Hybrid architecture safe intelligent elevator control system based on domestic operating system

Abstract

The invention relates to the technical field of intelligent elevator control, and discloses a hybrid architecture safe intelligent elevator control system based on a domestic operating system, which adopts a layered modular design and is specifically divided into a hardware layer, an operating system layer, a functional layer and an application layer from bottom to top. According to the invention, the system is equipped with a ShiWu MLU220 edge AI chip, supports the real-time analysis of 8 paths of 1080P video streams, and achieves the cross-modal fusion of video streams and depth data in combination with an improved YOLOv8 target detection model and a PointNetV2 geometric feature extraction network. Through an environment perception illumination compensation algorithm, the target recognition accuracy in a low-illumination environment is improved to 98.5%; and a cross-modal attention mechanism is adopted to carry out feature weighted fusion, and the missing report rate of dangerous behavior detection is lower than 0.5%. By means of the technology, the elevator system has the capacity of recognizing 23 types of abnormal scenes such as electric vehicle entering, crowding and article missing in real time, the response time is shortened to 150 ms, and the processing delay is obviously better than that of 500 ms or above of a traditional elevator system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent elevator control, in particular to a hybrid architecture safe intelligent elevator control system based on a domestic operating system. BACKGROUND

[0002] In the current elevator market, foreign brands account for more than two-thirds, and their control systems generally use non-domestic systems, which have two major problems: 1. Information security risk The existing foreign elevator control system has hidden dangers such as unencrypted communication protocol, remote maintenance backdoor, and lack of domestic security certification. According to the 2023 safety report of China Elevator Association, 32% of elevator network security incidents are caused by control system vulnerabilities.

[0003] 2. Insufficient intelligence Traditional elevator systems lack remote monitoring functions, lack predictive maintenance capabilities, lack early warning and prevention functions such as electric vehicles and dangerous goods entering the elevator, and the average maintenance response time is more than 45 minutes, which cannot meet the needs of modern smart buildings.

[0004] Although there are some domestic elevator control systems, most of them have poor compatibility, low intelligence, and imperfect security protection. Therefore, there is an urgent need for an elevator control system that ensures information security and has high intelligence. SUMMARY

[0005] The purpose of the present application is to provide a hybrid architecture safe intelligent elevator control system based on a domestic operating system to solve the problems raised in the background.

[0006] To achieve the above purpose, the present application provides the following technical solution: a hybrid architecture safe intelligent elevator control system based on a domestic operating system, which adopts a layered modular design and is specifically divided into a hardware layer, an operating system layer, a function layer, and an application layer from bottom to top; The hardware layer includes a main control unit, a sensing system, and an edge AI computing unit; The operating system layer is based on the open-source Hongmeng 4.0 LTS version as the system base, and creates an elevator-specific system ElevOS through deep customization; The function layer integrates a safety control module, a communication gateway, an AI inference engine, and a predictive maintenance engine; The application layer provides voice-controlled elevator service; each layer realizes cross-device collaboration through distributed soft bus technology; The main control unit uses a domestic high-performance chip and carries a TEE security module; The sensing system integrates a multimodal sensor array, including a high-precision vibration sensor, a triaxial accelerometer, a temperature sensor array, and a current and voltage monitoring module. The edge AI computing unit is equipped with the Cambricon MLU220 edge AI chip, and the operating system layer and the functional layer achieve resource scheduling through the distributed capabilities of the HarmonyOS kernel. The safety control module implements an ISO 13849 PLe level safety loop; The communication gateway supports conversion of multiple industrial protocols; The AI ​​inference engine integrates the lightweight TensorFlow Lite-Micro framework; The predictive maintenance engine is built on LSTM and Transformer models, and the application layer achieves cross-device interaction through HarmonyOS atomic services.

