Smart home device control method and system based on smart door lock

Abstract

The application belongs to the technical field of smart home, and particularly relates to a smart home device control method and system based on a smart door lock, which comprises the following steps: obtaining a plurality of historical data slices containing a leading event sequence; constructing a user behavior graph according to the historical data, and calculating a mode coincidence degree of a current behavior; meanwhile, learning a user environment preference according to the historical data, constructing a probability density model to calculate an imbalance degree of a current environment; finally, fusing the behavior mode coincidence degree and the environment imbalance degree into a decision factor through an adaptive nonlinear function, and the decision factor is used for dynamic matching in a historical successful scene library and driving the final device control. The application realizes adaptive, personalized and continuously optimized smart home control, and can improve user experience.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart home. More particularly, the present application relates to a smart home device control method and system based on a smart door lock. BACKGROUND

[0002] In a modern smart home system, a smart door lock is not only the first line of defense for home security, but also a key node for user interaction with the home environment. Triggering corresponding scene linkage through the unlocking and locking behaviors of the smart door lock, such as automatically turning on the lights, air conditioner and curtains in the home mode, has become an important way to improve the living experience.

[0003] In order to realize such scene linkage, the existing technology mostly adopts a control strategy based on fixed rules, such as pre-setting that if the door is unlocked within a certain time period, then the instructions of a certain device combination are executed. The advantage of this method is simple and intuitive, but its core defect is non-personalization. Therefore, the existing technology also introduces machine learning algorithms to analyze the user's historical behavior and assist in personalized automatic control of home devices. In related technologies, for example, the Chinese patent document with the authorization announcement number CN114237072B discloses a smart home door lock integrated control system and method applying cloud control technology, which discloses that the indoor environment is monitored and analyzed, and then the indoor environment is predicted, and the control state of the door lock is judged in advance according to the prediction result, and the corresponding judgment result is executed.

[0004] However, the existing technology mainly has the following limitations when implementing the above control: the rules are static and cannot adapt to the changing habits or atypical behaviors of users, resulting in low accuracy of scene execution, and as the number of devices increases, the configuration and maintenance of the rules become extremely complex; in addition, it is unable to deeply mine the complex and nonlinear behavior patterns of users under different time and environmental contexts, and it does not have the ability to self-optimize according to subsequent user feedback. SUMMARY

[0005] To solve the above technical problems of insufficient personalization and poor control effect of smart home device control, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present application provides a smart home device control method based on a smart door lock, comprising:

[0007] Obtain a plurality of historical data slices containing a sequence of leading events; construct a user behavior graph based on the event types in the historical data slices, the user behavior graph containing a plurality of event nodes and directed edges connecting the event nodes; obtain a current event sequence of a current data slice, calculate a behavior pattern fit degree of the current event sequence, the behavior pattern fit degree being positively correlated with an average edge weight of a corresponding path in the user behavior graph and an average centrality of a corresponding node; construct a preference probability density model for each internal environment dimension based on an internal environment vector in the historical data slices and a user manual adjustment record; obtain a current internal environment vector of the current data slice, calculate an imbalance degree of the current environment, the imbalance degree being positively correlated with a weighted sum of negative log likelihoods of each dimension of the current internal environment vector under the corresponding preference probability density model, the weight of the weighted sum being an adjustment frequency of each internal environment dimension in the user history; multiply the behavior pattern fit degree and the environment imbalance degree to obtain a decision factor, and select and execute a control instruction from a historical successful scene library based on the decision factor.

[0008] The application understands the dynamic changing behavior habits of the user by constructing a user behavior graph, and learns the individualized environment comfort of the user by constructing a preference probability density model, multiplies the behavior pattern fit degree representing the intention and the environment imbalance degree representing the demand to obtain a decision factor, and realizes that the scene is triggered only when the user behavior conforms to the habit and the environment really needs adjustment; the double condition judgment mechanism greatly improves the accuracy of scene triggering, effectively avoids the misoperation caused by atypical behavior, and realizes truly adaptive and individualized intelligent control.

