Router, gateway, camera environment sensing control method and system

Abstract

The application provides a kind of router, gateway, camera environment sensing control method and system in the technical field of network equipment control, method includes: step S1, create equipment control strategy generation model;Step S2, obtain a large number of historical environment monitoring data to build dataset;Step S3, the dataset is divided into training set, verification set and test set respectively to train, verify and test equipment control strategy generation model;Step S4, after the equipment control strategy generation model that passes the test is compressed, it is deployed to network equipment;Step S5, network equipment acquires real-time environment monitoring data by sensor array, after real-time environment monitoring data is preprocessed, input deployed equipment control strategy generation model, obtain real-time equipment control strategy;Step S6, network equipment dynamically adjusts operating parameters based on real-time equipment control strategy through preset adjustment period.The advantage of the application lies in: greatly improve the network equipment operating performance.

Description

Technical Field

[0001] This invention relates to the field of network device control technology, and in particular to a method and system for controlling the environment of routers, gateways, and cameras. Background Technology

[0002] With the rapid development of IoT, AI, and computer vision technologies, the demands for intelligent network devices in fields such as smart homes and smart security are increasing. Traditional network devices typically rely on preset fixed parameters to operate, making it difficult for them to adapt to environmental changes, which can lead to a decline in device performance under different environmental conditions.

[0003] In recent years, environmental sensing technology has begun to be applied in smart devices, attempting to collect environmental information through sensors and automatically adjust device parameters accordingly to adapt to diverse environmental conditions. However, the application of this technology in network devices such as routers, gateways, and cameras is still immature, especially in areas involving the perception of multiple environmental factors, adaptive adjustment strategies, and real-time performance guarantees. Specifically:

[0004] 1. Limited Perception Dimensions: Existing technologies typically only perceive a single environmental factor (such as light), lacking the ability to comprehensively perceive multiple key environmental factors.

[0005] 2. Lack of adaptive adjustment mechanism: After sensing changes in the environment, there is a lack of efficient and accurate parameter adjustment mechanism to dynamically optimize the equipment operating parameters.

[0006] 3. Insufficient real-time response capability: When faced with complex environmental changes, the processing often has a large delay, making it difficult to meet high real-time requirements.

[0007] Therefore, how to provide a method and system for environmental awareness control of routers, gateways, and cameras to improve the operating performance of network devices has become an urgent technical problem to be solved. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a method and system for environmental perception control of routers, gateways, and cameras, so as to improve the operating performance of network devices.

[0009] In a first aspect, the present invention provides a method for controlling the environmental awareness of a router, gateway, and camera, comprising the following steps:

[0010] Step S1: Create a device control strategy generation model and set the loss function of the device control strategy generation model;

[0011] Step S2: Obtain a large amount of historical environmental monitoring data, preprocess and label the historical environmental monitoring data, and then construct a dataset;

[0012] Step S3: Divide the dataset into a training set, a validation set, and a test set based on stratified sampling. Use the training set, validation set, and test set to train, validate, and test the device control strategy generation model, respectively.

[0013] Step S4: After compressing the device control policy generation model that has passed the test, deploy it to network devices of the device type, such as routers, gateways, or cameras.

[0014] Step S5: The network device collects real-time environmental monitoring data through a sensor array, preprocesses the real-time environmental monitoring data, and inputs it into the deployed device control strategy generation model to obtain a real-time device control strategy. Based on the real-time device control strategy, the operating parameters are dynamically adjusted through a preset adjustment cycle.

[0015] Step S6: The network device records the operation log in real time, and performs encrypted storage and distributed backup of the operation log.

[0016] Furthermore, step S1 specifically includes:

[0017] An equipment control strategy generation model is created based on an environmental feature extraction module, an environmental feature fusion module, and a parameter adjustment strategy generation module, and a loss function is set for the equipment control strategy generation model.

[0018] The environmental feature extraction module is constructed based on a light feature extraction unit, a temperature feature extraction unit, a humidity feature extraction unit, an electromagnetic feature extraction unit, a vibration feature extraction unit, an infrared feature extraction unit, a noise feature extraction unit, and a mode selection gating unit. The light feature extraction unit extracts light features from light data using a first 1D convolutional network; the temperature feature extraction unit extracts temperature features from temperature data using a second 1D convolutional network; the humidity feature extraction unit extracts humidity features from humidity data using a third 1D convolutional network; the electromagnetic feature extraction unit extracts electromagnetic features from electromagnetic interference intensity using a fourth 1D convolutional network; the vibration feature extraction unit extracts vibration features from physical vibration amplitude using a fifth 1D convolutional network; the infrared feature extraction unit extracts infrared features from infrared data using MobileNetV2; and the noise feature extraction unit extracts noise features from the environmental noise spectrum using Mel spectrum transformation and a BiLSTM network. The mode selection gating unit weights the light, temperature, humidity, electromagnetic, vibration, infrared, and noise features using a gating attention mechanism to obtain a multimodal feature vector group.

[0019] The environmental feature fusion module is constructed based on a multi-scale feature interaction unit and an adaptive feature compression unit. The multi-scale feature interaction unit is used to capture long-term cross-modal dependency features from a multi-modal feature vector group through a hierarchical Transformer structure. The adaptive feature compression unit is used to compress long-term dependency features through learnable gated convolution to obtain compressed fused features.

[0020] The parameter adjustment strategy generation module is constructed based on the strategy generation unit and the physical constraint unit. The strategy generation unit is used to infer the compressed fusion features through a 3-layer residual fully connected network to obtain the initial adjustment values ​​of the device operating parameters. The physical constraint unit is used to map the initial adjustment values ​​to the physical feasible region through a differential equation embedding layer and differentiable projection to obtain the suggested adjustment values ​​of the device operating parameters, and output the device control strategy carrying the suggested adjustment values.

[0021] The formula for the loss function is:

[0022] L=αL MSE +βL reg +γL phys ;

[0023] Where L represents the loss value of the loss function; L MSE L represents the mean square error of parameter prediction; reg L represents the policy smoothness regularization term; phys α represents the penalty term for violating physical constraints; α, β, and γ all represent weighting coefficients.

[0024] Furthermore, step S2 specifically includes:

[0025] Acquire a large amount of historical environmental monitoring data, including light intensity, temperature, humidity, electromagnetic interference intensity, physical vibration amplitude, infrared data, and environmental noise spectrum;

[0026] The historical environmental monitoring data are preprocessed, including at least data cleaning, data denoising, data normalization, and feature extraction. The preprocessed historical environmental monitoring data are then labeled with equipment control strategies to construct a dataset.

