User information management method based on big data of Internet of Things

Abstract

The invention relates to the technical field of Internet of Things data processing and user behavior analysis management, and discloses an Internet of Things big data-based user information management method, which comprises the following steps of: obtaining equipment information used by a user from a plurality of Internet of Things equipment; performing timestamp labeling and timestamp synchronization on the equipment information to form an event sequence containing an equipment identifier and a timestamp; analyzing a data interaction relationship between different Internet of Things devices in the event sequence through a time sequence association network algorithm, calculating time sequence similarity between events, constructing an association matrix, and forming graph structure representation; the behavior feedback of the user is analyzed from the graph structure representation, an instant reward is generated, and a personalized behavior management method is generated for the user through a Q value optimization strategy according to the current state of equipment use and the generated instant reward. The problems of data isomerism, time asynchronization and lack of deep behavior analysis and personalized management in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things data processing and user behavior analysis management, and in particular to a user information management method based on Internet of Things big data. Background Art

[0002] With the rapid development of the Internet of Things, the number of connected devices in the world has grown exponentially, covering all areas from smart home devices, wearable devices to industrial sensors and self-driving cars. These devices generate massive amounts of data, collectively referred to as IoT big data. How to effectively manage and analyze this data has become a hot topic in current research and application. However, the existing management methods for user usage information of IoT devices have obvious defects in many aspects.

[0003] First, IoT devices are diverse in types and functions, and their data formats and communication protocols are also different, which makes it difficult to integrate and process multi-source heterogeneous data. Devices produced by different manufacturers may use different communication standards and data formats, and the metadata, attributes and structures contained in the data are also different. This heterogeneity makes the standardization and unified management of data complicated, and it is difficult to build a comprehensive user information model. Existing methods often cannot effectively integrate data from different devices, resulting in one-sidedness and incompleteness of user behavior information, which hinders comprehensive and accurate user behavior analysis.

[0004] Secondly, inconsistent timestamps between devices is another problem that needs to be solved urgently. Since IoT devices may be distributed in different network environments, they are limited by device hardware performance, network latency, and different time synchronization mechanisms, resulting in deviations in the timestamps recorded by each device. This time synchronization problem will directly affect the accuracy of timing analysis and cannot correctly reconstruct the user's operation sequence and behavior trajectory on different devices. Traditional methods usually ignore the importance of time synchronization, or rely solely on simple time correction, which cannot meet the needs of high-precision timing analysis.

[0005] Thirdly, existing user behavior modeling methods are insufficient in mining the interaction relationship between users and multiple devices. Traditional user behavior analysis is mostly based on a single device or a single type of data, lacking a comprehensive understanding of cross-device and cross-scenario user behavior. There is a lack of effective algorithms to mine the interaction relationship between different devices and user behavior characteristics, and it is impossible to capture the comprehensive usage patterns of users on different devices. This leads to limited capabilities in building user portraits, personalized recommendations, and intelligent decision support, and it is impossible to provide users with accurate personalized services.

[0006] In addition, with the growth in the number of IoT devices and the scale of data, traditional data processing methods face challenges in efficiency and performance. Real-time processing and analysis of large-scale data requires efficient algorithms and powerful computing resources, but existing methods fail to fully consider the characteristics of IoT big data in algorithm design and system architecture, such as high frequency, low latency and data streaming processing requirements. As a result, data processing has high latency and large resource usage, making it difficult to meet real-time and efficiency requirements, affecting user experience and system responsiveness.

[0007] In summary, the existing management methods for user usage information of IoT devices have obvious shortcomings in terms of difficulty in data fusion, inaccurate time synchronization, insufficient user behavior modeling, low processing efficiency, and lack of security and privacy protection. In order to solve the above problems, a method is needed that can efficiently manage and analyze user usage information on multiple IoT devices. Summary of the invention

[0008] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract of the specification and the title of the invention of this application to avoid blurring the purpose of this section, the abstract of the specification and the title of the invention, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0009] In view of the above existing problems, the present invention is proposed.

[0010] In order to solve the above technical problems, the present invention provides the following technical solutions: obtaining device information used by a user from multiple IoT devices, wherein the device information includes user identity information and device usage behavior information, and each device is identified by a unique device identifier;

[0011] Performing timestamp marking and timestamp synchronization on the device information to form an event sequence including a device identifier and a timestamp;

[0012] The data interaction relationship between different IoT devices in the event sequence is analyzed by using a temporal association network algorithm, the temporal similarity between events is calculated, an association matrix is ​​constructed, and a graph structure representation is formed;

[0013] The user's behavioral feedback is analyzed from the graph structure representation to generate an instant reward, and a personalized management method is generated for the user through a Q-value optimization strategy based on the current state of device use and the generated instant reward.

[0014] As a preferred solution of the user information management method based on IoT big data described in the present invention, the IoT device used by the user is bound to the corresponding user ID, and the device information is collected through sensor technology and network communication, and the device status and user operation behavior are recorded; wherein:

[0015] The user identity information includes a user name and a user ID;

[0016] The device usage behavior information includes the device on / off status, usage frequency and operation mode;

[0017] The device identifier is a unique identifier for each device, which is assigned by the hardware manufacturer of the device or generated by a unique algorithm.

[0018] As a preferred solution of the user information management method based on Internet of Things big data described in the present invention, the device on / off status records the switch status of the device at different time points, the usage frequency records the number of times the device is used or the active time of the device in a certain period of time, and the operation mode records the interaction method between the user and the device.

[0019] As a preferred solution of the user information management method based on IoT big data described in the present invention, the collected device information is formed into an initial data set including device identifier, user identity information and device behavior information, and structured storage is performed. The mathematical expression of the initial data set is:

[0020] D i =(ID device ,{Data behavior ,Data user})

[0021] Among them, D i is the data of the ith device, ID device is the device identifier, Data behavior Data is the device usage behavior data. user The user's identity data.

[0022] As a preferred solution of the user information management method based on IoT big data of the present invention, the event sequence is obtained, including:

[0023] Synchronize the local time of each device to a unified standard time through the PTP protocol;

[0024] After adding the synchronized timestamp to each device data record, an event sequence including the device identifier, timestamp and device behavior data is formed. Each event record includes the device identifier, synchronized timestamp and device behavior information.

