Smart building system

By introducing a multi-layered closed-loop security system into the smart building system, and utilizing multi-dimensional data feature vectors and advanced intrusion detection models, the problems of single protection and insufficient computing power in existing technologies are solved, enabling effective identification and blocking of complex threats, and improving the system's security and responsiveness.

CN123122849APending Publication Date: 2026-11-17QINGDAO HISENSE BOSCH AIR CONDITIONING SYSTEM CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510585154.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-05-07
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing intrusion detection technologies in smart building systems suffer from limitations such as single-layer protection, lack of multi-dimensional comprehensive protection, reliance on static rules, insufficient intelligence, and inadequate collaborative protection. They are unable to cope with complex dynamic attacks and lack sufficient computing power to effectively deal with zero-day attacks and complex threats.

Method used

It adopts a multi-layered closed-loop security system, including collaborative protection at the device end, edge end, and cloud end. It utilizes multi-dimensional data feature vectors and advanced intrusion detection models, and identifies anomalies through time series models and graph neural networks to achieve dynamic identification and blocking of security threats.

Benefits of technology

It enhances the ability to identify unknown threats, improves the ability to resist zero-day attacks and complex dangers, ensures real-time response and big data analysis support, and meets the complex environmental requirements of smart building systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123122849A_ABST
    Figure CN123122849A_ABST
Patent Text Reader

Abstract

This invention discloses a smart building system, comprising a data acquisition unit, first and second intrusion detection units, and first and second response units. The data acquisition unit generates multi-dimensional data feature vectors based on multi-dimensional data from the device, edge, and cloud. The first intrusion detection unit, deployed at the edge, identifies anomalies and triggers alarms using a first intrusion detection model built based on a time-series model. The first response unit, also deployed at the edge, filters out and isolates abnormal nodes after the first intrusion detection unit triggers an alarm. The second intrusion detection unit, deployed in the cloud, locates abnormal nodes and / or edges using a second intrusion detection model built based on a graph neural network. The second response unit, deployed in the cloud, establishes abnormal links based on a subgraph composed of abnormal nodes and / or edges, generates and outputs protection commands to block the abnormal links. This application achieves close cooperation between the device, edge, and cloud to form a complete closed-loop security system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to a smart building system. Background Technology

[0002] A smart building system is a comprehensive management platform composed of electronic devices and computer networks. It enables centralized control and coordinated management of multiple subsystems within a building (such as lighting, air conditioning, ventilation, water supply and drainage, elevators and escalators, fire protection, and power supply systems), thereby improving building efficiency and comfort. In practical applications, smart building administrators, maintenance engineers, and other relevant personnel interact with the smart building system through computer terminals or remote terminals to monitor and adjust the operational status of each subsystem in real time, enabling automated and intelligent system operation.

[0003] Because smart building systems create a vast Internet of Things (IoT) environment with diverse devices and varying levels of security, they are easy targets for attacks. Network intrusions can steal sensitive data, user privacy, or trade secrets, and paralyze critical building facilities.

[0004] Existing intrusion detection technologies for smart building systems mainly fall into three categories: rule-based, feature-based, and deep learning and behavior analysis-based. Rule-based methods identify known attacks through pre-defined rules, offering simplicity and intuitiveness but struggling to detect novel threats. Feature-based methods detect anomalies by analyzing network traffic and user behavior, enabling the identification of more attack types, but still have limitations in handling complex zero-day attacks. Deep learning-based intelligent detection can automatically learn and adapt to new attacks, offering greater flexibility and becoming a research focus. However, these technologies generally suffer from problems such as single-layer protection, lack of multi-dimensional comprehensive protection, reliance on static rules, insufficient intelligence, inadequate collaborative protection, and low processing efficiency. Furthermore, they are mostly concentrated on a single security level, making it difficult to cope with constantly evolving attacks; moreover, the lack of collaboration in multi-layer protection limits the effectiveness of protection. As the number of integrated devices and data volumes in smart building systems increases, traditional methods also face bottlenecks in performance and computing power. Security technology development lags behind the rapid evolution of smart building systems, and traditional protection concepts are too static and singular, with intrusion detection relying on expert experience, making it difficult to adapt to dynamic and complex attack environments.

[0005] The information disclosed in this background section is only intended to enhance the understanding of the background technology of this application, and therefore may include prior art that is not known to those skilled in the art. Summary of the Invention

[0006] The first aspect of this application provides a smart building system, including a data acquisition unit, a first intrusion detection unit, a first response unit, a second intrusion detection unit, and a second response unit; wherein, the data acquisition unit is configured to acquire multidimensional data from device terminals, edge terminals, and the cloud, and generate multidimensional data feature vectors based on the multidimensional data; the multidimensional data includes: user behavior data, device behavior data, and network traffic data; the first intrusion detection unit is deployed at the edge terminal, and is configured to identify anomalies based on the multidimensional data feature vectors using a first intrusion detection model, and trigger an alarm when an anomaly is identified; the first intrusion detection model is based on... A time-series model is constructed; a first response unit is deployed at the edge and configured to filter out abnormal nodes after the first intrusion detection unit triggers an alarm; and isolate the abnormal nodes; a second intrusion detection unit is deployed in the cloud and configured to respond to the alarm triggered by the first intrusion detection unit, locate abnormal nodes and / or abnormal edges based on the second intrusion detection model, which is constructed based on a graph neural network; a second response unit is deployed in the cloud and configured to establish abnormal links based on the subgraph formed by abnormal nodes and / or abnormal edges; and generate and output protection commands to block the abnormal links.

[0007] The above technical solution has the following advantages or beneficial effects: This application closely coordinates the device end, edge end, and cloud end to form a complete closed-loop security system. Each layer can identify and block security threats at the corresponding level, and further enhance the overall protection effect through multi-layer linkage; the first intrusion detection model and the second intrusion detection model can dynamically identify unknown threats and dynamically update the strategy, no longer relying solely on static rules, enhancing the ability to resist zero-day attacks and complex dangers. The rapid response mode at the edge end and the global training mode in the cloud end can not only ensure real-time requirements, but also obtain the support of big data analysis, and can cope with physical, network, privilege escalation and distributed attacks, meeting the actual needs of intelligent building systems with dense equipment, complex user roles and diverse communication modes.

[0008] In some embodiments of this application, user behavior data includes one or more of the following: user information, operation details, time information, geographical location, operation frequency, and abnormal information; wherein, user information includes one or more of the following: user ID, account type, and account permissions; operation details include one or more of the following: operation type, operation parameters, and operation result; time information includes one or more of the following: access time and operation time; geographical location includes one or more of the following: GPS coordinates and IP address; and abnormal information includes one or more of the following: error code and alarm identifier.

[0009] The above technical solution has the following advantages or beneficial effects: user behavior data can be used to collect user behavior information from multiple dimensions, which helps to locate abnormal behavior or security incidents.

[0010] In some embodiments of this application, the device end includes: a device end node; the device end node includes: a sensor, an execution unit, and / or a field control device; the edge end includes: an edge end node; the edge end node includes an edge end control device; the cloud end includes: a cloud end node; the cloud end node includes a cloud control device; device behavior data includes one or more of the following: execution unit status, execution unit events, sensor data, sensor status, sensor time, operation log, configuration log, control command type, and control command frequency; wherein, the execution unit status includes one or more of the following: power on, power off, operating mode, temperature, operating time, and energy consumption; the execution unit events include one or more of the following: execution unit alarm, Execution unit errors and execution unit maintenance records; operation logs include one or more of the following: operation logs of field control devices, operation logs of edge control devices, and operation logs of cloud control devices; configuration logs include one or more of the following: configuration logs of field control devices, configuration logs of edge control devices, and configuration logs of cloud control devices; control command types include one or more of the following: control command types of field control devices, control command types of edge control devices, and control command types of cloud control devices; control command frequencies include one or more of the following: control command frequencies of field control devices, control command frequencies of edge control devices, and control command frequencies of cloud control devices.

