Safety production monitoring method based on edge collaboration

Abstract

The invention relates to the technical field of edge computing, in particular to a safety production monitoring method based on edge collaboration, which comprises the following steps: collecting multi-node clock pulses, calculating adjacent time differences to form a pulse difference parameter set, carrying out statistical distribution, setting a dynamic tolerance boundary, and constructing an effective deviation parameter set; the method comprises the following steps of: generating a weighted smooth deviation result by combining clock stability and link delay weighted normalization, sorting monitoring data according to a timestamp to identify duplication abrupt change exception, constructing a cross-node exception time sequence by associating nodes and time information, and judging a safety state transition result of equipment. A time difference parameter is constructed through multi-node reference pulse arrival time, a dynamic deviation constraint is formed by combining distribution characteristics, clock stability and link measurement weighted smoothing are introduced, cross-node time reference consistency is kept, operation data unified time sequence rearrangement is carried out, a sudden change event reliably reappears in a time window, and an abnormal identifier is associated with a node identity. And forming a cross-node event evolution link.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, and in particular to a method for monitoring safe production based on edge collaboration. Background Technology

[0002] Edge computing technology primarily involves a series of technologies and methods for data processing, storage, and computation at the network edge. The core idea of ​​edge computing is to move data processing from traditional cloud computing centers to the network edge, closer to the data source, to reduce data transmission latency and bandwidth pressure, and improve real-time performance and response speed. Edge computing not only involves the distributed management of computing resources but also includes localized data processing, storage, analysis, and decision-making to improve system efficiency and reliability. Its applications cover multiple fields such as smart manufacturing, smart cities, the Internet of Things, and autonomous driving, and it demonstrates enormous potential, especially in real-time data processing and low-latency scenarios.

[0003] Traditional edge-based collaborative safety monitoring methods involve multiple edge devices working together to monitor the safety status of the production environment in real time. This method relies on processing and analyzing data acquired by monitoring devices through edge computing nodes and providing immediate warnings of anomalies on the production site. Traditional methods use a single edge node to independently complete data acquisition and analysis, making it difficult to fully utilize the collaborative advantages between multiple devices, thus limiting data processing and response speed. Such methods place high demands on the computing power, data transmission bandwidth, and processing capabilities of edge devices, and in complex production environments, the efficiency of collaborative management and real-time feedback needs improvement.

[0004] Existing edge-collaborative safety production monitoring relies on single nodes to complete data collection and analysis. There is a lack of unified constraints on the time base between nodes, and cross-device data alignment depends on post-event correction. When faced with multi-source asynchronous data, time drift is likely to occur. Anomaly judgment is based on local data fluctuations, making it difficult to form cross-node event correlations. Under complex operating conditions, link latency and hardware jitter are superimposed, leading to deviations in the location of sudden events. The evolution of safety status lacks continuous characterization capabilities, and the stability and traceability of monitoring results are insufficient. Summary of the Invention

[0005] To address the technical problems of existing edge-collaborative safety production monitoring, which relies on single nodes for data collection and analysis, lacks unified constraints on time benchmarks between nodes, depends on post-event correction for cross-device data alignment, is prone to time-series drift when dealing with multi-source asynchronous data, relies on local data fluctuations for anomaly judgment, makes it difficult to establish cross-node event correlations, and suffers from the superposition of link latency and hardware jitter under complex operating conditions, leading to deviations in the location of sudden events, lack of continuous characterization of safety state evolution, and insufficient stability and traceability of monitoring results, this invention provides an edge-collaborative safety production monitoring method.

[0006] To achieve the above objectives, this invention employs a security production monitoring method based on edge collaboration, comprising the following steps: S1: Collect multi-node clock pulse signals from distributed industrial sites, detect the arrival time of the reference pulse for each edge node, sort the pulse timestamps between nodes according to the network topology at the edge computing layer, calculate the difference between adjacent node timestamp pairs, and generate a set of pulse difference parameters. S2: Call the pulse difference parameter set, statistically analyze the distribution characteristic parameters of the difference values ​​and dynamically calculate the deviation tolerance boundary, filter the difference values ​​within the deviation tolerance boundary, and construct an effective deviation parameter set; S3: Call the set of effective deviation parameters, obtain the hardware clock stability index and link delay measurement value of the edge node, calculate the corresponding weight coefficients through principal component analysis, perform weighted processing with the deviation parameters and normalize them to generate a weighted smooth deviation result; S4: Call the timestamp information corresponding to the weighted smoothing deviation result, arrange the operation status monitoring data in time sequence, identify the time when the numerical mutation event occurs, determine whether the numerical mutation event recurs within the preset time window, and generate an abnormal status identifier for the equipment.