[0025] The behavior data of the device at least includes the device switch status, usage frequency and operation mode.

[0026] As a preferred solution of the user information management method based on IoT big data described in the present invention, the mathematical expression formula of the event sequence is:

[0027] E t ={(ID device ,timestamp t ,Data t )}

[0028] Among them, E t represents the event sequence at time t, ID device is the unique identifier of the device, timestamp t It is the synchronized timestamp, indicating the standard time when the device behavior occurs. t The device behavior data includes device behavior information.

[0029] As a preferred solution of the user information management method based on IoT big data described in the present invention, based on the timestamps and device behavior information between devices, the temporal similarity between two devices is calculated, and an event correlation matrix is ​​constructed. Each element of the event correlation matrix represents the similarity between two events. The higher the similarity between events, the stronger the correlation between the behaviors of the two devices.

[0030] The calculation formula of the time series similarity is:

[0031]

[0032] ΔT=|Timestamp i -Timestamp j |

[0033] Where ΔT represents the time difference between device i and device j, Adj i,j is the association matrix element between device i and device j, α is a hyperparameter that adjusts the influence of time difference, and Behavior Similarity is the behavior similarity, which is 1 if the behaviors are the same and 0 if the behaviors are different.

[0034] As a preferred solution of the user information management method based on IoT big data described in the present invention, a fusion analysis of device data is performed through graph convolution operation to construct a graph structure representation, wherein device identifiers are used as nodes of the graph, and temporal association relationships are used as edges of the graph to connect the nodes. The weight of each edge represents the temporal similarity between devices. The higher the similarity of each pair of devices, the greater the weight of the edge, indicating that the behaviors are more similar.

[0035] As a preferred solution of the user information management method based on IoT big data described in the present invention, the personalized management method includes device switch control, device mode switching and configuration adjustment of the user-bound device.

[0036] Beneficial effects of the present invention:

[0037] 1. It can accurately associate users with the devices they use, avoiding data inaccuracy caused by device confusion or data duplication, and achieving the beneficial effect of providing a high-quality and reliable data foundation for subsequent data analysis and user behavior modeling;

[0038] 2. It can accurately reflect the interaction sequence and timing relationship between devices, avoiding data analysis deviation caused by inconsistent device time, achieving the beneficial effect of improving the accuracy of timing analysis and ensuring data synchronization;

[0039] 3. The modeling of complex interactive relationships between multiple devices is realized, and the time series similarity is quantified into the form of matrix and graph, making the data relationship more intuitive and computable, achieving the beneficial effect of revealing hidden associations between devices and enhancing the depth of user behavior analysis;

[0040] 4. We have introduced reinforcement learning into user behavior management. By analyzing actual user feedback, we have continuously optimized the management strategy of the equipment, achieving the beneficial effects of improving user experience, optimizing resource utilization, and realizing personalized services. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0042] Figure 1 It is a flow chart of the user information management method based on the Internet of Things big data shown in the present invention;

[0043] Figure 2 This is a schematic diagram of comparing the accuracy of behavior recognition shown in the present invention;

[0044] Figure 3 It is a schematic diagram of energy consumption comparison shown in the present invention;

[0045] Figure 4 This is a schematic diagram for comparing data processing efficiency shown in the present invention. DETAILED DESCRIPTION

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0047] Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without making any creative work should fall within the scope of protection of the present invention.

[0048] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0049] According to an embodiment of the present invention, Figure 1 The flowchart shown is a user information management method based on IoT big data, which specifically includes the following steps:

[0050] S1. Obtain device information used by the user from multiple IoT devices, where the device information includes user identity information and device usage behavior information, and each device is identified by a unique device identifier; what needs to be explained in this step is:

[0051] Bind the IoT devices used by users to the corresponding user IDs, for example, user A uses device 1 and user B uses device 2;

[0052] The device identifier is a unique identifier for each device, which is assigned by the device's hardware manufacturer or generated by a unique algorithm. For example, the identifier of device 1 is 001 and the identifier of device 2 is 002;

[0053] User identity information includes user name and user ID;

[0054] As an example, user A's user name is "Alice", user ID = 001, user B's user name is "Bob", user ID = 002, and the binding relationship is: device 1 is bound to user A (ID 001), and device 2 is bound to user B (ID 002);

[0055] Collect device information through sensor technology and network communication, and record device status and user operation behavior;

[0056] As an example, the behavior records of device 1 (used by user A) and device 2 (used by user B) during a certain period of time are as follows:

[0057] Device 1 (user A) behavior record:

[0058] Device on / off status: On at 2024-11-01 08:00, Off at 2024-11-01 10:00, On at 2024-11-01 14:00;

[0059] Frequency of use: Device 1 was turned on 3 times on November 1, 2024;

[0060] Operation mode: Device 1 is turned on and off by user A by clicking a button;

[0061] Device 2 (user B) behavior record:

[0062] Device on / off status: On at 2024-11-01 07:30, Off at 2024-11-01 12:30;

[0063] Frequency of use: Device 2 was turned on twice on November 1, 2024;

[0064] Operation method: Device 2 is turned on and off by voice control;

[0065] The recorded device usage behavior information includes the device on / off status, usage frequency, and operation mode. The device on / off status records the on / off status of the device at different time points. The usage frequency records the number of times the device is used, or the length of time the device is active within a certain period of time. The operation mode records the way the user interacts with the device, such as button clicks, voice control, and automatic mode.

