A wireless communication security transmission method based on an internet of things gateway

Abstract

The application discloses a wireless communication security transmission method based on an Internet of Things gateway, which comprises the following steps: collecting original data and communication identification, and analyzing and generating standardized device data and connection relationship data; cleaning the standardized device data, extracting features, and constructing a communication behavior history sequence; collecting link data to generate link state parameters and evaluation results; inputting the communication behavior history sequence, data features, and link state evaluation results into a liquid time constant network to generate a dynamic security state; performing covariance evolution optimization to update parameters based on the dynamic security state and the link state parameters; generating a security transmission strategy based on the dynamic security state; and performing encryption and wireless transmission control based on the security transmission strategy to complete security transmission. The application realizes security control and communication collaborative optimization on the gateway side, and improves transmission stability and security.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) communication and network security technology, and in particular to a secure wireless communication transmission method based on an IoT gateway. Background Technology

[0002] In IoT networks, numerous terminal devices connect to gateways via wireless links and exchange data. Gateways handle protocol conversion, data forwarding, and some edge processing functions. Regarding secure wireless transmission, existing technologies primarily perform data encryption at the terminal side or centralized security control in the cloud. This involves encrypting, authenticating, and controlling access to the raw data, then scheduling transmission based on link quality indicators. Some solutions introduce link state awareness mechanisms to monitor received signal strength, signal-to-noise ratio, packet loss rate, and transmission latency, adjusting transmission rates and retransmission strategies based on rules or simple models. A few solutions incorporate neural networks to model communication behavior, combining anomaly detection methods to identify abnormal communication activity, and then updating security policies accordingly.

[0003] Existing technologies still have significant shortcomings. Gateways primarily handle data forwarding and basic processing, lacking a comprehensive security control structure for the entire communication process, and data lacks a unified protection mechanism during access and transmission. Link state awareness and security policy generation are often handled independently, lacking a collaborative mechanism to dynamically adjust security policies based on link state parameter sets, resulting in a lack of correlation between security policies and changes in the wireless environment. Communication behavior modeling often employs discrete-time models or fixed-window processing, making it difficult to express the evolutionary characteristics of communication behavior over continuous time, and insufficient in jointly characterizing short-term fluctuations and long-term trends. Internal model parameters often use fixed or simple update methods, lacking an adaptive optimization process based on distributed modeling and covariance evolution, limiting the model's adaptability to complex wireless environments. Abnormal communication handling mostly remains at the detection level, lacking an adjustment mechanism linked to the state evolution process, and the overall secure transmission process lacks closed-loop update capabilities.

[0004] Therefore, how to provide a secure wireless communication transmission method based on IoT gateways is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This invention proposes a secure transmission method for wireless communication of IoT gateways based on liquid time constant networks and covariance evolution optimization. It continuously models communication behavior and drives the dynamic generation of security policies, realizing coordinated security control and communication scheduling on the gateway side. It has the advantages of strong environmental adaptability, high security and good transmission stability.

[0006] A secure wireless communication transmission method based on an Internet of Things (IoT) gateway according to an embodiment of the present invention includes: Collect raw data and device communication identification data from multi-source heterogeneous terminal devices, perform protocol parsing and format unification processing on the raw data, and generate standardized device data and device connection relationship data; Perform data cleaning and feature extraction on standardized equipment data, construct a data feature set, and build a historical sequence of communication behavior based on equipment connection relationship data; Collect wireless link operation data and perform state awareness and assessment processing to generate a set of link state parameters and link state assessment results; The historical sequence of communication behavior, data feature set and link state evaluation results are input into the liquid time constant network of time constant modulation unit, multi-scale liquid memory unit and transition triggering unit. Time constant dynamic modulation is performed in time constant modulation unit, hierarchical liquid recursion is performed in multi-scale liquid memory unit, and anomaly judgment and state transition update are performed in transition triggering unit to generate dynamic security state. Based on the dynamic security state, link state parameter set and data feature set, a time-varying covariance evolution strategy is constructed. The parameters of the liquid time constant network are modeled for parameter distribution, adaptively updated for covariance and optimized for distribution sampling. During the iterative update of parameter distribution, candidate parameter evaluation and group update are completed, and the optimized model parameters are generated and the liquid time constant network is updated. Based on the dynamic security state output by the updated liquid time constant network, a set of secure transmission strategy parameters is generated. Based on the set of secure transmission strategy parameters, data encryption and wireless transmission control are performed on standardized equipment data to achieve secure transmission of standardized equipment data.

[0007] Optionally, the original data includes business data content, data packet length, data generation time, data transmission frequency, and the business type to which the data belongs. The device communication identification data includes a unique device identifier, device access type, communication protocol type, device address information, channel access identifier, and communication port identifier.

[0008] Optionally, the generation of standardized device data and device connection relationship data includes: The raw data and device communication identification data are time-aligned to generate the raw data sequence. The original data sequence is processed by protocol identification. The original data sequence is matched based on a preset protocol feature set to obtain the corresponding protocol type identifier. The original data sequence is then processed by protocol parsing to generate a parsed data sequence. The parsed data sequence is processed by field extraction. According to the unified data structure definition, the data content field, timestamp field, device identifier field and communication parameter field are extracted and encoded and mapped to generate a structured data sequence. Perform data normalization processing on structured data sequences to generate standardized equipment data; Based on the device communication identification data, correlation analysis is performed to construct device connection pairs according to the order of communication occurrence, and weights are assigned to the connection pairs according to the communication frequency to generate device connection relationship data.

[0009] Optionally, the construction of a data feature set and the construction of a communication behavior history sequence based on device connection relationship data includes: Perform outlier detection and processing on standardized equipment data to generate a cleaned data sequence; Missing value processing is performed on the cleaned data sequence, and the missing field data is filled in according to the mean of the time neighborhood to generate a complete data sequence; Feature extraction processing is performed on the completed data sequence. Service type features and data length features are extracted from the data content field, time interval features and transmission frequency features are extracted from the timestamp field, and channel occupancy features and transmission strength features are extracted from the communication parameter field to generate an initial feature set. The initial feature set is subjected to feature encoding processing, the discrete features are subjected to number mapping processing, and the continuous features are subjected to interval partitioning processing to generate an encoded feature set. The encoded feature set is weighted according to a preset weight coefficient to generate a data feature set; Based on device connection relationship data, the device connection pairs are arranged in the order of communication occurrence time. The data feature set corresponding to each device connection pair is spliced ​​in chronological order to form a sequence structure, and the features of each time step in the sequence are weighted to generate a historical sequence of communication behavior.

[0010] Optionally, the generation of the link state parameter set and link state evaluation results includes: Collect wireless link operation data, including received signal strength, signal-to-noise ratio, channel bandwidth, interference intensity, number of lost packets, total number of transmitted data packets, number of retransmissions, and transmission delay; Perform time window partitioning on the wireless link operation data to generate a window link data sequence; Perform parameter calculations on the window link data sequence and normalize it to generate a standardized set of link state parameters; The standardized set of link state parameters is weighted and fused according to preset weight coefficients to generate a link state evaluation value. The link state assessment value is used to perform hierarchical mapping processing to generate the link state assessment result.

[0011] Optionally, generating a dynamic security state includes: The historical sequence of communication behavior is divided into multiple consecutive time step sequences according to time order, and the data feature set in the time step sequence is vectorized to generate a time step input vector sequence; A liquid time constant network structure is constructed, which includes a time constant modulation unit, a multi-scale liquid memory unit, and a transition trigger unit. Different units are connected in a directed manner according to the network connection relationship to form a liquid recursive structure. A state inertia processing path and a nonlinear mapping processing path are set in the improved liquid time constant network. In the time constant modulation unit, the set of link state parameters is mapped to time constant modulation coefficients, the time constant of each neuron in the liquid time constant network is updated, and the original time constant and the time constant modulation coefficients are weighted and summed to obtain the modulated time constant; In the multi-scale liquid memory unit, the time step input vector sequence is input into the liquid recursive structure, and the state update processing is performed on each time step according to the recursive update rule. The state propagation is performed in the short-time memory channel and the long-time memory channel respectively to generate short-time feature sequences and long-time feature sequences. The fusion feature sequence is generated by weighted summation according to the preset fusion weight. The fused feature sequence is continuously evolved through a state inertia processing path and a nonlinear mapping processing path to obtain a continuously evolving state vector. In the transition triggering unit, anomaly detection and state transition update processing are performed on the continuously evolving state vector, and the corresponding feature value is adjusted to the sum of the preset transition amplitude value and the original feature value to generate the transition state vector. Normalize the transition state vector to generate a dynamic safe state.