[0027] Furthermore, step S3 specifically includes:

[0028] Based on stratified sampling, the dataset is divided into a training set, a validation set, and a test set in a ratio of 8:1:1. The device control strategy generation model is trained using the training set and a loss function. During the training process, the hyperparameters of the device control strategy generation model are continuously optimized until the loss value of the loss function is less than a preset loss threshold or a preset early stop condition is met.

[0029] The accuracy of the strategy is calculated using the validation set to validate the trained device control strategy generation model. If the validation fails, the training set is expanded and training continues; if the validation passes, then:

[0030] The F1 score and confidence level are calculated using the test set to test the validated device control strategy generation model. If the test fails, the training set is expanded and training continues; if the test passes, training ends.

[0031] Furthermore, step S4 specifically includes:

[0032] The device control strategy generation model that has passed the test is compressed by combining quantization technology and dynamic pruning technology. After the performance of the compressed device control strategy generation model is verified, it is deployed to network devices of device type router, gateway or camera through containerization technology.

[0033] Step S6 specifically involves:

[0034] Network devices record at least the following operational logs in real time: basic device information, user operation records, configuration change records, system events, and hardware status.

[0035] The log fingerprint of the running log is calculated using the HMAC algorithm. The running log and the log fingerprint are then encrypted into Level 1 encrypted data using the AES algorithm. The Level 1 encrypted data is converted into hexadecimal data. The numbers 7 and A, and 9 and B in the hexadecimal data are swapped to obtain Level 2 encrypted data. The Level 2 encrypted data is then encrypted into Level 3 encrypted data using the RSA algorithm. Each character in the Level 3 encrypted data is shifted 8 bits to the left to obtain Level 4 encrypted data. The Level 4 encrypted data is then encrypted into an encrypted log using the RC6 algorithm. The encrypted log is stored in a preset path and backed up in a distributed manner.

[0036] Secondly, the present invention provides a router, gateway, and camera environmental perception control system, comprising the following modules:

[0037] The device control strategy generation model creation module is used to create a device control strategy generation model and set the loss function of the device control strategy generation model.

[0038] The dataset construction module is used to acquire a large amount of historical environmental monitoring data, and to construct the dataset after preprocessing and labeling the historical environmental monitoring data.

[0039] The equipment control strategy generation model training module is used to divide the dataset into a training set, a validation set, and a test set based on stratified sampling, and to train, validate, and test the equipment control strategy generation model using the training set, validation set, and test set, respectively.

[0040] The device control policy generation model deployment module is used to compress the tested device control policy generation model and deploy it to network devices of the device type, such as routers, gateways, or cameras.

[0041] The environmental perception and control module is used by network devices to collect real-time environmental monitoring data through a sensor array, preprocess the real-time environmental monitoring data and input it into the deployed device control strategy generation model to obtain a real-time device control strategy, and dynamically adjust the operating parameters based on the real-time device control strategy through a preset adjustment cycle.

[0042] The operation log management module is used to record operation logs of network devices in real time, and to encrypt, store, and distribute the operation logs.

[0043] Furthermore, the device control strategy generation model creation module is specifically used for:

[0044] An equipment control strategy generation model is created based on an environmental feature extraction module, an environmental feature fusion module, and a parameter adjustment strategy generation module, and a loss function is set for the equipment control strategy generation model.

[0045] The environmental feature extraction module is constructed based on a light feature extraction unit, a temperature feature extraction unit, a humidity feature extraction unit, an electromagnetic feature extraction unit, a vibration feature extraction unit, an infrared feature extraction unit, a noise feature extraction unit, and a mode selection gating unit. The light feature extraction unit extracts light features from light data using a first 1D convolutional network; the temperature feature extraction unit extracts temperature features from temperature data using a second 1D convolutional network; the humidity feature extraction unit extracts humidity features from humidity data using a third 1D convolutional network; the electromagnetic feature extraction unit extracts electromagnetic features from electromagnetic interference intensity using a fourth 1D convolutional network; the vibration feature extraction unit extracts vibration features from physical vibration amplitude using a fifth 1D convolutional network; the infrared feature extraction unit extracts infrared features from infrared data using MobileNetV2; and the noise feature extraction unit extracts noise features from the environmental noise spectrum using Mel spectrum transformation and a BiLSTM network. The mode selection gating unit weights the light, temperature, humidity, electromagnetic, vibration, infrared, and noise features using a gating attention mechanism to obtain a multimodal feature vector group.

[0046] The environmental feature fusion module is constructed based on a multi-scale feature interaction unit and an adaptive feature compression unit. The multi-scale feature interaction unit is used to capture long-term cross-modal dependency features from a multi-modal feature vector group through a hierarchical Transformer structure. The adaptive feature compression unit is used to compress long-term dependency features through learnable gated convolution to obtain compressed fused features.

[0047] The parameter adjustment strategy generation module is constructed based on the strategy generation unit and the physical constraint unit. The strategy generation unit is used to infer the compressed fusion features through a 3-layer residual fully connected network to obtain the initial adjustment values ​​of the device operating parameters. The physical constraint unit is used to map the initial adjustment values ​​to the physical feasible region through a differential equation embedding layer and differentiable projection to obtain the suggested adjustment values ​​of the device operating parameters, and output the device control strategy carrying the suggested adjustment values.

[0048] The formula for the loss function is:

[0049] L=αL MSE +βL reg +γL phys ;

[0050] Where L represents the loss value of the loss function; L MSE L represents the mean square error of parameter prediction; reg L represents the policy smoothness regularization term; phys α represents the penalty term for violating physical constraints; α, β, and γ all represent weighting coefficients.

[0051] Furthermore, the dataset construction module is specifically used for:

[0052] Acquire a large amount of historical environmental monitoring data, including light intensity, temperature, humidity, electromagnetic interference intensity, physical vibration amplitude, infrared data, and environmental noise spectrum;

[0053] The historical environmental monitoring data are preprocessed, including at least data cleaning, data denoising, data normalization, and feature extraction. The preprocessed historical environmental monitoring data are then labeled with equipment control strategies to construct a dataset.

[0054] Furthermore, the device control strategy generation model training module is specifically used for:

[0055] Based on stratified sampling, the dataset is divided into a training set, a validation set, and a test set in a ratio of 8:1:1. The device control strategy generation model is trained using the training set and a loss function. During the training process, the hyperparameters of the device control strategy generation model are continuously optimized until the loss value of the loss function is less than a preset loss threshold or a preset early stop condition is met.