[0066] The collected device information is formed into an initial data set including device identifiers, user identity information, and device behavior information for structured storage;

[0067] As an example, each device information data record contains a device identifier (Device ID), user identity information (such as user ID, user name, device association information), and device behavior information (such as on / off status, usage frequency, and operation method). The mathematical expression of the initial data set is as follows:

[0068] D i =(ID device ,{Data behavior ,Data user})

[0069] Among them, D i is the data of the ith device, ID device is the device identifier, Data behavior Data is the device usage behavior data. user The user's identity data;

[0070] For example, the data of device 1 (user A) is:

[0071] D1=(ID device =001,{Data behavior ={state={2024-11-01 08:00:00:ON,2024-11-01 10:00:00:OFF,2024-11-01 14:00:00:ON},frequency=3,operation=buttonclick},Data user ={uesrID=001,username=Alice,deviceAssociation={device1}}})

[0072] Device 2 (User B) data:

[0073] D2=(ID device =002,{Data behavior ={state={2024-11-01 07:30:00:ON,2024-11-01 12:30:00:OFF},frequency=2,operation=voicecontrol},Datau ser ={userID=002,username=Bob,deviceAssociation={device2}}})

[0074] The above data structure is used to store and query the behavior information of users and devices. In this way, each record includes the device identifier, device behavior data and user identity information, forming a complete structured data set. Its mathematical expression formula is:

[0075] D={D1,D2,…,D n}

[0076] Among them, each D n They are all data records that contain device behavior data and user identity information, ensuring that the interaction between users and devices is recorded in detail to facilitate analysis of subsequent steps and the generation of personalized management strategies.

[0077] S2. Each device generates a local timestamp when collecting data, and synchronizes the time of all devices to a unified standard time through the PTP protocol. Devices adjust local time through synchronization message exchange, and add a synchronized timestamp to each device data record to form an event sequence containing device identifiers and timestamps. Among them, what needs to be explained in this step is:

[0078] When IoT devices collect data, each device generates a local timestamp, which indicates the specific time when the device collects the data. Since there is a deviation between the clocks of the devices, the embodiment of the present invention synchronizes the local time of each device to a unified standard time (e.g., UTC) through the PTP protocol (Precision Time Protocol);

[0079] For example, it is assumed that device 1, device 2, and device 3 collect data at different time points, and their local timestamps are as follows:

[0080] Device 1: local timestamp = 2024-11-01 08:00:00 (local time of device 1);

[0081] Device 2: local timestamp = 2024-11-01 08:00:10 (local time of device 2);

[0082] Device 3: local timestamp = 2024-11-01 08:00:20 (local time of device 3);

[0083] Through the PTP protocol, devices exchange synchronization messages and adjust their respective clocks so that the local time of all devices is synchronized to a unified standard time (for example, UTC time). The synchronized standard time is:

[0084] Standard time (UTC) = 2024-11-01 08:00:00

[0085] Device 1, Device 2, and Device 3 will adjust their local time according to the PTP protocol so that their timestamps are aligned as follows:

[0086] Device 1: synchronized timestamp = 2024-11-01 08:00:00 (UTC);

[0087] Device 2: synchronized timestamp = 2024-11-01 08:00:00 (UTC);

[0088] Device 3: synchronized timestamp = 2024-11-01 08:00:00 (UTC);

[0089] After adding the synchronized timestamp, each device data record forms an event sequence containing the device identifier, timestamp and device behavior data. Each event record contains the device identifier, synchronized timestamp and device behavior data (such as device switch status, usage frequency, and operation mode).

[0090] As an example, the behavior data collected by device 1, device 2, and device 3 at the standard time after synchronization are as follows:

[0091] Device 1: The device identifier is 001, the synchronized timestamp is 2024-11-01 08:00:00 (UTC), and the device behavior data includes device startup and button click operations;

[0092] Device 2: The device identifier is 002, the synchronized timestamp is 2024-11-01 08:00:00 (UTC), and the device behavior data includes device shutdown and voice control operations;

[0093] Device 3: The device identifier is 003, the synchronized timestamp is 2024-11-01 08:00:00 (UTC), and the device behavior data includes device on and automatic mode;

[0094] Each device record contains a device identifier, a synchronized timestamp, and device behavior data. The mathematical expression of the event sequence is:

[0095] E t ={(ID device ,timestam p t,Data t )}

[0096] Among them, E t represents the event sequence at time t, ID device is the unique identifier of the device, timestamp t It is the synchronized timestamp, indicating the standard time when the device behavior occurs. t It is the behavior data of the device, including the device behavior information (such as device status and operation mode);

[0097] In an optional implementation manner, the device behavior data record collected at time point t=1 is defined as follows:

[0098] Event log of device 1:

[0099] E1={(ID device =001, timestamp1 = 2024-11-01 08:00:00UTC, Data1 = {state = ON, operation = buttonclick})}

[0100] Event log of device 2:

[0101] E2={(ID device =002, timestamp2 = 2024-11-01 08:00:00UTC, Data2 = {state = OFF, operation = voicecontrol})}

[0102] Event log of device 3:

[0103] E3={(ID device =003, timestamp3 = 2024-11-01 08:00:00UTC, Data3 = {state = ON, operation = automode})}

[0104] Ultimately, the event records of all devices will be aggregated into a structured event sequence, sorted and stored according to device identifiers and timestamps for subsequent query and processing;

[0105] The summarized sequence of events is:

[0106] E={E1,E2,E3}

[0107] Among them, E1, E2, and E3 represent event records of each device at different time points. Each event record contains a device identifier, a synchronized timestamp, and device behavior data.

[0108] Furthermore, the event sequence E is stored in the database for analysis, processing and optimization of subsequent steps. The event record storage format is shown in Table 1:

[0109] Table 1. Event record table

[0110] Device identifier Timestamp (UTC) Device behavior information 001 2024-11-01 08:00:00UTC Turn on the device and click the button 002 2024-11-01 08:00:00UTC Turn off the device and voice control operation 003 2024-11-01 08:00:00UTC Turn on the device, automatic mode

[0111] S3. Analyze the data interaction relationship between different IoT devices in the event sequence through the temporal association network algorithm, calculate the temporal similarity between events, construct the association matrix, and form a graph structure representation. Among them, what needs to be explained in this step is:

[0112] Based on the timestamps and device behavior data between devices, the time series similarity between the two devices is calculated, and an event correlation matrix is ​​constructed. Each element of the matrix represents the similarity between two events. The higher the similarity between events, the stronger the correlation between the behaviors of the two devices.