[0012] Optionally, generating optimized model parameters and updating the liquid time constant network includes: The parameters of the liquid time constant network are vectorized to generate a parameter distribution vector; A covariance matrix is ​​constructed based on the parameter distribution vector. Initialization processing is performed on the covariance matrix by calculating the variance of each parameter dimension in the parameter distribution vector and filling it with diagonal elements to obtain the initial covariance matrix. The initial covariance matrix is ​​scaled based on the link state parameter set. The parameter values ​​in the link state parameter set are multiplied by the preset adjustment coefficient, and the corresponding dimensions in the covariance matrix are scaled to generate a time-varying covariance matrix. Sampling processing is performed on the parameter distribution vector based on the time-varying covariance matrix to generate a candidate parameter set according to the multidimensional normal distribution; The fitness evaluation process is performed on the candidate parameter set. The candidate parameter set is input into the liquid time constant network respectively. The fitness value is calculated by adding the reciprocal of the sum of the absolute values ​​of the differences between the dynamic security state and the link state evaluation results to a preset constant, and a candidate parameter fitness sequence is generated. Based on the candidate parameter fitness sequence, update the parameter distribution vector and covariance matrix to generate updated parameter distribution vector and updated covariance matrix; Based on the updated parameter distribution vector, the parameters of the liquid time constant network are updated in groups. The parameter distribution vector is segmented according to the parameter type, and the parameters of each segment are assigned to each unit of the model to generate the optimized model parameters and complete the update of the liquid time constant network.

[0013] Optionally, the set of parameters for generating secure transmission strategies includes: Perform policy mapping on dynamic security states, and calculate the mean, maximum, minimum and high-risk state components of the state components. Security level mapping is performed based on the mean of state components and the number of high-risk state components to generate security level results; Based on the security level results, parameter mapping is performed to generate a set of secure transmission policy parameters, including encryption strength parameters, key update cycle parameters, and retransmission control parameters.

[0014] Optionally, the secure transmission of standardized device data includes: Perform block processing on standardized equipment data to generate a sequence of blocks to be encrypted; Multiple rounds of encryption operations are performed on the sequence of blocks to be encrypted based on the encryption strength parameter, and the number of encryption rounds is adjusted based on the link state evaluation results to generate an encrypted data sequence. A time-series key set is generated based on the key update cycle parameter, and the time-series key set is matched to the corresponding data block according to the transmission time information of each data block in the encrypted data sequence. Based on the retransmission control parameters and link status evaluation results, the encrypted data sequence is processed for transmission scheduling, a transmission queue is generated, and retransmission control is performed on the data blocks that fail to be transmitted. Perform wireless transmission processing on the data blocks in the transmission queue to complete the secure transmission of standardized device data.

[0015] The beneficial effects of this invention are: By introducing a unified processing flow of communication behavior history sequence, data feature set and link state parameter set on the IoT gateway side, the original data of terminal devices, device connection relationship data and wireless link operation data are incorporated into the same modeling framework. The continuous time feature expression is completed through liquid time constant network, which improves the problem of insufficient expression of communication behavior evolution features caused by static data processing and discrete time modeling in the existing technology. It enables the short-term fluctuations and long-term trends of communication behavior to be characterized in a unified process, and improves the completeness and continuity of communication state representation.

[0016] By constructing a liquid time constant network structure that incorporates time constant modulation units, disturbance suppression units, and state transition mechanisms, the dynamic modulation of the network's internal time constant by the link state parameter set is achieved. The security state evolution is completed during the state inertia and nonlinear mapping process, and state transition updates are performed under abnormal fluctuation conditions. This improves the problems of separation between security policies and link states in existing technologies, as well as the difficulty in handling sudden interference and attack behaviors. It enables the dynamic security state to simultaneously reflect changes in the communication environment and abnormal disturbance information, thereby improving the adaptability and responsiveness of security policy generation.

[0017] By constructing a time-varying covariance evolution strategy based on parameter distribution modeling, covariance adaptive updating, and distributed sampling optimization, and combining the link state parameter set with the dynamic security state to form a parameter update closed loop, the liquid time constant network parameters are continuously iteratively optimized during candidate parameter evaluation and group update. This improves the problem of insufficient adaptability to complex wireless environments caused by fixed or single-update model parameters in existing technologies, enabling model parameters to be continuously adjusted with changes in the communication environment, thereby improving the stability and overall optimization capability of the secure transmission process. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They explain the invention together with the embodiments of the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a secure wireless communication transmission method based on an Internet of Things gateway proposed in this invention; Figure 2 This is a schematic diagram of an improved liquid time constant network structure for a secure wireless communication transmission method based on an Internet of Things gateway proposed in this invention. Figure 3 This is a schematic diagram illustrating the optimization process of the time-varying covariance evolution strategy for a wireless communication secure transmission method based on an Internet of Things gateway proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figure 1 , Figure 2 and Figure 3 A secure wireless communication transmission method based on an IoT gateway includes: Collect raw data and device communication identification data from multi-source heterogeneous terminal devices, perform protocol parsing and format unification processing on the raw data, and generate standardized device data and device connection relationship data; Perform data cleaning and feature extraction on standardized equipment data, construct a data feature set, and build a historical sequence of communication behavior based on equipment connection relationship data; Collect wireless link operation data and perform state awareness and assessment processing to generate a set of link state parameters and link state assessment results; The historical sequence of communication behavior, data feature set and link state evaluation results are input into the liquid time constant network of time constant modulation unit, multi-scale liquid memory unit and transition triggering unit. Time constant dynamic modulation is performed in time constant modulation unit, hierarchical liquid recursion is performed in multi-scale liquid memory unit, and anomaly judgment and state transition update are performed in transition triggering unit to generate dynamic security state. Based on the dynamic security state, link state parameter set and data feature set, a time-varying covariance evolution strategy is constructed. The parameters of the liquid time constant network are modeled for parameter distribution, adaptively updated for covariance and optimized for distribution sampling. During the iterative update of parameter distribution, candidate parameter evaluation and group update are completed, and the optimized model parameters are generated and the liquid time constant network is updated. Based on the dynamic security state output by the updated liquid time constant network, a set of secure transmission strategy parameters is generated. Based on the set of secure transmission strategy parameters, data encryption and wireless transmission control are performed on standardized equipment data to achieve secure transmission of standardized equipment data.

[0021] In this embodiment, the original data includes business data content, data packet length, data generation time, data transmission frequency, and the business type to which the data belongs. The device communication identification data includes the device unique identifier, device access type, communication protocol type, device address information, channel access identifier, and communication port identifier.

[0022] In this embodiment, generating standardized device data and device connection relationship data includes: The raw data and device communication identification data are time-aligned to generate a raw data sequence. Specifically, the time alignment process involves: The timestamp of each record in the original data is compared with the timestamp of the corresponding device communication identifier. The order of the original data records is adjusted by calculating the time difference to ensure that all data of the same device are arranged in chronological order. The time window is set to 1 second. Records exceeding the window are adjusted by forward or backward interpolation to generate a continuous time series. Each record in the generated original data series contains the original data field and the corresponding device identifier information. The original data sequence undergoes protocol identification processing. Based on a preset protocol feature set, the original data sequence is matched to obtain the corresponding protocol type identifier. Then, protocol parsing processing is performed on the original data sequence to generate a parsed data sequence. Specifically, the protocol identification processing on the original data sequence includes: The protocol header identifier, function code, data length, and checksum are extracted from each raw data record. Each record is then matched against a preset protocol feature set. Records with a matching degree greater than 0.8 are determined to be of the corresponding protocol type; otherwise, an exception handling process is initiated. The protocol parsing process converts the binary or hexadecimal raw data into structured field values ​​based on the determined protocol type, including data content fields, command fields, and control information fields. This generates a parsed data sequence. Each record contains the parsed field values, the original timestamp, and device identification information. The parsed data sequence undergoes field extraction processing. Data content fields, timestamp fields, device identifier fields, and communication parameter fields are extracted according to a unified data structure definition, and encoding mapping processing is performed to generate a structured data sequence. Specifically, the field extraction processing for the parsed data sequence includes: The data content field, timestamp field, device identifier field, and communication parameter field are extracted from the parsed data sequence. Each field is arranged according to a unified data structure. The data content field is numerically encoded, mapping text or enumeration type to integer numbers from 1 to 256. The timestamp field is converted into a sequence of seconds starting from zero. The device identifier field is mapped to integer numbers from 1 to 128. The channel occupancy value in the communication parameter field is mapped to the range of 0 to 1. The transmission strength field is scaled to 0 to 1. After encoding and mapping, a structured data sequence is generated. Perform data normalization processing on structured data sequences to generate standardized equipment data; Based on device communication identification data, association analysis is performed to construct device connection pairs according to the communication sequence. Weights are then assigned to these connection pairs based on communication frequency to generate device connection relationship data. Specifically, the weight assignment process based on communication frequency is as follows: Based on the source and destination device information in the device communication identification data, device connection pairs are constructed in the order of communication occurrence time. Each connection pair records the start and end time of communication and the number of communications. The connection pairs are assigned weights based on the statistical communication frequency, with the weight range being 0 to 1. The maximum communication frequency corresponds to a weight of 1, the minimum communication frequency corresponds to a weight of 0, and the rest are calculated according to a linear ratio to generate device connection relationship data.