[0007] As a further aspect of the present invention, the pulse difference parameter set includes node time offset, relative synchronization error, and time series dispersion index; the effective deviation parameter set includes reliable time offset interval, stable deviation sample set, and offset feature quantity after anomaly suppression; the weighted smoothing deviation result includes global time calibration deviation value, node comprehensive reliable weight value, and time series consistency characterization index; and the equipment abnormal status identifier includes anomaly occurrence time marker, anomaly repeatability discrimination result, and equipment operation anomaly level.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Collects multi-node clock pulse signals from distributed industrial sites, detects the arrival time of the reference pulse for each edge node, records the arrival time of the trigger edge of the pulse waveform for each node based on the node identifier, and performs calibration and solidification operations using timestamp values ​​to obtain the edge node reference pulse timestamp sequence. S102: Based on the edge node reference pulse timestamp sequence, establish a timestamp index mapping relationship based on the network topology adjacency relationship, sort the mapped node timestamp pairs according to the timestamp values, perform index continuity verification and missing data interpolation completion operations, and generate an ordered node timestamp vector. S103: Based on the ordered timestamp vector of the nodes, perform a difference calculation operation on the timestamp values ​​of adjacent index positions on the network topology, record the node index interval corresponding to each pair of differences, and perform vectorized aggregation on the differences based on the node index interval to generate a set of pulse difference parameters.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Call the pulse difference parameter set, iterate through the difference value index item by item, count the number of times the difference value appears according to the index mapping, divide according to the difference value value range, aggregate the frequency results in the range in sequence, and generate difference value distribution characteristic parameters. S202: Based on the difference value distribution characteristic parameters, perform interval comparison on the frequency change of the distribution interval, determine the distribution cutoff position according to the point with the maximum frequency change rate, perform interval boundary movement and recombination around the cutoff position, and perform consistency verification on the recombination interval according to the basic frequency threshold to obtain the deviation tolerance boundary parameters; S203: Based on the deviation tolerance boundary parameter, call the difference values ​​in the pulse difference parameter set one by one, perform interval judgment between the difference values ​​and the deviation tolerance boundary, mark the index of the difference value that meets the condition, and sequentially aggregate the corresponding difference values ​​to establish a valid deviation parameter set.

[0010] As a further aspect of the present invention, the step of performing interval judgment between the difference value and the deviation tolerance boundary refers to calculating the relative distance between the difference value and the deviation tolerance boundary based on the deviation tolerance boundary parameter. If the difference value is within the interval range of the deviation tolerance boundary, the difference value index is marked, and the corresponding difference values ​​are aggregated in order. The difference value aggregation process includes sorting the difference values ​​that meet the conditions in order, merging adjacent difference values ​​and updating the corresponding frequency statistics information to form a set of effective deviation parameters.

[0011] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Call the set of effective deviation parameters, calculate the frequency offset of the clock crystal oscillator of the edge node, perform sampling and recording on the rate of change of the offset, extract the round-trip delay of data packets between nodes and classify them according to the transmission path, and generate a node performance index matrix. S302: Based on the node performance index matrix, extract the node processor utilization rate and bandwidth consumption rate and perform weighted processing, map the clock frequency offset and inter-node delay to the normalized interval respectively, calculate the node weight coefficient, perform boundary value limiting on the weight coefficient, and obtain the edge collaboration weight coefficient group. S303: Based on the edge collaboration weight coefficient group, extract the deviation parameter values ​​from the effective deviation parameter set item by item, perform multiplication and weighted operation on the weight coefficient and the deviation parameter, perform sequential accumulation on the combination result, and establish a weighted smooth deviation result.

[0012] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Call the timestamp information corresponding to the weighted smoothing deviation result, sort and reorganize the running status monitoring data according to the timestamp, perform alignment mapping on the monitoring values ​​under the same time index, perform difference operation on the values ​​corresponding to adjacent time indices, and generate a time-series difference value sequence. S402: Based on the time-series differential numerical sequence, compare each differential value in the sequence with a preset numerical mutation judgment threshold, and perform annotation and aggregation processing on the time indexes that exceed the threshold to generate a set of numerical mutation times. S403: Based on the set of numerical mutation times, perform calculations on the time intervals corresponding to adjacent mutation times, perform interval judgment with the preset time window parameters, perform counts on the mutation indexes that meet the time window conditions, and if the count exceeds the alarm threshold, generate an abnormal device status identifier.

[0013] As a further aspect of the present invention, the preset numerical mutation judgment threshold is obtained by performing offline statistical analysis on the time-series difference numerical sequence corresponding to the operating status monitoring data within the normal operating range, obtaining the upper limit of the fluctuation of the difference value, and using the upper limit of the fluctuation as the preset numerical mutation judgment threshold. The preset time window parameter is determined by statistically classifying the occurrence time intervals of adjacent abnormal events, selecting a time interval range whose occurrence frequency meets a set ratio, and then using the start time value and end time value corresponding to the time interval range as the preset time window parameter.

[0014] As a further aspect of the present invention, the method further includes step S5: S5: Call the device abnormal state identifier, extract the timestamp information and node identifier, associate and map the timestamp information and node identifier through edge collaborative processing, construct a cross-node abnormal event time sequence association table, determine the state transition trajectory of the safe production equipment, and output the safe state transition determination result. The safety state transition determination results include the equipment state evolution sequence, cross-node anomaly correlation, and safety operation state change trend.