[0056] The accuracy of the strategy is calculated using the validation set to validate the trained device control strategy generation model. If the validation fails, the training set is expanded and training continues; if the validation passes, then:

[0057] The F1 score and confidence level are calculated using the test set to test the validated device control strategy generation model. If the test fails, the training set is expanded and training continues; if the test passes, training ends.

[0058] Furthermore, the device control strategy generation model deployment module is specifically used for:

[0059] The device control strategy generation model that has passed the test is compressed by combining quantization technology and dynamic pruning technology. After the performance of the compressed device control strategy generation model is verified, it is deployed to network devices of device type router, gateway or camera through containerization technology.

[0060] The operation log management module is specifically used for:

[0061] Network devices record at least the following operational logs in real time: basic device information, user operation records, configuration change records, system events, and hardware status.

[0062] The log fingerprint of the running log is calculated using the HMAC algorithm. The running log and the log fingerprint are then encrypted into Level 1 encrypted data using the AES algorithm. The Level 1 encrypted data is converted into hexadecimal data. The numbers 7 and A, and 9 and B in the hexadecimal data are swapped to obtain Level 2 encrypted data. The Level 2 encrypted data is then encrypted into Level 3 encrypted data using the RSA algorithm. Each character in the Level 3 encrypted data is shifted 8 bits to the left to obtain Level 4 encrypted data. The Level 4 encrypted data is then encrypted into an encrypted log using the RC6 algorithm. The encrypted log is stored in a preset path and backed up in a distributed manner.

[0063] The advantages of this invention are:

[0064] 1. By creating a device control strategy generation model and setting its loss function, a large amount of historical environmental monitoring data is acquired. This data is preprocessed and labeled to construct a dataset. The dataset is divided into training, validation, and test sets using stratified sampling. The device control strategy generation model is then trained, validated, and tested using these sets. The tested model is compressed and deployed to network devices such as routers, gateways, or cameras. The network devices collect real-time environmental monitoring data via sensor arrays and input it into the device control strategy generation model to obtain real-time device control strategies. Operating parameters are dynamically adjusted based on the real-time device control strategies according to a preset adjustment cycle, and operating logs are recorded in real-time. These logs are encrypted and stored in a distributed backup. In short, by deploying the compressed device control strategy generation model in routers, gateways, or cameras, they can utilize built-in sensor arrays to collect multi-dimensional environmental data (real-time environmental monitoring data) in real time. Real-time device control strategies are then dynamically generated directly on the network devices based on the device control strategy generation model, and operating parameters are adaptively adjusted according to a preset adjustment cycle, thereby significantly improving the operating performance of the network devices.

[0065] 2. By integrating data from seven heterogeneous sensors (light / temperature / humidity / electromagnetic / vibration / infrared / noise), dynamic weighted fusion is achieved through modal selection gating units, solving the problem of weight allocation for multi-source data. By adopting a dedicated feature extraction architecture (such as MobileNetV2 for infrared data and BiLSTM for noise spectrum), the optimal algorithm is customized for each data type, breaking through the limitations of traditional single environmental parameter perception. Through cross-modal feature interaction (hierarchical Transformer), the correlation of complex environments is captured, significantly improving the accuracy of environmental state recognition (such as the coupling effect analysis of electromagnetic interference and vibration amplitude).

[0066] 3. By introducing a differential equation embedding layer and differentiable projection into the policy generation module, the initial adjustment value is mapped to the physical feasible region (e.g., the router's transmission power does not exceed the hardware threshold). By creatively adding a physical constraint violation penalty term to the loss function, the feasibility of the policy is guaranteed from the training source, avoiding dangerous operations caused by model overfitting (e.g., the camera still turns on high power mode in an overheated environment), thus solving the key pain point of "safety and reliability" of AI control policies in industrial scenarios.

[0067] 4. By combining quantization and dynamic pruning compression techniques, the model size is reduced while maintaining model accuracy. Through a containerized deployment architecture, the model can be quickly loaded into resource-constrained network devices such as routers, gateways, and cameras, breaking through the computing power bottleneck of deep learning models on embedded devices and achieving millisecond-level environmental response.

[0068] 5. By combining four-level hybrid encryption (AES→RSA→RC6) with character substitution obfuscation technology (such as 7 / A swap), log fingerprinting and distributed storage, HMAC is used to ensure data integrity, prevent single points of failure, and resist common threats such as replay attacks / man-in-the-middle attacks. It is especially suitable for privacy-sensitive devices such as cameras.

[0069] 6. Through an innovative multimodal environmental perception architecture (integrating dynamic fusion of data from seven types of sensors: light, temperature, humidity, electromagnetic, vibration, infrared, and noise) and a physical constraint embedding mechanism (differential equation mapping to ensure policy feasibility), combined with quantization compression and containerized deployment, millisecond-level response of edge devices is achieved. Under the premise of ensuring a four-level hybrid encryption security system (AES-RSA-RC6 + character obfuscation), the dynamic environmental adaptability of network devices such as routers, gateways, and cameras is significantly improved, resolving the triangular contradiction of "accuracy-latency-security" in edge AI control, and providing a highly reliable and self-evolving environmental perception and control closed loop for smart cities, industrial IoT, and other scenarios. Attached Figure Description

[0070] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0071] Figure 1 This is a flowchart of an environmental perception control method for a router, gateway, and camera according to the present invention.

[0072] Figure 2 This is a schematic diagram of the structure of a router, gateway, and camera environmental perception control system according to the present invention. Detailed Implementation

[0073] The overall idea of ​​the technical solution in this application embodiment is as follows: By deploying a compressed device control policy generation model in a router, gateway, or camera, it can collect multi-dimensional environmental data in real time using a built-in sensor array, and directly generate real-time device control policies on the network device end based on the device control policy generation model, and adaptively adjust the operating parameters according to a preset adjustment cycle to improve the operating performance of the network device.

[0074] Please refer to Figures 1 to 2 As shown, a preferred embodiment of the environmental perception control method for a router, gateway, and camera according to the present invention includes the following steps:

[0075] Step S1: Create a device control strategy generation model and set the loss function of the device control strategy generation model;

[0076] Step S2: Obtain a large amount of historical environmental monitoring data, preprocess and label the historical environmental monitoring data, and then construct a dataset;

[0077] Step S3: Divide the dataset into a training set, a validation set, and a test set based on stratified sampling. Use the training set, validation set, and test set to train, validate, and test the device control strategy generation model, respectively.