[0023] In this embodiment, the construction of a data feature set and the construction of a communication behavior history sequence based on device connection relationship data include: Perform outlier detection and processing on standardized equipment data to generate a cleaned data sequence; Missing value processing is performed on the cleaned data sequence, and the missing field data is filled in according to the mean of the time neighborhood to generate a complete data sequence; Feature extraction processing is performed on the completed data sequence. Service type and data length features are extracted from the data content field; time interval and transmission frequency features are extracted from the timestamp field; and channel occupancy and transmission strength features are extracted from the communication parameter field, generating an initial feature set. Specifically, the feature extraction processing on the completed data sequence includes: From the data content field of the completed data sequence, extract the business type feature and data length feature. The business type feature is generated into an integer number according to the classification coding method, and the data length feature is retained as the original value. Calculate the time interval feature and transmission frequency feature from the timestamp field. The time interval is the time difference between adjacent data records, and the transmission frequency is the number of data packets per unit time. Extract the channel occupancy feature and transmission strength feature from the communication parameter field. The channel occupancy is calculated by the ratio of occupancy duration, and the transmission strength is obtained by signal strength normalization. Generate an initial feature set, and form an 8-dimensional feature vector for each record. Feature encoding is performed on the initial feature set, number mapping is performed on discrete features, and interval partitioning is performed on continuous features to generate an encoded feature set. Specifically, the feature encoding process on the initial feature set is as follows: Perform number mapping on discrete features in the initial feature set, mapping each type of discrete feature to an integer number from 1 to 256. Perform interval division processing on continuous features, dividing the continuous numerical field into 10 intervals and mapping them to integer numbers to generate an coded feature set. The feature dimension of each record is expanded to 16 dimensions, while maintaining the consistency of the feature order. The encoded feature set is subjected to feature weighting processing according to preset weight coefficients to generate a data feature set. Specifically, the feature weighting processing of the encoded feature set according to preset weight coefficients is as follows: The data feature set is generated by weighting and summing each dimension according to the preset weight coefficients. The weight of the service type feature is set to 0.25, the weight of the data length feature is set to 0.15, the weight of the time interval feature is set to 0.2, the weight of the transmission frequency feature is set to 0.1, the weight of the channel occupancy feature is set to 0.15, and the weight of the transmission strength feature is set to 0.15. The feature value of each record is multiplied by the corresponding weight and then summed to generate the data feature set. Based on device connection relationship data, device connection pairs are arranged according to the order of communication occurrence. The data feature sets corresponding to each device connection pair are concatenated in chronological order to form a sequence structure, and weighted processing is performed on the features at each time step in the sequence to generate a communication behavior history sequence. Specifically, generating the communication behavior history sequence involves: Based on the device connection relationship data, the device connection pairs are arranged in the order of communication occurrence time. The data feature sets corresponding to the connection pairs are spliced ​​in time order to form a sequence structure with a sequence length of 480 to 960 time steps. The feature vector of each time step in the sequence has a dimension of 16. The feature vector of each time step is multiplied by a time weight, which is linearly distributed from 1 to 0.1 in decreasing proportion, to generate the final communication behavior history sequence.

[0024] In this embodiment, the generation of the link state parameter set and the link state evaluation result includes: Collect wireless link operation data, including received signal strength, signal-to-noise ratio, channel bandwidth, interference intensity, number of packet losses, total number of transmitted data packets, number of retransmissions, and transmission delay. Specifically, the collection of wireless link operation data includes: The wireless link of each terminal device connected to the gateway is monitored in real time. The received signal strength is recorded in negative decibels and milliwatts, the signal-to-noise ratio is recorded in decibels, the channel bandwidth is recorded in MHz, the interference intensity is recorded as the signal-to-interference ratio, the number of lost packets is recorded as the number of data packets lost per second, the total number of transmitted data packets is recorded as the total number of data packets transmitted per second, the number of retransmissions is recorded as the number of data packets retransmitted per second, and the transmission delay is recorded as the average time from data packet transmission to reception, in milliseconds. Each record is accompanied by a timestamp and device identification information. The wireless link operation data is divided into time windows to generate a window link data sequence. Specifically, the time window division process for the wireless link operation data is as follows: The continuous wireless link operation data is divided into fixed-length time windows, with the window length set to 1 second. Each window contains the collection records of all devices within the time period, arranged in chronological order to generate a window link data sequence. Each sequence record includes the start and end time of the window, the mean and variance statistics of each indicator. The window link data sequence is processed by parameter calculation and normalization to generate a standardized set of link state parameters. Specifically, the parameter calculation process for the window link data sequence involves: For each window, calculate the channel quality parameter, which is the ratio of received signal strength to interference strength. Calculate the packet loss rate parameter, which is the number of lost packets divided by the total number of transmitted data packets. Calculate the retransmission rate parameter, which is the number of retransmissions divided by the total number of transmitted data packets. Calculate the average transmission delay, which is the arithmetic mean of the delays of all data packets within the window. Normalize the calculated parameters to the interval between 0 and 1 by subtracting the minimum and maximum values ​​to generate a standardized set of link state parameters. Each record contains the normalized channel quality, packet loss rate, retransmission rate, and average delay. The standardized link state parameter set is weighted and fused according to preset weight coefficients to generate a link state evaluation value. Specifically, the weighted fusion process of the standardized link state parameter set according to preset weight coefficients is as follows: The parameters in the standardized link state parameter set are weighted and summed according to preset weight coefficients. The weight coefficients are set to 0.4 for channel quality, 0.2 for packet loss rate, 0.2 for retransmission rate, and 0.2 for average latency. The parameter values ​​of each record are multiplied by their corresponding weights and then summed to obtain a single fused value, which generates the link state evaluation value. A hierarchical mapping process is performed based on the link state assessment values ​​to generate link state assessment results. Specifically, the hierarchical mapping process based on the link state assessment values ​​involves: The link status assessment value is divided into five levels according to preset thresholds: level 1 represents the best link and level 5 represents the worst link. The thresholds are 0.0 to 0.2, 0.2 to 0.4, 0.4 to 0.6, 0.6 to 0.8, and 0.8 to 1.0, respectively. The fusion value corresponding to each record is mapped to the corresponding level to generate the link status assessment result, which is used for dynamic security status evolution and security policy adjustment.

