Edge gateway protocol data gradient collection and transmission method based on relative entropy

Abstract

A kind of edge gateway protocol data gradient acquisition and transmission method based on relative entropy belongs to the field of industrial internet of things edge computing.The method includes: S1, extracting five-dimensional protocol feature vector such as byte information entropy, message length standard deviation, realizing protocol type automatic identification;S2, by fingerprint similarity, semantic accuracy and historical frequency weighted fusion to obtain suitable configuration confidence, complete channel automatic configuration;S3, independently calculate each channel communication reliability and time deviation penalty factor, evaluate data quality and time alignment;S4, use relative entropy to measure parameter distribution deviation, combine change rate and alarm gain factor to calculate information value index, according to adaptive threshold grading cache;S5, according to Arrhenius acceleration model, convert temperature stress into equivalent failure rate, determine transmission urgency by integrating information value and state danger, dynamically allocate transmission time slot.

Description

Technical Field

[0001] This invention belongs to the field of industrial Internet of Things edge computing, specifically involving a method for data gradient acquisition and transmission based on relative entropy edge gateway protocol. Background Technology

[0002] As the core hub connecting field electromechanical equipment and the cloud platform, the industrial IoT edge gateway undertakes key functions such as protocol adaptation, data acquisition, edge computing, and uplink transmission. With the increasing diversification of industrial field equipment, the coexistence of Modbus, OPC UA, CAN bus, Profibus, and various proprietary serial port protocols in the same workshop has become the norm. Edge gateways need to have the ability to blindly identify and adaptively adapt to multiple protocols.

[0003] Regarding protocol identification, Chinese invention patent application CN115988100A, published on August 18, 2023, discloses a device intelligent sensing IoT gateway management platform based on multi-protocol adaptive design. This patent application adopts an active detection method, sending protocol adaptation test information to terminal devices and performing feature extraction and adaptation scoring based on the response information. However, this patent application has the following shortcomings: First, it relies on the active response of terminal devices, and the adaptation success rate drops significantly in scenarios where devices malfunction, communication links are unstable, or proprietary protocols cannot respond correctly. Second, its feature extraction is limited to the response content level, such as sequential features, numerical features, textual semantic features, and image features, lacking in-depth analysis of message byte-level statistical features, such as byte distribution randomness, frame length stability, and communication periodicity. It is difficult to characterize the essential fingerprint of the protocol from the physical layer and link layer, resulting in poor adaptability in proprietary protocols and non-standard communication scenarios. Third, its adaptation decision is based on only a single scoring mechanism, without integrating multi-dimensional decision-making criteria such as fingerprint similarity, semantic accuracy, and historical statistical priors of the same model of devices, leading to insufficient quantification of adaptation reliability and a high risk of misadaptation.

[0004] In terms of data acquisition and quality assessment, existing technologies typically employ a fixed-period polling method to read data from each channel's registers, fusing multi-source heterogeneous data into a unified time series before processing. This approach results in parameters with different dimensions and physical meanings being represented in a mixed manner, distorting subsequent calculations of changes and failing to accurately reflect the true state evolution of each channel. More critically, existing technologies lack independent quality assessment mechanisms for the communication reliability and time synchronization accuracy of each channel, and fail to establish a correlation between data quality indices and time series alignment, making it difficult to preserve the original data format and dimensions for comparable analysis within a unified time coordinate system.

[0005] In terms of information value quantification and transmission scheduling, existing technologies have two fundamental flaws: First, information value density assessment is usually based on the rate of change of parameter values ​​or simple threshold exceedance judgments, lacking theoretical support from information theory. There is no necessary physical connection between information value density and parameter distribution changes, and the assessment results are highly subjective. Second, transmission scheduling decisions often adopt first-in-first-out or fixed-priority queues, failing to fully consider the dangerous evolution of equipment operating status and the real-time value differences of information content. Furthermore, there is a lack of unified physical correlation between the three elements of information urgency, equipment status danger, and bandwidth fairness, resulting in insufficient basis for setting scheduling weights. For example, the CN115988100A patent application for a device intelligent sensing IoT gateway management platform based on multi-protocol adaptive technology only focuses on the protocol adaptation stage and does not extend to subsequent links such as data quality assessment, information value quantification, hierarchical caching, and dynamic bandwidth allocation. Furthermore, it does not organically unify the physical aspects of device reliability, such as the accelerated failure rate effect caused by temperature stress, with transmission scheduling decisions. It cannot make the urgency of transmission have both the physical interpretability of information content value and device state danger, and it is difficult to achieve gradient control of prioritizing the transmission of high-value urgent data and edge caching of stable data under bandwidth-limited conditions. Summary of the Invention

[0006] This invention aims to at least solve the technical problems existing in the prior art, and in particular, it innovatively proposes a method for edge gateway protocol data gradient acquisition and transmission based on relative entropy.

[0007] To achieve the above-mentioned objectives of this invention, this invention provides a method for edge gateway protocol data gradient acquisition and transmission based on relative entropy, comprising the following steps:

[0008] S1. The edge gateway extracts the original communication byte stream of electromechanical equipment through mirror port or serial listening method, and extracts the protocol feature vector from the message sequence in the sliding window to realize automatic identification of protocol type.

[0009] S2. Compare the protocol feature vector with the built-in standard protocol library of the edge gateway, calculate the adaptation confidence, and complete the automatic channel configuration based on the adaptation threshold.

[0010] S3. Establish a unified synchronization cycle on the edge side, calculate the communication reliability and time deviation penalty factor for each channel, and evaluate the data quality index according to the time sequence alignment.

[0011] S4. Calculate the information value index by using the relative entropy of parameter distribution, combined with the relative entropy change rate and alarm gain factor, and cache in a graded manner according to the caching decision threshold.

[0012] S5. Convert the equipment operating temperature stress into an equivalent failure rate, determine the transmission urgency by normalized information value, relative increment of equivalent failure rate and bandwidth margin factor, and dynamically allocate transmission time slots.

[0013] In the preferred embodiment of the above technical solution, the multidimensional protocol feature vector in S1 includes five dimensions: byte information entropy, message length standard deviation, message arrival interval autocorrelation coefficient, control character density, and timing regularity. These dimensions respectively characterize the statistical features of the protocol message from five independent dimensions: byte distribution randomness, frame length stability, communication periodicity, control character density, and timing regularity.

[0014] In the preferred embodiment of the above technical solution, the appropriate configuration confidence in S2 is obtained by weighted fusion of fingerprint similarity, semantic accuracy, and the frequency of occurrence of protocols of the same model in history. The fingerprint similarity maps the geometric distance between the protocol feature vector and the standard template to a similarity score through an exponential kernel function. The semantic accuracy is evaluated by the field out-of-bounds, checksum errors, and semantic anomalies in the parsed message of the protocol template. The frequency of occurrence of protocols of the same model in history provides an independent decision-making basis using the statistical prior of the same model of equipment.