[0007] Preferably, the implementation method of the multimodal AI recognition algorithm in the system includes the following steps: The video stream from the elevator car is captured by domestically produced cameras, and three-dimensional geometric data is obtained by combining it with a depth sensor. The video stream was initially detected using an improved YOLOv8 object detection model, while point cloud features were generated by processing depth data using PointNetV2. An environment-aware illumination compensation algorithm is used to dynamically adjust the video stream, as expressed by the formula: ; in The pixel depth value. It is a scaling factor used to adjust the intensity of ambient light. It is an attenuation factor, multiplied by the depth value, indicating that the influence of ambient light decreases as distance increases. It is a constant offset used to further adjust the final pixel intensity. It is the original pixel intensity; After extracting RGB visual features using MobileNetV3 and geometric features using PointNetV2, a cross-modal attention mechanism is employed for feature fusion. The fusion weights are calculated as follows: ; in It is the Sigmoid activation function. The weight matrix for visual features. Visual feature vectors or matrices, The weight matrix for geometric features. These are geometric eigenvectors or matrices. It is the bias vector; The target threat level is calculated based on the fusion characteristics, and a security response mechanism is triggered when the confidence level exceeds the threshold.

[0008] Preferably, the method for implementing the predictive maintenance function of the system includes the following steps: Deploy a domestically produced MEMS sensor network to collect 12 types of operating parameters, including vibration, temperature, and humidity. Extracting time-domain features such as root mean square value and frequency-domain features such as wavelet packet energy entropy, the formula is expressed as: ; in For the first The probability of wavelet packet coefficients is calculated; a multi-model fusion prediction architecture is constructed, with a temporal convolutional network (TCN) for short-term fault warning and a RUL model based on Transformer for remaining lifetime prediction. The loss function is defined as: ; in These are the weighting coefficients. It is the loss function of the TCN model. It is the loss function of the Transformer model; A maintenance plan is generated by combining the equipment fatigue model, and a maintenance work order is automatically triggered when the predicted remaining life is lower than the safety threshold.

[0009] Preferably, the method for implementing the system information security protection mechanism includes the following steps: The physical layer achieves optical isolation from foreign PLC systems through a domestically produced security gateway, and the protocol layer deploys a MODBUS-CANopen conversion module; The security layer is configured with a whitelist and SM4 real-time encryption. The encryption process satisfies the following: ; in For plain text, For session key, This refers to encryption operations using the SM4 algorithm; the HarmonyOS security platform implements end-to-end signature verification, and the dynamic trust measurement module monitors the memory integrity of critical processes in real time, triggering a security circuit breaker when abnormal code injection is detected; zero-trust access control adopts an attribute-based access control model, and the decision logic is represented as follows: ; in Input parameters for the main body, Input parameters for the object, Input parameters for the environment; environment attributes include time and location parameters.

[0010] Preferably, the system is compatible with the implementation method of non-domestic elevator control system, including the following steps: Develop an interface compatible layer supporting MODBUS, CANopen, Profibus; Design a special signal acquisition module to realize current loop / relay / opto-isolating conversion; Deploy a domestic edge computing box with protocol conversion engine and instruction interception module inside to perform format verification and security filtering on non-standard instructions; Adopt hybrid cloud architecture, edge nodes handle real-time control instructions, domestic cloud platform stores operation data, edge-cloud data synchronization is realized through HTTPS protocol to ensure data sovereignty meets the requirements of China Network Security Law.

[0011] Preferably, the specific method of visual and sensor data fusion of the system includes the following steps: Establish an environment perception adjustment layer to dynamically correct video stream exposure parameters according to depth sensor data, the correction formula is: ; Where is the adjusted exposure time, is the number of depth sensor data, is the average depth value, is the depth sensor data point, is the basic exposure time; The double-flow feature extraction layer runs MobilenetV3 and PointNetV2 models respectively to generate visual feature vectors with dimension 256 and geometric feature vectors with dimension 128; The cross-modal attention mechanism layer calculates the attention weight of visual features on geometric features, the formula is: ; represents a weight matrix that maps visual features to the same space as geometric features; The final fused features are time-series modeled by a gated recurrent unit to output the target classification result.

[0012] Preferably, the deployment scheme of lightweight artificial intelligence technology of the system includes the following steps: The edge AI computing unit is equipped with the Cambrian MLU220 chip, 4TOPS computing power is provided, 8-way 1080P video stream analysis is supported, an AI inference engine integrates a TensorFlow Lite-Micro framework, a model compression technology is adopted to compress the YOLOv8 model parameter quantity from 31M to 1.8M, and the quantization accuracy is kept at 98%, an OTA algorithm update pipeline is deployed, no-sense upgrading is realized through an A / B dual-partition mechanism of the Hongmeng system, and the update process meets the following conditions: ; When the model accuracy decreases by more than a threshold value, the system is automatically rolled back to the previous version; the edge node cooperates with the cloud platform for training, and a federal learning framework is adopted to protect data privacy.