[0011] The above technical solution has the following advantages or beneficial effects: it can monitor and manage the status of the entire smart building system through device behavior data, and build a hierarchical architecture to realize distributed data collection and improve the traceability of operations.

[0012] In some embodiments of this application, network traffic data includes one or more of the following: source IP / domain name, destination IP / domain name, port number, communication protocol type, traffic size, traffic direction, connection duration, connection status, session ID, connection timestamp, abnormal communication identifier, and error communication identifier; wherein the connection status includes one or more of the following: number of connections, success, failure, and abnormal disconnection.

[0013] The above technical solution has the following advantages or beneficial effects: potential attacks can be detected in a timely manner through network traffic data, enhancing the ability to track the communication process.

[0014] In some embodiments of this application, the acquisition unit is configured to perform the following steps to generate a multidimensional data feature vector based on multidimensional data: acquiring device behavior data according to a first predetermined period; acquiring network traffic data according to a second predetermined period; slicing user behavior data, device behavior data, and network traffic data into data blocks according to the time period corresponding to the time window, using a preset time window as the unit; calculating the average value of device behavior data within each time window as a device behavior window feature; calculating the average value of network traffic data within each time window as a network traffic window feature; determining whether user behavior data exists within the time window; if user behavior data exists within the time window, counting the number of times user behavior data occurs within the time window as a user behavior window feature; if user behavior data does not exist within the time window, filling in missing values ​​using interpolation to generate user behavior window features; and concatenating the user behavior window features, device behavior window features, and network traffic window features along the feature dimensions to generate a multidimensional data feature vector.

[0015] The above technical solution has the following advantages or beneficial effects: By using the above method, the problem that user behavior data, device behavior data and network traffic data are misaligned on the time axis, making it difficult to directly integrate and analyze the data, is solved.

[0016] In some embodiments of this application, the first intrusion detection unit is configured to perform the following steps: generating time window features based on multidimensional data feature vectors; using the time window features as input to the first intrusion detection model, the first intrusion detection model outputs the anomaly probability corresponding to the current time window features; and triggering an alarm when the output anomaly probability exceeds a preset security threshold.

[0017] The above technical solution has the following advantages or beneficial effects: it enables rapid detection, alarm and response at the edge through the first intrusion detection unit.

[0018] In some embodiments of this application, the second intrusion detection unit is configured to perform the following steps to construct a second intrusion detection model: generating a global graph node set of device-side nodes, edge-side nodes, and cloud-side nodes; determining node features based on multi-dimensional data feature vectors; and establishing a node feature matrix based on the node feature vectors; generating a global graph edge set; determining edge features based on multi-dimensional data feature vectors; and establishing an edge feature matrix based on the edge feature vectors; constructing an adjacency matrix to form the topology of the global graph; constructing a second intrusion detection model based on GCN, GraphSAGE, or an enhanced GraphSAGE architecture, wherein the second intrusion detection model takes node features and edge features as input and outputs the anomaly probability or classification result of each node or each edge; training the second intrusion detection model based on supervised learning; and deploying the trained second intrusion detection model in the cloud.

[0019] The above technical solution has the following advantages or beneficial effects: the second intrusion detection model can further explore the impact of neighboring nodes and the entire network, discover potential attack paths, and identify abnormal nodes and abnormal edges.

[0020] In some embodiments of this application, when training the second intrusion detection model based on supervised learning, the cross-entropy loss function is used as the loss function, and the cross-entropy loss is multiplied by a weight factor based on the predicted probability.

[0021] The above technical solution has the following advantages or beneficial effects: by introducing focus loss, it can better address the problems of insufficient intrusion event data coefficients and abnormal sample types.

[0022] In some embodiments of this application, the smart building system further includes a partitioning unit; the partitioning unit is deployed at the edge and / or in the cloud, and is configured to divide into multiple logical areas, and set access control rules for each logical area, and only authorize subjects whose identities have been verified by the access control rules to access the corresponding logical area.

[0023] The above technical solution has the following advantages or beneficial effects: the partitioning department uses a configuration system to perform logical partitioning, which can logically isolate key equipment.

[0024] In some embodiments of this application, the smart building system further includes an access unit; the access unit is deployed at the edge and is configured to verify the permissions of the device end nodes based on their identity identifiers and allow the authenticated device end nodes to establish connections.

[0025] The above technical solution has the following advantages or beneficial effects: the access unit is used to defend against the access of counterfeit devices and illegal device end nodes, which can effectively prevent the implantation of malicious devices.

[0026] Other features and advantages of the present invention will become clearer after reading the detailed embodiments of the invention in conjunction with the accompanying drawings. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic block diagram of the structure of a smart building system provided in some embodiments of the present invention;

[0029] Figure 2This is a schematic block diagram of the structure of a smart building system provided in some embodiments of the present invention;

[0030] Figure 3 This is a schematic block diagram of the structure of a device-end node in a smart building system provided in some embodiments of the present invention;

[0031] Figure 4 This is a schematic block diagram of the structure of a smart building system provided in some embodiments of the present invention;

[0032] Figure 5 This is a schematic block diagram of the structure of a smart building system provided in some embodiments of the present invention;

[0033] Figure 6 A flowchart of a smart building system provided in some embodiments of the present invention;

[0034] Figure 7 A flowchart of a smart building system provided in some embodiments of the present invention;

[0035] Figure 8 A flowchart of a smart building system provided in some embodiments of the present invention;

[0036] Figure 9 This is a schematic block diagram of the structure of a smart building system provided in some embodiments of the present invention;

[0037] Figure 10 This is a schematic block diagram of the structure of a smart building system provided in some embodiments of the present invention;

[0038] In the diagram: 10. Smart building system;

[0039] 101. Lighting Subsystem; 102. Air Conditioning Subsystem; 103. Ventilation Subsystem; 104. Water Supply and Drainage Subsystem; 105. Elevator Subsystem; 106. Fire Protection Subsystem; 107. Power Supply Subsystem;

[0040] 111. Cloud; 112. Edge; 113. Device;

[0041] 114. Cloud node; 115. Edge node; 116. Device node;

[0042] 1141. Cloud-based control equipment; 1151. Edge-based control equipment; 1161. Sensors; 1162. Actuation units; 1163. Field control equipment;

[0043] 1142. Cloud platform; 1152. Service integration platform;

[0044] 121. Acquisition Unit; 122. First Intrusion Detection Unit; 123. First Response Unit; 124. Second Intrusion Detection Unit; 125. Second Response Unit; 126. Partition Unit; 127. Access Unit. Detailed Implementation

[0045] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0046] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0047] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "multiple" means two or more.

[0048] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0049] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0050] The following disclosure provides many different embodiments or examples for implementing various structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. In addition, examples of various specific processes and materials are provided in this invention, but those skilled in the art will recognize the application of other processes and / or the use of other materials.

[0051] A smart building system is a comprehensive management platform for automated building management, comprising multiple subsystems. These subsystems connect to a network using pre-configured communication standards or protocols, enabling efficient interconnection and collaborative control. The smart building system can further connect to the internet. However, due to the vast Internet of Things (IoT) environment formed within a smart building system, the diverse range of devices and varying levels of security make it an easy target for attacks. Network intrusions can steal sensitive data, user privacy, or trade secrets, and paralyze critical building facilities.