[0015] In the preferred embodiment of the above technical solution, the communication reliability in S3 is the ratio of the number of successful response frames to the number of request frames, reflecting the communication stability of the physical layer and the link layer; the time deviation penalty factor is calculated based on the deviation between the data timestamp and the unified synchronization time, according to the exponential decay law, to determine the timeliness score; the data quality is a comprehensive evaluation of communication reliability and time synchronization accuracy, and the data quality is assessed by a weighted combination of the time deviation penalty factor and communication reliability, with the time deviation penalty factor directly reflecting the impact of time synchronization accuracy on data quality.

[0016] In the preferred embodiment of the above technical solution, the relative entropy in S4 is the deviation of the sampling frequency distribution of key monitoring parameters within the current window from the prior probability distribution under steady-state operating conditions, reflecting the degree to which the equipment operating state deviates from the steady state; the relative entropy change rate is the rate of change of relative entropy over time, reflecting the severity of the equipment state evolution; the alarm gain factor maps discrete alarm levels to continuous value gain, realizing the exponential amplification relationship between alarm levels and value gain.

[0017] In the preferred embodiment of the above technical solution, the information value index in S4 integrates four dimensions: relative entropy information content, state evolution rate, alarm level, and service level. The larger the value, the richer the information content of the current data of the device, the more drastic the state change, the higher the alarm level, or the greater the commercial value. The cache decision threshold is an adaptive threshold based on the average information value of all current devices, which is dynamically adjusted with the change of device state. Data exceeding the threshold is marked as real-time upload, and data below the threshold enters the edge cache queue.

[0018] In the preferred embodiment of the above technical solution, the equivalent failure rate in S5 is based on the Arrhenius acceleration model, which transforms the difference between the equipment operating temperature and the reference temperature into an exponential growth effect of the failure rate, thereby activating energy values ​​that cover typical failure modes of industrial electronic equipment.

[0019] In the preferred embodiment of the above technical solution, the transmission urgency in S5 is obtained by a weighted combination of three dimensions: information urgency, state hazard, and bandwidth fairness. The information urgency is measured by normalized information value, the state hazard is measured by the relative increment of the equivalent failure rate, and the relative increment of the equivalent failure rate is the Arrhenius acceleration factor minus one, which is zero when the equipment has no temperature anomaly. The bandwidth fairness is achieved by using a bandwidth margin factor to prevent a few high-priority devices from monopolizing bandwidth resources for a long time. The normalized weighting of the three factors gives the transmission scheduling decision a unified physical meaning and engineering interpretability.

[0020] In the preferred embodiment of the above technical solution, the method repeats S3 to S5 at the beginning of each synchronization cycle, and executes S1 and S2 once every fixed number of message cycles to update the protocol adaptation status, forming a closed-loop gradient acquisition and transmission control. The data quality index of each channel output by S3 is used for bandwidth allocation weight correction in S5, and the priority of channels with lower quality is appropriately reduced in bandwidth contention.

[0021] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0022] This invention achieves automatic blind identification and adaptive adaptation of multiple protocols through five-dimensional protocol feature vectors, solving the problems of low efficiency and poor adaptability of manual configuration in industrial scenarios with multiple protocols coexisting. It avoids information distortion caused by multi-source data fusion through independent channel quality assessment and time sequence alignment. It establishes a strict physical connection between information value and parameter distribution changes through relative entropy, overcoming the shortcomings of traditional threshold judgment with strong subjectivity. It unifies equipment reliability physics and transmission scheduling under the same framework through the Arrhenius failure rate acceleration factor, making the transmission urgency combine information urgency, state danger, and bandwidth fairness. It realizes gradient control of priority transmission of high-value urgent data and edge buffering of stable data, which significantly improves the real-time performance of key information and the engineering interpretability of transmission decisions under bandwidth-constrained conditions.

[0023] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0024] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0025] Figure 1This is a flowchart of S1 of the present invention;

[0026] Figure 2 This is a flowchart of S2 of the present invention;

[0027] Figure 3 This is a flowchart of S5 of the present invention. Detailed Implementation

[0028] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0029] like Figures 1 to 3 As shown, this invention discloses a method for edge gateway protocol data gradient acquisition and transmission based on relative entropy.

[0030] To address the aforementioned issues, this invention achieves blind identification of protocol types through five-dimensional protocol feature vectors, establishes a physical connection between information value and parameter distribution changes through relative entropy, and unifies the physical meaning of transmission urgency through the Arrhenius failure rate acceleration factor, thereby realizing end-to-end gradient optimization from protocol adaptation, quality assessment, value quantification to bandwidth allocation.

[0031] To achieve the above objectives, this invention provides a method for edge gateway protocol data gradient acquisition and transmission based on relative entropy, comprising the following steps performed sequentially:

[0032] S1, Raw communication byte stream feature extraction and automatic protocol type identification. The edge gateway collects the raw communication byte stream of electromechanical equipment through mirrored ports or serial monitoring. A sliding window step size is used. Extracting a sequence of messages from N consecutive messages, with the total number of messages N within the sliding window being 100, corresponds to approximately 100 synchronization cycles. The communication data is selected based on the statistical stability requirements of industrial communication to ensure sample moment convergence. In industrial settings, 100 messages are sufficient to bring the sample mean and variance to stable values, satisfying the application conditions of the central limit theorem and ensuring the reliability of subsequent statistical feature calculations. The sliding window step size is 50 messages, with a 50% window overlap to balance identification real-time performance and computational overhead. This design ensures that the protocol identification result is updated every 50 new messages, while retaining 50% of the historical data, so that the detection delay for protocol type switching does not exceed 50 message cycles. This value is selected based on the statistical stability requirements of industrial communication to ensure sample moment convergence. Specifically, the byte information entropy is calculated. To measure the randomness of byte distribution, calculate the standard deviation of message length. To measure frame length stability, the first-order autocorrelation coefficient of the message arrival interval is calculated. Revealing the periodicity of communication and calculating the control character density To measure the degree of structuring, calculate the temporal regularity. Measuring the regularity of communication timing. Five-dimensional features constitute the protocol feature vector. This is used for subsequent protocol template comparison.

[0033] In serial monitoring mode, message boundaries are identified using general heuristic rules, including fixed-length truncation, inter-frame timeout, and specific byte pattern matching. The inter-frame timeout threshold is set based on typical industrial communication intervals, taking 3.5 character transmission times. This value originates from the timeout frameout specifications of standard protocols such as Modbus RTU, ensuring that the inter-frame interval is at least 3.5 character times at common baud rates, avoiding misidentifying consecutive frames as single frames. The specific byte pattern matching identifies common frame header identifiers, such as 0x02 and 0xAA55.