[0025] In this embodiment, generating a dynamic security state includes: The historical sequence of communication behavior is divided into multiple consecutive time-step sequences according to chronological order, and the data feature sets in the time-step sequences are vectorized to generate a time-step input vector sequence. Specifically, the historical sequence of communication behavior is divided into multiple consecutive time-step sequences according to chronological order as follows: The historical sequence of communication behavior is divided into a continuous sequence of time steps of fixed length according to the time order. The length of each time step is set to 1 second. Each time step contains data records of all terminal devices within the time period. The data is arranged in ascending order according to the timestamp. Each record retains information such as service type, data length, time interval, transmission frequency, channel occupancy and transmission strength to ensure that the data of the same device is arranged continuously in the sequence. The data feature set in the time step sequence is vectorized and represented as follows: The data feature set within each time step is numerically processed, the service type is mapped to an integer number from 1 to 256, the data length is normalized to the range of 0 to 1 according to the actual byte value, the time interval and the transmission frequency are normalized to the range of 0 to 1 respectively, and the channel occupancy and transmission strength are scaled to the range of 0 to 1. Each processed record is concatenated into a 16-dimensional vector to form the input vector of the time step. All vectors of each time step are arranged in chronological order to generate a sequence of time step input vectors. The sequence length varies between 480 and 960 time steps depending on the device communication frequency. A liquid time constant network structure is constructed, comprising time constant modulation units, multi-scale liquid memory units, and transition triggering units. Different units are directionally connected according to network connectivity relationships, forming a liquid recursive structure. State inertia processing paths and nonlinear mapping processing paths are set in the improved liquid time constant network. Specifically, the liquid time constant network structure is constructed as follows: The traditional liquid time constant network retains the main branches of input mapping, continuous time state recursion, and output mapping. The input interface of time step input vector sequence is retained at the original input end. The original single liquid recursive state update structure is transformed into a main processing structure composed of time constant modulation unit, multi-scale liquid memory unit and transition trigger unit. State inertia processing path and nonlinear mapping processing path are set in the main processing structure. The traditional liquid time constant network structure oriented towards single continuous state evolution is transformed into an improved liquid time constant network structure oriented towards link state drive, short-time and long-time dual-scale memory, and abnormal state transition update. The time step input vector sequence is first input to the time constant modulation unit, which outputs the modulated time constant. Then, the time step input vector sequence and the modulated time constant are input together to the multi-scale liquid memory unit. The multi-scale liquid memory unit outputs short-time feature sequences and long-time feature sequences. The short-time feature sequences and long-time feature sequences are fused to form a fused feature sequence. The fused feature sequence enters the state inertia processing region along the state inertia processing path, and then enters the nonlinear mapping region along the nonlinear mapping processing path. The nonlinear mapping region outputs a continuously evolving state vector. The continuously evolving state vector is input to the transition trigger unit, which outputs a transition state vector. The transition state vector is used as the dynamic safety state output. The time constant modulation unit includes: Link parameter input area: Receives channel quality parameters, packet loss rate parameters, retransmission rate parameters, and delay parameters from the link status parameter set; Weighted mapping region: The channel quality parameter is assigned a modulation weight of 0.4, the packet loss rate parameter is assigned a modulation weight of 0.2, the retransmission rate parameter is assigned a modulation weight of 0.2, and the delay parameter is assigned a modulation weight of 0.2. Modulation coefficient generation area: Perform product calculation on the four types of link parameters and their corresponding modulation weights respectively, and then sum the four product results to obtain the time constant modulation coefficient; Time constant update region: The original time constant and the time constant modulation coefficient are weighted and summed at 0.6 and 0.4 respectively to obtain the modulated time constant; The time constant modulation unit is set in series at the front end of the liquid recursive backbone. The link parameter input area first reads the link state parameter set, the weight mapping area completes the correspondence between the link parameters and the modulation weights, the modulation coefficient generation area forms the time constant modulation coefficient, and the time constant update area completes the original time constant update. This transforms the fixed time constant update method inside the traditional liquid time constant network into a dynamic time constant update method driven by the link state. Multi-scale liquid memory units include: Input distribution area: Receives the time-step input vector sequence and the modulated time constant, and distributes the input to the short-time memory channel and long-time memory channel according to the uniform time-step number; Short-term memory channel: Perform liquid recursive state update on the most recent 8 time steps to form a short-term feature sequence. The state of each time step in the short-term memory channel is calculated by the short-term state vector of the previous time step and the input vector of the current time step. Long-term memory channel: Perform liquid recursive state update on the most recent 64 time steps to form a long-term feature sequence. The state of each time step in the long-term memory channel is calculated by the long-term state vector of the previous time step and the input vector of the current time step. Fusion output region: The short-time feature sequence is assigned a fusion weight of 0.6, the long-time feature sequence is assigned a fusion weight of 0.4, and the features at the corresponding time steps are weighted and summed to form a fused feature sequence; The multi-scale liquid memory unit is set after the time constant modulation unit. The input distribution area first completes the unified time step distribution. The short-time memory channel forms a short-time fluctuation representation, and the long-time memory channel forms a long-time trend representation. The fusion output area combines the two types of representations into a unified fusion feature sequence, transforming the traditional liquid time constant network single-channel recursive memory structure into a short-time and long-time dual-channel parallel liquid memory structure. The transition trigger unit includes: Difference detection region: Performs element-by-element absolute value calculation of the difference between the continuously evolved state vector of the current time step and the continuously evolved state vector of the previous time step; Threshold comparison zone: The absolute value of the difference is compared with a transition threshold of 0.25 to filter out abnormal state components; Amplitude Mapping Area: The transition amplitude is set according to the corresponding level of the link status assessment result. When the link level is 1, the transition amplitude is set to 0.05; when the link level is 2, the transition amplitude is set to 0.1; when the link level is 3, the transition amplitude is set to 0.2; when the link level is 4, the transition amplitude is set to 0.3; and when the link level is 5, the transition amplitude is set to 0.4. Transition Update Region: The selected abnormal state components are superimposed with transition amplitudes to form a transition state vector; The transition triggering unit is set in series after the nonlinear mapping processing path. The difference detection area first forms the state change amplitude information, the threshold comparison area determines the abnormal state component, the amplitude mapping area determines the transition amplitude under the corresponding link level, and the transition update area outputs the transition state vector, thus transforming the traditional liquid time constant network continuous smooth state evolution structure into a state update structure that combines continuous evolution and abnormal transition. The state inertia processing path is as follows: The state inertial processing path is set between the multi-scale liquid memory unit and the transition trigger unit. The starting point is the fusion feature sequence output by the multi-scale liquid memory unit, and the ending point is the input end of the nonlinear mapping processing path. The state inertial processing path consists of an inertial input area, a historical state buffer area, and a weighted accumulation area. The inertial input area receives the fusion feature vector of the current time step, the historical state buffer area stores the safe state vector of the previous time step, and the weighted accumulation area assigns an inertial weight of 0.6 to the safe state of the previous time step, assigns an input weight of 0.4 to the fusion feature of the current time step, and performs element-wise weighted summation on the two types of vectors to form an inertial state vector. In this way, the state inertial processing path links the short-term fluctuation information and long-term trend information output by the multi-scale liquid memory unit with the historical safe state, so that the state update process retains both the state continuation feature and the current input feature, forming the basis of the inertial state in continuous time recursion. The nonlinear mapping processing path is as follows: The nonlinear mapping processing path is set after the state inertia processing path. The starting point is the inertial state vector output by the state inertia processing path, and the ending point is the input of the transition triggering unit. The nonlinear mapping processing path consists of a mapping input area, a slope adjustment area, a bias superposition area, and a nonlinear mapping area. The mapping input area receives the inertial state vector. The slope adjustment area multiplies each state component by a slope coefficient of 1.20. The bias superposition area adds a bias value of 0.1 to each state component. The nonlinear mapping area inputs a hyperbolic tangent function element by element to the adjusted state components and outputs a continuously evolving state vector. Through this setting, the nonlinear mapping processing path converts the inertial state obtained by linear accumulation into a continuous state representation with nonlinear evolution capability, so that different risk levels, different link states, and different communication behavior characteristics form separable evolution trajectories in the state space, and provides an input basis for the transition triggering unit to perform anomaly judgment and state transition update. In the time constant modulation unit, the link state parameter set is mapped to time constant modulation coefficients. Each time constant modulation coefficient is the sum of the products of each parameter value in the link state parameter set and its corresponding modulation weight. An update process is performed on the time constant of each neuron in the liquid time constant network. The original time constant and the time constant modulation coefficients are weighted and summed to obtain the modulated time constant. Specifically, mapping the link state parameter set to time constant modulation coefficients involves: The channel quality, packet loss rate, retransmission rate, and average delay in the link state parameter set are mapped to time constant modulation coefficients. The channel quality is multiplied by 0.4, the packet loss rate by 0.2, the retransmission rate by 0.2, and the average delay by 0.2. The summation is then used to generate the modulation coefficients for each neuron. The time constants of each neuron in the liquid time constant network are updated, specifically as follows: The original time constant and the modulation coefficient are weighted and summed in proportions of 0.6 and 0.4 to generate the modulated time constant. The time constant controls the neuron's response speed to the input vector and the memory retention time. In the multi-scale liquid memory unit, the time-step input vector sequence is input into the liquid recursive structure. State update processing is performed on each time step according to the recursive update rule. The current time-step state is calculated from the previous time-step state and the current time-step input vector through a nonlinear mapping function. State propagation is performed in the short-time memory channel and the long-time memory channel respectively, generating short-time feature sequences and long-time feature sequences. These are then weighted and summed according to preset fusion weights to generate a fused feature sequence. Specifically, the state update processing for each time step according to the recursive update rule is as follows: For each time step t in the input vector sequence, the current time step vector is combined with the state vector of the previous time step t-1. The combination method is to multiply the previous time step state vector by the inertia coefficient 0.6 and the current time step input vector by the input weight 0.4, and then sum them to generate an initial state vector. A nonlinear mapping function is applied to the initial state vector to map each element to the interval -1 to 1. The mapping function is the hyperbolic tangent function. The mapping result is recorded for the nonlinear state vector of each time step. The short-term memory channel uses the nonlinear state vectors of the most recent 8 time steps. The vector elements are weighted and accumulated element by element according to the exponential decay weight from 1 to 0.3 to generate a short-term feature vector. Each short-term feature vector has a dimension of 16. The long-term memory channel uses the nonlinear state vectors of the most recent 64 time steps. The vector elements are weighted and accumulated element by element according to the linearly decreasing weight from 1 to 0.1 to generate a long-term feature vector. Each long-term feature vector has a dimension of 16. The short-term feature sequence and the long-term feature sequence are output separately. Short-term memory channel state propagation, specifically: The time step input vector sequence is input into the short-time liquid memory channel. The short-time window length is 8 time steps. The current time step state is calculated by mapping the previous time step state and the current input vector through the hyperbolic tangent function. The short-time feature sequence is recorded as a 16-dimensional vector, retaining the change information of the last 8 time steps. Long-term memory channel state propagation, specifically: The time step input vector sequence is input into the long-term liquid memory channel. The long-term window length is 64 time steps. The current time step state is calculated by mapping the previous time step state and the current input vector through the hyperbolic tangent function, generating a 16-dimensional vector of long-term feature sequence to retain long-term trend information. The weighted summation is performed according to the preset fusion weights, specifically as follows: The short-term feature sequence and the long-term feature sequence are weighted and summed with weights of 0.6 and 0.4 respectively. The short-term feature vector is multiplied by 0.6 and the long-term feature vector is multiplied by 0.4. The elements are added together to generate a fused feature vector. The fusion result retains the sensitivity to short-term changes and the stability of long-term trends. The fused feature sequence is continuously evolved through a state inertia processing path and a nonlinear mapping processing path to obtain a continuously evolving state vector. In the transition triggering unit, anomaly detection and state transition update processing are performed on the continuously evolving state vector. The corresponding eigenvalues ​​are adjusted to the sum of the preset transition amplitude value and the original eigenvalues ​​to generate the transition state vector. Specifically, the anomaly detection processing on the continuously evolving state vector is as follows: The current time step 48-dimensional continuously evolving state vector is compared element-by-element with the previous time step 48-dimensional continuously evolving state vector. The absolute value of the difference of the same dimension component is calculated. The 48 absolute values ​​of the difference are compared with the anomaly threshold of 0.25. When the absolute value of the difference of any dimension is greater than 0.25, the corresponding dimension is marked as an anomaly dimension. When the link level component in the link state evaluation result is greater than or equal to 4 and the absolute value of the difference is greater than 0.2, the corresponding dimension is also marked as an anomaly dimension. All anomaly dimensions are combined into an anomaly trigger set for performing state transition updates. The state transition update process is as follows: The transition amplitude value is set according to the link level in the link state assessment result. When the link level is 1, the transition amplitude value is set to 0.05; when the link level is 2, the transition amplitude value is set to 0.1; when the link level is 3, the transition amplitude value is set to 0.2; when the link level is 4, the transition amplitude value is set to 0.3; and when the link level is 5, the transition amplitude value is set to 0.4. For each abnormal dimension in the abnormal trigger set, the original continuous evolution state component is added to the corresponding transition amplitude value to obtain the component value after the transition. For non-abnormal dimensions, the original continuous evolution state components remain unchanged. After the 48 components are updated, a 48-dimensional transition state vector is formed. The abnormal dimension in the transition state vector records the component value before the transition, the transition amplitude value, and the component value after the transition. Normalize the transition state vector to generate a dynamic safe state.

