is sent to the corresponding intelligent home device.

[0146] Embodiment 2:

[0147] Please refer to Figure 4 The present application provides an embodiment: a big data life habit and behavior analysis system for smart home, the system includes a data acquisition module, a data modeling module, a behavior mode management module and a control strategy generation module;

[0148] The data acquisition module is used for real-time acquisition of user behavior data, environment data and device state data, and pre-processing;

[0149] The data modeling module is used for unified modeling of the collected multi-dimensional data, generation of a correlation matrix and extraction of behavior characteristics, identification of the life behavior mode and periodic rule of the user;

[0150] The behavior mode management module is used for outputting the life behavior mode of the user;

[0151] The control strategy generation module is configured to generate a scenario operation suggestion and a specific device control instruction according to a matching result of the real-time data and the generated life behavior mode.

[0152] The data modeling module comprises:

[0153] The correlation modeling unit is configured to uniformly model the preprocessed multi-dimensional data, generate a correlation matrix, and capture an interaction relationship between user behaviors and time, environment, and device states.

[0154] The feature extraction unit is configured to extract behavior features of the user from the correlation matrix, including time regularity features, device linkage features, and environment adaptation features.

[0155] Although the embodiments of the present application have been shown and described above, it should be understood that the above-described embodiments are exemplary and are not to be construed as limiting the present application, and those ordinarily skilled in the art can make changes, modifications, replacements, and variations to the above-described embodiments within the scope of the present application.

Claims

1. A big data life habit and behavior analysis method for smart home, characterized in that, The big data life habit and behavior analysis method comprises: Collecting multi-dimensional data of a user using a smart home; Processing the multi-dimensional data to output a behavior preference label; Inputting the behavior preference label into a preset behavior analysis model to output a user's life behavior mode through the behavior analysis model; Generating a smart home device control strategy according to the life behavior mode in combination with real-time data; The behavior analysis model comprises: A feature input layer for receiving a user behavior feature vector generated by the behavior preference label; A time series modeling layer for capturing time dependence and periodicity of behavior according to time series data in the user behavior feature vector to output a user behavior trend; A behavior mode analysis layer for obtaining a user's scenario-based behavior mode according to the output of the time series modeling layer and the behavior preference label; A behavior mode output layer for enhancing and classifying the scenario-based behavior mode through an adjacency matrix and a mask and outputting a life behavior mode through semantic conversion; The time series modeling layer comprises: A time window division sub-layer for dividing time series data in the user behavior feature vector according to a dynamic time window to extract key behaviors in a time period, wherein the dynamic time window is calculated according to volatility of the user behavior feature vector; A time dependence analysis sub-layer for generating a time dependence relationship graph through a recurrent neural network; A rule identification sub-layer for extracting periodicity features of behavior through frequency spectrum analysis. 2.The big data life habit and behavior analysis method for smart home of claim 1, wherein, The processing of the multi-dimensional data comprises: Uniformly modeling the multi-dimensional data to generate a correlation matrix; Performing mode analysis on the correlation matrix to extract behavior features; Generating a behavior preference label through a clustering algorithm according to the behavior features. 3.The big data life habit and behavior analysis method for smart home of claim 2, wherein, The uniformly modeling of the multi-dimensional data comprises: Constructing a time-environment bidirectional index structure; Analyzing an interaction relationship of the time-environment bidirectional index structure according to a predefined correlation rule to obtain a behavior correlation matrix; Updating the behavior correlation matrix through an attention mechanism to generate a correlation matrix. 4.The big data life habit and behavior analysis method for smart home of claim 2, wherein, The mode analysis of the correlation matrix comprises: Analyzing data distribution in the correlation matrix through a mode mining algorithm to extract time regularity mode, device linkage mode and environment adaptation mode of user behavior; Reducing dimensions of the regularity mode, device linkage mode and environment adaptation mode through feature embedding technology to generate behavior features. 5.The big data life habit and behavior analysis method for smart home of claim 2, wherein, The generation of the behavior preference label through the clustering algorithm comprises: Processing the behavior features according to DBSCAN and K-means clustering algorithms to generate clustering results according to similarity measurement of user behavior features; Generating a behavior preference label for each clustering result. 6.The big data life habit and behavior analysis method for smart home of claim 1, wherein, The generation of the smart home device control strategy in combination with real-time data comprises: Matching real-time data with a user's life behavior mode to determine whether a current scene conforms to an existing behavior mode; Generating a scenario-based operation suggestion according to a matching result to determine a linked smart home device and an operation sequence thereof; Generating a control instruction and sending the control instruction to a corresponding smart home device.

7. A big data life habit and behavior analysis system for smart home for implementing the big data life habit and behavior analysis method for smart home according to any one of claims 1 to 6, characterized in that, The system comprises a data acquisition module, a data modeling module, a behavior pattern management module and a control strategy generation module; The data acquisition module is configured to acquire user behavior data, environment data and device state data in real time and perform preprocessing; The data modeling module is configured to uniformly model the acquired multi-dimensional data, generate a correlation matrix and extract behavior characteristics, and identify the user's life behavior pattern and periodicity; The behavior pattern management module is configured to output the user's life behavior pattern; The control strategy generation module is configured to generate scenario-based operation suggestions and device control instructions according to the matching result of real-time data and the generated life behavior pattern. 8.The big data lifestyle and behavior analytics system for smart home of claim 7, wherein, The data modeling module comprises: An association modeling unit configured to uniformly model the preprocessed multi-dimensional data, generate a correlation matrix, and capture the interaction relationship between user behavior and time, environment and device state; A feature extraction unit configured to extract the user's behavior characteristics from the correlation matrix, including time regularity characteristics, device linkage characteristics and environment adaptation characteristics.

Images