[0009] Preferably, the weight of the directed edge is obtained, including:

[0010] Obtaining a directed edge connecting the event nodes and , obtaining a set of event transitions from to in the historical data slices, denoted as ;

[0011] Multiplying a difference between a timestamp of the current data slice and a timestamp corresponding to a historical data slice in which the event transition occurs for the i-th time in by a time decay constant to obtain a freshness of the event transition occurring for the i-th time in ; and accumulating the freshness of all the event transitions occurring in the history to obtain the weight of the directed edge from the event node to .

[0012] ​​​The application introduces a freshness concept based on timestamp difference and time decay constant, and accumulates so that recent and more frequent behavior conversion obtains higher weight in the graph, and can dynamically reflect the evolution process of user habits; therefore, the system can more accurately identify the current core behavior mode of the user, rather than being misled by old habits that have long been outdated, thereby improving the accuracy of the behavior mode fitting degree.

[0013] Preferably, the time decay constant is positively correlated with the inverse of the average value of the interval length of adjacent timestamps in all historical data slices.

[0014] Preferably, the centrality of the node is calculated by a graph centrality algorithm, and the graph centrality algorithm is a PageRank algorithm.

[0015] Preferably, the preference probability density model is a KDE model based on kernel density estimation.

[0016] Preferably, the behavior mode fitting degree is multiplied by the environmental imbalance degree to obtain a decision factor, including: performing first nonlinear mapping on the behavior mode fitting degree to obtain behavior intention potential energy; performing second nonlinear mapping on the environmental imbalance degree to obtain environmental demand potential energy; and multiplying the behavior intention potential energy and the environmental demand potential energy to obtain the decision factor.

[0017] The application first converts the behavior fitting degree and the environmental imbalance degree into behavior intention potential energy and environmental demand potential energy through nonlinear mapping, and then multiplies them, realizes the logic that only when the potential energy of both the behavior intention and the environmental demand is high, the final decision factor will significantly increase; ensures the prudence of system decision, avoids triggering scene linkage when there is only weak signal in any dimension, and makes the decision process more robust.

[0018] Preferably, the function curve shape of the first nonlinear mapping is determined by the mean and variance of all historical behavior mode fitting degrees; and the function curve shape of the second nonlinear mapping is determined by the mean and variance of all historical environmental imbalance degrees.

[0019] The application optimizes the nonlinear mapping function, so that its shape is dynamically determined by the historical mean and variance of the corresponding input indicators, i.e., behavior fitting degree or environmental imbalance degree, so that the judgment threshold is no longer fixed, but is dynamically adjusted according to the unique data distribution of each user, improving the sensitivity and adaptability of the system to individual user state changes.

[0020] Preferably, the internal environment vector contains at least the following two items: indoor temperature; indoor air quality index; and human existence state perceived by millimeter wave radar.

[0021] Preferably, further comprising: monitoring whether a manual intervention behavior of the user exists within a preset time window after executing the control instruction; if the manual intervention behavior exists, storing a result of the manual intervention behavior as a new successful scenario in the historical successful scenario library, and triggering online updating of the user behavior graph and the preference probability density model.

[0022] The present application regards the manual intervention behavior of the user after the system automatically executes the instruction as a feedback signal, and such intervention is not only stored as a new successful scenario, but also triggers online updating of the behavior graph and the preference model, so that the system has the ability of continuous learning and self-correction, can learn from errors, and its decision will be more and more in line with the real needs of the user with the growth of the use time.

[0023] In a second aspect, the present application provides a smart home device control system based on an intelligent door lock, comprising a processor and a memory, and the memory stores computer program instructions, which realize the above-mentioned smart home device control method based on an intelligent door lock when executed by the processor.

[0024] By adopting the above technical solution, the above-mentioned smart home device control method based on an intelligent door lock is generated into a computer program and stored in the memory to be loaded and executed by the processor, so that a terminal device is made according to the memory and the processor, and convenient use is achieved.