[0026] In this embodiment, generating a dynamic security state includes: In this embodiment, generating optimized model parameters and updating the liquid time constant network includes: The parameters of the liquid time constant network are vectorized to generate a parameter distribution vector. Specifically, the vectorization of the parameters of the liquid time constant network is as follows: The parameters of the time constant modulation unit, multi-scale liquid memory unit, perturbation suppression unit, state inertia unit, nonlinear mapping unit, and transition triggering unit in the liquid time constant network are expanded into a one-dimensional parameter sequence in a fixed order. The time constant modulation unit extracts 16 neuron time constant parameters and 4 modulation weight parameters. The multi-scale liquid memory unit extracts 64 short-term channel recursive weight parameters, 64 long-term channel recursive weight parameters, and 32 input mapping parameters. The perturbation suppression unit extracts 16 threshold parameters and 16 amplitude reduction parameters. The state inertia unit extracts 16 inertia weight parameters and 16 input weight parameters. The nonlinear mapping unit extracts 16 slope parameters and 16 bias parameters. The transition triggering unit extracts 16 transition threshold parameters and 16 transition amplitude parameters. All parameters are concatenated to form a 292-dimensional parameter distribution vector. The parameter value of each dimension is mapped to the interval from -1 to 1. In the vector, the first to 20 dimensions represent the time constant modulation unit parameters, the 21st to 180th dimensions represent the multi-scale liquid memory unit parameters, the 181st to 212th dimensions represent the disturbance suppression unit parameters, the 213th to 244th dimensions represent the state inertia unit parameters, the 245th to 276th dimensions represent the nonlinear mapping unit parameters, and the 277th to 292nd dimensions represent the transition triggering unit parameters. The covariance matrix is ​​constructed based on the parameter distribution vector. Initialization is performed on the covariance matrix by calculating the variance of each parameter dimension in the parameter distribution vector and filling it with diagonal elements. Specifically, the construction of the covariance matrix based on the parameter distribution vector is as follows: A 292-row, 292-column covariance matrix is ​​constructed based on a 292-dimensional parameter distribution vector. The diagonal elements in the covariance matrix represent the variance values ​​of the corresponding parameter dimensions, while the off-diagonal elements represent the degree of co-variance between any two parameter dimensions. The variance values ​​are obtained by reading the average values ​​of the same dimension parameters during the five historical optimization rounds, calculating the sum of squares of the differences between the parameter values ​​in each round and the average value, and then dividing by 5. Initially, all off-diagonal elements of the covariance matrix are set to 0, indicating that the parameter dimensions are independent of each other in the initial stage of optimization. The covariance matrix is ​​initialized as follows: The variance values ​​were filled one by one for each of the 292 main diagonal elements. The initial variance of the time constant modulation unit parameter dimension was set to 0.12, the initial variance of the multi-scale liquid memory unit parameter dimension was set to 0.1, the initial variance of the disturbance suppression unit parameter dimension was set to 0.08, the initial variance of the state inertia unit parameter dimension was set to 0.09, the initial variance of the nonlinear mapping unit parameter dimension was set to 0.07, and the initial variance of the transition triggering unit parameter dimension was set to 0.11. Minimum value protection was performed on all main diagonal elements. When any variance value was less than 0.01, the variance value was fixed to 0.01 to form the initial covariance matrix. The initial covariance matrix is ​​scaled based on the link state parameter set. This involves multiplying each parameter value in the link state parameter set with a preset adjustment coefficient, and then scaling the corresponding dimensions of the covariance matrix to generate a time-varying covariance matrix. Specifically, the scaling process based on the link state parameter set is as follows: Read four parameters from the link state parameter set: channel quality, packet loss rate, retransmission rate, and average delay. Multiply each of these four parameters by an adjustment coefficient: channel quality adjustment coefficient is set to 0.35, packet loss rate adjustment coefficient to 0.25, retransmission rate adjustment coefficient to 0.2, and average delay adjustment coefficient to 0.2. Sum the four products to obtain the scale adjustment value. If the scale adjustment value is less than 0.8, fix it at 0.8; if the scale adjustment value is greater than 1.50, fix it at 1.50. Apply the scale adjustment value to the corresponding parameters of the 292-dimensional parameter distribution vector. Six sets of parameter intervals are used. The parameter intervals of the time constant modulation unit are multiplied by a scale adjustment value of 1.20, the parameter intervals of the multi-scale liquid memory unit are multiplied by a scale adjustment value of 1.10, the parameter intervals of the disturbance suppression unit are multiplied by a scale adjustment value of 0.9, the parameter intervals of the state inertia unit are multiplied by a scale adjustment value of 1, the parameter intervals of the nonlinear mapping unit are multiplied by a scale adjustment value of 0.85, and the parameter intervals of the transition triggering unit are multiplied by a scale adjustment value of 1.25. The main diagonal elements of the corresponding intervals of the covariance matrix are scaled by the same factor to generate a time-varying covariance matrix. Sampling processing is performed on the parameter distribution vector based on the time-varying covariance matrix to generate a candidate parameter set according to a multidimensional normal distribution. Specifically, the sampling processing on the parameter distribution vector based on the time-varying covariance matrix involves: The 292-dimensional parameter distribution vector is used as the mean vector of the multidimensional normal distribution, and the time-varying covariance matrix is ​​used as the covariance input of the multidimensional normal distribution. The number of samples per round is set to 64 groups. 64 292-dimensional candidate parameter vectors are randomly selected from the multidimensional normal distribution. The mean offset information and sampling offset information of each candidate parameter vector are retained. Boundary clipping is performed on the candidate parameter vectors. When the parameter value of any dimension is greater than 1, it is fixed to 1. When the parameter value of any dimension is less than -1, it is fixed to -1. The 64 groups of candidate parameter vectors are combined into a candidate parameter set with 64 rows and 292 columns. A fitness evaluation process is performed on the candidate parameter set. The candidate parameter sets are input into the liquid time constant network, and the fitness value is calculated based on the sum of the absolute values ​​of the differences between the dynamic security state and link state evaluation results, plus the reciprocal of a preset constant. This generates a candidate parameter fitness sequence. Specifically, the fitness evaluation process for the candidate parameter set involves: The 64 candidate parameter vectors are mapped back to the 6 units of the liquid time constant network one by one, and a single round of network operation is completed to obtain 64 candidate dynamic security state outputs. The difference between each candidate dynamic security state output and the link state evaluation result at the same time step is calculated. The difference is calculated by subtracting the 48 components of the link state evaluation result from each of the 48 components in the dynamic security state vector, and then taking the absolute value of the difference of each component. The sum of the 48 absolute values ​​is obtained to get the total difference. The total difference is added to the preset constant 1, and the reciprocal is taken to obtain the fitness value of a single candidate parameter. When the total difference is small, it indicates that the dynamic security state and the link state are highly matched, and the corresponding fitness value is large. After all 64 candidate parameters are evaluated, 64 fitness values ​​are obtained. The 64 fitness values ​​are arranged in descending order to form the candidate parameter fitness sequence. The parameter distribution vector and covariance matrix are updated based on the candidate parameter fitness sequence to generate updated parameter distribution vectors and updated covariance matrices. Specifically, the update process based on the candidate parameter fitness sequence is as follows: The top 16 candidate parameter vectors with the highest fitness values ​​from the candidate parameter fitness sequence are selected as the preferred parameter groups. The 16 fitness values ​​are divided by the sum of the 16 fitness values ​​to obtain normalized weights. The sum of the 16 normalized weights is 1. The 16 preferred parameter vectors are multiplied by the corresponding normalized weights in each dimension. The 16 products in the same dimension are then summed to obtain 292 dimension values ​​of the updated parameter distribution vector. The updated parameter distribution vector is then smoothed and corrected. The smoothing correction ratio is set to 0.3 for the old mean and 0.7 for the new mean. The original parameter distribution vector is multiplied by 0.3, the new sum is multiplied by 0.7, and then the results are added in each dimension to form the final updated parameter distribution vector. Read the difference between the 16 sets of preferred parameter vectors and the updated parameter distribution vector. Calculate a 292-dimensional deviation vector for each set of preferred parameter vectors. Perform an outer product operation between the deviation vector and its transpose to obtain a 292-row, 292-column deviation matrix. Multiply each of the 16 deviation matrices by its corresponding normalized weight, and then sum them element-wise to obtain a new covariance estimation matrix. Multiply the old covariance matrix by a retention factor of 0.4, multiply the new covariance estimation matrix by an update factor of 0.6, and then sum them element-wise to form the updated covariance matrix. Perform a lower bound constraint on all diagonal elements of the updated covariance matrix, fixing any diagonal element to 05 when it is less than 05. The parameters of the liquid time constant network are updated in groups based on the updated parameter distribution vector. The parameter distribution vector is segmented according to parameter type, and the parameters of each segment are assigned to the respective units of the model, generating optimized model parameters and completing the update of the liquid time constant network. Specifically, the group update process based on the updated parameter distribution vector is as follows: The updated parameter distribution vector is segmented into fixed-dimensional intervals and mapped back to the six units of the liquid time constant network. Dimensions 1 to 20 are mapped to the time constant modulation unit, dimensions 21 to 180 are mapped to the multi-scale liquid memory unit, dimensions 181 to 212 are mapped to the perturbation suppression unit, dimensions 213 to 244 are mapped to the state inertia unit, dimensions 245 to 276 are mapped to the nonlinear mapping unit, and dimensions 277 to 292 are mapped to the transition triggering unit. Within each unit, the parameters are assigned one by one in order. The time constant parameter is mapped to the corresponding neuron, the recursive weight parameter is mapped to the short-time channel and the long-time channel, the threshold parameter is mapped to the anomaly detection position, and the amplitude parameter is mapped to the transition update position, thus completing the full assignment of 292 parameters. The liquid time constant network update was completed, specifically as follows: The six units that have completed grouping and assignment are reassembled into a liquid time constant network. The original connection order between the time constant modulation unit, multi-scale liquid memory unit, disturbance suppression unit, state inertia unit, nonlinear mapping unit, and transition triggering unit remains unchanged. One round of verification is performed on the updated liquid time constant network. The verification input uses the safe state input vector of the most recent 32 time steps, and the verification output uses the dynamic safe state vector and the state representation vector. Boundary pruning is performed again for parameter out-of-bounds values ​​that occur during the verification process. The optimized model parameters are written and the liquid time constant network is updated.