[0015] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Call the device abnormal status identifier, extract the timestamp information and node identifier, perform the association mapping between the timestamp and the node identifier, classify and reorganize the mapping results according to the node identifier, perform the continuity check of the timestamp sequence under the same node identifier, mark the interruption position, and generate the node abnormal time sequence. S502: Based on the node anomaly time series, perform cross-node timestamp alignment, establish a unified global time base, reorder to resolve timestamp conflicts, splice the sequence according to the sorting order, perform order consistency judgment on the spliced ​​sequence, separate and label inconsistent sequences, and establish a cross-node anomaly event time series association table. S503: Based on the cross-node abnormal event time sequence association table, perform calculations on the state changes of adjacent event nodes, determine the transition relationship based on the state identifiers before and after the event, perform trajectory encoding and summarization on the determination results, and generate a safe state transition determination result.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, time difference parameters are constructed by using the arrival time of multi-node reference pulses, and dynamic deviation constraints are formed by combining distribution characteristics. Clock stability and link measurement are introduced to participate in weighted smoothing, so that the time base across nodes remains consistent, the running data is rearranged in a unified time sequence, the mutation event is reliably reproduced within the time window, the anomaly identifier is mapped to the node identity, a cross-node event evolution link is formed, the device state transition trajectory has continuity and verifiability, and the stability and accuracy of the safety judgment result are improved simultaneously. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the accompanying drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0020] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0021] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0022] In this embodiment of the invention, sometimes the subscript such as W1 is written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0023] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0024] Please see Figure 1 This invention provides a method for secure production monitoring based on edge collaboration, comprising the following steps: S1: Collect multi-node clock pulse signals from distributed industrial sites, detect the arrival time of the reference pulse for each edge node, sort the pulse timestamps between nodes according to the network topology at the edge computing layer, calculate the difference between adjacent node timestamp pairs, and generate a set of pulse difference parameters. S2: Call the pulse difference parameter set, statistically analyze the distribution characteristic parameters of the difference values ​​and dynamically calculate the deviation tolerance boundary, filter the difference values ​​within the deviation tolerance boundary, and construct an effective deviation parameter set; S3: Call the effective deviation parameter set, obtain the hardware clock stability index and link delay measurement value of the edge node, calculate the corresponding weight coefficient through principal component analysis, perform weighted processing with the deviation parameters and normalize, and generate a weighted smooth deviation result. S4: Call the timestamp information corresponding to the weighted smoothing deviation result, arrange the operation status monitoring data in time sequence, identify the time when the numerical mutation event occurs, determine whether the numerical mutation event recurs within the preset time window, and generate an abnormal equipment status identifier. S5: Call the device abnormal status identifier, extract the timestamp information and node identifier, associate and map the timestamp information and node identifier through edge collaborative processing, construct a cross-node abnormal event time sequence association table, determine the state transition trajectory of safe production equipment, and output the safe state transition determination result. The pulse difference parameter set includes node time offset, relative synchronization error, and time series dispersion index. The effective deviation parameter set includes reliable time offset interval, stable deviation sample set, and offset feature quantity after anomaly suppression. The weighted smoothing deviation result includes global time calibration deviation value, node comprehensive reliable weight value, and time series consistency characterization index. The equipment abnormal state identifier includes anomaly occurrence time marker, anomaly repeatability discrimination result, and equipment operation anomaly level. The safety state transition judgment result includes equipment state evolution sequence, cross-node anomaly correlation relationship, and safety operation state change trend.

[0025] Please see Figure 2 The specific steps of S1 are as follows: S101: Collects multi-node clock pulse signals from distributed industrial sites, detects the arrival time of the reference pulse for each edge node, records the arrival time of the trigger edge of the pulse waveform for each node based on the node identifier, and performs calibration and solidification operations using timestamp values ​​to obtain the edge node reference pulse timestamp sequence. This system collects multi-node clock pulse signals from distributed industrial sites. In safety monitoring scenarios targeting chemical industrial parks or large manufacturing workshops, sensors deployed in areas such as reactors, pressure pipelines, and storage tanks act as edge nodes. These sensors receive periodic clock pulses from a central synchronization server via hardwired or wireless sensor networks, triggering high-frequency counters to poll the signal input port levels in real time. A voltage comparison reference value is set. The reference value is taken as the logic high-level voltage value of the node I / O port. of times, for example in In TTL level systems, setting When the input signal voltage is detected to be from Ascend and cross Instantly, upon detecting the arrival of the pulse trigger edge, the hardware interrupt service routine is immediately triggered to read the current counter value driven by the local high-precision crystal oscillator. Simultaneously, it reads the device's unique identifier stored in the node register. For example, reading a node The counter value is ,node The counter value is Based on local clock frequency (Set as) Convert the counter value to a time unit using the following formula: Then the node The initial time is ,node for Then the value of that moment With node identifier The data is combined and written to specific sectors of non-volatile memory in hexadecimal format to complete the hardening operation, preventing data loss due to reset. A check bit is added to the timestamp value, ultimately forming an edge node reference pulse timestamp sequence containing node identity and precise arrival time.

[0026] S102: Based on the reference pulse timestamp sequence of edge nodes, establish a timestamp index mapping relationship based on the network topology relationship, sort the mapped node timestamp pairs according to the timestamp value, perform index continuity verification and interpolation completion of missing data, and generate an ordered timestamp vector of nodes. Based on the edge node reference pulse timestamp sequence, a hash table structure is constructed by traversing the node data records stored in memory, using the node topology index defined in the network topology. As the key, with the corresponding timestamp value Establish a one-to-one mapping relationship as values, for example, establish a mapping: The timestamp values ​​are extracted from the mapping to form an array to be sorted. Using quicksort logic, the first element of the array is selected as the pivot. The remaining elements are compared to the pivot; if a value is less than the pivot, it is placed in the left partition; otherwise, it is placed in the right partition. This process is recursively repeated until the timestamps are sorted in ascending order from earliest to latest. For example, the sorted data above would be in the following order: (correspond ), (correspond ), (correspond Then, consecutive integer indices are assigned to the sorted sequence. ( ), check if there are any gaps in the index sequence, if the index and If the corresponding physical storage addresses are not contiguous or the logical tags are missing, memory reorganization is performed to relink the discrete storage blocks in the logical address space. As shown in Table 1, some data states after sorting and index rearrangement are displayed. For conflicting items with identical timestamp values ​​in the sorting results, secondary sorting is performed according to the predefined topological hierarchy order or link distance weight value in the network topology relationship to ensure the uniqueness and determinism of the sequence. Finally, the structure array containing strict order relationship and corresponding node information is output to generate an ordered timestamp vector of nodes.