[0025] The present application has the following beneficial effects:

[0026] (1) The present application can accurately capture the unique comfortable interval of each user for environmental factors such as temperature and brightness by constructing a preference probability density model based on user historical manual adjustment for each environmental dimension, and quantifies the sensitivity of the user to different dimensions, so that the device control is no longer to execute general instructions, but to meet the personalized comfort needs of specific users as the goal, which significantly improves the living experience;

[0027] (2) The user behavior graph constructed by the present application can dynamically reflect the evolution of user habits by introducing a weight mechanism based on time decay, and give higher weight to recent behaviors; combined with the graph centrality algorithm to evaluate the importance of events, so that the behavior pattern consistency can accurately judge whether the current behavior is in line with the recent and core life rules of the user, so as to effectively distinguish different intentions such as regular homecoming and temporary outing, and avoid false triggering. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 is a flow chart schematically showing a smart home device control method based on an intelligent door lock in the present application;

[0029] Figure 2is a schematic diagram of a preferred probability density distribution model of air quality index;

[0030] Figure 3 is a schematic diagram of a preferred probability density distribution model of temperature. DETAILED DESCRIPTION

[0031] The embodiment of the application discloses a smart home device control method based on a smart door lock, referring to Figure 1 , comprising steps S1-S4:

[0032] S1: acquiring a plurality of historical data slices of the smart door lock, the data slices containing interaction data, internal environment vectors, external environment vectors, and user state data.

[0033] It should be noted that, in order to optimize the control effect of the smart home devices controllable through the smart door lock, the interaction data of the door lock and the behaviors of the user related to the control of the home devices, such as the behaviors with explicit intentions such as coming home or leaving home, should be fully acquired, which are not isolated instantaneous actions but a process composed of a series of events with time sequence correlation, for example, coming home may start with the mobile phone entering the home geographic enclosure, and then the door lock is opened. Therefore, it is not enough to only capture the state snapshot of the door lock interaction moment, and the leading event sequence with time sequence relationship causing the interaction should also be captured. Therefore, the application uniformly manages all related events recorded by the system, and constructs a more structured historical data slice with the interaction event of the smart door lock as an anchor point.

[0034] Specifically, an event log is established in the door lock system, and all events generated from the smart door lock since the smart door lock is used are continuously recorded, the sources include the smart door lock, various sensors, and user personal devices, and a time stamp is determined for each event to form an event log sorted by time.

[0035] The interaction data of the door lock interaction event is acquired, the interaction data containing a time stamp, a user identity, an interaction type, and an authentication method, the interaction type at least including unlocking, locking, and relocking, and the authentication method at least including fingerprint, password, and NFC.

[0036] The internal environment vector corresponding to the time stamp of the door lock interaction event is acquired, the data source of the internal environment being the smart home devices in data transmission with the smart door lock, the internal environment being, for example, temperature, air quality index, and human existence state sensed by a millimeter wave radar, the human existence state using 0 to represent no one, 1 to represent stillness, and 2 to represent activity, and all the internal environments constituting the internal environment vector.

[0037] The timestamp of the door lock interaction event corresponds to the external environment vector. The data source of the external environment is the third-party meteorological and environmental service data obtained in real time through network API based on the home geographic location information. The external environment includes, for example, weather type, outdoor temperature, ultraviolet index, and outdoor air quality index.

[0038] The internal environment vector and the external environment vector jointly constitute a scenario snapshot corresponding to the door lock interaction event.

[0039] User state data is obtained. The data source of the user state data is information obtained from the user's smart device or cloud service after authorization by the user, including phone geofencing state and event tags overlapping with the timestamp of the interaction data. The phone geofencing state includes, for example, entering the home area and being in the home area. The event tags include, for example, work and vacation.

[0040] The events in a preset time window on the left side of the timestamp of each door lock interaction data are obtained, and a leading event sequence of an interaction event is constructed in chronological order. For example, the preset time window is 10 minutes.

[0041] The timestamp of the historical data slice and the corresponding door lock interaction data, scenario snapshot, and leading event sequence jointly constitute a complete historical data slice.

[0042] At this point, multiple historical data slices of the smart door lock are obtained. The historical data slice contains a timestamp, door lock interaction data, a scenario snapshot, and a leading event sequence. The set of historical data slices is denoted as wherein is the current data slice, is the Nth historical data slice.

[0043] S2: According to the interaction data in the historical data slice, a user behavior graph is constructed. Based on the graph and the current data slice, the behavior pattern consistency degree is obtained.