[0113] As an example, three devices i, j and k are defined, and their behaviors at different time points are shown in Table 2:

[0114] Table 2. Equipment usage behavior record

[0115] equipment Timestamp Behavior i 2024-11-01 08:00:00 Turn on the device j 2024-11-01 08:00:05 Turn on the device k 2024-11-01 08:00:10 Turn off the device

[0116] The calculation formula for time series similarity is:

[0117]

[0118] ΔT=|Timestamp i -Timestamp j |

[0119] Where ΔT represents the time difference between device i and device j, Adj i,j is the association matrix element between device i and device j, α is a hyperparameter that adjusts the effect of time difference, BehaviorSimilarity is the behavior similarity, which is 1 if the behaviors are the same and 0 if the behaviors are different;

[0120] Exemplarily, the calculation process of device i and device j is as follows:

[0121] Time difference:

[0122] ΔT ij =|2024-11-0108:00:00-2024-11-0108:00:05|=5 seconds

[0123] Behavior similarity: Turning on the device and turning on the device are the same, so the behavior similarity is 1;

[0124] Timing similarity:

[0125]

[0126] Exemplarily, the calculation process of device i and device k is as follows:

[0127] Time difference:

[0128] ΔT ik =|2024-11-0108:00:00-2024-11-0108:00:10|=10 seconds

[0129] Behavior similarity: Turning on a device is different from turning off a device, so the behavior similarity is 0;

[0130] Timing similarity:

[0131]

[0132] Exemplarily, the calculation process of device j and device k is as follows:

[0133] Time difference:

[0134] ΔT jk =|2024-11-0108:00:05-2024-11-0108:00:10|=5 seconds

[0135] Behavior similarity: Turning on a device is different from turning off a device, so the behavior similarity is 0;

[0136] Timing similarity:

[0137]

[0138] Specifically, based on the calculation results of the above time series similarity, the event association matrix Adj is constructed:

[0139]

[0140] Each element in the matrix Adj i,j represents the timing similarity between device i and device j, for example, Adj i,j =0.952, indicating that the behavior similarity between device i and device j is high, while Adj i,k = 0 means there is no similar behavior between device i and device k;

[0141] Furthermore, the device data is fused and analyzed through graph convolution operations to construct a graph structure representation, in which the device identifier is used as the node of the graph. For example, device i is defined as ID i 、Device j is ID j 、Device k is ID k , and then use the time series association relationship as the edge of the graph to connect the nodes. The weight of each edge represents the time series similarity between devices. For example, from ID i To ID j The edge with weight 0.952 is from ID i To ID k The edge with weight 0 (no edge connection) is from ID j To ID k The edge weight is 0 (no edge connection), then the final adjacency matrix Adj is:

[0142]

[0143] This matrix not only represents the connection relationship between devices, but also contains the similarity information of device behaviors. The higher the similarity of each pair of devices, the greater the weight of the edge, indicating that their behaviors are more similar;

[0144] It is also necessary to explain in detail that the comprehensive feature representation of each device is a high-dimensional vector, which contains the behavioral characteristics of the device and the fusion information of its behavior with neighboring devices. These features are learned from the temporal similarity and behavioral data between the device and its neighboring devices through graph convolution operations (GCN), where:

[0145] Node representation (device feature vector): The representation of each node (device) obtained through graph convolution operation is a vector containing the timing behavior data of the device and its relationship with neighboring devices. For example, the feature representation of device i can be represented by a vector hi express:

[0146] h i =[Behavioral characteristics, influence of timing similarity of neighboring devices]

[0147] Fusion of neighbor information: The feature representation of a device not only considers its own behavior, but also incorporates the feature information of neighbor devices. This process is achieved through the weighted neighbor aggregation operation at each layer in the graph convolutional network.

[0148] Hierarchical information: After multiple layers of graph convolution, the representation of a node aggregates neighbor information from different layers. For example, the device feature representation at the kth layer contains information from neighbors at the kth layer, thus reflecting deeper timing and behavioral characteristics.

[0149] As an example, obtaining a comprehensive feature representation of a device may be achieved by:

[0150] Assuming that the first layer of the graph convolutional network uses a simplified weighted sum aggregation method, and the initial feature value of each device is only its own behavior data (such as device status, usage frequency), the graph convolution operation will adjust the representation of each device according to the characteristics of neighboring devices;

[0151] There are three devices i, j and k defined, and the behavior data has been included in the event correlation matrix, as shown in Table 3:

[0152] Table 3. Event association record table

[0153]

[0154]

[0155] The initial feature matrix is ​​the initial behavioral feature vector of each device. Let the initial feature matrix be X, where each row is the feature vector of the device, and its mathematical expression is:

[0156]

[0157] The feature vector of device i is [1,0], which means that the state of device i is "device turned on" and the usage frequency is 1;

[0158] The feature vector of device j is [1,0], which means that the state of device j is "on device" and the usage frequency is 1;

[0159] The feature vector of device k is [0,1], which means that the status of device k is "off device" and the usage frequency is 1;

[0160] The calculation of graph convolution is based on the adjacency matrix Adj and the initial feature matrix X, and the first layer of graph convolution operation is performed:

[0161]

[0162] in, is the representation of device i after the first convolution layer, N(i) is the set of neighbor nodes of device i, and W (1) is the weight matrix of the first layer, b (1) is the bias term of the first layer;

[0163] The weight matrix and bias terms are defined as follows:

[0164]

[0165] b (1) =[0,0]

[0166] The neighbor node of device i is j, so N(i) = {j}. The similarity between device i and device j is 0.952. According to the graph convolution formula, the first layer convolution result of device i is:

[0167]

[0168]

[0169] Therefore, the first-layer feature representation of device i is [0.622, 0.622];

[0170] It is not difficult to understand that the new feature representations of device j and device k are also calculated through the graph convolution operation shown above, which will not be described in detail in this embodiment. The feature representation of device j will be updated because it is connected to device i, and the feature representation of device k will be updated because of its connection with i and j.