[0027] Table 1: Sorting and Indexing of Pulse Timestamps for Edge Nodes

[0028] As shown in Table 1, after sorting, nodes are assigned new logical indices based on the absolute time of pulse arrival, establishing a correspondence between physical nodes and temporal positions based on network topology.

[0029] S103: Based on the ordered timestamp vector of the nodes, perform difference calculation on the timestamp values ​​of adjacent index positions on the network topology, record the node index interval corresponding to each pair of differences, and perform vectorized aggregation on the differences according to the node index interval to generate a set of pulse difference parameters. Based on the ordered timestamp vector of the nodes, initialize a vector of length [length missing]. Difference result cache, setting loop variable from Traversal to Extract the index position in each iteration. timestamp value With the current index position timestamp value Perform subtraction operation For example, referencing the data in Table 1 above, when At that time, calculate ,when At that time, calculate The calculated difference index range of corresponding nodes Perform associative storage to form tuples A threshold determination is performed on the calculated difference, and the upper limit of the normal transmission delay fluctuation range is set as follows: ,like In Within the range, it is marked as "normal jitter". If the difference is not found, it is marked as "abnormal delay". The calculated difference values ​​are then filled into the vector space in index order to construct the difference feature vector. Simultaneously, the node topology location information and device type label corresponding to each pair of differences are aggregated, and the numerical vector and metadata description information are encapsulated into the same data packet to generate a set of pulse difference parameters.

[0030] Please see Figure 3 The specific steps of S2 are as follows: S201: Call the pulse difference parameter set, iterate through the difference value index item by item, count the number of times the difference value appears according to the index mapping, divide according to the difference value value range, aggregate the frequency results in the range in sequence, and generate the difference value distribution characteristic parameters. The pulse difference parameter set is invoked, and a hash mapping structure or multidimensional array for storing statistical results is initialized. The memory pointer is then set to point to the difference feature vector generated in the previous stage. The starting address, read the vector length ,For example For each data point, set the statistical step size for the difference values. for This serves as the smallest granularity for interval division, establishing the lower limit of the statistical range. and upper limit Through formula Calculate the total number of numerical intervals that need to be divided. Construct an index mapping table to map each numerical range. Assign a unique range index ( Then, the traversal loop is started to extract items one by one. elements in For each Through calculation Determine the range index to which it belongs, if ,but Then access the corresponding index. The counter address is used to increment the value stored at that location by one. After traversal is complete, based on the range index The order of the statistical frequencies of the intervals and the center value of the interval A one-to-one correspondence is established, and invalid intervals with a frequency of zero are removed. The remaining non-zero frequency data are linearly arranged and aggregated according to the interval values ​​in ascending order to construct a structured data sequence containing interval identifiers, center values, and corresponding frequencies, and to generate the difference value distribution characteristic parameters.

[0031] S202: Based on the characteristic parameters of the difference value distribution, perform interval comparison on the frequency change of the distribution interval, determine the distribution cutoff position according to the point with the maximum frequency change rate, perform interval boundary movement and recombination around the cutoff position, and perform consistency verification on the recombination interval according to the basic frequency threshold to obtain the deviation tolerance boundary parameter; Based on the characteristic parameters of the difference value distribution, call the function containing the interval center value. and frequency Structured sequences, setting a baseline frequency threshold The threshold value is the maximum frequency in the sequence. of ,For example Next, then Next, calculate the rate of change of frequency between adjacent intervals. As shown in Table 2, through monitoring To find the inflection point in the trend of change, when the frequency of two consecutive intervals is lower than... or the frequency of a single interval is lower than And its frequency change rate greater than When the current interval position is determined to be the starting point of the "long tail" of the distribution, i.e., the truncation position. For example, determining the index in Table 2 This is the initial cutoff point, and then the data is analyzed around this cutoff location. Perform boundary expansion and set safety margin coefficients. Calculate physical boundary values Based on this, the interval Perform a continuity check to check if there are any void intervals with a frequency of zero within the range. If so, shrink the boundary back to the end of the largest continuous interval before the void to ensure the compactness of the tolerance area. The final locked boundary value is shown in the final judgment results in Table 2, and the deviation tolerance boundary parameters are obtained.

[0032] Table 2: Frequency Distribution and Boundary Determination Table of Difference Values

[0033] As shown in Table 2, the cutoff position was determined based on the inflection point of the sudden drop in frequency (index 4 to index 5), and the final deviation tolerance boundary was calculated accordingly for subsequent anomaly screening.

[0034] S203: Based on the deviation tolerance boundary parameter, call the difference values ​​in the pulse difference parameter set one by one, perform interval judgment between the difference values ​​and the deviation tolerance boundary, mark the index of the difference value that meets the condition, and sequentially aggregate the corresponding difference values ​​to establish a valid deviation parameter set. Based on the deviation tolerance boundary parameters, read the determined boundary values. ,For example (Taken from the calculation results of the previous stage) ), and reload the original set of pulse difference parameters. Initialize the valid difference cache queue and iterate through each difference value in the set. The execution interval contains decision logic and calculates the relative distance coefficient between the difference value and the boundary. ,like ,show Located within the allowable deviation tolerance range Inside, the original index of the difference value Mark as "valid" and set the value Store in a cache queue, for example ,calculate If it is deemed valid; otherwise... If a value is found to be "abnormal," it is marked as "abnormal" and removed. After the traversal is complete, the valid differences in the cache queue are bubble sorted according to the original index order. For consecutive difference values ​​(such as indexes...), the difference is further sorted. and (All are valid), perform vector merging operation, accumulate and merge continuous small deviations into a cumulative deviation block, and synchronously update the frequency weight information corresponding to the block. Finally, encapsulate the data after filtering, sorting and merging to establish a set of valid deviation parameters.