[0078] Step S4: After compressing the device control policy generation model that has passed the test, deploy it to network devices of the device type, such as routers, gateways, or cameras.

[0079] Step S5: The network device collects real-time environmental monitoring data through a sensor array, preprocesses the real-time environmental monitoring data, and inputs it into the deployed device control strategy generation model to obtain a real-time device control strategy. Based on the real-time device control strategy, the operating parameters are dynamically adjusted through a preset adjustment cycle.

[0080] By integrating data from seven heterogeneous sensors (light / temperature / humidity / electromagnetic / vibration / infrared / noise), dynamic weighted fusion is achieved through modal selection gating units, solving the problem of weight allocation for multi-source data. By employing a dedicated feature extraction architecture (such as MobileNetV2 for infrared data and BiLSTM for noise spectrum), and customizing optimal algorithms for different data types, the limitations of traditional single-parameter perception are overcome. Through cross-modal feature interaction (hierarchical Transformer), complex environmental correlations are captured, significantly improving the accuracy of environmental state identification (such as the coupling effect analysis of electromagnetic interference and vibration amplitude).

[0081] Step S6: The network device records operation logs in real time, and encrypts, stores, and distributes these logs for backup. Specifically, in real time, the network device can provide a user interface to display real-time environmental information and device status, and allow users to manually adjust the device's operating parameters.

[0082] Examples of environmental awareness and control scenarios for network devices are as follows:

[0083] strong light interference Increase 5GHz channel power Enable HDR mode Switch to wired backhaul High electromagnetic interference Enable DFS channel Turn off WiFi Direct connection Start spectrum sensing High temperature and humidity Reduce CPU frequency Trigger the cooling fan Turn off unused radio frequency

[0084] Step S1 specifically involves:

[0085] An equipment control strategy generation model is created based on an environmental feature extraction module, an environmental feature fusion module, and a parameter adjustment strategy generation module, and a loss function is set for the equipment control strategy generation model.

[0086] The environmental feature extraction module is constructed based on a light feature extraction unit, a temperature feature extraction unit, a humidity feature extraction unit, an electromagnetic feature extraction unit, a vibration feature extraction unit, an infrared feature extraction unit, a noise feature extraction unit, and a mode selection gating unit. The light feature extraction unit extracts light features from light data using a first 1D convolutional network; the temperature feature extraction unit extracts temperature features from temperature data using a second 1D convolutional network; the humidity feature extraction unit extracts humidity features from humidity data using a third 1D convolutional network; the electromagnetic feature extraction unit extracts electromagnetic features from electromagnetic interference intensity using a fourth 1D convolutional network; the vibration feature extraction unit extracts vibration features from physical vibration amplitude using a fifth 1D convolutional network; the infrared feature extraction unit extracts infrared features from infrared data using MobileNetV2; and the noise feature extraction unit extracts noise features from the environmental noise spectrum using Mel spectrum transformation and a BiLSTM network. The mode selection gating unit weights the light, temperature, humidity, electromagnetic, vibration, infrared, and noise features using a gating attention mechanism to obtain a multimodal feature vector group.

[0087] The environmental feature fusion module is constructed based on a multi-scale feature interaction unit and an adaptive feature compression unit. The multi-scale feature interaction unit is used to capture long-term cross-modal dependency features from a multi-modal feature vector group through a hierarchical Transformer structure. The adaptive feature compression unit is used to compress long-term dependency features through learnable gated convolution to obtain compressed fused features.

[0088] The parameter adjustment strategy generation module is constructed based on the strategy generation unit and the physical constraint unit. The strategy generation unit is used to infer the compressed fusion features through a 3-layer residual fully connected network to obtain the initial adjustment values ​​of the device operating parameters. The physical constraint unit is used to map the initial adjustment values ​​to the physical feasible region through a differential equation embedding layer and differentiable projection to obtain the suggested adjustment values ​​of the device operating parameters, and output the device control strategy carrying the suggested adjustment values.

[0089] By introducing a differential equation embedding layer and differentiable projection into the policy generation module, the initial adjustment value is mapped to the physical feasible region (such as the router's transmission power not exceeding the hardware threshold). By creatively adding a physical constraint violation penalty term to the loss function, the feasibility of the policy is guaranteed from the training source, avoiding dangerous operations of the device due to model overfitting (such as the camera still turning on high power mode under overheating conditions), thus solving the key pain point of "safety and reliability" of AI control policies in industrial scenarios.

[0090] The formula for the loss function is:

[0091] L=αL MSE +βL reg +γL phys ;

[0092] Where L represents the loss value of the loss function; L MSE L represents the mean square error of parameter prediction; reg L represents the policy smoothness regularization term (first-order difference penalty); phys α represents the penalty term for violating physical constraints; α, β, and γ all represent weighting coefficients.

[0093] Step S2 specifically involves:

[0094] Acquire a large amount of historical environmental monitoring data, including light intensity, temperature, humidity, electromagnetic interference intensity, physical vibration amplitude, infrared data, and environmental noise spectrum;

[0095] The historical environmental monitoring data are preprocessed, including at least data cleaning, data denoising, data normalization, and feature extraction. The preprocessed historical environmental monitoring data are then labeled with equipment control strategies to construct a dataset.

[0096] Step S3 specifically involves:

[0097] Based on stratified sampling, the dataset is divided into a training set, a validation set, and a test set in a ratio of 8:1:1. The device control strategy generation model is trained using the training set and a loss function. During the training process, the hyperparameters of the device control strategy generation model are continuously optimized until the loss value of the loss function is less than a preset loss threshold or a preset early stop condition is met.

[0098] The accuracy of the strategy is calculated using the validation set to validate the trained device control strategy generation model. If the validation fails, the training set is expanded and training continues; if the validation passes, then:

[0099] The F1 score and confidence level are calculated using the test set to test the validated device control strategy generation model. If the test fails, the training set is expanded and training continues; if the test passes, training ends.

[0100] Step S4 specifically involves:

[0101] The device control strategy generation model that has passed the test is compressed by combining quantization technology and dynamic pruning technology. After the performance of the compressed device control strategy generation model is verified, it is deployed to network devices of device type router, gateway or camera through containerization technology.

[0102] By combining quantization and dynamic pruning compression techniques, the model size is reduced while maintaining model accuracy. Through a containerized deployment architecture, the model can be quickly loaded into resource-constrained network devices such as routers, gateways, and cameras, breaking through the computing power bottleneck of deep learning models on embedded devices and achieving millisecond-level environmental response.