[0034] The five-dimensional protocol feature vector is extracted from the message sequence within the sliding window. The calculation formulas for each dimension of the feature are as follows.

[0035] Byte Information Entropy The calculation formula: ;

[0036] In the formula, The frequency of byte value x within the sliding window is obtained by normalizing the histogram statistics of all message bytes within the window; X is the set of all possible byte values, i.e. This covers the entire value space of a single byte; It is a measure of the randomness of the distribution of message bytes, and its unit is nat, which is the information entropy unit with the natural logarithm as the base. The calculation formula is ,in This represents the number of times the byte value x appears within the window. Due to the regular distribution of function codes and the concentrated value range of address fields, the byte information entropy of structured binary protocols... The value is relatively low; due to the high randomness of characters and the dispersion of readable text characters, the byte information entropy of the ASCII text protocol is relatively low. The values ​​are relatively high. For example, the function codes of the Modbus RTU protocol are concentrated in a few values ​​such as 03, 04, and 06, and the information entropy is usually lower than 5.0 nat; while the JSON-based HTTP protocol, due to the large number of printable ASCII characters, usually has an information entropy higher than 6.0 nat. This feature can effectively distinguish between structured binary protocols and text-based protocols.

[0037] Message length standard deviation ;

[0038] In the formula, The length of the q-th message in bytes is obtained by locating the beginning and end identifiers of the message frame and counting. When locating, the frame boundary identified by the aforementioned general heuristic rule is used first. If the boundary is uncertain, the frame truncation point of the timeout judgment is used as the basis for length calculation. N is the total number of messages in the sliding window. The average message length within the window is calculated using the following formula: ; This is a measure of the stability of protocol message length, measured in bytes. Fixed-frame-length protocols include the Modbus RTU standard frame. The length of the frame header, function code, and CRC checksum is close to zero because, apart from the variable data segment, the lengths are fixed; while custom serial port protocols have a large range of data payload length variations. The value is relatively large. This feature is related to byte information entropy. They are independent of each other, but together they characterize the degree of structure of the protocol.

[0039] First-order autocorrelation coefficient of message arrival interval ;

[0040] In the formula, The arrival time interval between the q-th and (q+1)-th packets is obtained from the gateway hardware timestamp; N packets generate N-1 arrival time intervals, so the upper limit of the summation of the numerator is N-1, and the denominator is the sum of squared deviations of the arrival intervals of the N packets; t̄ is the average arrival interval, calculated using the formula: ; To reveal the periodic characteristics of protocol communication, dimensions are ignored. In master-slave polling protocols such as Modbus RTU, the master station periodically sends request frames, and the slave station responds within fixed time intervals; the intervals between adjacent messages are highly correlated. Approaching 1; in event-triggered protocols such as SNMP Trap, message transmission is driven by abnormal events, and adjacent interval statistics are independent. Approaching zero. This feature effectively distinguishes between polling and triggering communication modes.

[0041] Control character density ;

[0042] In the formula, The total number of occurrences of control characters within the window. Control characters include standard link layer control characters such as frame header identifier 0x02, frame trailer identifier 0x03, and escape character 0x10. The denominator is the sum of the byte lengths of N packets, i.e., the total number of bytes within the window. This is a dimensionless measure of the structured nature of a message. HDLC-like protocols such as PROFIBUS have a high density of control characters because each frame contains start and end delimiters; while raw TCP byte streams lack link-layer frame boundaries, resulting in a lower density of control characters. This feature is orthogonal to the three features mentioned above, supplementing the characterization of protocol type from the perspective of control characters.

[0043] Temporal regularity ;

[0044] In the formula, The standard deviation of message arrival intervals is calculated using the same formula as the standard deviation of message length. Same format; The average arrival interval is ε; ε is a very small positive number to prevent division by zero, taken as 10. -6 The value of seconds is much smaller than the typical cycle of industrial communication protocols in the millisecond range, ensuring that the denominator is not zero without having a substantial impact on the calculation result. To measure the regularity of communication timing, the value ranges from 0 to 1 and is dimensionless. In periodic polling protocols, the standard deviation of the message interval is much smaller than the mean. Approaching 1; in asynchronous event-triggered protocols, the standard deviation of the message interval is comparable to the mean. It approaches zero. This characteristic is related to the autocorrelation coefficient. Both describe the communication pattern from a temporal perspective, but Focusing on absolute temporal stability, It focuses on the relative correlation between adjacent intervals.

[0045] Five-dimensional protocol feature vector Defined as:

[0046] ;

[0047] This vector characterizes the physical and link layer statistical features of protocol messages from five independent dimensions: byte distribution randomness, frame length stability, communication periodicity, control character density, and timing regularity. These five features describe protocol attributes from different physical levels, exhibiting low correlation and high complementarity, thus forming a complete fingerprint representation of the protocol type.

[0048] S2 compares the protocol feature vector with the edge gateway's built-in standard protocol library, calculates the adaptation confidence level, and completes automatic channel configuration based on the adaptation threshold; the protocol feature vector extracted in S1... The reference feature vector of the k-th template in the built-in standard protocol library of the edge gateway Compare and calculate the Euclidean distance. Similarity to fingerprints Combined with semantic accuracy Frequency of occurrence of protocols with the same model of equipment in the past The appropriate configuration confidence is obtained by weighted summation. Select the maximum fit confidence level that exceeds the fit threshold. The protocol template serves as the basis for adaptation, dynamically generating adaptation channel parameters, including register mapping tables, byte order rules, and polling cycles. The standard protocol library pre-stores reference feature vectors for K types, including Modbus RTU, Modbus TCP, OPC UA, CAN 2.0B, Profibus DP, BACnet IP, and private serial port protocols. Each reference vector is obtained by statistical calibration of the standard message set of the corresponding protocol. The statistical calibration is constructed through the acquisition and statistical analysis of communication data from standard laboratory equipment. The library supports online expansion. When adding a new protocol type, the standard message set is manually entered and an automatic calibration process is executed to generate a new template.

[0049] Before calculating the Euclidean distance, each feature component is standardized to have a mean of 0 and a standard deviation of 1, thus eliminating the influence of differences in dimensions and numerical ranges on the distance calculation. The formula for calculating the standardized feature values ​​is as follows: , where μ is the mean of the feature across all templates, and σ is the standard deviation. This standardization method makes the mean of each feature component 0 and the standard deviation 1, eliminating the influence of differences in dimensions and numerical ranges on distance calculation and ensuring a balanced contribution of the five-dimensional features in Euclidean distance calculation.

[0050] The Euclidean distance is calculated as follows:

[0051] ;

[0052] In the formula, , , , , These are the standardized values ​​of each component of the reference feature vector of the k-th protocol template in the standard protocol library, which are statistically calibrated by the standard message set of this protocol type under large sample conditions; , , , , , is the value of the features extracted on-site after being standardized. The distance between the field protocol fingerprint and the standard template in the five-dimensional feature space is denoted as ...