[0035] Please see Figure 4 The specific steps of S3 are as follows: S301: Call the effective deviation parameter set, calculate the frequency offset of the clock crystal oscillator of the edge node, perform sampling and recording on the rate of change of the offset, extract the round-trip delay of data packets between nodes and classify them according to the transmission path, and generate a node performance index matrix. The valid deviation parameter set is invoked, and the metadata in the set header is parsed to obtain the logical address list of the edge nodes to be monitored. This is then achieved through the hardware status register interface embedded within the node. The current frequency value of the clock crystal oscillator is calculated in real time for the period. Combined with nominal frequency (like ), calculate instantaneous frequency offset and record continuously The offset values ​​for each sampling period are used to calculate the rate of change of the offset using the first-order finite difference method. Simultaneously, the master node sends ICMP probe packets with timestamped sequence numbers to each target node in the list, recording the round-trip time from the sender to the receiver's acknowledgment response. Repeat the process for each node. The system performs a second detection operation and calculates an average value to eliminate the impact of network jitter. Based on the physical connection medium properties, the transmission path is divided into three categories: "direct fiber optic connection," "industrial Ethernet," and "wireless sensor network." A latency baseline file is established for each category. Subsequently, the kernel counter of the node's operating system is read to obtain the current CPU time slice occupancy rate. and the real-time throughput of the network interface The collected frequency offset, rate of change, average round-trip delay after path classification, processor utilization rate and bandwidth consumption rate data are filled into a two-dimensional data structure according to the node ID order, as shown in Table 3, to construct a node performance index matrix containing multi-dimensional physical state characteristics.

[0036] Table 3: Edge Node Performance Monitoring Indicators

[0037] As shown in Table 3, the matrix summarizes the key performance data, providing the basic physical parameters for subsequent calculation of weight coefficients.

[0038] S302: Based on the node performance index matrix, extract the node processor utilization rate and bandwidth consumption rate and perform weighted processing. Map the clock frequency offset and inter-node latency to normalized intervals respectively, using the formula: ; Calculate the node weight coefficients, apply boundary numerical limits to the weight coefficients, and obtain the edge collaboration weight coefficient set; in, Representing the Each edge node corresponds to a weight coefficient. Representing the Normalized frequency offset of the hardware clock crystal oscillator at each node Representing the The node and the first Normalized round-trip latency of data packet transmission between nodes Represents the total number of edge nodes. The mean of the normalized clock frequency offset of the representative node. Represents the normalized mean delay between nodes. This represents the clock offset adjustment factor. This represents a stable value and a normal correction quantity; Based on the node performance index matrix, the extracted physical parameters are first normalized to eliminate dimensional differences, and a normalization benchmark for the frequency offset is set. Delay normalization benchmark For the nodes in Table 3 Calculate the normalized frequency offset Similarly, we can conclude , For latency data, assuming the total number of nodes (Only the first three items are shown in the demonstration), normalized latency between each pair of nodes Obtained through measured matrices, for example Calculate the normalized mean delay of nodes. Calculate the mean of the normalized frequency offset of the nodes. Set the clock offset adjustment coefficient Used to balance the impact of offset on weights, setting a stable value to correct the normal quantity. To prevent the denominator from being zero, the following formula is introduced: ; In this formula, Representing the Each edge node corresponds to a weight coefficient. Representing the Normalized frequency offset of the hardware clock crystal oscillator at each node Representing the The node and the first Normalized round-trip latency of data packet transmission between nodes Represents the total number of edge nodes. The mean of the normalized clock frequency offset of the representative node. Represents the normalized mean delay between nodes. This represents the clock offset adjustment factor. This represents a stable value and a normal correction quantity; The advantage of the formula lies in introducing the mean difference of the offset. With network latency distribution: ; A non-linear ratio relationship assigns greater weight to nodes whose clock skew deviates from the group average but whose network connectivity is high, thereby highlighting the influence of key deviation sources in collaborative computing; targeting nodes Perform specific calculations: Step 1, calculate the molecular part: ; The second step is to calculate the sum of squared time delay terms in the denominator, for each node. The associated link latency is: ; Take the square root ; The third step is to calculate the complete denominator: ; Step 4: Calculate the final weights: ; Similarly, the weights of the remaining nodes are calculated, and the calculated original weight coefficients are subjected to boundary value limiting processing, with the effective weight range set as follows: If the calculation result exceeds this range, it is forcibly set to the boundary value. This result indicates that the node... Since its frequency offset is close to the mean, its assigned collaborative weight is relatively low, which is consistent with the strategy of focusing only on larger deviations, and finally the edge collaborative weight coefficient group is obtained.

[0039] S303: Based on the edge collaboration weight coefficient group, extract the deviation parameter values ​​from the effective deviation parameter set item by item, perform a multiplication and weighted operation on the weight coefficients and deviation parameters, perform a sequence accumulation on the combined results, and establish a weighted smooth deviation result. Based on the edge collaborative weight coefficient set, an accumulation register with a length aligned to the time series is initialized, and the filtered deviation parameter values ​​are read item by item from the effective deviation parameter set. For example, extracting the deviation value at a certain moment. At the same time, retrieve the node to which the deviation value belongs. Corresponding weight coefficients (Taken from the aforementioned calculation results), perform weighted multiplication operation. ,Right now The above weighting operation is performed on the valid deviation data within the same time window. The weighted result sequence is then time-aligned using a sliding window accumulation algorithm, with the window size set to [value missing]. Calculate the weighted sum within a window for each data point.