[0103] Step S6 specifically involves:

[0104] Network devices record at least the following operational logs in real time: basic device information, user operation records, configuration change records, system events, and hardware status.

[0105] The log fingerprint of the running log is calculated using the HMAC algorithm. The running log and the log fingerprint are then encrypted into Level 1 encrypted data using the AES algorithm. The Level 1 encrypted data is converted into hexadecimal data. The numbers 7 and A, and 9 and B in the hexadecimal data are swapped to obtain Level 2 encrypted data. The Level 2 encrypted data is then encrypted into Level 3 encrypted data using the RSA algorithm. Each character in the Level 3 encrypted data is shifted 8 bits to the left to obtain Level 4 encrypted data. The Level 4 encrypted data is then encrypted into an encrypted log using the RC6 algorithm. The encrypted log is stored in a preset path and backed up in a distributed manner.

[0106] By combining four-level hybrid encryption (AES→RSA→RC6) with character substitution obfuscation technology (such as 7 / A swap), log fingerprinting and distributed storage, HMAC ensures data integrity, prevents single points of failure, and can resist common threats such as replay attacks / man-in-the-middle attacks. It is especially suitable for privacy-sensitive devices such as cameras.

[0107] Through an innovative multimodal environmental perception architecture (integrating dynamic fusion of data from seven types of sensors including light, temperature, humidity, electromagnetic fields, vibration, infrared, and noise) and a physical constraint embedding mechanism (differential equation mapping to ensure policy feasibility), combined with quantization compression and containerized deployment, millisecond-level response of edge devices is achieved. While ensuring a four-level hybrid encryption security system (AES-RSA-RC6 + character obfuscation), the dynamic environmental adaptability of network devices such as routers, gateways, and cameras is significantly improved. This resolves the triangular contradiction of "accuracy-latency-security" in edge AI control, providing a highly reliable and self-evolving environmental perception and control closed loop for smart cities, industrial IoT, and other scenarios.

[0108] A preferred embodiment of the environmental perception control system for a router, gateway, and camera of the present invention includes the following modules:

[0109] The device control strategy generation model creation module is used to create a device control strategy generation model and set the loss function of the device control strategy generation model.

[0110] The dataset construction module is used to acquire a large amount of historical environmental monitoring data, and to construct the dataset after preprocessing and labeling the historical environmental monitoring data.

[0111] The equipment control strategy generation model training module is used to divide the dataset into a training set, a validation set, and a test set based on stratified sampling, and to train, validate, and test the equipment control strategy generation model using the training set, validation set, and test set, respectively.

[0112] The device control policy generation model deployment module is used to compress the tested device control policy generation model and deploy it to network devices of the device type, such as routers, gateways, or cameras.

[0113] The environmental perception and control module is used by network devices to collect real-time environmental monitoring data through a sensor array, preprocess the real-time environmental monitoring data and input it into the deployed device control strategy generation model to obtain a real-time device control strategy, and dynamically adjust the operating parameters based on the real-time device control strategy through a preset adjustment cycle.

[0114] By integrating data from seven heterogeneous sensors (light / temperature / humidity / electromagnetic / vibration / infrared / noise), dynamic weighted fusion is achieved through modal selection gating units, solving the problem of weight allocation for multi-source data. By employing a dedicated feature extraction architecture (such as MobileNetV2 for infrared data and BiLSTM for noise spectrum), and customizing optimal algorithms for different data types, the limitations of traditional single-parameter perception are overcome. Through cross-modal feature interaction (hierarchical Transformer), complex environmental correlations are captured, significantly improving the accuracy of environmental state identification (such as the coupling effect analysis of electromagnetic interference and vibration amplitude).

[0115] The operation log management module is used by network devices to record operation logs in real time, and to encrypt, store, and distribute these logs. Specifically, in real-time, the network device can provide a user interface to display real-time environmental information and device status, and allow users to manually adjust the device's operating parameters.

[0116] Examples of environmental awareness and control scenarios for network devices are as follows:

[0117]

[0118]

[0119] The device control strategy generation model creation module is specifically used for:

[0120] An equipment control strategy generation model is created based on an environmental feature extraction module, an environmental feature fusion module, and a parameter adjustment strategy generation module, and a loss function is set for the equipment control strategy generation model.

[0121] The environmental feature extraction module is constructed based on a light feature extraction unit, a temperature feature extraction unit, a humidity feature extraction unit, an electromagnetic feature extraction unit, a vibration feature extraction unit, an infrared feature extraction unit, a noise feature extraction unit, and a mode selection gating unit. The light feature extraction unit extracts light features from light data using a first 1D convolutional network; the temperature feature extraction unit extracts temperature features from temperature data using a second 1D convolutional network; the humidity feature extraction unit extracts humidity features from humidity data using a third 1D convolutional network; the electromagnetic feature extraction unit extracts electromagnetic features from electromagnetic interference intensity using a fourth 1D convolutional network; the vibration feature extraction unit extracts vibration features from physical vibration amplitude using a fifth 1D convolutional network; the infrared feature extraction unit extracts infrared features from infrared data using MobileNetV2; and the noise feature extraction unit extracts noise features from the environmental noise spectrum using Mel spectrum transformation and a BiLSTM network. The mode selection gating unit weights the light, temperature, humidity, electromagnetic, vibration, infrared, and noise features using a gating attention mechanism to obtain a multimodal feature vector group.

[0122] The environmental feature fusion module is constructed based on a multi-scale feature interaction unit and an adaptive feature compression unit. The multi-scale feature interaction unit is used to capture long-term cross-modal dependency features from a multi-modal feature vector group through a hierarchical Transformer structure. The adaptive feature compression unit is used to compress long-term dependency features through learnable gated convolution to obtain compressed fused features.

[0123] The parameter adjustment strategy generation module is constructed based on the strategy generation unit and the physical constraint unit. The strategy generation unit is used to infer the compressed fusion features through a 3-layer residual fully connected network to obtain the initial adjustment values ​​of the device operating parameters. The physical constraint unit is used to map the initial adjustment values ​​to the physical feasible region through a differential equation embedding layer and differentiable projection to obtain the suggested adjustment values ​​of the device operating parameters, and output the device control strategy carrying the suggested adjustment values.