[0044] It should be noted that the user's home life is not a random combination of a series of isolated events, but there is internal logic and regularity, for example, the user goes home after work, which will be accompanied by a series of events with a chronological order such as the geographic fence state of the mobile phone changing to entering the home area, the interaction type being unlocking, etc. These event sequences collectively constitute the user's behavior pattern. Therefore, the present application constructs a weighted directed graph by all door lock interaction data and the leading event sequence in the historical data slice, and expresses the implicit behavior pattern in a structured form. The user's behavior is changing, so the behavior pattern is also dynamically changing. The recent behavior pattern is more representative of the user's current habits than the long-term behavior pattern. The higher the frequency of the behavior sequence, the higher the habit strength. Therefore, the present application introduces a time decay mechanism to the weight of the edge connecting the events when constructing the user behavior graph, so that the weight of the edge can reflect both the frequency of the sequence and the recency.

[0045] It should be further explained that evaluating whether the current behavior conforms to the habit needs to be considered from two dimensions. One is whether the current behavior sequence matches the high-weight path in the user behavior graph, that is, the strong habit path; the other is whether the events constituting the behavior sequence are the core and hub of the user's daily activities, which reflects the node importance of the behavior. A stable habit usually manifests as both a fixed behavior path and a path composed of some high-frequency and core event nodes. Therefore, the present application sets up a behavior pattern coincidence degree to quantitatively evaluate the compliance of the current behavior by comprehensively considering the path strength and node importance.

[0046] Specifically, the event types contained in all door lock interaction data and leading event sequences in the historical data slice are taken as nodes, and the chronological order of event occurrence is taken as directed edges to construct a user behavior graph: initialize an empty graph ; traverse each historical data slice in the historical data slice set, and form a complete event chain in chronological order by the events contained in the leading event sequence and the door lock interaction data; for each event type in the event chain, check whether it exists as a node in . If not, create the node in ; for each pair of adjacent events in the event chain, create a directed edge between the corresponding two nodes in . If the edge already exists, add one to the count of its occurrence frequency; obtain the user behavior graph .

[0047] Preferably, the weight of the edge connecting two event nodes in the user behavior graph is obtained according to the frequency and recency of the events in the historical data slice, including:

[0048] The weight of the edge connecting event nodes and directed edge from event node to event node . . .

[0049] The inverse of the average of the interval lengths of all adjacent timestamps in the set of historical data slices is denoted as the time decay constant. It is noted that the time decay constant is dynamically determined by the statistical characteristics of the sequence of timestamp intervals of all historical data slices of the user, and can be used as a coefficient to adjust the difference in edge weights.

[0050] The weight of the directed edge from event node to event node

[0051] satisfies the expression: .

[0052] In the expression, denotes the weight of the directed edge from event node to event node ; denotes the timestamp corresponding to the historical data slice in which the event transition occurs for the th time; denotes the timestamp of the current data slice; denotes the time decay constant. In the expression,

[0053] denotes the freshness of the event transition occurring for the th time in , which decreases exponentially with the increase of the time difference , indicating that the contribution of events that are more distant in time to the current habit is smaller; denotes the accumulation of the freshness of all occurrences of the event transition in history, and the larger the value, the more frequent and closer to the current the transition from to is, and the higher the corresponding habit strength is, and the larger the weight of the edge is.

[0054] Preferably, the behavior pattern consistency is obtained according to the path strength and node importance of the current data slice and the adjacent historical data slices in the user behavior graph, including:

[0055] The current event sequence is constructed in chronological order from the leading event sequence of the current data slice and the lock interaction data.

[0056] It should be noted that not all events in the user's behavior pattern have equal importance, and the unlocking event, as the pivot connecting the state inside and outside the home, is more important than the light-on event, and a graph centrality algorithm such as PageRank can calculate the influence of a node in the network, and a high centrality score of a node means that there are many important behavior paths pointing to it or from it.

[0057] The average importance of all nodes in the current event sequence in the user behavior graph is calculated using a graph centrality algorithm: run the graph centrality algorithm on all nodes in the user behavior graph to calculate the centrality score for each node; and The centrality scores of the nodes in the user behavior graph are calculated by taking the arithmetic mean of the centrality scores of the nodes in the user behavior graph, and the average importance of all nodes in the current event sequence in the user behavior graph is obtained.