[0040] ; In this way, instantaneous fluctuations are smoothed out. For continuous data streams, the accumulated results are updated in real time, and the smoothed values ​​are mapped to the coordinate system of the monitoring system to form a curve that reflects the trend of weighted deviation, thus establishing the weighted smoothed deviation result.

[0041] Please see Figure 5 The specific steps of S4 are as follows: S401: Call the timestamp information corresponding to the weighted smoothing deviation result, sort and reorganize the running status monitoring data according to the timestamp, perform alignment mapping on the monitoring values ​​under the same time index, perform difference operation on the values ​​corresponding to adjacent time indices, and generate a time-series difference value sequence. First, the circular buffer storing high-frequency time-series data is accessed. A data pointer is used to lock the weighted smoothed deviation data block for the latest monitoring period. The 64-bit Unix timestamp encapsulated in the header of each data record is read. Due to the randomness of distributed network transmission, the original data arrives out of order. A temporary index array is created to store the read timestamp values. As the sorting key, the merge sort algorithm is used to rearrange the data records, ensuring that the records are distributed strictly in chronological order, for example, by rearranging a disordered sequence. Reorganized into After sorting, set the standard sampling time grid and grid step size. Set as Traverse the sorted data sequence and check each timestamp. With standard grid points The deviation, if the deviation is less than If data is missing, direct anchoring is used; otherwise, linear interpolation is employed to utilize adjacent valid values. and Estimate the current grid point values, complete the alignment mapping operation, and establish a unified time index axis. ( Then, the differential operation engine is started, and the sliding registers are set to store the index of the current time. Deviation mapping value Index from the previous time step numerical value Perform first-order backward difference operation For example, when monitoring the pressure data of a chemical pump, the index... Corresponding time The mapping value is ,index Corresponding time The mapping value is The difference value is then calculated. The above subtraction operation is performed on each aligned time index, and the calculated changes with positive and negative polarities are stored in sequence into a double-precision floating-point array to generate a time-series difference numerical sequence.

[0042] S402: Based on the time-series differential numerical sequence, compare each differential value in the sequence with a preset numerical mutation judgment threshold, and perform annotation and aggregation processing on the time index that exceeds the threshold to generate a set of numerical mutation times. Entering the pre-detection stage of anomaly detection, the system first retrieves long-term archived data from storage that is marked as "stable operating state," with the time span set to the past. On that day, the same differential calculation process was performed on the monitoring data for that period to obtain a massive amount of background noise differential samples. The absolute value distribution of the samples was statistically analyzed, and the 99.9th percentile value was calculated as the baseline fluctuation upper limit. For example, statistical analysis shows that the maximum fluctuation range of the difference value under normal operating conditions is... Based on this, a preset threshold for determining numerical mutations is set. ,Right now This threshold setting reserves space for A safety margin is provided to filter out sporadic noise, followed by loading the current real-time time-series differential numerical sequence. Initialize an empty mutation event container, iterate through each difference value in the sequence, and perform absolute value comparison logic. Regarding the aforementioned sequence ,because If the index is reached, it is determined to be a normal fluctuation. At that time, it was discovered ,because The capture mechanism is immediately triggered, and the specific time index of that point is recorded. And the corresponding complete record of the differential amplitude, and perform deduplication and aggregation on continuously triggered indexes, that is, if the index and If the limit is exceeded consecutively, it is regarded as a continuation of the same mutation event. Only the start time or peak time is recorded. Finally, the selected time points that exceed the threshold are packaged into a list to generate a set of numerical mutation times.

[0043] S403: Based on the set of numerical mutation times, perform calculations on the time intervals corresponding to adjacent mutation times, perform interval judgment with the preset time window parameters, perform counts on mutation indexes that meet the time window conditions, and if the count exceeds the alarm threshold, generate an abnormal device status identifier. Read the mutation time points in the set arranged chronologically Initialize the interval calculator and extract two adjacent mutation times sequentially. and Perform subtraction operation To obtain the recurrence interval of mutation events, Table 4 lists some mutation times and their calculated interval data. Simultaneously, a pre-set fault mode feature library is retrieved. For specific "periodic oscillation faults" or "intermittent impact faults," the time difference distribution of adjacent abnormal peaks in the original fault cases is statistically analyzed, and the interval with the highest frequency of occurrence is selected as the judgment criterion. For example, the statistics show that the abnormal peak intervals caused by "centrifuge eccentric vibration" are concentrated in... to Between these, the preset time window parameters are set as follows: Then iterate through the calculated interval values. Determine whether it falls within the range of the window, for example, in Table 4. , in Within the interval, it is determined as a valid count. Events exceeding the specified range are considered isolated events and not included in the statistics; a counter is set up for each such event. The number of intervals falling into the window is accumulated, when The value reaches the set threshold (e.g.) When the device enters a specific abnormal mode, a corresponding fault code is generated based on the matching feature library index, and an abnormal status identifier for the device is generated.

[0044] Table 4: Calculation of Mutation Time Interval and Window Determination Table

[0045] As shown in Table 4, by calculating the time interval of mutations and comparing it with the window, continuous abnormal behaviors that meet specific frequency characteristics were identified.