[0053] fingerprint similarity ;

[0054] In the formula, To normalize the baseline distance, take This refers to the maximum distance between the current on-site fingerprint and all K templates in the database; when the on-site protocol differs significantly from all templates, such as a private protocol, Increase accordingly, all Approaching At this point, the similarity is evenly distributed, only affecting the relative ranking, not triggering automatic adaptation, and instead entering the manual confirmation process; the negative sign of the exponential function ensures that the greater the distance, the lower the similarity. To map geometric distance to an exponential similarity score, with a value ranging from 0 to 1 and dimensionless, this mapping conforms to the mapping principle of the Gaussian kernel function in pattern recognition. It exhibits a non-linear amplification effect on distance differences, making the similarity of closely spaced templates more prominent. When all... When they are equal, ,at this time The similarity is evenly distributed and does not affect the relative ranking.

[0055] semantic accuracy ;

[0056] In the formula, This represents the total number of field out-of-bounds errors, checksum errors, and semantic anomalies that occur when parsing N messages using the k-th protocol template. Field out-of-bounds errors refer to parsed data fields exceeding the predefined register range; checksum errors refer to CRC or LRC check failures; and semantic anomalies refer to parsed values ​​exceeding the predefined physical reasonable range for this device type, such as temperature values ​​exceeding the sensor range or negative values, which are physically impossible. N is the total number of messages within the sliding window. To measure the degree of semantic matching between protocols at the application layer, dimensionless, A value close to 1 indicates that the template can correctly parse the field message at both the syntactic and semantic levels.

[0057] Frequency of occurrence of protocols for the same model of equipment in history ;

[0058] In the formula, This represents the number of devices in the historical access records that have the same device model as the current device and have successfully adopted protocol k adaptation. Successful adoption of protocol k adaptation indicates the adaptation configuration reliability. Exceeding the adaptation threshold Furthermore, the adaptation records show that subsequent communication continues normally for more than a preset number of synchronization cycles; this data is stored in the non-volatile memory of the edge gateway and is updated after each successful adaptation; M is the total number of access attempts for this model of device; To leverage the statistical prior of protocol consistency among devices of the same model, which is dimensionless, devices of the same model from the same manufacturer typically use the same communication protocol, this frequency provides a third decision-making basis independent of feature fingerprints and semantic verification.

[0059] when At that time, that is, the first time this model of equipment is connected, Let 1 / K be a uniform distribution, indicating no prior preference.

[0060] The following provides the adaptation confidence level. The calculation formula:

[0061] ;

[0062] In the formula, , , For the weighting coefficients, satisfying The normalization constraints represent the relative importance of fingerprint similarity, semantic accuracy, and historical frequency, respectively. , , The values ​​were determined based on on-site debugging experience: fingerprint similarity directly reflects the degree of matching of underlying protocol features and contributes the most; semantic accuracy verifies the feasibility of application layer parsing and contributes the second most; historical frequency serves as an auxiliary reference and contributes the least. The sum of these three values ​​is strictly normalized to ensure... It has a probabilistic meaning and its value ranges from 0 to 1. The value of k represents the overall compatibility and reliability between the protocol and the field equipment, ranging from 0 to 1, and is dimensionless. The closer to 1, the higher the reliability of the fit.

[0063] choose And exceeds the adaptation threshold The protocol template serves as the basis for adaptation. Adaptation threshold. A value of 0.60 is used, meaning that automatic adaptation is only possible if the overall confidence level exceeds 60%. This threshold balances the success rate of automatic adaptation with the risk of misadaptation. If the adaptation fails, a manual confirmation process is triggered, whereby a field engineer selects the correct protocol from the candidate list. Upon successful adaptation, the adapted channel parameters are dynamically generated, including the register mapping table, byte order rules, and polling cycle, completing the automated configuration of the channel.

[0064] S3, Establish a unified synchronization cycle on the edge side , The value is set as an integer multiple of the industrial field control cycle, with a preferred range of 100 milliseconds to 1000 milliseconds. Data from each channel is... Time alignment is performed based on a benchmark to ensure that data from different protocol channels are comparable under a unified time coordinate system. The quality index is calculated independently for each channel, preserving the original data format and data dimensions, without performing any fusion or normalization operations.

[0065] The alignment operation employs a nearest-neighbor interpolation strategy, assigning the timestamps of each channel's data to the smallest synchronization time that is not less than the actual timestamp. This refers to the backward alignment principle; the aligned data retains the original timestamp. As metadata, but the time series reference used in subsequent calculations is unified as The interpolation process does not change the dimensions of the original data.

[0066] Communication reliability ;

[0067] In the formula, This represents the number of frames successfully responded to by the i-th channel within the most recent statistical period. A successful response means that a complete frame is received and the checksum is correct. This refers to the number of request frames sent to the i-th channel within the same statistical period; the statistical period is 10 synchronization periods, i.e., 10 This differs from the N=100 message cycles of the sliding window in S1; This is a measure of communication stability at the physical and data link layers, with a value ranging from 0 to 1, and is dimensionless. A value close to 1 indicates that the communication link in this channel is stable. This usually indicates problems such as line interference, terminal offline, or excessive load.

[0068] Time Deviation Penalty Factor ;

[0069] In the formula, , is the time offset of the actual timestamp of each channel data relative to the unified synchronization time. It can be positive or negative. A positive value indicates that the channel data is lagging behind the synchronization time, and a negative value indicates that it is ahead. This is the absolute value of the time deviation; The actual timestamp of the latest valid data for the i-th channel is obtained by capturing the system clock when the channel data acquisition is completed. To ensure unified synchronization; Let be the time decay constant, and take . Derived from the concept of information timeliness half-life, that is, when the time deviation reaches a synchronization period, the value of the exponent term is... The penalty factor decays to 0.5; It is the natural logarithm; the absolute value sign ensures that the direction of the deviation does not affect the degree of penalty. This represents the time deviation penalty factor, which is dimensionless. The larger the time deviation, the stronger the penalty. A positive value indicates that the channel data lags behind the synchronization time, while a negative value indicates that it is ahead. The absolute value sign ensures that the direction of the deviation does not affect the degree of penalty.

[0070] Data Quality Index ;

[0071] In the formula, Let Gaussian error function be defined as follows: , where u is the dummy variable of integration, representing the independent variable of the integrand, with a value range from 0 to Z, and the function maps the real number field to the interval (-1,1); Ensure the time deviation term is non-negative to avoid hour negative Anomaly; ω is the reliability weighting coefficient, which is 0.55. This reflects that the contribution of communication success rate to data quality in industrial sites is slightly higher than that of timestamp accuracy. This value is based on on-site debugging statistics. In most industrial environments, the impact of communication packet loss on data availability is greater than that of timestamp deviation. The standard deviation of the timestamp jitter of the i-th channel is obtained from the historical measurements of the most recent W=20 synchronization cycles. The value of W=20 is based on the short-term statistical stability requirement and covers a time span of approximately 2 to 20 seconds. These are normalization coefficients, designed to make the independent variable of the error function compatible with the standard normal distribution; The channel data quality metric is designed to comprehensively consider both communication reliability and time synchronization accuracy. Its value ranges from 0 to 1 and is dimensionless. The closer the value is to 1, the higher the reliability of the channel data.