[0027] In this embodiment, the generation of the secure transmission strategy parameter set includes: The dynamic security state is subjected to policy mapping, and the mean, maximum, minimum and high-risk state component values ​​are calculated. Specifically, the policy mapping process for the dynamic security state is as follows: Interval statistical processing is performed on the 48 components in the dynamic security state vector to calculate the mean, maximum, minimum, and number of high-risk components. High-risk components are defined as those with a value greater than 0.7, medium-risk components are defined as those with a value greater than 0.4 and not greater than 0.7, and low-risk components are defined as those with a value not greater than 0.4. The mean is used as the main mapping indicator, and the number of high-risk components is used as the correction indicator. When the mean is less than 0.3, it is mapped to a low security level; when the mean is greater than or equal to 0.3 and less than 0.55, it is mapped to a medium security level; when the mean is greater than or equal to 0.55 and less than 0.75, it is mapped to a high security level; when the mean is greater than or equal to 0.75, it is mapped to an enhanced security level. When the number of high-risk components is greater than or equal to 12, the level is increased by 1 level from the original level, with a maximum not exceeding the enhanced security level. Security level mapping is performed based on the mean of state components and the number of high-risk state components to generate security level results; Based on the security level results, parameter mapping is performed to generate a set of secure transmission policy parameters, including encryption strength parameters, key update cycle parameters, and retransmission control parameters. Specifically, the generated set of secure transmission policy parameters is as follows: When the security level is low, the encryption strength parameter is set to 2 rounds, the key update cycle parameter is set to 12 seconds, and the retransmission control parameter is set to 1 time. When the security level is medium, the encryption strength parameter is set to 3 rounds, the key update cycle parameter is set to 8 seconds, and the retransmission control parameter is set to 2 times. When the security level is high, the encryption strength parameter is set to 4 rounds, the key update cycle parameter is set to 5 seconds, and the retransmission control parameter is set to 3 times. When the security level is enhanced, the encryption strength parameter is set to 5 rounds, the key update cycle parameter is set to 3 seconds, and the retransmission control parameter is set to 4 times. These three parameters are written into the secure transmission policy parameter set, which is stored in the order of encryption strength parameter, key update cycle parameter, and retransmission control parameter.