[0046] Please see Figure 6The specific steps of S5 are as follows: S501: Call the device abnormal status identifier, extract the timestamp information and node identifier, perform the association mapping between timestamp and node identifier, classify and reorganize the mapping results according to the node identifier, perform the continuity check of the timestamp sequence under the same node identifier, mark the interruption position, and generate the node abnormal time series. Access the non-volatile database storing anomaly identifiers and extract the encapsulated metadata fields, primarily including 64-bit precision timestamps. A unique hardware identifier for the device , build a For a multi-bucket mapping structure of hash keys, iterate through the exception identifier records and push the exception timestamps generated by the same node into the corresponding dynamic array buckets in the order of writing. For example, for nodes... Extract the sequence After extraction, a continuity check scan is performed on the independent time series within each node bucket, with a standard heartbeat acquisition interval set. Set a tolerance threshold for continuity determination The threshold is set based on the statistical upper limit of network jitter and packet loss rate, and the difference between adjacent timestamps is calculated one by one. If detected in the sequence and The difference is and less than If the scan is determined to be continuous, continue scanning. If the scan reaches... and Calculate the difference as This value is greater than Immediately and Insert "breakpoint" markers between segments to cut the original long sequence into two independent abnormal sub-segments, and record the starting position index and duration of the interruption. After performing this operation on the node, repackage the segmented sequence and its breakpoint information to generate the node abnormal time series.

[0047] S502: Based on the node anomaly time series, perform cross-node timestamp alignment, establish a unified global time base, reorder to resolve timestamp conflicts, concatenate the sequence according to the sorting order, perform order consistency judgment on the concatenated sequence, separate and label inconsistent sequences, and establish a cross-node anomaly event time series association table. Based on node anomaly time series, a global timeline alignment engine is initialized, and a priority queue with automatic sorting functionality is created to sort nodes (such as...). , , The abnormal sub-fragment header pointer is pushed into the queue. A multi-way merge sort algorithm is executed based on the absolute value of the timestamp, popping the event record corresponding to the smallest timestamp in chronological order. A sequence concatenation operation is then performed to construct a globally unified timeline containing the abnormal behavior of the nodes. During the concatenation process, sequence consistency verification logic is initiated, and a time backtracking tolerance is set. ; ; If a node is found during the splicing process Event timestamp Appearing in the already spliced Then (i.e., earlier in physical time, but arriving in the queue later), calculate the time difference. If the sequence is determined to be out of order, the inconsistent record is removed from the main sequence and labeled as "out of order" and stored separately in the error correction cache. For compliant sequences that pass the verification, the node ID that triggered the anomaly at each time point and its corresponding event type are recorded. A structured form containing time index, node source, event attribute and sequence number is constructed, and a cross-node anomaly event time sequence association table is established.

[0048] S503: Based on the cross-node abnormal event time sequence association table, perform calculations on the state changes of adjacent event nodes, perform transition relationship determination based on the state identifiers before and after the event, perform trajectory encoding and summarization on the determination results, and generate safe state transition determination results; Read the event records in the association table line by line, as shown in Table 5. Set the state space set S = {00: Normal, 01: Slight fluctuation, 10: Critical alarm, 11: Fault shutdown}. For two adjacent events in the global sequence... and Read the current running state value of the associated node at the time of occurrence, calculate the Manhattan distance of the state vector or perform a bitwise XOR comparison to determine whether a state transition exists, for example, at time... ,node Status is ,node Status is , and at the moment ,node The state becomes ,node Induced to become , identify " "and" "Two simultaneous transition behaviors, based on a pre-defined causal logic rule base, encode this cross-node linkage change into hexadecimal trajectory codes. For example, 'the master node deteriorates, causing the slave node to fluctuate' is encoded as..." The trajectory code is filled into the result field, and the trajectory codes within the entire time window are summarized and statistically analyzed to form a complete state evolution link. Finally, the system-level security judgment conclusion after encoding is output, and a security state transition judgment result is generated.

[0049] Table 5: Cross-node anomaly event association and state transition table:

[0050] As shown in Table 5, the anomaly propagation process across nodes is recorded in detail, and discrete anomaly points are transformed into analyzable state transition chains through trajectory encoding.

[0051] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the described technical solutions.

Claims

1. A safety production monitoring method based on edge collaboration, characterized in that, Includes the following steps: S1: Collect multi-node clock pulse signals from distributed industrial sites, detect the arrival time of the reference pulse for each edge node, sort the pulse timestamps between nodes according to the network topology adjacency relationship at the edge computing layer, calculate the difference between adjacent node timestamp pairs, and generate a set of pulse difference parameters. S2: Call the pulse difference parameter set, statistically analyze the distribution characteristic parameters of the difference values ​​and dynamically calculate the deviation tolerance boundary, filter the difference values ​​within the deviation tolerance boundary, and construct an effective deviation parameter set; S3: Call the set of effective deviation parameters, obtain the hardware clock stability index and link delay measurement value of the edge node, calculate the corresponding weight coefficients through principal component analysis, perform weighted processing with the deviation parameters and normalize them to generate a weighted smooth deviation result; S4: Call the timestamp information corresponding to the weighted smoothing deviation result, arrange the operation status monitoring data in time sequence, identify the time when the numerical mutation event occurs, determine whether the numerical mutation event recurs within the preset time window, and generate an abnormal status identifier for the equipment.

2. The safety production monitoring method based on edge collaboration according to claim 1, characterized in that, The pulse difference parameter set includes node time offset, relative synchronization error, and time series dispersion index. The effective deviation parameter set includes reliable time offset interval, stable deviation sample set, and offset feature quantity after anomaly suppression. The weighted smoothing deviation result includes global time calibration deviation value, node comprehensive reliable weight value, and time series consistency characterization index. The equipment abnormal status identifier includes anomaly occurrence time marker, anomaly repeatability discrimination result, and equipment operation anomaly level.