[0052] Existing intrusion detection technologies for smart building systems mainly fall into three categories: rule-based, feature-based, and deep learning and behavior analysis-based. Rule-based methods identify known attacks through pre-defined rules, offering simplicity and intuitiveness but struggling to detect novel threats. Feature-based methods detect anomalies by analyzing network traffic and user behavior, enabling the identification of more attack types, but still have limitations in handling complex zero-day attacks. Deep learning-based intelligent detection can automatically learn and adapt to new attacks, offering greater flexibility and becoming a research focus. However, these technologies generally suffer from problems such as single-layer protection, lack of multi-dimensional comprehensive protection, reliance on static rules, insufficient intelligence, inadequate collaborative protection, and low processing efficiency. Furthermore, they are mostly concentrated on a single security level, making it difficult to cope with constantly evolving attacks; moreover, the lack of collaboration in multi-layer protection limits the effectiveness of protection. As the number of integrated devices and data volumes in smart building systems increases, traditional methods also face bottlenecks in performance and computing power. Security technology development lags behind the rapid evolution of smart building systems, and traditional protection concepts are too static and singular, with intrusion detection relying on expert experience, making it difficult to adapt to dynamic and complex attack environments.

[0053] To address the aforementioned issues, this application designs and provides a smart building system 10.

[0054] In some embodiments of this application, from a functional perspective, such as Figure 1 As shown, the subsystems include, but are not limited to, one or more of the following: lighting subsystem 101, air conditioning subsystem 102, ventilation subsystem 103, water supply and drainage subsystem 104, elevator subsystem 105, fire protection subsystem 106, and power supply subsystem 107. The subsystems may also include subsystems that perform other functions.

[0055] The lighting subsystem 101 is used for lighting control inside and outside the building, and achieves an energy-saving and comfortable lighting environment by intelligently adjusting the brightness and on / off status of the lights.

[0056] The air conditioning subsystem 102 is used to regulate the temperature and humidity of the air inside the building to ensure the comfort of the indoor environment.

[0057] The ventilation subsystem 103 is used for air circulation and exchange within the building to ensure that the indoor air quality reaches an ideal state.

[0058] Water supply and drainage systems are used to control the water supply and drainage within buildings, enabling the monitoring and automatic control of water flow.

[0059] The elevator subsystem 105 is used for scheduling and managing elevators and escalators to ensure safe operation.

[0060] The fire protection subsystem 106 is used to detect fires in the building and to perform alarm and fire suppression control.

[0061] The power supply subsystem 107 is used to distribute the power within the building, ensuring a stable and secure power supply.

[0062] In some embodiments of this application, the network connecting the various subsystems can be a wired network, a wireless network, or a combination of wired and wireless networks. The wireless network includes, but is not limited to, mature wireless networks in the prior art such as TCP / IP, WiFi, Zigbee, LoRa, NB-IoT, and Bluetooth.

[0063] like Figure 2 As shown, from a system architecture perspective, the smart building system 10 includes a device terminal 113, an edge terminal 112, and a cloud terminal 111. The device terminal 113, edge terminal 112, and cloud terminal 111 constitute a three-layer architecture. At each layer, multiple subsystems coexist and work collaboratively. Each subsystem is distributed across the device terminal 113, edge terminal 112, and cloud terminal 111, achieving data sharing and system control through the three-layer architecture and pre-configured communication mechanisms.

[0064] Device 113 is configured to operate as a perception layer. Device 113 is the layer in the smart building system 10 that is closest to the physical environment, used to detect the building's environmental and operational status, and execute control commands from the edge device 112 or the cloud 111. The hardware of device 113 is distributed throughout various areas of the building, directly interacting with the physical environment and converting the state of the physical environment into digital signals (e.g., sensors); it also converts digital control commands into physical actions (e.g., execution units).

[0065] For example, such as Figure 2 and Figure 3 As shown, from a hardware perspective, device 113 includes multiple device nodes 116; device nodes 116 include sensors 1161, execution units 1162, and field control devices 1163. Sensors 1161 include, but are not limited to, various types of sensors such as temperature sensors, humidity sensors, light sensors, smoke detectors, and current sensors; execution units 1162 include, but are not limited to, devices such as lighting fixtures, compressors, fans, electronic expansion valves, and sprinkler valves that perform switching, adjustment, start-up, or stop actions based on control signals; field control devices 1163 are hardware components that directly interact with sensors and execution units to achieve real-time monitoring and control of execution units.

[0066] The field control device 1163 includes components such as a processor, memory, input / output interface, communication interface, and power supply module; the processor is the core computing unit of the field control device 1163, and can be a microcontroller (MCU), digital signal processor (DSP), field programmable gate array (FPGA), or industrial-grade CPU; the memory can be read-only memory, random access memory, and non-volatile memory; the communication interface includes serial port, wireless communication module, industrial bus interface, etc.

[0067] For example, the field control device 1163 can be a wired controller for an indoor air conditioning unit; the wired controller is connected to the indoor air conditioning unit, receives user operation commands (such as power on / off, temperature setting, and fan speed adjustment), and drives the actuator (such as the fan) in the indoor unit to perform specific control actions.

[0068] Edge device 112, located between device 113 and cloud 111, is used to process, filter, and analyze data from device 113. Edge device 112 can alleviate the computing and storage pressure on cloud 111, enabling real-time control and rapid response. The hardware of edge device 112 is deployed in a computer room, equipment room, power distribution room, or dedicated control cabinet within a building. These areas ensure the stable operation of the hardware of edge device 112, avoid adverse effects from the external environment, and facilitate network connection with the hardware devices of device 113.

[0069] From a hardware perspective, edge device 112 includes multiple edge nodes 115. Edge nodes 115 include edge control devices 1151 and edge gateways. Edge control devices 1151 are used for real-time processing, analysis, and decision-making on data collected from the field. Edge control devices 1151 can not only perform control functions, such as turning execution units on or off, adjusting the parameters of execution units, and monitoring the status of execution units; they can also filter data, aggregate data, detect anomalies, and perform predictive analysis. Edge control devices 1151 support multiple communication protocols. On the one hand, they can communicate with sensors, execution units, and field control devices 1163 at the device end 113; on the other hand, they can upload processed data to the cloud 111 via a network channel, enabling bidirectional data flow and remote management.

[0070] The edge control device 1151 includes components such as a processor, memory, input / output interface, communication interface, and power module; wherein the processor is an industrial-grade processor or system-on-a-chip that supports multi-tasking and edge computing capabilities; the memory can be read-only memory, random access memory, non-volatile memory, and / or solid-state memory; the communication interface includes serial port, wireless communication module, industrial bus interface, etc.

[0071] For example, the edge control device 1151 can be a centralized controller supporting multiple communication protocols in an air conditioner, or it can be a combination of a centralized controller and an edge gateway. The centralized controller is configured to manage multiple indoor units and / or multiple outdoor units, aggregating data from multiple sensors to achieve centralized monitoring and coordinated control of the air conditioning system; the centralized controller is also configured to process real-time data, perform fault diagnosis and response. The edge gateway is configured to support multiple communication protocols to achieve communication connections with other subsystems.

[0072] Cloud 111 is the core of the smart building system 10, configured to store global data, perform in-depth analysis of the global data, and make intelligent decisions. The hardware of Cloud 111 is deployed on the infrastructure of a data center or cloud server provider. Cloud 111 manages the operational status of multiple subsystems within the building by receiving and aggregating large amounts of data from the edge device 112. Cloud 111 is configured to enable human-machine interaction, allowing administrators to remotely access it, view building operational status, adjust parameters, and respond to alarm events.