[0072] make , The normalized time quality index reflects the quality score after timestamp jitter standardization, which reflects the degree of deviation between the data timestamp of channel i and the synchronization time.

[0073] Timing alignment offset ;

[0074] In the formula This represents the time offset of each channel's data relative to the unified synchronization time, used for subsequent alignment compensation. The unit is seconds, and it can be positive or negative. A positive value indicates that the channel data lags behind the synchronization time, while a negative value indicates that it leads. The alignment operation uses a nearest neighbor interpolation strategy to align the timestamps of each channel's data to a value no less than [time value missing]. The smallest At integer multiples of time, the interpolation process does not change the dimensions and physical meaning of the original data.

[0075] Each channel's data retains its original format and dimensions, but includes a quality label. and alignment offset The role of S3 in relation to S4 is to provide independent data for each channel with quality scores, avoiding information loss caused by fusion and ensuring that subsequent information value assessment is based on the original measurements rather than the distorted data after fusion; the histogram statistics in S4 only use... High-quality data is obtained, with a sampling window W of 20 synchronization cycles, consistent with the timestamp jitter statistics window in S3. Each channel's... The value is also used for bandwidth allocation weight correction in S5, where lower-quality channels are given a lower priority in bandwidth contention.

[0076] S4 utilizes the relative entropy of the parameter distribution, combined with the rate of change of relative entropy and the alarm gain factor to calculate the information value index, and performs tiered caching based on the caching decision threshold; for the raw data of each channel output from S3, the concept of relative entropy in information theory is introduced to measure information value, and the KL divergence adopts an asymmetric form, i.e. Unlike the symmetric relative entropy form, this invention applies it to the quantitative evaluation of changes in the distribution of equipment operating parameters.

[0077] The key monitoring parameters are predefined in the equipment parameter configuration file and typically include core state variables such as temperature, pressure, and current. In multi-parameter scenarios, the relative entropy of each parameter is calculated independently, and the maximum value is taken as the equipment-level value. Histogram statistics are based on valid data after S3 time-series alignment, using only... High-quality data, with a sampling window W of 20 synchronization cycles.

[0078] Parameter distribution relative entropy ;

[0079] In the formula, The sampling frequency distribution of the j-th discrete interval of the key monitoring parameter of device i within the current window is obtained by performing histogram statistics and normalization on the W most recent sampled values ​​of the parameter. The histogram intervals are divided into equal widths to cover the full range of the parameter. The prior probability distribution of this parameter for device i under steady-state operating conditions is obtained from the device's factory calibration data or statistics from long-term stable operation and is stored in the device parameter library of the edge gateway; K is the total number of discrete intervals, and K=10 is taken. This value strikes a balance between distribution resolution and statistical stability. Too few intervals will result in insufficient resolution, while too many intervals will result in insufficient sample size for each interval, leading to statistical fluctuations. This represents the information deviation of the current parameter distribution relative to the steady-state distribution, conforming to the rigorous definition of KL divergence in information theory, with the unit being nat. A value of zero indicates that the current parameter distribution is completely consistent with the steady-state distribution, and the equipment is in a stable operating state. The larger the value, the more significantly the current operating state deviates from the steady state, the richer the information contained, and the more priority it needs to be transmitted for analysis by the host computer. The non-negativity of relative entropy is guaranteed by Gibbs' inequality, ensuring that the measure of information value is a non-negative real number.

[0080] when At that time, with Alternative To perform calculations, avoid the logarithmic terms approaching infinity. For the smallest positive number, take 10. -6 .

[0081] Rate of change of relative entropy ;

[0082] In the formula, Let be the relative entropy value at time t; The relative entropy value is the value at the previous sampling time; Δt is the sampling interval, which takes the value of the unified synchronization period. ; and The calculation window is W=20 synchronization cycles, and the adjacent windows overlap by 19 cycles, resulting in short-term autocorrelation. This design aims to smooth statistical fluctuations while preserving the state evolution trend. The absolute value in the numerator represents the magnitude of the change, without distinguishing between the direction of increase or decrease. The relative entropy is the rate of change of time, expressed in nat / s, reflecting the drastic degree of equipment state evolution. A larger value indicates that the device state is rapidly deviating from or returning to a steady state; even if... Even though the absolute value is not large, it should still be given a high information value evaluation because the state evolution process itself contains rich dynamic information.

[0083] Alarm gain factor ;

[0084] In the formula, This represents the current alarm level of device i. A value of 0 indicates no alarm, 1 indicates a general alarm such as parameter exceeding limits or communication timeout, and 2 indicates an emergency alarm such as over-temperature or over-voltage protection trigger. ξ is the alarm response threshold, set to 0.5, located at the arithmetic midpoint between no alarm level 0 and the lowest alarm level 1, ensuring a significant jump in the gain factor when the alarm level crosses this threshold; ξ is the alarm gain base, set to 0.5, controlling the baseline level of the gain amplitude; η is the alarm sensitivity coefficient, set to 2.0, this value ensures that the exponential term generates a significant gain when the alarm level crosses the threshold, when... hour ,when hour ,when hour This achieves an exponential amplification relationship between alarm level and value gain; This is used to map discrete alarm levels to continuous value gain factors. It is dimensionless, conforms to the principle of exponential amplification, and ensures that alarm information receives a significant priority boost in transmission scheduling.

[0085] Information Value Index ;

[0086] In the formula, κ is the weighting coefficient of the rate of change, which is taken as 0.1s / nat, so that... Dimensionless, when When the coefficient of increase in the rate of change is 1, The enhancement factor is 2 at nat / s. This value keeps the contribution of the state evolution rate to the information value within a moderately enhanced range, avoiding oversensitivity. The service level weights are derived from the equipment maintenance level agreement. Critical equipment is assigned a weight of 1.0, general equipment is assigned a weight of 0.6, and standard equipment is assigned a weight of 0.3. These weights are pre-configured by maintenance personnel when the equipment is first connected. This tiered strategy ensures that data from high-commercial-value equipment receives higher transmission priority under the same conditions. It is a measure of information value that integrates four dimensions: relative entropy information content, state evolution rate, alarm level, and business priority. The unit is nat. The larger the value, the richer the information content of the current data of the device, the more drastic the status change, the higher the alarm level, or the greater the commercial value, and the more it should be transmitted with priority.