[0058] It should be noted that the behavior structure of different users is different, and some users' behavior is highly modularized, for example, cooking-related events form a tight community with little intersection with audio-visual entertainment communities; the behavior of some users is more linear and programmatic in long chains; the average clustering coefficient is an indicator of the degree of community or modularity of the entire graph; by analyzing the overall structure of user behavior, the evaluation strategy is dynamically adjusted to determine whether to place more emphasis on the fixity of the path or the criticality of the node.

[0059] The local clustering coefficient of each node in the user behavior graph is calculated, and then the local clustering coefficients of all nodes are calculated by taking the arithmetic mean, and the average clustering coefficient of the user behavior graph is obtained, denoted as the balance factor.

[0060] The behavior pattern fit degree satisfies the expression:

[0061] ;

[0062] In the formula, represents the behavior pattern fit degree; represents the average edge weight of the corresponding path in the user behavior graph of the current event sequence; represents the average importance of all nodes in the user behavior graph in the current event sequence; represents the balance factor; represents the normalization function.

[0063] In the formula, represents the path strength of the corresponding path in the user behavior graph of the current event sequence, represents the overall importance of the nodes in the current event sequence, It represents the weighted fusion of the path strength and node importance, and the greater the value, the more the current event sequence conforms to a strong habit path composed of important events.

[0064] At this point, the behavior pattern coincidence degree is obtained.

[0065] S3: According to the context snapshot of the historical data slice and the user manual adjustment record, the environment imbalance degree is obtained.

[0066] It should be noted that the device linkage of the smart home should not only be triggered by the user's behavior, but also consider the actual needs of the environment. Only when the environment state deviates from the user's comfort zone, that is, the environment is in an imbalance state, is it meaningful to control the device and can improve the user experience. In addition, the user's comfort zone is not a fixed value. For example, the user may feel most comfortable at 23°C, but can also accept 22°C or 24°C, and will feel obviously uncomfortable at 18°C or 28°C. This preference feature can be accurately described by a probability distribution. The center of the distribution is the user's most preferred point, and gradually decays to both sides. Therefore, the present application calculates the imbalance degree of the environment state according to the probability of the environment state in the user's historical preference probability distribution. The lower the probability, the more the environment state deviates from the comfort zone, and the higher the imbalance degree.

[0067] It should be further noted that the sensitivity of the user to different environmental dimensions varies from individual to individual. For example, some people are very sensitive to temperature changes and need to frequently adjust the air conditioner, while others are more sensitive to light changes and need to adjust the light according to different indoor brightness. This sensitivity difference is directly reflected in the user's historical manual intervention frequency of various devices. Therefore, when calculating the overall environment imbalance degree, the deviation degree of different internal environment dimensions should be weighted, and the weight should be proportional to the user's historical sensitivity to the internal environment dimension. Therefore, the present application sets up the environment imbalance degree, which is obtained by weighting and integrating the deviation degree of each internal environment dimension and the corresponding preference probability distribution to obtain the overall improvement degree of the current environment.

[0068] Specifically, for each internal environment dimension, a preference probability density distribution model is learned and constructed according to the internal environment vector in the historical data slice and the associated user manual adjustment record: the historical data slice set is screened, and if no manual device adjustment event related to the environment occurs within a preset time range after the timestamp of any historical data slice, the historical data slice is marked as a class slice; for any internal environment dimension, the value of the internal environment dimension in the context snapshot of all class slices is extracted to form a class data set of the internal environment dimension.

[0069] The KDE method is a kernel density estimation method, which is a prior art and will not be described here. As shown in FIG. 2, the KDE method is used to process the internal environment dimension data set to generate a preference probability density distribution model of the internal environment dimension. Figure 2 FIG. 2 is a schematic diagram of a preference probability density distribution model of an air quality index. Figure 3 FIG. 3 is a schematic diagram of a preference probability density distribution model of temperature.

[0070] Preferably, the environmental imbalance degree is obtained according to the deviation of the current value of each internal environment dimension from the preference probability density distribution model thereof and the sensitivity of the user to each dimension, and includes:

[0071] The current value of the i-th dimension of the internal environment vector in the current data slice is obtained. The number of times that the user adjusts the device in the past is divided by the total number of adjustments of all devices to obtain the user dimension sensitivity weight.

[0072] The number of times that the user adjusts the device in the past is divided by the total number of adjustments of all devices to obtain the user dimension sensitivity weight.