[0028] In this embodiment, the secure transmission of standardized equipment data includes: Perform block processing on standardized equipment data to generate a sequence of blocks to be encrypted; Multiple rounds of encryption operations are performed on the sequence of blocks to be encrypted based on the encryption strength parameter, and the number of encryption rounds is adjusted according to the link state evaluation results to generate an encrypted data sequence. Specifically, the adjustment of the number of encryption rounds according to the link state evaluation results is as follows: Read the link level from the link status assessment result. When the link level is 1, subtract 1 round from the encryption strength parameter given by the security transmission strategy parameter set, with a minimum of 2 rounds. When the link level is 2 or 3, keep the original number of encryption rounds unchanged. When the link level is 4, add 1 round to the original number of encryption rounds. When the link level is 5, add 2 rounds to the original number of encryption rounds, with a maximum of 6 rounds. Use the adjusted number of rounds as the final number of encryption rounds for the current time window. Perform the same number of rounds of encryption operation on all data blocks within the current time window to form an encrypted data sequence. A time-series key set is generated based on the key update cycle parameter, and the time-series key set is matched to the corresponding data block according to the transmission time information of each data block in the encrypted data sequence. Specifically, the generation of the time-series key set based on the key update cycle parameter is as follows: A time-series key set is generated based on the key update cycle parameter. The initial key length is set to 16 bytes, and the initial key byte values ​​are set sequentially as 17, 29, 43, 58, 71, 86, 94, 103, 117, 126, 139, 148, 157, 169, 181, and 193. The current time step index, source device identifier hash value, and link level value are concatenated, and a 16-byte update vector is generated by summing the bytes and taking the remainder after dividing by 256. The initial key and the update vector are XORed byte by byte to obtain the current cycle key. When the key update cycle parameter is 12 seconds, a key set is generated every 12 seconds; when the key update cycle parameter is 8 seconds, a key set is generated every 8 seconds; when the key update cycle parameter is 5 seconds, a key set is generated every 5 seconds; and when the key update cycle parameter is 3 seconds, a key set is generated every 3 seconds. All cycle keys are written into the time-series key set in chronological order. The time-series key set is matched to the corresponding data block based on the transmission time information of each data block in the encrypted data sequence, specifically as follows: Read the sending timestamp corresponding to each data block in the encrypted data sequence, map the sending timestamp to the corresponding key period interval, bind the 16-byte period key corresponding to the key period interval to the data block identifier, use the same set of period keys for all data blocks generated in the same period, switch the period key corresponding to the new interval for data blocks sent across periods, form a matching table that corresponds one-to-one between data blocks and keys, and write the matching table into the sending control buffer. Based on retransmission control parameters and link state assessment results, transmission scheduling processing is performed on the encrypted data sequence to generate a transmission queue. Retransmission control processing is then performed on data blocks that fail to be transmitted. Specifically, the transmission scheduling processing based on retransmission control parameters and link state assessment results for the encrypted data sequence includes: Based on the retransmission control parameters and link status assessment results, the encrypted data sequence is prioritized for transmission. The basic scheduling weight is set to 0.4 for data blocks with link level 1 and 2, 0.7 for data blocks with link level 3, and 1.0 for data blocks with link level 4 and 5. Additional scheduling corrections are applied to the data block service types: the correction value is set to 0.5 for control services, 0.3 for monitoring services, and 0.1 for ordinary recording services. The basic scheduling weight and the service correction value are added together to obtain the scheduling priority value. The transmission order is formed by sorting the data blocks from high to low according to the scheduling priority value. For data blocks that fail to be sent, retransmission count control is performed, specifically as follows: For data blocks that fail to be transmitted, read the retransmission control parameters. When the retransmission control parameter is 1, the maximum number of retransmissions is 1. When the retransmission control parameter is 2, the maximum number of retransmissions is 2. When the retransmission control parameter is 3, the maximum number of retransmissions is 3. When the retransmission control parameter is 4, the maximum number of retransmissions is 4. After each transmission failure, the data block is reinserted into the head of its original transmission queue. When the number of transmission failures reaches the maximum number of retransmissions, the data block is transferred to the failure record area, and the data block identifier, failure count, link level, and transmission time are recorded. Perform wireless transmission processing on the data blocks in the transmission queue to complete the secure transmission of standardized device data.

[0029] Example 1: In a continuous IoT communication cycle, the gateway receives raw data streams from 128 terminal devices, totaling 3,200,000 records. Data packet lengths range from 256 to 1024 bytes, and the data transmission frequency is 5 to 12 times per second. The raw data includes business data content, data generation time, data transmission frequency, and the business type to which the data belongs. Device communication identification data includes a unique device identifier, communication protocol type, channel access identifier, and communication port identifier. In the initial acquisition phase, there is a significant imbalance in communication data, with high-frequency communication pairs accounting for 19% and low-frequency communication pairs accounting for 55%, with an average packet loss rate of 4.6%.

[0030] After the data enters the processing flow, time alignment processing is performed on the raw data and device communication identification data to control the time deviation within 1 millisecond. Through protocol identification and parsing, four types of communication protocols are identified, with a parsing success rate of 99.2%. During field extraction, a structured data sequence containing data content fields, timestamp fields, device identification fields, and communication parameter fields is generated. After normalization processing, the data values ​​are mapped to the 0–1 interval, and the standard deviation of the data distribution is reduced from the original 0.38 to 0.21.

[0031] When constructing device connection relationship data, the order of communication was statistically analyzed. Within one period, 2048 connection pairs were generated, with 16% of these pairs communicating more than 1000 times. During the data cleaning phase, outliers accounted for 11.3%. After outlier removal and replacement, the data effectiveness rate increased to 97.8%. Missing value imputation used a mean imputation method with a time window of 10 time steps, improving data continuity by approximately 18% after imputation.

[0032] In the feature extraction stage, service type features are extracted from the data content field, time interval features are extracted from the timestamp field, and channel occupancy features are extracted from the communication parameter field, forming an initial feature set. The feature dimension is 32-dimensional, which is expanded to 64-dimensional after encoding. In the feature weighting process, the weight coefficients are set to 0.2, 0.3, and 0.5 to perform weighted fusion of different features, forming a data feature set.

[0033] In the process of constructing the historical sequence of communication behavior, the device connection pairs are arranged according to the time sequence of communication occurrence, and the data feature set is spliced ​​into a time series structure. The length of a single communication behavior sequence ranges from 480 to 960 time steps, with a sequence feature mean of 0.47 and a variance of 0.16.

[0034] During link status acquisition, the received signal strength ranged from -75 to -53 dB / mW, the signal-to-noise ratio ranged from 8 to 26 dB, and the packet loss rate ranged from 0.8% to 6.4%. The link status parameter set was calculated, and its normalized mean was 0.52. The link status assessment value was calculated using weighted fusion, with an average value of 0.61, indicating that 12.5% ​​of the links were high-risk.

[0035] The historical sequence of communication behavior is input into a liquid time constant network. In the time constant modulation unit, the set of link state parameters participates in modulation, and the time constant is adjusted from the original value of 0.8 to the range of 0.5–1.6. In the multi-scale liquid memory unit, hierarchical recursive processing is performed on the sequence, with a short-time window of 8 time steps and a long-time window of 64 time steps. The mean of short-time features is 0.53, and the mean of long-time features is 0.49. In the disturbance suppression unit, abnormal fluctuations are compressed, reducing the proportion of abnormal features from 8.7% to 2.9%. Finally, a communication behavior memory vector is generated with a dimension of 128 and a mean of 0.51.

[0036] During the security state evolution phase, the communication behavior memory vector, data feature set, and link state evaluation results are concatenated to form a 192-dimensional input vector. In the state inertial unit, the state weight of the previous time step is set to 0.6, and the current input weight is 0.4, generating an inertial state vector. In the nonlinear mapping processing, the mean characteristic response increases from 0.51 to 0.63. During the transition triggering process, a total of 42 abnormal events are detected. After the transition, the peak characteristic value increases to 0.87, and the mean dynamic security state value is 0.66.

[0037] During the parameter optimization phase, distributional modeling was performed on the liquid time constant network parameters, with a parameter dimension of 256. The initial diagonal value of the covariance matrix was 0.12, which expanded to 0.08–0.23 after link state adjustment. Each round of sampling generated 64 candidate parameter sets, and the mean value of the optimal parameters after fitness calculation was 0.78. After 26 iterations, the error decreased from 0.41 to 0.07, and the model converged and stabilized.

[0038] During the secure transmission phase, a secure transmission strategy is generated based on the dynamic security status. The encryption strength is set to 3–5 rounds, and the key update cycle is 5–10 seconds. The data block size is 512 bytes, and the length of the encrypted data sequence increases by approximately 6%. During transmission scheduling, the data is divided into three levels of queues based on the link status assessment results, with the high-priority queue accounting for 21%. The average number of retransmission control operations is 28 per thousand packets.

[0039] In the comparative experiment, the same data scale and link environment were selected, and the traditional fixed strategy method was used for testing. The number of training samples was 3.2 × 10^6, and the number of test data samples was 8 × 10^5. The results are as follows: Transmission success rate: 94.3% for traditional methods, 99.1% for this invention. Average transmission latency: 61 milliseconds for traditional methods, 37 milliseconds for this invention; Packet loss rate: 4.8% for traditional methods, 1.6% for this invention; Anomaly detection accuracy: 69.5% for traditional methods, 90.8% for this invention; Security policy response time: 310 milliseconds for traditional methods, 92 milliseconds for this invention; Number of retransmissions: The traditional method requires 81 retransmissions per thousand packets, while this invention requires 28 retransmissions.

[0040] As can be seen from the embodiments, in actual operation, the present invention has shown good results in terms of communication behavior modeling accuracy, link state adaptability and security policy dynamic adjustment capability, verifying the engineering feasibility of the method in complex wireless environments.

[0041] Overall, this invention achieves stable improvements in continuous modeling of communication behavior, link state awareness fusion, dynamic evolution of security state, adaptive optimization of model parameter covariance, and collaborative generation of secure transmission strategies. The accuracy of communication state representation, security protection capability, transmission success rate, system stability, and overall optimization efficiency are continuously improved.