[0124] By introducing a differential equation embedding layer and differentiable projection into the policy generation module, the initial adjustment value is mapped to the physical feasible region (such as the router's transmission power not exceeding the hardware threshold). By creatively adding a physical constraint violation penalty term to the loss function, the feasibility of the policy is guaranteed from the training source, avoiding dangerous operations of the device due to model overfitting (such as the camera still turning on high power mode under overheating conditions), thus solving the key pain point of "safety and reliability" of AI control policies in industrial scenarios.

[0125] The formula for the loss function is:

[0126] L=αL MSE +βL reg +γL phys ;

[0127] Where L represents the loss value of the loss function; L MSE L represents the mean square error of parameter prediction; reg L represents the policy smoothness regularization term (first-order difference penalty); phys α represents the penalty term for violating physical constraints; α, β, and γ all represent weighting coefficients.

[0128] The dataset construction module is specifically used for:

[0129] Acquire a large amount of historical environmental monitoring data, including light intensity, temperature, humidity, electromagnetic interference intensity, physical vibration amplitude, infrared data, and environmental noise spectrum;

[0130] The historical environmental monitoring data are preprocessed, including at least data cleaning, data denoising, data normalization, and feature extraction. The preprocessed historical environmental monitoring data are then labeled with equipment control strategies to construct a dataset.

[0131] The device control strategy generation model training module is specifically used for:

[0132] Based on stratified sampling, the dataset is divided into a training set, a validation set, and a test set in a ratio of 8:1:1. The device control strategy generation model is trained using the training set and a loss function. During the training process, the hyperparameters of the device control strategy generation model are continuously optimized until the loss value of the loss function is less than a preset loss threshold or a preset early stop condition is met.

[0133] The accuracy of the strategy is calculated using the validation set to validate the trained device control strategy generation model. If the validation fails, the training set is expanded and training continues; if the validation passes, then:

[0134] The F1 score and confidence level are calculated using the test set to test the validated device control strategy generation model. If the test fails, the training set is expanded and training continues; if the test passes, training ends.

[0135] The device control strategy generation model deployment module is specifically used for:

[0136] The device control strategy generation model that has passed the test is compressed by combining quantization technology and dynamic pruning technology. After the performance of the compressed device control strategy generation model is verified, it is deployed to network devices of device type router, gateway or camera through containerization technology.

[0137] By combining quantization and dynamic pruning compression techniques, the model size is reduced while maintaining model accuracy. Through a containerized deployment architecture, the model can be quickly loaded into resource-constrained network devices such as routers, gateways, and cameras, breaking through the computing power bottleneck of deep learning models on embedded devices and achieving millisecond-level environmental response.

[0138] The operation log management module is specifically used for:

[0139] Network devices record at least the following operational logs in real time: basic device information, user operation records, configuration change records, system events, and hardware status.

[0140] The log fingerprint of the running log is calculated using the HMAC algorithm. The running log and the log fingerprint are then encrypted into Level 1 encrypted data using the AES algorithm. The Level 1 encrypted data is converted into hexadecimal data. The numbers 7 and A, and 9 and B in the hexadecimal data are swapped to obtain Level 2 encrypted data. The Level 2 encrypted data is then encrypted into Level 3 encrypted data using the RSA algorithm. Each character in the Level 3 encrypted data is shifted 8 bits to the left to obtain Level 4 encrypted data. The Level 4 encrypted data is then encrypted into an encrypted log using the RC6 algorithm. The encrypted log is stored in a preset path and backed up in a distributed manner.

[0141] By combining four-level hybrid encryption (AES→RSA→RC6) with character substitution obfuscation technology (such as 7 / A swap), log fingerprinting and distributed storage, HMAC ensures data integrity, prevents single points of failure, and can resist common threats such as replay attacks / man-in-the-middle attacks. It is especially suitable for privacy-sensitive devices such as cameras.

[0142] Through an innovative multimodal environmental perception architecture (integrating dynamic fusion of data from seven types of sensors including light, temperature, humidity, electromagnetic fields, vibration, infrared, and noise) and a physical constraint embedding mechanism (differential equation mapping to ensure policy feasibility), combined with quantization compression and containerized deployment, millisecond-level response of edge devices is achieved. While ensuring a four-level hybrid encryption security system (AES-RSA-RC6 + character obfuscation), the dynamic environmental adaptability of network devices such as routers, gateways, and cameras is significantly improved. This resolves the triangular contradiction of "accuracy-latency-security" in edge AI control, providing a highly reliable and self-evolving environmental perception and control closed loop for smart cities, industrial IoT, and other scenarios.

[0143] In summary, the advantages of this invention are:

[0144] 1. By creating a device control strategy generation model and setting its loss function, a large amount of historical environmental monitoring data is acquired. This data is preprocessed and labeled to construct a dataset. The dataset is divided into training, validation, and test sets using stratified sampling. The device control strategy generation model is then trained, validated, and tested using these sets. The tested model is compressed and deployed to network devices such as routers, gateways, or cameras. The network devices collect real-time environmental monitoring data via sensor arrays and input it into the device control strategy generation model to obtain real-time device control strategies. Operating parameters are dynamically adjusted based on the real-time device control strategies according to a preset adjustment cycle, and operating logs are recorded in real-time. These logs are encrypted and stored in a distributed backup. In short, by deploying the compressed device control strategy generation model in routers, gateways, or cameras, they can utilize built-in sensor arrays to collect multi-dimensional environmental data (real-time environmental monitoring data) in real time. Real-time device control strategies are then dynamically generated directly on the network devices based on the device control strategy generation model, and operating parameters are adaptively adjusted according to a preset adjustment cycle, thereby significantly improving the operating performance of the network devices.

[0145] 2. By integrating data from seven heterogeneous sensors (light / temperature / humidity / electromagnetic / vibration / infrared / noise), dynamic weighted fusion is achieved through modal selection gating units, solving the problem of weight allocation for multi-source data. By adopting a dedicated feature extraction architecture (such as MobileNetV2 for infrared data and BiLSTM for noise spectrum), the optimal algorithm is customized for each data type, breaking through the limitations of traditional single environmental parameter perception. Through cross-modal feature interaction (hierarchical Transformer), the correlation of complex environments is captured, significantly improving the accuracy of environmental state recognition (such as the coupling effect analysis of electromagnetic interference and vibration amplitude).

[0146] 3. By introducing a differential equation embedding layer and differentiable projection into the policy generation module, the initial adjustment value is mapped to the physical feasible region (e.g., the router's transmission power does not exceed the hardware threshold). By creatively adding a physical constraint violation penalty term to the loss function, the feasibility of the policy is guaranteed from the training source, avoiding dangerous operations caused by model overfitting (e.g., the camera still turns on high power mode in an overheated environment), thus solving the key pain point of "safety and reliability" of AI control policies in industrial scenarios.