[0087] Cache decision threshold ;

[0088] In the formula, ρ is the total number of connected devices; ρ is the threshold adjustment coefficient, which is 0.8. This value is based on the experience of the boundary between real-time data and cached data under the limited bandwidth conditions in industrial sites. That is, data with information value exceeding 80% of the average value of the entire site is regarded as high-value real-time data, and data below this threshold enters the cache queue. This is an adaptive threshold based on the average value of all current devices, dynamically adjusted as device status changes, to avoid insufficient adaptability of fixed thresholds in high-load or low-load scenarios. The unit is nat.

[0089] The tiered caching decision rules are as follows: When When data is marked as uploaded in real time, it will be prioritized for transmission to the host computer or cloud platform in the next transmission cycle; when At this time, data enters the edge cache queue and is stored using a differential compression algorithm. Differential compression achieves data compression by storing the difference between adjacent sampled values ​​rather than their absolute values. For slowly changing industrial parameters, the range of the difference values ​​is much smaller than the range of the original values, and the compression ratio can reach more than 10 to 1. Cache data is uploaded in batches when bandwidth is idle or sent in response to a query command.

[0090] S5 executes the Arrhenius model to convert the equipment's operating temperature stress into an equivalent failure rate, achieving a physical unification of equipment reliability status and transmission scheduling decisions. The Arrhenius model is suitable for predicting the failure of electronic components where thermal stress is the primary accelerating factor; for equipment with multiple failure mechanisms or significant electrical and mechanical stress, the model parameters need to be corrected based on field test data.

[0091] Equivalent failure rate ;

[0092] In the formula, This is the baseline failure rate of device i at a reference temperature, expressed in terms of failures per hour. Reliability data provided by the manufacturer is pre-stored in the local device reliability parameter library. The typical baseline failure rate range for electronic control equipment is... Every hour to per hour; The activation energy of device i is 0.3eV to 0.7eV, which corresponds to the activation energy range of typical failure mechanisms of electronic components. 0.3eV corresponds to low-energy barrier failure mechanisms such as interface state drift, and 0.7eV corresponds to high-energy barrier failure mechanisms such as electromigration and thermal oxidation. This range covers the typical failure modes of industrial electronic equipment. Determine based on the reliability manual or accelerated life test data provided by the equipment manufacturer. If such data is unavailable, use the default value according to the equipment type: 0.5eV for controllers, 0.4eV for sensors, and 0.6eV for actuators. Let be the Boltzmann constant, and take . eV / K is a fundamental constant in physics; For reference temperature, 298.15K, or 25℃, is taken. This temperature is the standard ambient temperature for reliability testing of electronic components. The current operating temperature of device i, in K, is obtained by the device's built-in temperature sensor or an external thermocouple; when The time-equivalent failure rate increases exponentially, when The equivalent failure rate decreases or approaches the baseline value. To account for the accelerated failure rate of equipment due to temperature, the unit is the number of failures per hour, which conforms to the Arrhenius acceleration model in reliability physics. When the equipment operating temperature is higher than the reference temperature, the equivalent failure rate increases exponentially; when it is lower than the reference temperature, the equivalent failure rate decreases. This model transforms the real-time monitorable physical quantity of thermal stress into the reliability indicator of failure rate, providing a reliability physics-level decision-making basis for transmission scheduling.

[0093] Specifically, in this step, the equipment operating temperature stress is converted into an equivalent failure rate, and the relative increment coefficient of the equivalent failure rate is used to measure the equipment's condition hazard. The relative increment coefficient P of the equivalent failure rate... i Defined as:

[0094] ;

[0095] in, Let i be the equivalent failure rate of device i at the current operating temperature. Let i be the reference failure rate of device i at the reference temperature. Let k be the activation energy of the failure mechanism corresponding to device i, and k be the Boltzmann constant. For reference temperature, 298.15K is used. This represents the current operating temperature of device i, with all temperatures measured in Kelvin. When... At that time, take The coefficient is 0. Numerically, this coefficient is equal to the Arrhenius acceleration factor minus one, representing the relative increment of the equivalent failure rate under the current temperature stress relative to the baseline failure rate. This coefficient is determined solely by temperature stress and is independent of the equipment's continuous operating time: when the equipment operating temperature is not higher than the reference temperature (i.e., there are no temperature anomalies), this coefficient remains zero regardless of the equipment's continuous operating time; when the equipment operating temperature is higher than the reference temperature, this coefficient monotonically increases with increasing temperature, thus reflecting only the incremental failure risk caused by temperature stress and not introducing deviations accumulated over time.

[0096] Normalized information value ;

[0097] In the formula, It is the maximum value of the information value index of all devices in the current period, that is ; As a relative measure of information value, it normalizes the information value of different devices to the same scale, which facilitates horizontal comparison and weighted combination among multiple devices. The value ranges from 0 to 1 and is dimensionless.

[0098] when At this time, all devices are in a steady state and there are no alarms. Pick It follows a uniform distribution to ensure fairness in bandwidth allocation.

[0099] When all All are less than the minimum threshold During nat, uniform distribution is also performed.

[0100] Bandwidth margin factor Defined as the deviation of a device's currently allocated bandwidth from its fair share. Fair share is defined as:

[0101] ;

[0102] The bandwidth margin factor is defined as:

[0103] ;

[0104] in, For a fair share, This represents the maximum total bandwidth of the link. This represents the total number of connected devices. This represents the bandwidth currently allocated to device i. A state of fair allocation is reached when the allocated bandwidth for each device equals its fair share. When the bandwidth margin factor is 0, it does not contribute to the transmission urgency; it is only considered underservice when device i has less allocated bandwidth than its fair share. It takes a positive value between 0 and 1, and the higher the level of underservice, the larger the value, so that it only makes a limited contribution to improving the transmission urgency of underservice devices and prevents a few high-priority devices from monopolizing bandwidth resources for a long time.

[0105] Transmission urgency of device i Value of Normalized Information The relative increment of equivalent failure rate and bandwidth margin factor The weighted combination yields:

[0106] ;

[0107] in, The weight of the information value element is set to 0.45; The weight of the condition hazard element is set to 0.35; The weight for the bandwidth fairness factor is set to 0.20. The transmission time slots of each device are allocated according to the proportion of its transmission urgency to the sum of the transmission urgency of all devices, and devices below the minimum bandwidth guarantee threshold are covered by the minimum bandwidth guarantee constraint.

[0108] The aforementioned weights were determined using a single-factor perturbation sensitivity analysis method. The specific process was as follows: A test dataset covering three typical scenarios—normal operating conditions, abnormal temperature operating conditions, and bandwidth preemption operating conditions—was constructed. For each element, the values ​​of the other two elements were fixed, and the target element was subjected to a positive and negative perturbation of 10% within its value range. The average relative rate of change of the time slot allocation results was recorded as the sensitivity of that element under that operating condition. The average sensitivity of the three operating conditions was taken, resulting in a sensitivity of 0.45 for the information value element, 0.35 for the state hazard element, and 0.20 for the bandwidth fairness element. These three values ​​were then normalized and used as weights. , , .