[0073] From a hardware perspective, cloud 111 includes cloud node 114; cloud node 114 includes cloud control device 1141. Cloud control device 1141 includes a high-performance processor and a large-capacity storage system, supporting large-scale parallel computing. Cloud control device 1141 can be, for example, a data center server, virtual machine, physical server, etc.

[0074] like Figure 4 As shown, from a software perspective, cloud node 114 also includes cloud platform 1142, which is deployed on cloud control device 1141. Cloud platform 1142 includes: service clusters, configuration editor, and other service modules; among which, the service cluster is a service cluster responsible for core business processing, configured to perform core business processing, such as data storage, data analysis, and user access control; the configuration editor is a visual configuration and design tool that allows users to configure the system through a graphical interface.

[0075] Cloud node 114 can interact with user terminal 117. User terminal 117 (e.g., smartphone or computer) initiates operation requests (e.g., checking building equipment status, adjusting air conditioning room temperature settings, etc.) through an APP or web application interface on a web browser; based on the HTTP protocol, the APP or web application sends the user's operation request to cloud node 114 via the Internet; after receiving the user's operation request, cloud node 114 performs authentication and permission verification; after passing authentication and permission verification, according to the content of the user's operation request, it calls the corresponding business logic module, accesses the database or data of edge control device 1151, and executes the operation corresponding to the user's operation request; cloud node 114 controls the device to encapsulate the processing result into response data, returns it to user terminal 117, and parses and displays it to the user through the APP or web application.

[0076] From a software perspective, the service integration platform 1152 is deployed on the edge control device 1151. The service integration platform 1152 can be based on open-source edge computing frameworks such as EdgeX Foundry. It includes business services for the edge device 112, a configuration editor, and subsystem services (e.g., the air conditioning subsystem 102 service). The business services for the edge device 112 include functional modules such as execution unit management, data preprocessing, real-time monitoring, alarm handling, and energy consumption management. The configuration editor is configured to allow maintenance personnel to control device parameters, design control logic, and build monitoring interfaces. Subsystem services are dedicated to the control and management of different subsystems on the edge device 112. For example, the air conditioning subsystem 102 service is used to monitor the status of air conditioning equipment, implement temperature conditions, perform fault diagnosis, and optimize energy consumption.

[0077] Edge device 112 can interact with web services. Edge control device 1151 uploads the collected and processed data to the cloud web service, centrally storing the data in the cloud 111; or maintenance personnel access edge control device 1151 through a web application to configure parameters, monitor status, and diagnose faults; or the cloud web service calls the API interface exposed by edge control device 1151 to send control commands to edge control device 1151, edge control device 1151 parses the command content and forwards the command to the subsystem service, the subsystem service drives device 113 to execute according to the received command; after executing the command, edge control device 1151 updates the status and uploads it to the cloud web service through the API interface.

[0078] like Figure 5As shown, the smart building system 10 provided in some embodiments of this application also includes a data acquisition unit 121, a first intrusion detection unit 122, a first response unit 123, a second intrusion detection unit 124, and a second response unit 125.

[0079] The acquisition unit 121 is configured to acquire multidimensional data from the device end node 116, the edge end node 115, and the cloud node 114, and generate multidimensional data feature vectors based on the multidimensional data.

[0080] In some embodiments of this application, multidimensional data includes user behavior data.

[0081] Taking the air conditioning subsystem 102 as an example, user behavior data includes user operation data; user operation data includes: operation data of users operating the air conditioner through the field control device 1163 (such as a wired controller, remote controller or control panel); operation data of maintenance personnel configuring the air conditioner (such as modifying compressor configuration parameters) through the edge control device 1151; or operation data of users remotely controlling the air conditioner through APP or Web application.

[0082] For example, the process of device node 116 collecting user behavior data includes: device node 116 collecting operation events and generating event logs (including parameters such as event type, timestamp, device ID, and operation parameters); device node 116 sending the collected user behavior data to edge node 115.

[0083] For example, the process of edge node 115 collecting user behavior data includes: edge node 115 collecting operation and maintenance events and generating operation and maintenance logs (such as event type, timestamp, device ID, operation parameters, etc.); edge node 115 uploading the user behavior data collected locally and the user behavior data collected by device node 116 to cloud node 114, and storing it in the local database at the same time.

[0084] For example, the process of cloud node 114 collecting user behavior data includes: cloud node 114 receiving user behavior data from user terminal 117, such as collecting it through SDK and sending it to a cloud web service; cloud node 114 receiving and storing user behavior data from edge node 115 and user terminal 117. Cloud node 114 can also push user behavior data to edge node 115 in real time or periodically. For example, cloud node 114 can filter user behavior data related to edge node 115 based on device ID, user ID, or other identifiers of edge node 115, and push or synchronize the filtered data to edge node 115. Cloud node 114 and edge node 115 adopt a two-level storage mechanism. Edge node 115 is responsible for short-term caching and fast response, while cloud node 114 is responsible for long-term storage and in-depth analysis. In this way, even if cloud node 114 is temporarily unavailable, edge node 115 can still save and process some user behavior data.

[0085] In some embodiments of this application, user behavior data includes, but is not limited to, one or more of the following: user information (e.g., one or more of the following: user ID, account type, and permissions), operation details (e.g., one or more of the following: operation type, operation parameters, and operation result), time information (e.g., one or more of the following: access time and operation time), geographical location (e.g., one or more of the following: GPS coordinates and IP address), and operation frequency and abnormal information (e.g., one or more of the following: error code and alarm identifier).

[0086] In some embodiments of this application, multidimensional data includes device behavior data.

[0087] In some embodiments of this application, device behavior data includes, but is not limited to, one or more of the following: execution unit status (e.g., one or more of the following: power on, power off, operating mode, temperature, operating time, and energy consumption), execution unit events (e.g., one or more of the following: alarms, errors, and maintenance records), sensor data, sensor status, sensor events (e.g., one or more of the following: alarms, errors, and maintenance records), operation logs (e.g., one or more of the following: operation logs of field control device 1163, operation logs of edge control device 1151, and operation logs of cloud control device 1141), and configuration logs (e.g., [example missing]). This may include one or more of the following: configuration logs of the field control device 1163, configuration logs of the edge control device 1151, and configuration logs of the cloud control device 1141; control command types (e.g., one or more of the following: control command types of the field control device 1163, control command types of the edge control device 1151, and control command types of the cloud control device 1141); and control command frequencies (e.g., one or more of the following: control command frequencies of the field control device 1163, control command frequencies of the edge control device 1151, and control command frequencies of the cloud control device 1141).

[0088] The process of device behavior data collection by device node 116 includes: device node 116 collects device behavior data and sends the collected device behavior data to edge node 115; edge node 115 receives the data from device node 116 and aggregates it, and uploads the aggregated device behavior data to cloud node 114 periodically or in real time, and can store the device behavior data locally to achieve two-level storage; cloud node 114 receives and stores the device behavior data from edge node 115.

[0089] In some embodiments of this application, multidimensional data includes network traffic data.

[0090] In some embodiments of this application, network traffic data includes network traffic data for external connections, which include, but are not limited to, software updates of sensors, execution units, and controllers, remote operation and maintenance access, and communications generated when users remotely control devices through an APP or Web interface.

[0091] In some embodiments of this application, in addition to network traffic data for external connections, network traffic data also includes network traffic data for internal connections, such as bidirectional data transmission between edge node 115 as a local data aggregation point and cloud node 114.

[0092] In some embodiments of this application, network traffic data includes, but is not limited to, one or more of the following: source IP / domain name, destination IP / domain name, port number, communication protocol type, traffic size, traffic direction, connection duration, connection status (e.g., including one or more of the following: number of connections, success, failure, abnormal disconnection), session ID, connection timestamp, abnormal communication identifier, and error communication identifier.