[0042] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A secure wireless communication transmission method based on an Internet of Things (IoT) gateway, characterized in that it includes: Collect raw data and device communication identification data from multi-source heterogeneous terminal devices, perform protocol parsing and format unification processing on the raw data, and generate standardized device data and device connection relationship data; Perform data cleaning and feature extraction on standardized equipment data, construct a data feature set, and build a historical sequence of communication behavior based on equipment connection relationship data; Collect wireless link operation data and perform state awareness and assessment processing to generate a set of link state parameters and link state assessment results; The historical sequence of communication behavior, data feature set and link state evaluation results are input into the liquid time constant network of time constant modulation unit, multi-scale liquid memory unit and transition triggering unit. Time constant dynamic modulation is performed in time constant modulation unit, hierarchical liquid recursion is performed in multi-scale liquid memory unit, and anomaly judgment and state transition update are performed in transition triggering unit to generate dynamic security state. Based on the dynamic security state, link state parameter set and data feature set, a time-varying covariance evolution strategy is constructed. The parameters of the liquid time constant network are modeled for parameter distribution, adaptively updated for covariance and optimized for distribution sampling. During the iterative update of parameter distribution, candidate parameter evaluation and group update are completed, and the optimized model parameters are generated and the liquid time constant network is updated. Based on the dynamic security state output by the updated liquid time constant network, a set of secure transmission strategy parameters is generated. Based on the set of secure transmission strategy parameters, data encryption and wireless transmission control are performed on standardized equipment data to achieve secure transmission of standardized equipment data.

2. The secure wireless communication transmission method based on an IoT gateway according to claim 1, characterized in that, The original data includes business data content, data packet length, data generation time, data transmission frequency, and the business type to which the data belongs. The device communication identification data includes the device unique identifier, device access type, communication protocol type, device address information, channel access identifier, and communication port identifier.

3. The secure wireless communication transmission method based on an IoT gateway according to claim 1, characterized in that, The generation of standardized device data and device connection relationship data includes: The raw data and device communication identification data are time-aligned to generate the raw data sequence. The original data sequence is processed by protocol identification. The original data sequence is matched based on a preset protocol feature set to obtain the corresponding protocol type identifier. The original data sequence is then processed by protocol parsing to generate a parsed data sequence. The parsed data sequence is processed by field extraction. According to the unified data structure definition, the data content field, timestamp field, device identifier field and communication parameter field are extracted and encoded and mapped to generate a structured data sequence. Perform data normalization processing on structured data sequences to generate standardized equipment data; Based on the device communication identification data, correlation analysis is performed to construct device connection pairs according to the order of communication occurrence, and weights are assigned to the connection pairs according to the communication frequency to generate device connection relationship data.

4. The secure wireless communication transmission method based on an IoT gateway according to claim 1, characterized in that, The construction of the data feature set and the construction of a communication behavior history sequence based on device connection relationship data include: Perform outlier detection and processing on standardized equipment data to generate a cleaned data sequence; Missing value processing is performed on the cleaned data sequence, and the missing field data is filled in according to the mean of the time neighborhood to generate a complete data sequence; Feature extraction processing is performed on the completed data sequence. Service type features and data length features are extracted from the data content field, time interval features and transmission frequency features are extracted from the timestamp field, and channel occupancy features and transmission strength features are extracted from the communication parameter field to generate an initial feature set. The initial feature set is subjected to feature encoding processing, the discrete features are subjected to number mapping processing, and the continuous features are subjected to interval partitioning processing to generate an encoded feature set. The encoded feature set is weighted according to a preset weight coefficient to generate a data feature set; Based on device connection relationship data, the device connection pairs are arranged in the order of communication occurrence time. The data feature set corresponding to each device connection pair is spliced ​​in chronological order to form a sequence structure, and the features of each time step in the sequence are weighted to generate a historical sequence of communication behavior.

5. The secure wireless communication transmission method based on an IoT gateway according to claim 1, characterized in that, The generated link state parameter set and link state evaluation results include: Collect wireless link operation data, including received signal strength, signal-to-noise ratio, channel bandwidth, interference intensity, number of lost packets, total number of transmitted data packets, number of retransmissions, and transmission delay; Perform time window partitioning on the wireless link operation data to generate a window link data sequence; Perform parameter calculations on the window link data sequence and normalize it to generate a standardized set of link state parameters; The standardized set of link state parameters is weighted and fused according to preset weight coefficients to generate a link state evaluation value. The link state assessment value is used to perform hierarchical mapping processing to generate the link state assessment result.

6. The secure wireless communication transmission method based on an IoT gateway according to claim 1, characterized in that, The generation of dynamic security state includes: The historical sequence of communication behavior is divided into multiple consecutive time step sequences according to time order, and the data feature set in the time step sequence is vectorized to generate a time step input vector sequence; A liquid time constant network structure is constructed, which includes a time constant modulation unit, a multi-scale liquid memory unit, and a transition trigger unit. Different units are connected in a directed manner according to the network connection relationship to form a liquid recursive structure. A state inertia processing path and a nonlinear mapping processing path are set in the improved liquid time constant network. In the time constant modulation unit, the set of link state parameters is mapped to time constant modulation coefficients, the time constant of each neuron in the liquid time constant network is updated, and the original time constant and the time constant modulation coefficients are weighted and summed to obtain the modulated time constant; In the multi-scale liquid memory unit, the time step input vector sequence is input into the liquid recursive structure, and the state update processing is performed on each time step according to the recursive update rule. The state propagation is performed in the short-time memory channel and the long-time memory channel respectively to generate short-time feature sequences and long-time feature sequences. The fusion feature sequence is generated by weighted summation according to the preset fusion weight. The fused feature sequence is continuously evolved through a state inertia processing path and a nonlinear mapping processing path to obtain a continuously evolving state vector. In the transition triggering unit, anomaly detection and state transition update processing are performed on the continuously evolving state vector, and the corresponding feature value is adjusted to the sum of the preset transition amplitude value and the original feature value to generate the transition state vector. Normalize the transition state vector to generate a dynamic safe state.

7. The secure wireless communication transmission method based on an IoT gateway according to claim 1, characterized in that, The process of generating optimized model parameters and updating the liquid time constant network includes: The parameters of the liquid time constant network are vectorized to generate a parameter distribution vector; A covariance matrix is ​​constructed based on the parameter distribution vector. Initialization processing is performed on the covariance matrix by calculating the variance of each parameter dimension in the parameter distribution vector and filling it with diagonal elements to obtain the initial covariance matrix. The initial covariance matrix is ​​scaled based on the link state parameter set. The parameter values ​​in the link state parameter set are multiplied by the preset adjustment coefficient, and the corresponding dimensions in the covariance matrix are scaled to generate a time-varying covariance matrix. Sampling processing is performed on the parameter distribution vector based on the time-varying covariance matrix to generate a candidate parameter set according to the multidimensional normal distribution; The fitness evaluation process is performed on the candidate parameter set. The candidate parameter set is input into the liquid time constant network respectively. The fitness value is calculated by adding the reciprocal of the sum of the absolute values ​​of the differences between the dynamic security state and the link state evaluation results to a preset constant, and a candidate parameter fitness sequence is generated. Based on the candidate parameter fitness sequence, update the parameter distribution vector and covariance matrix to generate updated parameter distribution vector and updated covariance matrix; Based on the updated parameter distribution vector, the parameters of the liquid time constant network are updated in groups. The parameter distribution vector is segmented according to the parameter type, and the parameters of each segment are assigned to each unit of the model to generate the optimized model parameters and complete the update of the liquid time constant network.

8. The secure wireless communication transmission method based on an IoT gateway according to claim 1, characterized in that, The set of parameters for generating secure transmission strategies includes: Perform policy mapping on dynamic security states, and calculate the mean, maximum, minimum and high-risk state components of the state components. Security level mapping is performed based on the mean of state components and the number of high-risk state components to generate security level results; Based on the security level results, parameter mapping is performed to generate a set of secure transmission policy parameters, including encryption strength parameters, key update cycle parameters, and retransmission control parameters.

9. A secure wireless communication transmission method based on an IoT gateway according to claim 1, characterized in that, The secure transmission of standardized equipment data includes: Perform block processing on standardized equipment data to generate a sequence of blocks to be encrypted; Multiple rounds of encryption operations are performed on the sequence of blocks to be encrypted based on the encryption strength parameter, and the number of encryption rounds is adjusted based on the link state evaluation results to generate an encrypted data sequence. A time-series key set is generated based on the key update cycle parameter, and the time-series key set is matched to the corresponding data block according to the transmission time information of each data block in the encrypted data sequence. Based on the retransmission control parameters and link status evaluation results, the encrypted data sequence is processed for transmission scheduling, a transmission queue is generated, and retransmission control is performed on the data blocks that fail to be transmitted. Perform wireless transmission processing on the data blocks in the transmission queue to complete the secure transmission of standardized device data.

Images