[0171] After the second layer of graph convolution, the feature representation of the device will further integrate the information of more neighboring devices. Through the second layer of convolution, the final feature representation of the device will contain more fusion features of the temporal similarity between devices, enhance the comprehensive behavioral features of each device, and make the similarity between devices more obvious. After two layers of graph convolution, the comprehensive feature representation of devices i, j, and k is obtained as follows:

[0172]

[0173] S4. Analyze the user's behavior feedback (such as device usage frequency) from the graph structure representation, generate instant rewards, update the Q value based on the current state of device use and the generated instant rewards, and generate a personalized behavior management method for the user through the Q value optimization strategy. Among them, what needs to be explained in this step is:

[0174] The Q-learning algorithm is used to analyze the user's behavioral feedback from the graph structure representation, and generate instant rewards by calculating the state space, action space and reward function;

[0175] The state space is a set that describes the user's current device usage behavior. In the embodiment of the present invention, the state space includes the current state of device usage, the user's operation history, and the working state of the device. Its mathematical expression formula is:

[0176] S={s1,s2,…,s n}

[0177] Among them, each state s i Including device status, device usage frequency and device historical behavior;

[0178] Device status: whether the device is turned on (0 or 1), the current mode of the device (e.g., energy-saving mode, normal mode, standby mode);

[0179] Frequency of device use: such as the number of times the device has been used or how long it has been used;

[0180] Historical device behavior: the user's previous behavior sequence, including whether the device was adjusted or optimized;

[0181] In the embodiment of the present invention, the action space represents the operations that a user can take on a device. Each action represents the user's adjustment of the device behavior, such as turning the device on and off, switching the device mode, and adjusting the configuration. The mathematical expression formula is:

[0182] A={a1,a2,…,a m}

[0183] Among them, action a i Including device on / off, device mode switching and configuration adjustments;

[0184] Device on / off: control the device on or off;

[0185] Device mode switching: switch to energy-saving mode, low power consumption mode or high-performance mode according to the current working status of the device;

[0186] Configuration adjustment: adjust the working configuration or parameters of the device, such as temperature setting and brightness adjustment;

[0187] In the embodiment of the present invention, the reward function is used to evaluate the effect of each action and provide feedback for the subsequent optimization strategy. Specifically, the rule of the reward function is:

[0188] Positive behavior: When the user's actions make the device more efficient (such as switching to energy-saving mode), give a higher reward;

[0189] Negative behavior: When the user's operation causes the device to waste resources or fail to achieve the expected goal, a lower reward is given;

[0190] Neutral behavior: When the operation does not significantly change the efficiency of the equipment, the reward is 0;

[0191] The mathematical expression of the reward function is:

[0192]

[0193] Assume that each state-action pair (s, a) has an immediate reward value R(s, a), which reflects the benefit of the action;

[0194] It should be noted that in the Q-learning algorithm, the Q value represents the cumulative reward obtained after performing a certain action in a certain state, and the Q value is updated using the following formula:

[0195]

[0196] Among them, Q(s,a) represents the Q value of performing action a in state s, α is the learning rate, which controls the update weight of new information to old information, R(s,a) is the immediate reward, which represents the reward obtained after taking action a, and γ is the discount factor, which represents the discount degree of future rewards. is in the new state s ′ Next, all possible actions a ′ The maximum value of the Q value is used to reflect the maximum expectation of future rewards;

[0197] According to the Q-learning process, the system will continuously update the Q value and select the optimal action through the following strategy:

[0198] Greedy Strategy: Choose the action with the largest Q value, that is:

[0199]

[0200] ε-greedy Strategy: In order to balance exploration and exploitation, the system selects a random action with a probability of ∈ and selects the action with the largest Q value with a probability of 1-∈, that is:

[0201] Random action with probability∈,

[0202]

[0203] The Q-value is continuously optimized through the Q-learning algorithm to generate personalized behavior management strategies. After the training is completed, the system will automatically adjust the device management strategy according to the user's behavior pattern to ensure that the user's needs are effectively met;

[0204] Specifically, the generated behavior management strategies include:

[0205] Device on / off: When the energy consumption of the device is too high or the user is no longer using the device, the device will be automatically turned off, and when the device needs to be enabled, it will be automatically turned on;

[0206] Device mode switching: If the current device is working in high-energy consumption mode and the user does not perform frequent operations, the system will automatically switch to energy-saving mode;

[0207] Energy saving and efficiency optimization: The system will automatically adjust the equipment usage time, operation mode and energy consumption based on historical usage data to maximize the efficiency of resource utilization;

[0208] For example, if the Q value for a state s i The next action a j If the action has a high return (such as efficient energy-saving mode), the system will automatically select the action in similar subsequent states to continuously optimize the user's device usage strategy;

[0209] In the embodiment of the present invention, the device behavior features extracted by the graph convolutional network (GCN) are combined with the Q-learning algorithm to generate a personalized device behavior management strategy. Through the Q-learning learning process, the system can automatically optimize the switch control, mode switching, and configuration adjustment operations of the device to improve energy efficiency.

[0210] For example, assume that there are two devices (device A and device B), collect the behavioral features of the devices, and represent them as a graph structure. The behavior of each device includes: the device's on / off state, usage frequency, working mode, and user's historical operations, as shown in Table 4:

[0211] Table 4. Device behavior table

[0212]

[0213] The timing similarity between device A and device B is calculated by the device behavior data. Assuming that the timestamp and device behavior data are used to measure the timing similarity between two devices, for example, the behavior similarity between device A and device B can be calculated by the following formula:

[0214] Adj A,B =f(Timestamp A ,Timestamp B ,BehaviorA ,Behavior B )

[0215] The above formula calculates that the timing similarity between device A and device B is 0.85 (indicating that their behaviors and timestamps are similar);

[0216] In the graph structure, nodes represent devices. Device A and device B correspond to nodes in the graph respectively. Edges represent the timing similarity between devices. The weight of the edge is 0.85, indicating that the behaviors of device A and device B are highly similar.