[0093] For example, traffic information is collected through the network monitoring interfaces of device node 116, edge node 115, and cloud node 114; the network traffic data collected by device node 116 is sent to edge node 115, edge node 115 receives the data from device node 116 and aggregates it, and uploads the aggregated network traffic data to cloud node 114 periodically or in real time, and can store the network traffic data locally to achieve two-level storage; cloud node 114 receives the network traffic data from edge node 115 and stores it.

[0094] On the one hand, the processing power and storage resources of edge node 115 and cloud node 114 are limited; on the other hand, the data change frequencies are different. For example, device behavior data usually changes discretely and is time-driven, with inconsistent operation intervals; while network traffic data changes relatively continuously, and the traffic volume, connection status, etc., may fluctuate in real time. In order to reasonably allocate computing and storage resources and avoid system overload, in some embodiments of this application, the acquisition unit 121 is configured to acquire device behavior data according to a first preset period and network traffic data (such as...) according to a second preset period. Figure 6 (as shown in step S101).

[0095] However, the different first and second set periods, as well as the random occurrence of user behavior, result in user behavior data being generated randomly. This causes user behavior data, device behavior data, and network traffic data to be misaligned on the timeline, making it difficult to directly integrate and analyze the data.

[0096] To solve this problem, the acquisition unit 121 is also configured to perform actions such as... Figure 6 Several other steps are shown:

[0097] Step S102: Using a preset time window as the unit, slice the user behavior data, the device behavior data, and the network traffic data into data blocks according to the time period corresponding to the time window.

[0098] This involves slicing user behavior data, device behavior data, and network traffic data into data blocks of the same time period, using a fixed-length time window as the unit.

[0099] The length of the time window is determined by the needs of anomaly detection, ensuring sufficient granularity while avoiding excessively sparse data within the window; for example, 1 minute.

[0100] Step S103: Calculate the average value of the device behavior data within each time window, and use it as the device behavior window feature.

[0101] Step S104: Calculate the average value of network traffic data within each time window, as the network traffic window feature.

[0102] Step S105: Determine whether user behavior data exists within the time window.

[0103] Step S106: If user behavior data exists within the time window, count the number of times the user behavior data occurs within the time window, and use this count as a feature of the user behavior window. The number of times the user behavior data occurs within the time window is converted into a numerical feature.

[0104] Step S107: If there is no user behavior data within the time window, interpolation is used to fill in the missing values ​​to generate user behavior window features.

[0105] For example, device behavior data is sampled every 10 seconds (first set period is 10 seconds), and network traffic data is sampled every 15 seconds (second set period is 15 seconds). After resampling using the above method, both device behavior data and network traffic data become an average value every minute, which is used as the device behavior window feature and network traffic window feature, respectively. User behavior data is not collected at a fixed period. By statistically analyzing the number of times user behavior data occurs within a time window and converting it into a numerical feature, the original irregular time is transformed into a numerical feature with a fixed time interval as the user behavior window feature. This adjusts data of different frequencies and time points to a unified window feature for each time window, thus completing time sequence alignment.

[0106] When user behavior data cannot be collected within a certain time window due to the lack of user activity, the user behavior window features are missing. Interpolation is used to fill in the missing user behavior window features to ensure that each time window has complete feature values.

[0107] Linear interpolation can be used to fill in missing data, ensuring data continuity and completeness.

[0108] The user behavior window features, device behavior window features, and network traffic window features are concatenated along the feature dimension to form a unified multidimensional data feature vector.

[0109] By setting a unified time window to slice user behavior data, device behavior data, and network traffic data, and adopting different resampling strategies for different types of data, unified alignment and fusion of multi-source time-series data are achieved. This solves the problems of asynchronous user behavior data, device behavior data, and network traffic data and differences in sampling frequency, and provides standardized and complete input features for subsequent identification, detection, and processing.

[0110] like Figure 5 As shown, the intelligent building system 10 also includes a first intrusion detection unit 122. The first intrusion detection unit 122 is deployed at the edge 112 and is configured to identify anomalies based on a multi-dimensional data feature vector using a first intrusion detection model, and trigger an alarm when an anomaly is detected. The first intrusion detection model is constructed based on the first intrusion detection model.

[0111] The first intrusion detection unit 122 is configured to perform, for example... Figure 7 The following steps are shown:

[0112] Step S201: Generate time window features based on multidimensional data feature vectors.

[0113] The length of the time window for generating time window features based on multidimensional data feature vectors can be set according to actual needs, and will not be further limited here.

[0114] Step S202: Using the time window features as input to the first intrusion detection model, the first intrusion detection model outputs the anomaly probability corresponding to the current time window features.

[0115] Step S203: Determine whether the anomaly probability exceeds the preset safety threshold.

[0116] Step S204: If the output anomaly probability exceeds the preset security threshold, the first intrusion detection model determines that the current time window features are abnormal, and the first intrusion detection unit 122 triggers an alarm.

[0117] If the output anomaly probability does not exceed the preset security threshold, then intrusion detection of time window features is maintained.

[0118] As shown in the figure, the intelligent building system 10 also includes a first response unit 123, which is deployed at the edge terminal 112. After the first intrusion detection model determines that the current time window features are abnormal and the first intrusion detection unit 122 triggers an alarm, the first response unit 123 is configured to filter out abnormal nodes based on traffic metadata within the corresponding time window and isolate the abnormal nodes (e.g., Figure 7 (as shown in step S205).

[0119] Traffic metadata includes, but is not limited to, one or more of the following: source IP address, destination IP address, port number, communication protocol type, connection duration, traffic volume, connection timestamp, and device ID. The first response unit 123 filters out abnormal nodes based on a preset rule engine. The rule engine can use pre-set rules such as statistical analysis, threshold filtering, cluster analysis, correlation analysis, and priority ranking to filter out abnormal nodes, which will not be elaborated upon here.

[0120] For example, edge node 115 can temporarily block the communication connection between the abnormal node (which may also include its neighboring nodes) and the local network through firewall rules or access control lists, or add the abnormal node (which may also include its neighboring nodes) to a temporary blacklist to prevent the attack from spreading vertically or horizontally.

[0121] The first intrusion detection model can be an RNN model, such as Vanilla RNN or LSTM. The first intrusion detection model can be trained offline: it is trained offline using historical data, and the trained first intrusion detection model is deployed on the edge for online inference.

[0122] like Figure 5 As shown, the intelligent building system 10 also includes a second intrusion detection unit 124. The second intrusion detection unit 124 is deployed in the cloud 111 and is configured to respond to alarms triggered by the first intrusion detection unit 122, and to locate abnormal nodes and / or abnormal edges based on a second intrusion detection model. The second intrusion detection model is constructed based on a graph neural network.

[0123] In some embodiments of this application, the second intrusion detection model is built on the GCN architecture.

[0124] The second intrusion detection model based on the GCN architecture uses a message passing mechanism to exchange information between nodes. It analyzes the relationships between nodes and edges by aggregating and transmitting node and edge features. Abnormal node features propagate along edges, while each central node collects node and edge features from its neighbors, updating its own state or representation. After multiple rounds of information transmission and aggregation, the central node's representation contains comprehensive information about itself and its neighbors. The GCN-based graph neural network model can then use these representations to progressively determine whether nodes exhibit abnormal behavior or whether edges have abnormal connections. The GCN-based graph neural network model classifies or scores each node or edge to indicate whether it is abnormal, thereby locating abnormal nodes or edges.