[0109] Thus, the five steps complete a full data acquisition and transmission cycle. The edge gateway completes each synchronization cycle... Initially, steps S3 to S5 are executed repeatedly. S3 to S5 form a fast-change control loop with a period of [period missing]. The preferred value is 100ms. S1 and S2 are executed once every N message cycles to update the protocol adaptation state. S1 to S2 form a slow-varying control loop with a period of N message cycles, where N=100, corresponding to approximately 100 This means a 10-second interval, forming a closed-loop gradient acquisition and transmission control. The dual-timescale structure is designed based on the following: protocol characteristics have short-term statistical stability, eliminating the need for re-identification in each synchronization cycle; while data quality assessment, information value calculation, and bandwidth allocation require real-time responses to device status changes. S4's caching decision only affects whether data enters the cache queue, not S5's bandwidth allocation calculation; all devices participate. The calculations and actual transmissions of the cache device are performed when the bandwidth is idle.

[0110] The study conducted a high-temperature scenario verification of four devices to verify the data acquisition and transmission process of four devices operating under high temperature and high load conditions in an energy and power scenario. The focus was on verifying the failure rate acceleration effect and dynamic response of the Arrhenius model under extreme temperature conditions.

[0111] The scenario is set as follows: the edge gateway connects to four devices via a hybrid RS485 bus and Ethernet, with a unified synchronization cycle. Take 0.1 seconds, total link bandwidth It is 300kbps.

[0112] Equipment 1 is a critical motor controller, with a maintenance level of critical. Take 1.0, operating temperature The alarm level is 85℃, or 358.15K. A value of 2 indicates an emergency alarm, representing the baseline failure rate. for Activation energy per hour Taking 0.6 eV, the remaining expected runtime The number of frames requested in the last 100 periods over a 72-hour period. The number of successful response frames is 100. 88, timestamp The standard deviation of the timestamp jitter is 0.088 seconds. It takes 0.015 seconds;

[0113] Device 2 is a general pump set controller. Take 0.6, operating temperature The alarm level is 35℃, or 308.15K. =0, for per hour, Take 0.3 eV, It lasts for 72 hours. It is 100. It is 99. It takes 0.099 seconds. It takes 0.003 seconds;

[0114] Device 3 is a standard valve positioner. Take 0.3, operating temperature The alarm level is 72℃, or 345.15K. A value of 1 indicates a general alarm. for per hour, Take 0.5 eV, It lasts for 72 hours. It is 100. It is 93. It takes 0.094 seconds. It takes 0.008 seconds;

[0115] Device 4 is a standard temperature and humidity sensor. Take 0.3, operating temperature The alarm level is 45℃, or 318.15K. =0, for per hour, Take 0.4 eV, It lasts for 72 hours. It is 100. It is 96. It takes 0.097 seconds. It takes 0.006 seconds.

[0116] The verification process for S3 is as follows, regarding communication reliability. Perform the calculation:

[0117] Device 1 ;

[0118] Device 2 ;

[0119] Device 3 ;

[0120] Device 4 .

[0121] Time deviation penalty factor Perform the calculation: Second.

[0122] Device 1 Second, ;

[0123] Device 2 Second, ;

[0124] Device 3 Second, ;

[0125] Device 4 Second, .

[0126] Data quality index Perform the calculation:

[0127] Device 1 , , ;

[0128] Device 2 , , ;

[0129] Device 3 , , ;

[0130] Device 4 , , .

[0131] The verification process for S4 is as follows: The steady-state distribution is obtained from calibration data and operational statistics.

[0132] Within the current window, histogram statistics show K=10 discrete intervals, using only... High-quality data.

[0133] The motor current parameters of device 1 deviate significantly from steady state. nat, nat / s;

[0134] The outlet pressure of pump 2 in the equipment is stable. nat, nat / s;

[0135] The valve opening of device 3 deviates moderately. nat, nat / s;

[0136] The ambient temperature and humidity of Equipment 4 fluctuate slightly. nat, nat / s.

[0137] Alarm gain factor Perform the calculation:

[0138] Device 1 ;

[0139] Device 2 ;

[0140] Device 3 ;

[0141] Device 4 Same as device 2.

[0142] Information Value Index Perform the calculation:

[0143] nat;

[0144] nat;

[0145] nat;

[0146] nat.

[0147] Cache decision threshold nat.

[0148] Tiered caching decision results:

[0149] Device 1 Marked as real-time upload;

[0150] Device 2 Enter the edge cache;

[0151] Device 3 Enter the edge cache;

[0152] Device 4 Enter the edge cache.

[0153] Only the data from device 1 was marked as being uploaded in real time, a decision that reasonably reflects the highest information value of device 1 under high-temperature emergency alarm conditions.

[0154] The verification process for S5 is as follows; for the equivalent failure rate Perform the calculation:

[0155] The unified synchronization period is 0.1 seconds, and the total link bandwidth is [missing information]. 300kbps, total number of connected devices The fair share is 4. kbps. Boltzmann constant. eV / K, reference temperature Take 298.15K.

[0156] Device 1 is a key motor controller, which activates the motor. The value is 0.6 eV, the current operating temperature T is 358.15 K, or 85 degrees Celsius, and the exponent term is:

[0157] ;

[0158] Arrhenius Acceleration Factor The relative increment of equivalent failure rate Reference failure rate Equivalent failure rate per hour Hourly; bandwidth allocated in the previous cycle was 60kbps, bandwidth margin factor Normalized information value .

[0159] Device 2 is a general pump set controller, with activation energy. The value is 0.3 eV, the current operating temperature T is 308.15 K, the exponential term is 0.0379, and the acceleration factor is... , Reference failure rate Equivalent failure rate per hour 100kbps bandwidth allocated per hour. , .

[0160] Device 3 is a standard valve positioner, with activation energy. The value is 0.5 eV, the current operating temperature T is 345.15 K, the exponential term is 0.2650, and the acceleration factor is... , The baseline failure rate is The equivalent failure rate per hour is Hourly bandwidth allocated: 80kbps , .

[0161] Device 4 is a standard temperature and humidity sensor, with activation energy. The value is 0.4 eV, the current operating temperature T is 318.15 K, the exponential term is 0.0979, and the acceleration factor is... , The baseline failure rate is Equivalent failure rate per hour Hourly bandwidth allocated: 40kbps , .