[0013] Preferably, the system generates and executes a safety verification process of the elevator control instruction, including the following steps: After the application layer generates the control instruction, the communication gateway layer filters illegal instruction formats through a regular expression, and the compliance rate is calculated as follows: ; The safety instruction is encrypted by SM4 and executed in the Hongmeng security sandbox, the sandbox adopts Linux namespaces and cgroups technology to realize resource isolation; the operation log is stored through a blockchain, and the hash value is calculated as follows: ; Wherein SHA-256 is a secure hash algorithm, which can convert data of any length into a fixed length (256 bits) hash value, represents the timestamp of the event, is a unique identifier of the user performing the operation, is a specific operation or behavior described, such as "login", "transfer"; The dynamic trust measurement module samples the key process memory image every 500ms, compares it with the benchmark hash value, and triggers system reset when the difference is more than 3%; the safety fuse mechanism immediately cuts off the power supply and locks the car when a malicious instruction is detected.

[0014] The application provides a hybrid architecture safety intelligent elevator control system based on a domestic operating system. The system has the following beneficial effects: (1), The present application constructs a defense-in-depth system through three mechanisms of physical isolation layer, protocol conversion layer and encryption verification layer. Among them, the domestic security gateway adopts optical coupling isolation circuit to realize complete isolation of electrical signal with foreign capital PLC system, block the hardware level attack path; deploy MODBUS-CANopen dual protocol stack conversion module, support protocol whitelist filtering mechanism, the filtering efficiency reaches 99.9%; integrate SM4 national encryption algorithm encryption engine, the encryption throughput reaches 500Mbps, ensure the confidentiality of data transmission; the dynamic trustworthiness measurement module is built-in in the Hongmeng system, the memory integrity of the key process is monitored in real time, and the response time of abnormal behavior is less than 200ms. The system is verified through GJB 8114-2013 Military Equipment Safety Design Requirements, realizes the whole link security protection from the physical layer to the application layer, effectively resists illegal instruction injection, protocol vulnerability attack and other security threats.

[0015] (2), The present application carries the Cambrian MLU220 edge AI chip on the system, supports 8-way 1080P video stream real-time analysis, combines the improved YOLOv8 target detection model and PointNetV2 geometric feature extraction network, realizes the cross-modal fusion of video stream and depth data. Through the environment perception light compensation algorithm, the target recognition accuracy is improved to 98.5% in low light environment; the cross-modal attention mechanism is used for feature weighted fusion, and the false alarm rate of dangerous behavior detection is less than 0.5%. The technology enables the elevator system to have the ability to identify 23 types of abnormal scenes such as electric vehicle entering the elevator, personnel congestion and article leaving in real time, and the response time is shortened to 150ms, which is significantly better than the processing delay of more than 500ms of traditional elevator system.

[0016] (3), The present application deploys 12 types of domestic MEMS sensor network, collects vibration, current harmonic and other operation parameters, combines wavelet packet energy entropy feature extraction and time sequence convolution network model, realizes 92.3% fault prediction accuracy within 7 days; the remaining life prediction model based on Transformer optimizes the key component replacement cycle by 30%. The hybrid cloud architecture adopts edge node real-time processing control instruction, domestic cloud platform storage structured data, realizes edge-cloud collaborative training through federated learning framework, and the model update cycle is shortened to 24 hours. The scheme reduces the average annual unplanned downtime of the elevator from 12 times to 3 times, reduces the maintenance cost by 45%, and prolongs the service life of the equipment to 18 years. BRIEF DESCRIPTION OF DRAWINGS

[0017] Fig. 1 It is the system overall architecture view of the present application; Fig. 2 It is the implementation method step flow view of the multi-modal AI recognition algorithm in the system of the present application; Fig. 3 It is the implementation method step flow view of the predictive maintenance function of the system of the present application. Fig. 4 The implementation method step flow view of the system information security protection mechanism of the application; Fig. 5 The implementation method step flow view of the system compatible with non-domestic elevator control system of the application; Fig. 6 The specific method step flow view of the system vision and sensor data fusion of the application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the application.