[0217] Exemplarily, the graph structure is represented as follows:

[0218] Node 1 (device A): contains the device switch status, usage frequency, current mode and historical behavior characteristics;

[0219] Node 2 (device B): contains the device switch status, usage frequency, current mode and historical behavior characteristics;

[0220] Edge (0.85): The similarity edge between device A and device B, indicating that the behavioral characteristics of the two devices are highly similar;

[0221] In the Q-learning algorithm, the state space (S) defines the various possible states of the device, and each state s i Features include:

[0222] Device status: whether the device is turned on (0 or 1), the device's operating mode (e.g., energy-saving mode, normal mode);

[0223] Device usage frequency: how often the device is used, such as how many times a day;

[0224] Device working time: the working time of the device in the current mode;

[0225] Device history: the device’s previous operation history (e.g., whether the user has adjusted the temperature or switched modes);

[0226] Similarity features in graph structure: Temporal similarity features between devices extracted through graph convolutional networks;

[0227] For example, define the state of device A (s1): device A is currently turned on, the frequency of use is 5 times / day, the working mode is normal mode, and the historical behavior is temperature adjustment. Define the state of device B (s2): device B is currently turned off, the frequency of use is 2 times / day, the working mode is energy-saving mode, and the historical behavior is mode switching. The state space: S = {s1, s2};

[0228] Action space (A) defines the operations that users can perform on the device. The action space includes turning the device on / off (controlling the device to turn on or off), mode switching (switching to energy-saving mode, low power mode or high-performance mode according to the current working state of the device), and configuration adjustment (adjusting the temperature and brightness configuration of the device);

[0229] For example, action a1 is defined as turning on / off a device, action a2 is defined as switching a device mode (e.g., switching from normal mode to energy-saving mode), and action a3 is defined as adjusting a device configuration (e.g., adjusting a temperature setting). Then, the action space is:

[0230] A={a1: turn on / off device, a2: switch device mode, a3: adjust device configuration};

[0231] Exemplarily, the reward function R(s,a) is used to evaluate the effect of each action, and defines that when switching to energy-saving mode, the reward is +1, when switching to high-energy consumption mode, the reward is -1, and when adjusting the temperature does not significantly change the energy efficiency, the reward is 0;

[0232] Define the Q value update parameters: learning rate α = 0.1, discount factor γ = 0.9, current state s1 (device A is currently turned on and the mode is normal), current action a2 (switch the device mode to energy saving mode), immediate reward R (s1, a2) = +1, and maximum Q value (Assuming that device B currently has no significant Q value), the Q value update process is:

[0233] Q(s1,a2)←Q(s1,a2)+0.1[1+0.9×0-0]=Q(s1,a2)+0.1

[0234] When Q(s1,a2)=0, after updating:

[0235] Q(s1,a2)=0+0.1=0.1

[0236] Through the above Q-learning update process, the system will select the optimal action according to the Q value. For example, in the subsequent state, the system will select a2 (switch to energy-saving mode) as the optimal action because Q(s1,a2) is the largest;

[0237] When the system selects "switch to energy-saving mode" as the optimal action, it means that the current state of the device (e.g., high-power consumption mode, less operation, or infrequent use) does not require high-performance operation, but can optimize resource usage by reducing power consumption;

[0238] Exemplarily, based on the optimization of the above Q-learning algorithm, the following personalized management strategy is generated:

[0239] (1) Equipment switch control strategy

[0240] Threshold setting: If the usage frequency of device A (or device B) is less than 2 times / day, the system will automatically shut down the device;

[0241] For example, if the frequency of use of device B is 2 times / day, and the system determines that the energy consumption and current status of device B are not worth keeping on, then the system can choose to shut down the device;

[0242] Control strategy: When the device is used less than 2 times per day, the system will automatically shut down the device. If the device is running in energy-saving mode and the usage time is less than the set threshold (for example, 4 hours), the system will shut down the device.

[0243] For example, device A is used 5 times / day, and device B is used 2 times / day. According to the threshold setting, the device will be shut down when the device usage frequency is ≤ 2 times / day. For device A, the frequency is higher (5 times / day), so the device will not be shut down. For device B, the frequency is lower (2 times / day), so the system will automatically shut down the device.

[0244] (2) Device Mode Switching Strategy

[0245] Threshold setting: If the current mode of the device is "normal mode" or "high energy consumption mode", and the device usage time (for example, more than 6 hours) does not reach a certain threshold, the system considers that the device does not need high energy consumption operation and switches to energy saving mode;

[0246] When the usage frequency is low and the device is working for a long time (for example, the usage time of device A is ≥ 6 hours and the operation frequency is low), the system will choose to switch to energy-saving mode. That is, if the device is in high-energy consumption mode and the user does not perform frequent operations, the system will switch the device to energy-saving mode;

[0247] For example, if device A is used 5 times / day and works for 10 hours, the system may consider that the device consumes too much energy and automatically switches to energy-saving mode. In state s1 (device A is used for 10 hours), after the Q value is updated, the system considers that switching to energy-saving mode can reduce energy consumption, so it will select action a2 (switch to energy-saving mode);

[0248] (3) Device configuration adjustment strategy

[0249] Threshold setting: If the current configuration of the device does not reach the user's set preference value (such as the temperature is too high or too low), the system will automatically adjust the device configuration. For example, when the device is at a higher temperature and the ambient temperature is lower, the system will adjust the device temperature to lower;

[0250] If the device does not adjust the configuration according to the user's preferences (for example, the temperature is too high or the brightness is too low), the system will automatically adjust the device configuration;

[0251] For example, if the temperature of the defined device exceeds 22°C, the temperature is automatically adjusted to the set value. If the current temperature of device A is 25°C and the user's preferred temperature is 22°C, the system will automatically adjust the configuration to reach the temperature value set by the user.

[0252] In an optional embodiment, the device switch control includes: if the device is used less than 2 times / day, the device is automatically turned off; if the device is used more than 2 times / day, the device is kept turned on.

[0253] In an optional embodiment, device mode switching includes: if the device is in a high energy consumption mode for a long time and is used less frequently, switching to an energy saving mode; if the device is used frequently and has high energy efficiency requirements, maintaining the current mode (eg, a high energy mode).

[0254] In an optional embodiment, the configuration adjustment includes automatically adjusting the device configuration (such as temperature or brightness) if it does not meet the user's preference.

[0255] Preferably, the processing of step 1 of the method of the present invention ensures the comprehensiveness and accuracy of data collection, lays the foundation for subsequent data processing, and combined with the time synchronization of step 2, solves the problem of inconsistent timestamps of different devices and improves the accuracy of timing analysis. Through the implementation of step 3, the correlation between different devices is excavated, overcoming the deficiency of traditional methods that cannot effectively model the interaction relationship between devices. By using the operation of step 4, the management system can continuously optimize the management strategy according to the actual behavior and feedback of users, and provide more intelligent and personalized services.