[0125] The node-wise representation of the graph convolutional layer in the second intrusion detection model based on the GCN architecture is shown in the following equation:

[0126]

[0127] in, It is the feature representation of the central node v at the k-th layer. It is the set of neighboring nodes of the central node v. W is the feature representation of neighbor node u at the (k-1)th layer. (k) is the weight matrix of the k-th layer, and are the trainable parameters; c vu σ is the normalization coefficient used to balance the influence of the number of neighbors of different nodes, and σ is a non-linear activation function, such as ReLU. For example, layer 0 is the original input, and each subsequent layer calculates a new central node feature representation based on the features of the current central node and its neighboring nodes through aggregation, linear transformation, and activation functions. That is, the central node feature of layer 1 is obtained by performing a graph convolution operation based on the features of layer 0; the central node feature of layer 2 is obtained by performing another graph convolution operation based on the node features of layer 1, and so on.

[0128] In some embodiments of this application, the second intrusion detection model is built on the GraphSAGE architecture.

[0129] The second intrusion detection model built on the GraphSAGE architecture performs intrusion detection by sampling and aggregating the feature information of neighboring nodes, enabling information exchange between nodes. Each central node randomly samples a portion of its neighboring nodes, uses an aggregation function to summarize the features of the sampled neighboring nodes, and then combines the aggregation result with its own features to update the node's representation. Through multiple rounds of information sampling, aggregation, and updating, the representation of the central node not only includes its own features but also incorporates the local structure and behavioral information of its neighbors. Based on this comprehensive representation, the second intrusion detection model built on the GraphSAGE architecture can effectively capture abnormal patterns of nodes or edges, classify or score nodes or edges, and identify abnormal behaviors or connections.

[0130] The node-wise representation of the graph convolutional layer of the second intrusion detection model based on the GraphSAGE architecture is shown in the following equation:

[0131]

[0132] in, It is the feature representation of the current central node v at the k-th layer. It is the feature representation of the current central node v at the (k-1)th layer. It is a subset of neighboring nodes sampled from the current central node v, AGGREGATE k It is the neighbor aggregation function of the k-th layer, CONCAT is the concatenation function, and W is the neighbor aggregation function of the k-th layer. (k)σ is the weight matrix of the k-th layer, and σ is a non-linear activation function, such as ReLU. The aggregation function can be mean aggregation, LSTM aggregation, pooling aggregation, or other aggregation forms.

[0133] In some embodiments of this application, the second intrusion detection model is built on an enhanced GraphSAGE architecture, in which an attention mechanism is introduced.

[0134] The second intrusion detection model, built upon the enhanced GraphSAGE architecture, performs intrusion detection by sampling neighboring nodes and calculating attention weights between them and the current central node, thus achieving weighted aggregation of neighboring node features. The current central node randomly samples a subset of neighboring nodes, dynamically assigns importance weights to these nodes using an attention mechanism, weights and aggregates the sampled neighboring node features, and then combines the weighted aggregation result with its own features to update the representation of the current central node. Through multiple rounds of information sampling, weighted aggregation, and updating, the representation of the current central node not only includes its own features but also incorporates local structural and behavioral information from neighboring nodes, while also emphasizing attention to key neighboring nodes.

[0135] The node-wise representation of the graph convolutional layer of the second intrusion detection model built on the enhanced GraphSAGE architecture is shown in the following equation:

[0136]

[0137] The following method can be used to calculate, taking LeakyReLU activation as an example:

[0138]

[0139] Where v is the current central node; u and w are the neighboring nodes of the current central node; W is a subset of neighboring nodes sampled from the central node v. (k) Let a be the weight matrix of the k-th layer. (k) These are the weight vectors in the attention mechanism, and CONCAT is the vector concatenation function. It is the normalized attention weight of the current central node v to its neighbor node u at the k-th layer, satisfying The attention scores between the central node v and its neighbor node u at layer k. Let V represent the sum of attention score exponents of all neighbor nodes in the subset of neighbor nodes sampled from the central node v.

[0140] The second intrusion detection model, built on the enhanced GraphSAGE architecture, is sensitive to abnormal propagation between nodes when performing intrusion detection.

[0141] The following is for reference only. Figure 8 The second intrusion detection unit 124 is configured to perform the following steps to construct a second intrusion detection model.

[0142] Step S301: Generate a global graph node set of device-side node 116, edge-side node 115 and cloud-side node 114, determine node features based on multi-dimensional data feature vectors, and establish a node feature matrix based on node feature vectors.

[0143] The key features that best reflect the node's attributes and behavior from the multidimensional data feature vector are used as node features, where each node represents an entity in the smart building system 10. For example, node features include: device node features, user node features, and service node features; where device node features include execution unit status (including temperature, running time, and energy consumption) and instruction type; user node features include: number of user actions; and service node features include: traffic volume and connection status (including the number of connections).

[0144] Step S302: Generate the edge set of the global graph, determine the edge features based on the multidimensional data feature vector, and establish the edge feature matrix based on the edge feature vector.

[0145] Edge features are derived from multidimensional data feature vectors and represent the relationships between nodes. These edge features depict the interactions and attributes between nodes. Examples of edge features include: control relationships, communication frequency, dependencies, and abnormal communication identifiers. The control relationship includes control command type and control command frequency; communication frequency includes data exchange frequency or connection duration; where data exchange frequency is the number of times a connection is established between the source node and the target node per unit time, reflecting the activity level of communication; connection duration is the duration of the connection, which can be average duration, maximum duration, or total duration, reflecting the stability and continuity of the connection; dependency is specifically determined based on data exchange frequency and duration to determine whether there is a dependency between nodes; when the data exchange frequency is higher than a preset data exchange frequency threshold and the average connection duration exceeds a preset average connection duration threshold, it is determined that there is a dependency between nodes; or it is determined based on topology to determine whether there is a dependency between nodes, for example, sensor data is aggregated through edge node 115, and edge node 115 is then uploaded to cloud node 114, the sensor, edge node 115, and cloud node 114 have a dependency, forming a chain dependency; abnormal communication identifiers can be directly used to analyze abnormal traffic patterns and identify abnormal dependencies or potential attack paths.

[0146] For categorical data, such as control command types, one-hot encoding or feature representation (Embedding) can be used to convert them into numerical values.

[0147] Step S303: Construct an adjacency matrix to form the topology of the global graph.

[0148] Step S304: Encapsulate the adjacency matrix, node feature matrix, and edge feature matrix.

[0149] The adjacency matrix, node feature matrix, and edge feature matrix are each encapsulated into the format required by the graph neural network framework and used as model input. For example, the adjacency matrix is ​​converted into a sparse matrix format to save storage space; the node feature matrix is ​​converted into a two-dimensional tensor, and the edge feature matrix is ​​converted into a two-dimensional tensor.

[0150] Step S305: Construct a second intrusion detection model based on GCN, GraphSAGE, or an enhanced GraphSAGE architecture. The second intrusion detection model takes node features and edge features as input and outputs the anomaly probability or classification result of each node or edge.

[0151] Further, a second intrusion detection model is trained based on supervised learning, and the trained second intrusion detection model is deployed on the cloud. The specific steps include:

[0152] Step S306: Divide the node features and edge features into training set, validation set and test set, and generate node anomaly labels and edge anomaly labels for supervised learning.

[0153] Step S307: Define the loss function and optimizer.

[0154] For example, a focus loss can be introduced into the cross-entropy loss function by using the cross-entropy loss function as the loss function and multiplying the cross-entropy loss by a weight factor based on the prediction probability.