[0147] 4. By combining quantization and dynamic pruning compression techniques, the model size is reduced while maintaining model accuracy. Through a containerized deployment architecture, the model can be quickly loaded into resource-constrained network devices such as routers, gateways, and cameras, breaking through the computing power bottleneck of deep learning models on embedded devices and achieving millisecond-level environmental response.

[0148] 5. By combining four-level hybrid encryption (AES→RSA→RC6) with character substitution obfuscation technology (such as 7 / A swap), log fingerprinting and distributed storage, HMAC is used to ensure data integrity, prevent single points of failure, and resist common threats such as replay attacks / man-in-the-middle attacks. It is especially suitable for privacy-sensitive devices such as cameras.

[0149] 6. Through an innovative multimodal environmental perception architecture (integrating dynamic fusion of data from seven types of sensors: light, temperature, humidity, electromagnetic, vibration, infrared, and noise) and a physical constraint embedding mechanism (differential equation mapping to ensure policy feasibility), combined with quantization compression and containerized deployment, millisecond-level response of edge devices is achieved. Under the premise of ensuring a four-level hybrid encryption security system (AES-RSA-RC6 + character obfuscation), the dynamic environmental adaptability of network devices such as routers, gateways, and cameras is significantly improved, resolving the triangular contradiction of "accuracy-latency-security" in edge AI control, and providing a highly reliable and self-evolving environmental perception and control closed loop for smart cities, industrial IoT, and other scenarios.

[0150] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for controlling the environment of a router, gateway, and camera, characterized in that: Includes the following steps: Step S1: Create a device control strategy generation model and set the loss function of the device control strategy generation model; Step S2: Obtain historical environmental monitoring data, preprocess and label the historical environmental monitoring data, and then construct a dataset; Step S3: Divide the dataset into a training set, a validation set, and a test set based on stratified sampling. Use the training set, validation set, and test set to train, validate, and test the device control strategy generation model, respectively. Step S4: Combining quantization and dynamic pruning techniques, the tested device control strategy generation model is compressed. After performance verification of the compressed device control strategy generation model, it is deployed to network devices of the device type, such as routers, gateways, or cameras, using containerization technology. Step S5: The network device collects real-time environmental monitoring data through a sensor array, preprocesses the real-time environmental monitoring data, and inputs it into the deployed device control strategy generation model to obtain a real-time device control strategy. Based on the real-time device control strategy, the operating parameters are dynamically adjusted through a preset adjustment cycle. Step S6: The network device records the operation log in real time, and encrypts and stores the operation log and performs distributed backup. Step S1 specifically involves: An equipment control strategy generation model is created based on an environmental feature extraction module, an environmental feature fusion module, and a parameter adjustment strategy generation module, and a loss function is set for the equipment control strategy generation model. The environmental feature extraction module is constructed based on a light feature extraction unit, a temperature feature extraction unit, a humidity feature extraction unit, an electromagnetic feature extraction unit, a vibration feature extraction unit, an infrared feature extraction unit, a noise feature extraction unit, and a mode selection gating unit. The light feature extraction unit extracts light features from light data using a first 1D convolutional network; the temperature feature extraction unit extracts temperature features from temperature data using a second 1D convolutional network; the humidity feature extraction unit extracts humidity features from humidity data using a third 1D convolutional network; the electromagnetic feature extraction unit extracts electromagnetic features from electromagnetic interference intensity using a fourth 1D convolutional network; the vibration feature extraction unit extracts vibration features from physical vibration amplitude using a fifth 1D convolutional network; the infrared feature extraction unit extracts infrared features from infrared data using MobileNetV2; and the noise feature extraction unit extracts noise features from the environmental noise spectrum using Mel spectrum transformation and a BiLSTM network. The mode selection gating unit weights the light, temperature, humidity, electromagnetic, vibration, infrared, and noise features using a gating attention mechanism to obtain a multimodal feature vector group. The environmental feature fusion module is constructed based on a multi-scale feature interaction unit and an adaptive feature compression unit. The multi-scale feature interaction unit is used to capture long-term cross-modal dependency features from a multi-modal feature vector group through a hierarchical Transformer structure. The adaptive feature compression unit is used to compress long-term dependency features through learnable gated convolution to obtain compressed fused features. The parameter adjustment strategy generation module is constructed based on the strategy generation unit and the physical constraint unit. The strategy generation unit is used to infer the compressed fusion features through a 3-layer residual fully connected network to obtain the initial adjustment values ​​of the device operating parameters. The physical constraint unit is used to map the initial adjustment values ​​to the physical feasible region through a differential equation embedding layer and differentiable projection to obtain the suggested adjustment values ​​of the device operating parameters, and output the device control strategy carrying the suggested adjustment values. The formula for the loss function is: ; in, This represents the loss value of the loss function; This represents the mean square error of parameter prediction; This represents the policy smoothness regularization term; This indicates a penalty for violating physical constraints; and All of these represent weighting coefficients.

2. The environmental perception and control method for a router, gateway, and camera as described in claim 1, characterized in that: Step S2 specifically involves: Historical environmental monitoring data is acquired, including light intensity, temperature, humidity, electromagnetic interference intensity, physical vibration amplitude, infrared data, and environmental noise spectrum. The historical environmental monitoring data are preprocessed, including at least data cleaning, data denoising, data normalization, and feature extraction. The preprocessed historical environmental monitoring data are then labeled with equipment control strategies to construct a dataset.

3. The environmental perception and control method for a router, gateway, and camera as described in claim 1, characterized in that: Step S3 specifically involves: Based on stratified sampling, the dataset is divided into a training set, a validation set, and a test set in a ratio of 8:1:

1. The device control strategy generation model is trained using the training set and a loss function. During the training process, the hyperparameters of the device control strategy generation model are continuously optimized until the loss value of the loss function is less than a preset loss threshold or a preset early stop condition is met. The accuracy of the strategy is calculated using the validation set to validate the trained device control strategy generation model. If the validation fails, the training set is expanded and training continues; if the validation passes, then: The F1 score and confidence level are calculated using the test set to test the validated device control strategy generation model. If the test fails, the training set is expanded and training continues; if the test passes, training ends.