[0073] The environmental imbalance degree satisfies the expression:

[0074]

[0075] In the expression, the environmental imbalance degree is represented by D, the total number of internal environment dimensions is represented by N, the probability density value of the current value of the i-th internal environment dimension under the preference probability density distribution model thereof is represented by f(x i), the user dimension sensitivity weight is represented by w i, the normalization function is represented by φ, and the minimum value is represented by min.

[0076] In the expression, the negative log-likelihood of the current value is represented by L(x i), which is the deviation of the dimension. If x i is at the center of the user comfort zone, the value of L(x i) is large, and the deviation value is small. Conversely, if x i is far away from the comfort zone, the value of L(x i) is small, and the deviation value is large.

[0077] The weighted deviation of the dimension is represented by w i L(x i), and the total deviation degree is obtained by accumulating the weighted deviations of all environment dimensions. The larger the value is, the more the overall environment deviates from the user's comfort zone, and the higher the imbalance degree is. ​​​​​​​​​​​​​​​​​​​

[0078] At this point, the environmental imbalance degree is obtained.

[0079] S4: Based on the behavior pattern coincidence degree and the environmental imbalance degree, a scene activation potential is generated, and scene matching and decision execution are performed.

[0080] It should be noted that an ideal smart home device control decision should be triggered when both the user's behavior conforms to their inherent habits and the home environment indeed needs to be adjusted, and the historical data distribution of the user's behavior habits and environmental imbalance degree is different for different people, for example, the historical value of the behavior pattern coincidence degree of a user with a regular life is always a large value, at this time, a small decline should be considered as an exception, and the value of the behavior pattern coincidence degree of a user with a random life may fluctuate greatly. Therefore, a fixed mapping function cannot adapt to such individual differences.

[0081] It should be further noted that in order to achieve more intelligent fusion, the present application designs two adaptive nonlinear mapping functions, which are used to process the behavior pattern coincidence degree and the environmental imbalance degree respectively before they are fused. The shape of the two nonlinear mapping functions is dynamically determined by the distribution of the values calculated from the historical data slices of the corresponding input indicators, which enables the mapping process to automatically adapt to the individual characteristics of each user. Thus, the present application sets up a decision factor as the final decision basis for triggering scene linkage.

[0082] Specifically, according to the distribution of the behavior pattern coincidence degree and the environmental imbalance degree in the historical data slices, the decision factor is obtained, including:

[0083] The decision factor satisfies the expression:

[0084] ;

[0085] In the formula, represents the scene activation potential; represents the behavior pattern coincidence degree; represents the environmental imbalance degree; , is a nonlinear mapping function corresponding to the behavior pattern coincidence degree and the environmental imbalance degree, which is usually a S-shaped function, and the nonlinear mapping function is a Gaussian function.

[0086] In the formula, represents the potential of the behavior intention, represents the potential of the environmental demand, and the product of the two realizes the logical relationship, that is, only when the potential of the behavior intention and the potential of the environmental demand are both high, the final will be high, thereby effectively triggering the subsequent scene.

[0087] It is to be noted that, for example, the key parameters of its mapping function, such as the center point and shape parameter, are dynamically determined by the historical mean and historical variance of C calculated from all historical data slices, if the historical values of are generally high and concentrated, the curve of will be relatively steep, so that a small numerical decline will lead to a sharp decrease in output potential.

[0088] Preferably, according to the decision factor, the scene matching and decision execution include:

[0089] Take as a dynamic search threshold, search all historical successful scenes in the historical successful scene library whose situational snapshots have a similarity greater than with the situational snapshot of the current data slice, forming a candidate scene set.

[0090] If the candidate scene set is not empty, select the scene most similar to the current situation and execute its corresponding device control instruction set. If the candidate scene set is empty, do not perform any operation.

[0091] After executing the instructions, monitor whether the user performs manual intervention within a preset time window. If intervention occurs, take the user's operation result as a new successful scene sample and store it in , and trigger online update of the user behavior graph of S2 and the preference probability density model of S3.

[0092] At this point, the intelligent home device control based on the intelligent door lock is completed.

[0093] The embodiment of the application also discloses an intelligent home device control system based on an intelligent door lock, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the intelligent home device control method based on the intelligent door lock is realized.