[0019] Examples of the described embodiments are shown in the drawings, in which the same or similar notations represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are intended to explain the application, and cannot be understood as a limitation of the application.

[0020] Embodiment 1 A preferred embodiment of a hybrid architecture security intelligent elevator control system based on a domestic operating system provided by the application is as follows Figs. 1-6 As shown in the figure: a hybrid architecture security intelligent elevator control system based on a domestic operating system adopts a layered modular design and is specifically divided into a hardware layer, an operating system layer, a function layer and an application layer from bottom to top; The hardware layer includes a master control unit, a sensing system and an edge AI computing unit; The operating system layer is based on the open source HongMeng 4.0 LTS version as a system base, and an elevator dedicated system ElevOS is created through deep customization; the following features are provided: Real-time performance: microkernel architecture optimization, interrupt response ≤1 microsecond, task scheduling jitter <30 μs, meeting the real-time requirement of the elevator safety standard EN81-20 for the control system.

[0021] Distributed architecture: based on soft bus technology to realize multi-device collaboration, supporting seamless linkage with access control, fire control and building control systems, with a synchronization error <5 ms.

[0022] Security features: Build a four-layer protection system - kernel-level security isolation (CC EAL5+ certified), hardware TEE protection of critical data, full-stack support for national encryption algorithms (SM3 digest / SM4 storage encryption), dynamic intrusion detection system (behavior analysis-based IDS); Functional layer integrates security control module, communication gateway, AI inference engine and predictive maintenance engine. Application layer integrates Multi-modal recognition: Fusion of visual and sensor data for electric vehicle recognition, dangerous behavior detection, etc.

[0023] Intelligent maintenance: Based on the fault prediction model, automatically generate maintenance work orders to optimize response efficiency.

[0024] Voice interaction: Integrates elevator-specific language model ElevGPT, supports natural language control and fault diagnosis; Each layer realizes cross-device collaboration through distributed soft bus technology. The main control unit uses domestic high-performance chips and carries TEE security modules. The sensor system integrates a multi-modal sensor array, including high-precision vibration sensors (monitoring rail wear), three-axis accelerometers (detecting abnormal vibration), temperature sensor arrays (real-time monitoring of motor temperature rise), and current and voltage monitoring modules (capturing electrical abnormalities). The edge AI computing unit is equipped with Cambrian MLU220 edge AI chips, and the operating system layer and the functional layer realize resource scheduling through the distributed capabilities of the Hongmeng kernel. The security control module realizes ISO 13849 PLe level safety circuit. The communication gateway supports multiple industrial protocol conversion. The AI inference engine integrates the lightweight TensorFlow Lite-Micro framework. The predictive maintenance engine is based on LSTM and Transformer models, and the application layer realizes cross-device interaction through Hongmeng atomization services.

[0025] Embodiment 2 Please refer to Figs. 1-6 , and on the basis of Embodiment 1, further obtained: the implementation method of the multi-modal AI recognition algorithm in the system, comprising the following steps: Collect elevator car video stream through domestic camera, and obtain three-dimensional geometric data combined with depth sensor; Use the improved YOLOv8 target detection model to preliminarily detect the video stream, and simultaneously process the depth data through PointNetV2 to generate point cloud features; Use the environment perception light compensation algorithm to dynamically adjust the video stream, which can be expressed by the formula: ; wherein is the pixel depth value, is a scaling factor to adjust the intensity of ambient light, is a decay factor multiplied by the depth value, indicating that the influence of ambient light will weaken as the distance increases, is a constant offset to further adjust the final pixel intensity, is the original pixel intensity; After extracting the RGB visual features by MobilenetV3 and the geometric features by PointNetV2, cross-modal attention mechanism is used for feature fusion, and the fusion weight calculation formula is: ; wherein is the Sigmoid activation function, is the weight matrix of visual features, is the visual feature vector or matrix, is the weight matrix of geometric features, is the geometric feature vector or matrix, is the bias vector; According to the fusion features, the target threat level is calculated, and when the confidence exceeds the threshold, the safety response mechanism is triggered.