[0162] The urgency of transmission for each device is

[0163] ;

[0164] ;

[0165] ;

[0166] ;

[0167] The total transmission urgency of the four devices is 0.9440. Based on the proportion of transmission urgency, time slots are allocated as follows: Device 1 receives approximately 69.66%, Device 2 receives approximately 2.19%, Device 3 receives approximately 13.40%, and Device 4 receives approximately 14.75%, totaling 100%. The corresponding allocated bandwidth is: Device 1 approximately 209.0 kbps, Device 2 approximately 6.6 kbps (with minimum bandwidth guarantee as a safety net), Device 3 approximately 40.2 kbps, and Device 4 approximately 44.2 kbps.

[0168] It is evident that equipment 1 is operating under an abnormal high-temperature condition of 85 degrees Celsius. The effective failure rate is 0.4788, which is about 47.9% higher than the baseline. Temperature stress has a significant impact on reliability indicators. Based on this, the present invention prioritizes the transmission urgency of equipment 4, allowing its status data to be transmitted first, providing timely support for predictive maintenance. Although equipment 4 has no high temperature anomaly, its allocated bandwidth is lower than the fair share, and the bandwidth margin factor provides it with appropriate compensation. Equipment 2 and equipment 3 are in a fair allocation steady state or a bandwidth over-share state, with a bandwidth margin factor of zero. Their transmission urgency is determined only by the information value and the state hazard, verifying that the mechanism of action of each modified element is consistent with the design expectation.

[0169] In summary, this invention achieves end-to-end optimization through five interconnected steps: blind protocol identification, adaptive reliability quantification, channel quality assessment, information value measurement, and dynamic bandwidth gradient allocation. S1 and S2 address the automatic identification and adaptation issues in multi-protocol scenarios, eliminating the need for manual pre-configuration. S3 avoids information distortion caused by multi-source data fusion through independent channel quality assessment. S4 introduces the concept of relative entropy from information theory, establishing a strict physical link between information value and parameter distribution changes. S5 unifies equipment reliability physics and transmission scheduling within the same framework using the Arrhenius failure rate acceleration factor, giving transmission urgency a clear physical meaning. Numerical verification results from two embodiments demonstrate that this invention can achieve reasonable data hierarchical transmission decisions under complex industrial conditions, prioritizing the transmission of high-value, urgent data, buffering stable data at the edge, and balancing urgency, risk, and fairness in bandwidth allocation.

[0170] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for data gradient acquisition and transmission based on relative entropy in an edge gateway protocol, characterized in that, Includes the following steps: S1. The edge gateway extracts the original communication byte stream of electromechanical equipment through mirror port or serial listening method, and extracts the protocol feature vector from the message sequence in the sliding window to realize automatic identification of protocol type. S2. Compare the protocol feature vector with the built-in standard protocol library of the edge gateway, calculate the adaptation confidence, and complete the automatic channel configuration based on the adaptation threshold. S3. Establish a unified synchronization cycle on the edge side, calculate the communication reliability and time deviation penalty factor for each channel, and evaluate the data quality index according to the time sequence alignment. S4. Calculate the information value index by using the relative entropy of parameter distribution, combined with the relative entropy change rate and alarm gain factor, and cache in a graded manner according to the caching decision threshold. S5. Convert the equipment operating temperature stress into an equivalent failure rate, determine the transmission urgency by normalized information value, relative increment of equivalent failure rate and bandwidth margin factor, and dynamically allocate transmission time slots.

2. The method according to claim 1, characterized in that, The multidimensional protocol feature vector in S1 includes five dimensions: byte information entropy, message length standard deviation, message arrival interval autocorrelation coefficient, control character density, and timing regularity. These dimensions characterize the statistical features of the protocol message from five independent dimensions: byte distribution randomness, frame length stability, communication periodicity, control character density, and timing regularity.

3. The method according to claim 1, characterized in that, The appropriate configuration confidence in S2 is obtained by weighted fusion of fingerprint similarity, semantic accuracy, and the frequency of occurrence of protocols of the same model in history. The fingerprint similarity maps the geometric distance between the protocol feature vector and the standard template to a similarity score through an exponential kernel function. The semantic accuracy is evaluated by parsing the field out of bounds, checksum errors, and semantic anomalies in the protocol template parsing message. The frequency of occurrence of protocols of the same model in history provides an independent decision basis using the statistical prior of the same model of equipment.

4. The method according to claim 1, characterized in that, In S3, communication reliability is the ratio of successful response frames to request frames, reflecting the communication stability of the physical layer and the link layer. The time deviation penalty factor is calculated based on the deviation between the data timestamp and the unified synchronization time, according to an exponential decay law, to determine the timeliness score. The data quality is a comprehensive evaluation of communication reliability and time synchronization accuracy, assessed by a weighted combination of the time deviation penalty factor and communication reliability. The time deviation penalty factor directly reflects the impact of time synchronization accuracy on data quality.

5. The method according to claim 1, characterized in that, In S4, the relative entropy is the deviation of the sampling frequency distribution of key monitoring parameters within the current window from the prior probability distribution under steady-state operating conditions, reflecting the degree to which the equipment operating state deviates from the steady state; the relative entropy change rate is the rate of change of relative entropy over time, reflecting the severity of the equipment state evolution; the alarm gain factor maps discrete alarm levels to continuous value gain, realizing the exponential amplification relationship between alarm levels and value gain.

6. The method according to claim 1, characterized in that, The information value index in S4 integrates four dimensions: relative entropy information content, state evolution rate, alarm level, and service level. The larger the value, the richer the information content of the current data of the device, the more drastic the state change, the higher the alarm level, or the greater the commercial value. The cache decision threshold is an adaptive threshold based on the average information value of all devices. It is dynamically adjusted with the change of device state. Data exceeding the threshold is marked as real-time upload, and data below the threshold enters the edge cache queue.

7. The method according to claim 1, characterized in that, The S5 equivalent failure rate is based on the Arrhenius acceleration model, which transforms the difference between the equipment operating temperature and the reference temperature into an exponential growth effect of the failure rate, and activates energy values ​​that cover typical failure modes of industrial electronic equipment.

8. The method according to claim 1, characterized in that, In S5, the transmission urgency is obtained by a weighted combination of three dimensions: information urgency, state hazard, and bandwidth fairness. The information urgency is measured by normalized information value, the state hazard is measured by the relative increment of the equivalent failure rate, and the relative increment of the equivalent failure rate is the Arrhenius acceleration factor minus one, which is zero when the equipment has no temperature anomaly. The bandwidth fairness is prevented by the bandwidth margin factor to prevent a few high-priority devices from monopolizing bandwidth resources for a long time.

9. The method according to claim 1, characterized in that, At the beginning of each synchronization cycle, S3 to S5 are executed repeatedly. S1 and S2 are executed once every fixed number of message cycles to update the protocol adaptation status, forming a closed loop of gradient acquisition and transmission control. The data quality index of each channel output by S3 is used to correct the bandwidth allocation weight in S5. The priority of channels with lower quality is appropriately reduced in bandwidth contention.

Images