[0094] The above system also comprises a communication bus and a communication interface and other components familiar to those skilled in the art, the settings and functions of which are known in the art, so they will not be described here.

[0095] Although the present specification has shown and described multiple embodiments of the present application, it will be apparent to those skilled in the art that such embodiments are provided only by way of example. Those skilled in the art will think of many changes, changes and alternatives without departing from the idea and spirit of the present application. It should be understood that various alternatives to the embodiments of the application described herein can be employed in practicing the present application.​

Claims

1. A method for controlling smart home devices based on a smart door lock, characterized in that, include: Obtain multiple historical data slices containing a sequence of preceding events; A user behavior graph is constructed based on the event types in the historical data slices. The user behavior graph contains multiple event nodes and directed edges connecting the event nodes. Obtain the current event sequence of the current data slice, and calculate the behavioral pattern matching degree of the current event sequence. The behavioral pattern matching degree satisfies the expression: ； Indicates the degree of consistency in behavioral patterns; This represents the average edge weight of the corresponding path in the user behavior graph for the current event sequence; This represents the average importance of all nodes in the current event sequence within the user behavior graph; Indicates the balance factor; Represents the normalization function; Based on the internal environment vectors in the historical data slices and the user's manual adjustment records, a preference probability density model is constructed for each internal environment dimension; the current internal environment vector of the current data slice is obtained, and the imbalance of the current environment is calculated. The imbalance is positively correlated with the weighted sum of the negative log-likelihoods of the values ​​of each dimension of the current internal environment vector under the corresponding preference probability density model. The weight of the weighted sum is the frequency of the user's historical adjustment to each internal environment dimension. The decision factor is obtained by multiplying the behavioral pattern fit degree by the environmental imbalance degree, and control instructions are selected and executed from the historical successful scenario library based on the decision factor.

2. The smart home device control method based on a smart door lock according to claim 1, characterized in that, The weights of the directed edges satisfy: ; Get connection event nodes and Directed edge The interaction type in the historical data slice is obtained by arrive The set of event transitions, denoted as ; Indicates from the event node arrive The weight of the directed edge; express The Middle The timestamp corresponding to the historical data slice that the event transformation occurs for the first time; Indicates the timestamp of the current data slice; This represents the time decay constant.

3. The smart home device control method based on a smart door lock according to claim 2, characterized in that, The time decay constant is positively correlated with the reciprocal of the average interval between adjacent timestamps in all historical data slices.

4. The smart home device control method based on a smart door lock according to claim 1, characterized in that, The importance of the node is calculated using a graph centrality algorithm, specifically the PageRank algorithm.

5. A smart home device control method based on a smart door lock according to claim 1, characterized in that, The preference probability density model is a KDE model based on kernel density estimation.

6. A smart home device control method based on a smart door lock according to claim 1, characterized in that, The decision factor is obtained by multiplying the behavioral pattern fit degree by the environmental imbalance degree, including: performing a first nonlinear mapping on the behavioral pattern fit degree to obtain behavioral intention potential energy; performing a second nonlinear mapping on the environmental imbalance degree to obtain environmental demand potential energy; and multiplying the behavioral intention potential energy by the environmental demand potential energy to obtain the decision factor.

7. A smart home device control method based on a smart door lock according to claim 6, characterized in that, The shape of the function curve of the first nonlinear mapping is determined by the mean and variance of the fit of all historical behavior patterns; the shape of the function curve of the second nonlinear mapping is determined by the mean and variance of the imbalance of all historical environments.

8. A smart home device control method based on a smart door lock according to claim 1, characterized in that, The internal environment vector includes at least the following two items: indoor temperature; indoor air quality index; and the presence of the human body as sensed by millimeter-wave radar.

9. A smart home device control method based on a smart door lock according to claim 1, characterized in that, Also includes: After executing the control command, monitor whether the user engages in any manual intervention within a preset time window; If manual intervention occurs, the result of the manual intervention is stored as a new successful scenario in the historical successful scenario library, and an online update of the user behavior graph and preference probability density model is triggered.

10. A smart home device control system based on a smart door lock, characterized in that, include: A processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement a smart home device control method based on a smart door lock according to any one of claims 1-9.

Images