[0026] The implementation method of the system predictive maintenance function comprises the following steps: Deploy domestic MEMS sensor network to collect 12 types of operation parameters such as vibration, temperature and humidity; Extract time domain features such as root mean square value and frequency domain features such as wavelet packet energy entropy, and the formula is: ; wherein is the th wavelet packet coefficient probability; Construct a multi-model fusion prediction architecture, use time series convolution network (TCN) for short-term fault early warning, and use RUL model based on Transformer for remaining life prediction, and the loss function is defined as: ; wherein is the weight coefficient, is the loss function of the TCN model, is the loss function of the Transformer model; Combine the equipment fatigue model to generate maintenance plans, and automatically trigger maintenance work orders when the predicted remaining life is lower than the safety threshold.

[0027] The implementation method of the system information security protection mechanism comprises the following steps: The physical layer realizes optical coupling isolation with the foreign capital PLC system through the localized security gateway, and the protocol layer deploys a MODBUS-CANopen conversion module; The security layer configures an instruction white list and SM4 real-time encryption, and the encryption process meets: ; Wherein is a plaintext, is a session key, is an encryption operation using the SM4 algorithm; The Hongmeng security base realizes full-link signature verification, the dynamic trust measurement module monitors the memory integrity of the key process in real time, and triggers the security fuse when detecting abnormal code injection; The zero-trust access control adopts an attribute-based access control model, and the decision logic is represented as: ; Wherein is a subject input parameter, is an object input parameter, is an environment input parameter, and the environment attributes include time and location parameters.

[0028] The implementation method of the system compatible with the non-domestic elevator control system comprises the following steps: Develop an interface compatible layer supporting MODBUS, CANopen and Profibus; Design a special signal acquisition module to realize current loop / relay / optocoupler isolation conversion; Deploy a localized edge computing box with a built-in protocol conversion engine and instruction interception module to perform format verification and security filtering on non-standard instructions; Adopt a hybrid cloud architecture, the edge node processes real-time control instructions, the domestic cloud platform stores operation data, and the edge-cloud data synchronization is realized through the HTTPS protocol to ensure that the data sovereignty meets the requirements of the Chinese network security law.

[0029] The specific method of system vision and sensor data fusion comprises the following steps: Establish an environment perception adjustment layer to dynamically correct the video stream exposure parameters according to the depth sensor data, and the correction formula is: ; Wherein is the adjusted exposure time, is the number of depth sensor data, is the average depth value, is the depth sensor data point, is the basic exposure time; The double-flow feature extraction layer runs the MobilenetV3 and PointNetV2 models respectively to generate visual feature vectors with a dimension of 256 and geometric feature vectors with a dimension of 128; The cross-modal attention mechanism layer calculates the attention weight of the visual feature on the geometric feature, and the formula is: ; represents a weight matrix for mapping the visual feature to the same space as the geometric feature; The final fused feature is time-series modeled by a gated recurrent unit, and the target classification result is output.

[0030] The deployment scheme of the system lightweight artificial intelligence technology includes the following steps: The edge AI computing unit is equipped with the Cambrian MLU220 chip, providing 4TOPS computing power support for 8-way 1080P video stream analysis; the AI inference engine integrates the TensorFlow Lite-Micro framework, adopts model compression technology to compress the YOLOv8 model parameter from 31M to 1.8M, and the quantization accuracy is maintained at 98%; the OTA algorithm update pipeline is deployed to realize non-sensing upgrade through the A / B dual-partition mechanism of the Hongmeng system, and the update process meets: ; When the model accuracy decreases by more than the threshold value, it automatically rolls back to the previous version; the edge node cooperates with the cloud platform for training, and adopts the federated learning framework to protect data privacy.

[0031] The safety verification process of the system elevator control instruction generation and execution includes the following steps: After the application layer generates the control instruction, the communication gateway layer filters illegal instruction formats through regular expressions, and the compliance rate calculation formula is: ; The security instruction is encrypted by SM4 and executed in the Hongmeng security sandbox, and the sandbox uses Linux namespaces and cgroups technology to realize resource isolation; the operation log is stored through blockchain, and the hash value calculation formula is: ; where is a secure hash algorithm that can convert data of any length to a fixed length (256-bit) hash value, represents the timestamp of the event, is the unique identifier of the user performing the operation, is the specific operation or behavior described, such as "login" or "transfer"; The dynamic credibility measurement module samples the key process memory image every 500 ms, compares with the benchmark hash value, and triggers system reset when the difference exceeds 3%; the security fuse mechanism immediately cuts off the power supply and locks the car when detecting malicious instructions.