[0256] Preferably, the method of the present invention solves the problems existing in the prior art such as difficulty in integrating multi-source heterogeneous data, inaccurate time synchronization, insufficient user behavior modeling, and lack of personalized management strategies. By introducing time synchronization, temporal correlation network algorithms, and reinforcement learning strategies, efficient management and analysis of user usage information on multiple IoT devices is achieved, and personalized behavior management solutions can be provided for users, thereby improving the real-time and efficiency of data processing and enhancing user experience.

[0257] It is also necessary to further explain in the embodiments of the present invention that the conventional method uses a device information management method based on simple statistics and rules to process and manage user data of IoT devices, which has the following defects:

[0258] Lack of time synchronization and correlation analysis: Due to the lack of time synchronization, the timing relationship between different devices cannot be accurately analyzed, resulting in an inaccurate understanding of the interaction between devices and the user's comprehensive behavior;

[0259] Low data processing efficiency: With the increase in the number of IoT devices and the amount of data, traditional statistical methods cannot efficiently process massive data and cannot meet real-time requirements;

[0260] Lack of personalization and adaptive capabilities: The device management strategy cannot be dynamically adjusted according to the user's actual usage and preferences, lacks personalized services, and the user experience is poor;

[0261] Low energy efficiency: Failure to effectively optimize the operating status of equipment leads to energy waste and lack of intelligent management of energy consumption;

[0262] This experiment aims to verify the effectiveness of the proposed user information management method based on IoT big data in the information management of users using IoT devices in a smart home environment. The experimental comparison indicators include:

[0263] User behavior pattern recognition accuracy: evaluate the accuracy of the proposed method in identifying and predicting user behavior on IoT devices, and verify its advantages in capturing complex user behavior patterns and the interaction relationship between devices;

[0264] Energy efficiency improvement effect: verify the optimization effect of personalized behavior management strategy on energy consumption, and evaluate the role of the method in reducing energy waste and improving equipment operation efficiency;

[0265] Data processing and analysis efficiency: Compare the efficiency differences between the method of the present invention and traditional methods in processing and analyzing IoT big data, and verify its advantages in real-time and high efficiency;

[0266] The design experiment was conducted in a typical smart home environment, including:

[0267] Living space: A three-bedroom, two-living-room apartment, about 120 square meters;

[0268] IoT devices: 10 in total, distributed in different rooms, including:

[0269] Device A: Smart lighting (Device ID: Device001);

[0270] Device B: Smart thermostat (Device ID: Device002);

[0271] Device C: Smart TV (Device ID: Device003);

[0272] Device D: Smart speaker (Device ID: Device004);

[0273] Device E: Smart curtain (Device ID: Device005);

[0274] Device F: Smart mattress (Device ID: Device006);

[0275] Device G: Smart refrigerator (Device ID: Device007);

[0276] Device H: Smart oven (Device ID: Device008);

[0277] Device I: Smart water heater (Device ID: Device009);

[0278] Device J: Smart door lock (Device ID: Device010);

[0279] Number of users: Three permanent users, User1 (adult), User2 (adult) and User3 (child);

[0280] User characteristics:

[0281] User 1: I leave home early and come back late on weekdays, prefer energy-saving mode, and like to use smart speakers to relax at night;

[0282] User 2: works from home, frequently uses smart thermostats and lighting equipment, and has high requirements for environmental comfort;

[0283] User3: After school, she is at home, likes to watch smart TV and use smart mattress, and goes to bed early;

[0284] Experimental period: 28 consecutive days, covering weekdays and weekends;

[0285] Network environment: high-speed and stable Wi-Fi network to ensure reliable data transmission of IoT devices;

[0286] Time synchronization: The Precision Time Protocol (PTP) is used to synchronize the time of all devices, with a time accuracy of milliseconds;

[0287] Data storage and processing platform: using a cloud server, configured with an 8-core CPU, 16GB memory and high-speed SSD storage, to run a comparative test between the algorithm of the method of the present invention and the traditional method;

[0288] Experimental operation of traditional methods: collect user operation data from each IoT device independently, including device ID, user ID, operation time (local timestamp), device status and operation mode, conduct independent statistical analysis on the data of each device, calculate the number of daily usage and average usage time, and use the moving average method to predict the future usage of the device based on the historical usage data of the device. Generate corresponding management strategies based on the prediction results and timed switches, apply the generated strategies to the automatic control of the device, and the device operates according to the preset rules;

[0289] The experimental operation of the method of the present invention is as follows: obtaining device information used by users from all IoT devices, including device ID, user ID, operation time (unified UTC timestamp), device status and operation mode, using the PTP protocol to synchronize the time of all devices to ensure the time consistency of the event sequence, calculating the time series similarity between different devices, considering the time proximity and the correlation of user operations, constructing an association matrix, quantifying the similarity between devices, forming a graph structure representation, taking the device as a node and the time series similarity as the weight of the edge, extracting the user's behavior pattern from the graph structure, generating an instant reward value r for evaluating the effect of the current strategy, using the Q-learning algorithm in reinforcement learning, continuously updating the Q value table according to the instant reward and the current status of the device, generating a personalized device management strategy, and realizing intelligent control of the device;

[0290] The experimental comparison indicators include behavior pattern recognition accuracy, energy efficiency improvement rate, and data processing and analysis efficiency, among which:

[0291] In the experimental operation of the behavior pattern recognition accuracy index, the traditional method predicts the user's next device operation based on the historical data of a single device. The method of the present invention uses the temporal association network and Q-learning algorithm to comprehensively consider the association between devices and the user's behavior pattern to predict the user's next device operation. The calculation formula is as follows:

[0292] Behavior pattern recognition accuracy =

[0293] (Number of correctly predicted next device actions of users / Total number of predictions) × 100%

[0294] In the experimental operation of the energy efficiency improvement rate indicator, the daily energy consumption of all equipment under the two methods was recorded, the energy efficiency improvement rate was calculated, and the energy-saving effect of the method of the present invention was evaluated. The calculation formula is as follows:

[0295] Average daily energy consumption = total daily energy consumption of each device (unit: kWh)

[0296] Energy efficiency improvement rate =

[0297] (Energy consumption of traditional method - energy consumption of the method of the present invention) / energy consumption of traditional method × 100%

[0298] In the experimental operation of the data processing and analysis efficiency index, the data processing time of the two methods under the same data volume was recorded to evaluate the efficiency of the method of the present invention in big data processing. The calculation formula is as follows:

[0299] Data processing time = the time required to process and analyze all device data per day (in minutes)

[0300] Data volume = number of data items collected daily

[0301] The experimental results obtained are as follows Figure 2 , Figure 3 , Figure 4 shown.