[0155] Intrusion events constitute a very small percentage of overall network traffic, resulting in a far smaller number of anomalous samples compared to normal samples. Furthermore, the diversity of anomalous samples makes it difficult to learn effective anomalous features. If the cross-entropy loss is contributed by a large number of normal samples, the influence of anomalous samples will be diluted, making the secondary intrusion detection model prone to predicting normal samples and leading to false negatives. By introducing a focusing loss function, an adjustment factor is introduced on top of the cross-entropy loss function. This multiplies the cross-entropy loss with a weighting factor based on the predicted probability, dynamically adjusting the sample weights. For majority class samples that the secondary intrusion detection model can classify correctly, the focusing loss reduces the loss contribution, preventing it from dominating gradient updates. For minority class samples that are difficult to classify, the focusing loss amplifies the loss contribution, alleviating the class imbalance problem. In this way, cross-entropy ensures the basic probability fitting ability, while the focusing loss improves the sensitivity to anomalous samples, thus increasing the accuracy of intrusion detection.

[0156] For example, Adam can be used as the optimizer to initialize the parameters of the second intrusion detection model.

[0157] Step S308: Input the node feature matrix, edge feature matrix, and adjacency matrix, perform forward propagation, calculate the anomaly probability of nodes and edges respectively, calculate the loss function, and backpropagate to update the parameters of the second intrusion detection model; after multiple rounds of iterative training, optimize the anomaly detection performance of the second intrusion detection model for nodes and edges.

[0158] Step S309: Validate and tune the second intrusion detection model on the validation set, and test and evaluate the second intrusion detection model on the test set.

[0159] Step S310: Deploy the second intrusion detection model in the cloud 111. The second intrusion detection unit 124 responds to the alarm triggered by the first intrusion detection unit 122 and locates abnormal nodes and / or abnormal edges based on the second intrusion detection model.

[0160] As shown in the figure, the intelligent building system 10 also includes a second response unit 125, which is deployed in the cloud 111. After the second intrusion detection model locates abnormal nodes and / or abnormal edges, the second response unit 125 extracts the subgraph formed by the abnormal nodes and / or abnormal edges, and establishes an abnormal link set or abnormal link set based on the subgraph formed by the abnormal nodes and / or abnormal edges; the second response unit 125 further generates and outputs protection commands to block the abnormal links, especially blocking their communication paths in the cloud 111, the edge terminal 112, and the device terminal 113.

[0161] In some embodiments of this application, the smart building system 10 also includes a communication unit. The communication unit is deployed at the edge 112. After the first intrusion detection model determines that the current time window features are abnormal and triggers an alarm, the first response unit 123 is configured to filter out abnormal nodes based on the traffic metadata within the corresponding time window; the communication unit uploads the abnormal nodes and their paths to the cloud 111.

[0162] The following describes the construction process of the second intrusion detection model in some other embodiments of this application.

[0163] Step S401: Generate a set of nodes for the abnormal subgraph, determine the features of the abnormal nodes based on the multidimensional data feature vector, and establish an abnormal node feature matrix based on the abnormal node feature vector.

[0164] Step S402: Generate the edge set of the abnormal subgraph, determine the abnormal edge features based on the multidimensional data feature vector, and establish the edge feature matrix based on the abnormal edge feature vector.

[0165] Step S403: Construct an adjacency matrix to form the topological structure of the abnormal subgraph.

[0166] Step S404: Encapsulate the adjacency matrix, node feature matrix, and edge feature matrix.

[0167] Step S405: Construct a second intrusion detection model based on a graph autoencoder or a variational graph autoencoder.

[0168] Step S406: Train the second intrusion detection model with the goal of minimizing the adjacency matrix reconstruction error.

[0169] Step S407: Identify the abnormal link set using the trained second intrusion detection model. The second response unit 125 generates and outputs a protection command to block the communication path of any abnormal link in the abnormal link set between the cloud 111, the edge terminal 112, and the device terminal 113.

[0170] The abnormal link further identifies the propagation path of the intrusion behavior in the smart building system 10. Based on the identification and analysis of the overall abnormal pattern, it can reflect the timing and direction of the intrusion behavior and effectively track and locate it.

[0171] Since the size of the anomaly subgraph is much smaller than that of the whole graph, the computational resource consumption is significantly reduced, making it particularly suitable for real-time or near-real-time anomaly detection needs. The anomaly subgraph focuses on the identified anomaly nodes and edges, avoiding interference from a large number of normal nodes in the whole graph, making unsupervised learning more feasible and accurate.

[0172] like Figure 9 As shown, in some embodiments of this application, the smart building system 10 further includes a partitioning unit 126. The partitioning unit 126 is deployed at the edge 112 and / or the cloud 111. The partitioning unit 126 is configured to divide the smart building system 10 into multiple logical areas, set access control rules for each logical area, and only authorize entities whose identities have been verified by the access control rules to access the corresponding logical area.

[0173] Here, "subject" refers to an entity capable of actively initiating operations or requests, including but not limited to users, processes, programs, or devices—that is, the object performing the operation. The partition 126 can be implemented using a configuration editor in the cloud 111 or at the edge 112. For example, it can divide multiple logical areas according to different functions, locations, or security levels, configure access permissions and access control policies, and establish a mapping relationship between logical areas and subjects. The configuration editor can also manage the identity information and permissions of entities and display the logical areas and permission allocation through a graphical interface.

[0174] like Figure 10 As shown, in some embodiments of this application, the smart building system 10 further includes an access unit 127. The access unit 127 is deployed at the edge terminal 112 and is used to verify the permissions of the device terminal node 116 based on its unique identity, and to allow the authenticated device terminal node 116 to establish a connection.

[0175] The unique identifier of device node 116 can be a MAC address, device serial number, or digital certificate, used for device node 116 identification and authentication. When device node 116 establishes a communication connection for the first time, it initiates a "registration" or "handshake" with edge node 115, sending its unique identifier. Edge node 115 is pre-configured with a list of device nodes 116 allowed to access the network. This list contains the unique identifiers of all authorized and trusted device nodes 116. If device node 116's authentication is successful, edge node 115 allows it to access the network; if authentication fails, edge node 115 denies it access, preventing unauthorized device nodes 116 from entering the network.

[0176] By implementing identity authentication and access control for device end nodes 116 through configuration technology, it is ensured that only trusted devices exist in the network of the smart building system 10, preventing security risks caused by unauthorized device end nodes 116 accessing the network.

[0177] For example, the smart building system 10 includes an air conditioning subsystem 102 and a lighting subsystem 101. Taking the air conditioning subsystem 102 being subjected to DoS as an example, the protection process of the smart building system 10 is explained.

[0178] The acquisition unit 121 acquires multidimensional data from the device end node 116, the edge end node 115, and the cloud node 114, and generates multidimensional data feature vectors based on the multidimensional data.

[0179] The first intrusion detection unit 122 identifies anomalies based on multi-dimensional data feature vectors and uses a first intrusion detection model, and triggers an alarm when an anomaly is identified.

[0180] For example, the first intrusion detection model can output the anomaly probability corresponding to the current time window. When the output anomaly probability exceeds the preset security threshold, it is considered that the smart building system 10 has an anomaly, such as the data packet frequency being much higher than normal, or the processor and storage unit running under continuous high load, etc. The first intrusion detection unit 122 triggers an alarm.

[0181] The first response unit 123 filters out abnormal nodes based on traffic metadata within the corresponding time window and isolates the abnormal nodes (such as an indoor unit or a wired controller) to reduce the risk of abnormal requests continuing to spread to other device nodes.

[0182] The second intrusion detection unit 124 responds to the alarm triggered by the first intrusion detection unit 122, performs anomaly detection on nodes and edges based on the second intrusion detection model, and locates abnormal nodes and / or abnormal edges.