[0032] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another, without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0033] Finally, it should be noted that the above only describes the preferred embodiments of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art will still be able to modify the technical solutions described in the foregoing embodiments or make equivalent replacements to some technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A hybrid architecture based secure intelligent elevator control system based on indigenous operating system characterized by, It adopts a layered and modular design, which is divided into hardware layer, operating system layer, functional layer and application layer from bottom to top; The hardware layer includes a main control unit, a sensing system, and an edge AI computing unit. The operating system layer is based on the open-source HarmonyOS 4.0 LTS version as the system foundation, and the elevator-specific system ElevOS is created through deep customization; The functional layer integrates a security control module, a communication gateway, an AI inference engine, and a predictive maintenance engine. The application layer integrates multimodal recognition, intelligent maintenance, and voice interaction; The main control unit uses a domestically produced high-performance chip and is equipped with a TEE security module. The sensing system integrates a multimodal sensor array, including a high-precision vibration sensor, a triaxial accelerometer, a temperature sensor array, and a current and voltage monitoring module. The edge AI computing unit is equipped with the Cambricon MLU220 edge AI chip, and the operating system layer and the functional layer achieve resource scheduling through the distributed capabilities of the HarmonyOS kernel. The safety control module implements an ISO 13849 PLe level safety loop; The communication gateway supports conversion of multiple industrial protocols; The AI ​​inference engine integrates the lightweight TensorFlow Lite-Micro framework; The predictive maintenance engine is built on LSTM and Transformer models, and the application layer achieves cross-device interaction through HarmonyOS atomic services.

2. The hybrid architecture based secure intelligent elevator control system based on homegrown operating system as claimed in claim 1 wherein: The implementation method of the multimodal AI recognition algorithm in the system includes the following steps: The video stream from the elevator car is captured by domestically produced cameras, and three-dimensional geometric data is obtained by combining it with a depth sensor. The video stream was initially detected using an improved YOLOv8 object detection model, while point cloud features were generated by processing depth data using PointNetV2. An environment-aware illumination compensation algorithm is used to dynamically adjust the video stream, as expressed by the formula: ； wherein is a pixel depth value, is a scaling factor to adjust the intensity of ambient light, is a decay factor multiplied by the depth value, indicating that the influence of ambient light will weaken as the distance increases, is a constant offset to further adjust the final pixel intensity, is the original pixel intensity; After extracting RGB visual features using MobileNetV3 and geometric features using PointNetV2, a cross-modal attention mechanism is employed for feature fusion. The fusion weights are calculated as follows: ； wherein is a Sigmoid activation function, is a weight matrix for visual features, is a vector or matrix of visual features, is a weight matrix for geometric features, is a vector or matrix of geometric features, is a bias vector; The target threat level is calculated based on the fusion characteristics, and a security response mechanism is triggered when the confidence level exceeds the threshold.

3. The hybrid architecture based secure intelligent elevator control system based on homegrown operating system as claimed in claim 1 wherein: The method for implementing the predictive maintenance function of the system includes the following steps: Deploy a domestically produced MEMS sensor network to collect 12 types of operating parameters, including vibration, temperature, and humidity. Extracting time-domain features such as root mean square value and frequency-domain features such as wavelet packet energy entropy, the formula is expressed as: ； wherein is the first wavelet packet coefficient probability; The short-term fault early warning adopts a time sequence convolution network (TCN), and the remaining useful life prediction is based on a RUL model of a Transformer. A loss function is defined as follows: ； wherein is a weight coefficient, is a loss function of the TCN model, is a loss function of the Transformer model; A maintenance plan is generated by combining the equipment fatigue model, and a maintenance work order is automatically triggered when the predicted remaining life is lower than the safety threshold.