4. The environmental perception and control method for a router, gateway, and camera as described in claim 1, characterized in that: Step S6 specifically involves: Network devices record operational logs in real time, including at least basic device information, user operation records, configuration change records, system events, and hardware status. The log fingerprint of the running log is calculated using the HMAC algorithm. The running log and the log fingerprint are then encrypted into Level 1 encrypted data using the AES algorithm. The Level 1 encrypted data is converted into hexadecimal data. The numbers 7 and A, and 9 and B in the hexadecimal data are swapped to obtain Level 2 encrypted data. The Level 2 encrypted data is then encrypted into Level 3 encrypted data using the RSA algorithm. Each character in the Level 3 encrypted data is shifted 8 bits to the left to obtain Level 4 encrypted data. The Level 4 encrypted data is then encrypted into an encrypted log using the RC6 algorithm. The encrypted log is stored in a preset path and backed up in a distributed manner.

5. A router, gateway, and camera environmental perception control system, characterized in that: Includes the following modules: The device control strategy generation model creation module is used to create a device control strategy generation model and set the loss function of the device control strategy generation model. The dataset construction module is used to acquire historical environmental monitoring data, preprocess and label the historical environmental monitoring data, and then construct the dataset. The equipment control strategy generation model training module is used to divide the dataset into a training set, a validation set, and a test set based on stratified sampling, and to train, validate, and test the equipment control strategy generation model using the training set, validation set, and test set, respectively. The device control strategy generation model deployment module is used to compress the tested device control strategy generation model by combining quantization technology and dynamic pruning technology. After performance verification of the compressed device control strategy generation model, it is deployed to network devices of device type router, gateway or camera through containerization technology. The environmental perception and control module is used by network devices to collect real-time environmental monitoring data through a sensor array, preprocess the real-time environmental monitoring data and input it into the deployed device control strategy generation model to obtain a real-time device control strategy, and dynamically adjust the operating parameters based on the real-time device control strategy through a preset adjustment cycle. The operation log management module is used for network devices to record operation logs in real time, and to encrypt, store, and distribute the operation logs. The device control strategy generation model creation module is specifically used for: An equipment control strategy generation model is created based on an environmental feature extraction module, an environmental feature fusion module, and a parameter adjustment strategy generation module, and a loss function is set for the equipment control strategy generation model. The environmental feature extraction module is constructed based on a light feature extraction unit, a temperature feature extraction unit, a humidity feature extraction unit, an electromagnetic feature extraction unit, a vibration feature extraction unit, an infrared feature extraction unit, a noise feature extraction unit, and a mode selection gating unit. The light feature extraction unit extracts light features from light data using a first 1D convolutional network; the temperature feature extraction unit extracts temperature features from temperature data using a second 1D convolutional network; the humidity feature extraction unit extracts humidity features from humidity data using a third 1D convolutional network; the electromagnetic feature extraction unit extracts electromagnetic features from electromagnetic interference intensity using a fourth 1D convolutional network; the vibration feature extraction unit extracts vibration features from physical vibration amplitude using a fifth 1D convolutional network; the infrared feature extraction unit extracts infrared features from infrared data using MobileNetV2; and the noise feature extraction unit extracts noise features from the environmental noise spectrum using Mel spectrum transformation and a BiLSTM network. The mode selection gating unit weights the light, temperature, humidity, electromagnetic, vibration, infrared, and noise features using a gating attention mechanism to obtain a multimodal feature vector group. The environmental feature fusion module is constructed based on a multi-scale feature interaction unit and an adaptive feature compression unit. The multi-scale feature interaction unit is used to capture long-term cross-modal dependency features from a multi-modal feature vector group through a hierarchical Transformer structure. The adaptive feature compression unit is used to compress long-term dependency features through learnable gated convolution to obtain compressed fused features. The parameter adjustment strategy generation module is constructed based on the strategy generation unit and the physical constraint unit. The strategy generation unit is used to infer the compressed fusion features through a 3-layer residual fully connected network to obtain the initial adjustment values ​​of the device operating parameters. The physical constraint unit is used to map the initial adjustment values ​​to the physical feasible region through a differential equation embedding layer and differentiable projection to obtain the suggested adjustment values ​​of the device operating parameters, and output the device control strategy carrying the suggested adjustment values. The formula for the loss function is: ; in, This represents the loss value of the loss function; This represents the mean square error of parameter prediction; This represents the policy smoothness regularization term; This indicates a penalty for violating physical constraints; and All of these represent weighting coefficients.

6. The router, gateway, and camera environmental perception control system as described in claim 5, characterized in that: The dataset construction module is specifically used for: Historical environmental monitoring data is acquired, including light intensity, temperature, humidity, electromagnetic interference intensity, physical vibration amplitude, infrared data, and environmental noise spectrum. The historical environmental monitoring data are preprocessed, including at least data cleaning, data denoising, data normalization, and feature extraction. The preprocessed historical environmental monitoring data are then labeled with equipment control strategies to construct a dataset.

7. The router, gateway, and camera environmental perception control system as described in claim 5, characterized in that: The device control strategy generation model training module is specifically used for: Based on stratified sampling, the dataset is divided into a training set, a validation set, and a test set in a ratio of 8:1:

1. The device control strategy generation model is trained using the training set and a loss function. During the training process, the hyperparameters of the device control strategy generation model are continuously optimized until the loss value of the loss function is less than a preset loss threshold or a preset early stop condition is met. The accuracy of the strategy is calculated using the validation set to validate the trained device control strategy generation model. If the validation fails, the training set is expanded and training continues; if the validation passes, then: The F1 score and confidence level are calculated using the test set to test the validated device control strategy generation model. If the test fails, the training set is expanded and training continues; if the test passes, training ends.

8. The router, gateway, and camera environmental perception control system as described in claim 5, characterized in that: The operation log management module is specifically used for: Network devices record operational logs in real time, including at least basic device information, user operation records, configuration change records, system events, and hardware status. The log fingerprint of the running log is calculated using the HMAC algorithm. The running log and the log fingerprint are then encrypted into Level 1 encrypted data using the AES algorithm. The Level 1 encrypted data is converted into hexadecimal data. The numbers 7 and A, and 9 and B in the hexadecimal data are swapped to obtain Level 2 encrypted data. The Level 2 encrypted data is then encrypted into Level 3 encrypted data using the RSA algorithm. Each character in the Level 3 encrypted data is shifted 8 bits to the left to obtain Level 4 encrypted data. The Level 4 encrypted data is then encrypted into an encrypted log using the RC6 algorithm. The encrypted log is stored in a preset path and backed up in a distributed manner.

Images