[0183] The second response unit 125 extracts the subgraph formed by abnormal nodes and / or abnormal edges, establishes abnormal links or sets of abnormal links, determines which nodes have been affected by the attack, and clearly determines the affected range through the set of abnormal links; the second response unit 125 generates and outputs protection instructions to block the communication path of any abnormal link in the set of abnormal links in the cloud 111, edge terminal 112 and device terminal 113.

[0184] For example, immediately implement network isolation for nodes clearly under attack, disconnect from the abnormal communication source, remove the abnormal device node 116 from the list of allowed network access devices, and prevent the spread of abnormal data; activate protection mode for neighboring nodes (not severely affected but potentially dangerous), lower the communication threshold and activate protection mechanisms. At the same time, activate backup control policies, enable backup nodes, and gradually restore normal operation of the affected area.

[0185] This application integrates the device 113, edge 112, and cloud 111 to form a complete closed-loop security system. Each layer can identify and block security threats at its corresponding level, and the overall protection effect is further enhanced through multi-layer linkage. The first and second intrusion detection models can dynamically identify unknown threats and dynamically update strategies, no longer relying solely on static rules, thus enhancing the ability to resist zero-day attacks and complex dangers. The rapid response mode of the edge 112 and the global training mode of the cloud 111 can ensure real-time requirements and obtain the support of big data analysis. It can deal with physical, network, privilege escalation, and distributed attacks, meeting the actual needs of smart building systems with dense devices, complex user roles, and diverse communication modes.

[0186] It should be noted that any one of the acquisition unit 121, the first intrusion detection unit 122, the first response unit 123, the second intrusion detection unit 124, the second response unit 125, the partitioning unit 126, the access unit 127, and the communication unit can be implemented by a processor running a program. The processor can be the processor in the field control device 1163, the processor in the edge control device 1151, the processor in the cloud control device 1141, or a combination of several of them.

[0187] In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0188] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A smart building system, characterized in that, include: The acquisition unit is configured to acquire multidimensional data from the device end, the edge end, and the cloud, and generate multidimensional data feature vectors based on the multidimensional data; The multidimensional data includes: user behavior data, device behavior data, and network traffic data; A first intrusion detection unit is deployed at the edge. The first intrusion detection unit is configured to identify anomalies based on the multidimensional data feature vector and a first intrusion detection model, and to trigger an alarm when an anomaly is identified. The first intrusion detection model is constructed based on a time series model. A first response unit, deployed at the edge, is configured to filter out abnormal nodes after the first intrusion detection unit triggers an alarm; and to isolate the abnormal nodes. A second intrusion detection unit, deployed in the cloud, is configured to respond to alarms triggered by the first intrusion detection unit, and locate abnormal nodes and / or abnormal edges based on a second intrusion detection model constructed based on a graph neural network; and The second response unit, deployed in the cloud, is configured to establish an abnormal link based on the subgraph formed by the abnormal nodes and / or abnormal edges; and to generate and output protection commands to block the abnormal link.

2. The intelligent building system according to claim 1, characterized in that: The user behavior data includes one or more of the following: User information, operation details, time information, geographical location, operation frequency, and exception information; in, The user information includes one or more of the following: user ID, account type, and account permissions; The operation details include one or more of the following: operation type, operation parameters, and operation result; The time information includes one or more of the following: access time and operation time; The geographic location includes one or more of the following: GPS coordinates and IP address; The abnormal information includes one or more of the following: error codes and alarm identifiers.

3. The intelligent building system according to claim 1, characterized in that: The device end includes: a device end node; the device end node includes: a sensor, an execution unit and / or a field control device; The edge end includes: an edge end node; the edge end node includes an edge end control device; The cloud includes: cloud nodes: the cloud nodes include cloud control devices; The device behavior data includes one or more of the following: Execution unit status, execution unit events, sensor data, sensor status, sensor time, operation log, configuration log, control command type, and control command frequency; The execution unit status includes one or more of the following: power on, power off, operating mode, temperature, operating time, and energy consumption; The execution unit events include one or more of the following: execution unit alarms, execution unit errors, and execution unit maintenance records; The operation log includes one or more of the following: the operation log of the field control device, the operation log of the edge control device, and the operation log of the cloud control device; The configuration log includes one or more of the following: the configuration log of the field control device, the configuration log of the edge control device, and the configuration log of the cloud control device; The control command type includes one or more of the following: the control command type of the field control device, the control command type of the edge control device, and the control command type of the cloud control device; The control command frequency includes one or more of the following: the control command frequency of the field control device, the control command frequency of the edge control device, and the control command frequency of the cloud control device.

4. The intelligent building system according to claim 1, characterized in that: The network traffic data includes one or more of the following: Source IP / domain name, destination IP / domain name, port number, communication protocol type, traffic size, traffic direction, connection duration, connection status, session ID, connection timestamp, abnormal communication identifier, and error communication identifier; The connection status includes one or more of the following: number of connections, success, failure, and abnormal disconnection.

5. The intelligent building system according to any one of claims 1 to 4, characterized in that: The acquisition unit is configured to perform the following steps to generate a multidimensional data feature vector based on the multidimensional data: The device behavior data is collected according to the first predetermined cycle; The network traffic data is collected according to the second predetermined cycle; Using a preset time window as a unit, the user behavior data, the device behavior data, and the network traffic data are sliced ​​into data blocks according to the time period corresponding to the time window; Calculate the average value of device behavior data within each time window, and use it as the feature of the device behavior window; Calculate the average value of network traffic data within each time window, and use it as the network traffic window feature; Determine if user behavior data exists within the time window; If user behavior data exists within the time window, the number of times the user behavior data occurs within the time window is used as the feature of the user behavior window. If there is no user behavior data within the time window, interpolation is used to fill in the missing values ​​to generate user behavior window features; The user behavior window features, device behavior window features, and network traffic window features are concatenated along the feature dimension to generate the multidimensional data feature vector.

6. The intelligent building system according to claim 5, characterized in that: The first intrusion detection unit is configured to perform the following steps: Time window features are generated based on the multidimensional data feature vectors; Using the time window features as input to the first intrusion detection model, the first intrusion detection model outputs the anomaly probability corresponding to the current time window features; When the output anomaly probability exceeds a preset security threshold, the first intrusion detection unit triggers an alarm.

7. The intelligent building system according to claim 6, characterized in that: The second intrusion detection unit is configured to perform the following steps to construct a second intrusion detection model: Generate a global graph node set of the device-side nodes, edge-side nodes, and cloud-side nodes; determine node features based on the multi-dimensional data feature vectors; and establish a node feature matrix based on the node feature vectors. Generate the edge set of the global graph, determine the edge features based on the multidimensional data feature vector, and establish the edge feature matrix based on the edge feature vector; Construct an adjacency matrix to form the topology of the global graph; A second intrusion detection model is constructed based on GCN, GraphSAGE, or an enhanced GraphSAGE architecture. The second intrusion detection model takes the node features and edge features as input and outputs the anomaly probability or classification result of each node or edge. The second intrusion detection model is trained using supervised learning, and the trained second intrusion detection model is deployed in the cloud.

8. The intelligent building system according to claim 7, characterized in that: When training the second intrusion detection model based on supervised learning, the cross-entropy loss function is used as the loss function, and the cross-entropy loss is multiplied by the weight factor based on the predicted probability.

9. The intelligent building system according to any one of claims 1 to 4, characterized in that, Also includes: The partitioning unit, deployed at the edge and / or in the cloud, is configured to divide into multiple logical regions, set access control rules for each logical region, and authorize only subjects whose identities have been verified by the access control rules to access the corresponding logical region.

10. The intelligent building system according to any one of claims 1 to 4, characterized in that, Also includes: An access unit, deployed at the edge, is configured to verify the permissions of device nodes based on their identity identifiers and allow authenticated device nodes to establish connections.