4. The hybrid architecture secure intelligent elevator control system based on a domestic operating system according to claim 1, characterized in that: The method for implementing the system information security protection mechanism includes the following steps: The physical layer achieves optical isolation from foreign PLC systems through a domestically produced security gateway, and the protocol layer deploys a MODBUS-CANopen conversion module; The security layer is configured with a whitelist and SM4 real-time encryption. The encryption process satisfies the following: ； wherein is a plaintext, is a session key, is an encryption operation using the SM4 algorithm; the Hongmeng security base implements full-link signature verification, and the dynamic trust measurement module monitors the memory integrity of the key process in real time and triggers the security fuse when detecting abnormal code injection; the zero-trust access control adopts an attribute-based access control model, and the decision logic is represented as: ； wherein are subject input parameters, are subject input parameters, are environment input parameters, the environment attributes including time and location parameters.

5. The hybrid architecture secure intelligent elevator control system based on a domestic operating system according to claim 1, characterized in that: The method for implementing a system compatible with non-domestic elevator control systems includes the following steps: Develop an interface compatibility layer that supports MODBUS, CANopen, and Profibus; A special signal acquisition module is designed to realize current loop / relay / opto-isolator conversion; A localized edge computing box is deployed, which is equipped with a protocol conversion engine and an instruction interception module to perform format verification and security filtering on non-standard instructions; A hybrid cloud architecture is adopted, with edge nodes processing real-time control instructions and a domestic cloud platform storing operation data. The edge-cloud data synchronization is realized through the HTTPS protocol to ensure data sovereignty compliance with the requirements of the Chinese Network Security Law.

6. The hybrid architecture secure intelligent elevator control system based on a domestic operating system according to claim 1, characterized in that: The specific method of the system for fusing visual and sensor data includes the following steps: An environmental perception adjustment layer is established to dynamically correct the video stream exposure parameters based on depth sensor data, with the correction formula being: ； wherein is the adjusted exposure time, is the number of depth sensor data, is the average depth value, is the i th depth sensor data point, is the base exposure time; A double-flow feature extraction layer runs the MobilenetV3 and PointNetV2 models to generate visual feature vectors with a dimension of 256 and geometric feature vectors with a dimension of 128; A cross-modal attention mechanism layer calculates the attention weight of visual features on geometric features, with the formula being: ； denotes a weight matrix for mapping the visual features to the same space as the geometric features; The final fused features are subjected to time series modeling through a gated recurrent unit to output the target classification result.

7. The hybrid architecture secure intelligent elevator control system based on a domestic operating system according to claim 1, characterized in that: The deployment scheme of the system's lightweight artificial intelligence technology includes the following steps: The edge AI computing unit is equipped with a Cambrian MLU220 chip, providing 4TOPS computing power support for 8-way 1080P video stream analysis. The AI inference engine integrates the TensorFlow Lite-Micro framework, and uses model compression technology to compress the YOLOv8 model parameter from 31M to 1.8M, with a quantization accuracy of 98%. The OTA algorithm update pipeline is deployed to realize non-sensing upgrade through the A / B dual-partition mechanism of the HarmonyOS system, and the update process meets the following requirements: ； When the model accuracy decreases by more than the threshold value, it automatically rolls back to the previous version; the edge node and the cloud platform cooperate in training, and the federal learning framework is used to protect data privacy.

8. The hybrid architecture secure intelligent elevator control system based on a domestic operating system according to claim 1, characterized in that: The safety verification process of the system for generating and executing elevator control instructions includes the following steps: After the application layer generates control instructions, the communication gateway layer filters illegal instruction formats through regular expressions, with the compliance rate calculation formula being: ； The security instructions are encrypted by SM4 and executed in the HarmonyOS security sandbox. The sandbox uses Linux namespaces and cgroups technology to realize resource isolation; the operation log is stored through blockchain, with the hash value calculation formula being: ； wherein is a secure hash algorithm that can convert data of arbitrary length into a fixed length (256-bit) hash value, is a timestamp indicating when the event occurred, is a unique identifier for the user performing the operation, is a specific operation or action being described, such as "login," "transfer money"; The dynamic trustworthiness measurement module samples the key process memory image every 500ms and compares it with the benchmark hash value. When the difference exceeds 3%, the system is reset; the safety fuse mechanism immediately cuts off the power supply and locks the car when malicious instructions are detected.

Images