[0302] Reference Figure 2 , the dotted line (circular mark) represents the traditional method, and the accuracy fluctuates between 58% and 62%. Since the traditional method only relies on the historical data of a single device, it cannot capture the user's cross-device behavior pattern, resulting in a low accuracy rate and slow improvement; the solid line (square mark) represents the method of the present invention, and the accuracy rate increases from 80% to 96%, and tends to be stable after the 15th day. The method of the present invention uses a temporal association network and a reinforcement learning algorithm to quickly learn the user's behavior habits, and the accuracy rate is significantly improved, and reaches a stable high level after the 15th day;

[0303] Reference Figure 3 The dotted line (circle mark) is the traditional method, and the energy consumption is between 150-126 kWh, showing a slow downward trend. Due to the lack of personalized management strategies, the energy consumption of the traditional method decreases slowly; the solid line (square mark) is the method of the present invention, and the energy consumption decreases from 140 kWh to 85 kWh, with a greater decrease. The method of the present invention effectively reduces unnecessary energy consumption through intelligent equipment management strategies, and the energy-saving effect is significant;

[0304] Reference Figure 4 The dotted line (circular mark) is the traditional method. The processing time increases linearly with the increase of data volume, and the efficiency decreases. The traditional method uses simple statistical analysis, so it is inefficient when processing large amounts of data. The solid line (square mark) is the method of the present invention. The processing time increases slowly, showing good scalability. The method of the present invention uses optimized algorithms and data structures (such as association matrices and graph structures) to improve the efficiency of big data processing.

[0305] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A user information management method based on Internet of Things big data, characterized in that: include: Obtain device information used by a user from multiple IoT devices, wherein the device information includes user identity information and device usage behavior information, and each device is identified by a unique device identifier; Performing timestamp marking and timestamp synchronization on the device information to form an event sequence including a device identifier and a timestamp; The data interaction relationship between different IoT devices in the event sequence is analyzed by using a temporal association network algorithm, the temporal similarity between events is calculated, an association matrix is ​​constructed, and a graph structure representation is formed; The user's behavioral feedback is analyzed from the graph structure representation to generate an instant reward, and a personalized management method is generated for the user through a Q-value optimization strategy based on the current state of device use and the generated instant reward.

2. The user information management method based on Internet of Things big data according to claim 1 is characterized in that: Bind the IoT devices used by users to the corresponding user IDs, collect device information through sensor technology and network communication, and record device status and user operation behaviors; among which: The user identity information includes a user name and a user ID; The device usage behavior information includes the device on / off status, usage frequency and operation mode; The device identifier is a unique identifier for each device, which is assigned by the hardware manufacturer of the device or generated by a unique algorithm.

3. The user information management method based on Internet of Things big data according to claim 2 is characterized in that: The device on / off status records the on / off status of the device at different time points, the usage frequency records the number of times the device is used or the active time of the device in a certain period of time, and the operation mode records the interaction method between the user and the device.

4. The user information management method based on Internet of Things big data according to claim 1 or 2 is characterized in that: The collected device information is formed into an initial data set including device identifier, user identity information and device behavior information, and structured storage is performed. The mathematical expression of the initial data set is: D i =(ID device ,{Data behavior ,Data user }) Among them, D i is the data of the ith device, ID device is the device identifier, Data behavior Data is the device usage behavior data. user The user's identity data.

5. The user information management method based on Internet of Things big data according to claim 1 is characterized in that: The event sequence is obtained, including: Synchronize the local time of each device to a unified standard time through the PTP protocol; After adding the synchronized timestamp to each device data record, an event sequence including the device identifier, timestamp and device behavior data is formed. Each event record includes the device identifier, synchronized timestamp and device behavior information. The behavior data of the device at least includes the device switch status, usage frequency and operation mode.

6. The user information management method based on Internet of Things big data according to claim 5 is characterized in that: The mathematical expression formula of the event sequence is: E t ={(ID device ,timestamp t ,Data t )} Among them, E t represents the event sequence at time t, ID device is the unique identifier of the device, timestamp t It is the synchronized timestamp, indicating the standard time when the device behavior occurs. t The device behavior data includes device behavior information.

7. The user information management method based on Internet of Things big data according to claim 1 is characterized in that: Based on the timestamps and device behavior information between the devices, the time series similarity between the two devices is calculated, and an event correlation matrix is ​​constructed, wherein each element of the event correlation matrix represents the similarity between two events. The higher the similarity between the events, the stronger the correlation between the behaviors of the two devices. The calculation formula of the time series similarity is: ΔT=|Timestamp i -Timestamp j | Where ΔT represents the time difference between device i and device j, Adj i,j is the association matrix element between device i and device j, α is a hyperparameter that adjusts the influence of time difference, and BehaviorSimilarity is the behavior similarity, which is 1 if the behaviors are the same and 0 if the behaviors are different.

8. The user information management method based on Internet of Things big data according to claim 1 is characterized in that: Through graph convolution operations, a fusion analysis of device data is performed to construct a graph structure representation, in which the device identifiers are used as nodes of the graph, and the temporal association relationships are used as edges to connect the nodes. The weight of each edge represents the temporal similarity between devices. The higher the similarity of each pair of devices, the greater the weight of the edge, indicating that the behaviors are more similar.

9. The user information management method based on Internet of Things big data according to claim 1 is characterized in that: The personalized management method includes device switch control, device mode switching and configuration adjustment of the device bound to the user.

Images