3. The safety production monitoring method based on edge collaboration according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Collects multi-node clock pulse signals from distributed industrial sites, detects the arrival time of the reference pulse for each edge node, records the arrival time of the trigger edge of the pulse waveform for each node based on the node identifier, and performs calibration and solidification operations using timestamp values ​​to obtain the edge node reference pulse timestamp sequence. S102: Based on the edge node reference pulse timestamp sequence, establish a timestamp index mapping relationship based on the network topology adjacency relationship, sort the mapped node timestamp pairs according to the timestamp values, perform index continuity verification and missing data interpolation completion operations, and generate an ordered node timestamp vector. S103: Based on the ordered timestamp vector of the nodes, perform a difference calculation operation on the timestamp values ​​of adjacent index positions on the network topology, record the node index interval corresponding to each pair of differences, and perform vectorized aggregation on the differences based on the node index interval to generate a set of pulse difference parameters.

4. The safety production monitoring method based on edge collaboration according to claim 3, characterized in that, The specific steps of S2 are as follows: S201: Call the pulse difference parameter set, iterate through the difference value index item by item, count the number of times the difference value appears according to the index mapping, divide according to the difference value value range, aggregate the frequency results in the range in sequence, and generate difference value distribution characteristic parameters. S202: Based on the difference value distribution characteristic parameters, perform interval comparison on the frequency change of the distribution interval, determine the distribution cutoff position according to the point with the maximum frequency change rate, perform interval boundary movement and recombination around the cutoff position, and perform consistency verification on the recombination interval according to the basic frequency threshold to obtain the deviation tolerance boundary parameters; S203: Based on the deviation tolerance boundary parameter, call the difference values ​​in the pulse difference parameter set one by one, perform interval judgment between the difference values ​​and the deviation tolerance boundary, mark the index of the difference value that meets the condition, and sequentially aggregate the corresponding difference values ​​to establish a valid deviation parameter set.

5. The safety production monitoring method based on edge collaboration according to claim 4, characterized in that, The process of performing interval judgment between the difference value and the deviation tolerance boundary refers to calculating the relative distance between the difference value and the deviation tolerance boundary based on the deviation tolerance boundary parameter. If the difference value is within the interval range of the deviation tolerance boundary, the difference value index is marked, and the corresponding difference values ​​are aggregated in order. The difference value aggregation process includes sorting the difference values ​​that meet the conditions in order, merging adjacent difference values ​​and updating the corresponding frequency statistics information to form a set of effective deviation parameters.

6. The safety production monitoring method based on edge collaboration according to claim 4, characterized in that, The specific steps for S3 are as follows: S301: Call the set of effective deviation parameters, calculate the frequency offset of the clock crystal oscillator of the edge node, perform sampling and recording on the rate of change of the offset, extract the round-trip delay of data packets between nodes and classify them according to the transmission path, and generate a node performance index matrix. S302: Based on the node performance index matrix, extract the node processor utilization rate and bandwidth consumption rate and perform weighted processing, map the clock frequency offset and inter-node delay to the normalized interval respectively, calculate the node weight coefficient, perform boundary value limiting on the weight coefficient, and obtain the edge collaboration weight coefficient group. S303: Based on the edge collaboration weight coefficient group, extract the deviation parameter values ​​from the effective deviation parameter set item by item, perform multiplication and weighted operation on the weight coefficient and the deviation parameter, perform sequential accumulation on the combination result, and establish a weighted smooth deviation result.

7. The safety production monitoring method based on edge collaboration according to claim 6, characterized in that, The specific steps of S4 are as follows: S401: Call the timestamp information corresponding to the weighted smoothing deviation result, sort and reorganize the running status monitoring data according to the timestamp, perform alignment mapping on the monitoring values ​​under the same time index, perform difference operation on the values ​​corresponding to adjacent time indices, and generate a time-series difference value sequence. S402: Based on the time-series differential numerical sequence, compare each differential value in the sequence with a preset numerical mutation judgment threshold, and perform annotation and aggregation processing on the time indexes that exceed the threshold to generate a set of numerical mutation times. S403: Based on the set of numerical mutation times, perform calculations on the time intervals corresponding to adjacent mutation times, perform interval judgment with the preset time window parameters, perform counts on the mutation indexes that meet the time window conditions, and if the count exceeds the alarm threshold, generate an abnormal device status identifier.

8. The safety production monitoring method based on edge collaboration according to claim 7, characterized in that, The preset numerical mutation judgment threshold is obtained by performing offline statistical analysis on the time-series difference value sequence corresponding to the operating status monitoring data within the normal operating range, and the fluctuation upper limit of the difference value is used as the preset numerical mutation judgment threshold. The preset time window parameter is determined by statistically classifying the occurrence time intervals of adjacent abnormal events, selecting a time interval range whose occurrence frequency meets a set ratio, and then using the start time value and end time value corresponding to the time interval range as the preset time window parameter.

9. The safety production monitoring method based on edge collaboration according to claim 1, characterized in that, The method further includes step S5: S5: Call the device abnormal state identifier, extract the timestamp information and node identifier, associate and map the timestamp information and node identifier through edge collaborative processing, construct a cross-node abnormal event time sequence association table, determine the state transition trajectory of the safe production equipment, and output the safe state transition determination result. The safety state transition determination results include the equipment state evolution sequence, cross-node anomaly correlation, and safety operation state change trend.

10. The safety production monitoring method based on edge collaboration according to claim 9, characterized in that, The specific steps of S5 are as follows: S501: Call the device abnormal status identifier, extract the timestamp information and node identifier, perform the association mapping between the timestamp and the node identifier, classify and reorganize the mapping results according to the node identifier, perform the continuity check of the timestamp sequence under the same node identifier, mark the interruption position, and generate the node abnormal time sequence. S502: Based on the node anomaly time series, perform cross-node timestamp alignment, establish a unified global time base, reorder to resolve timestamp conflicts, splice the sequence according to the sorting order, perform order consistency judgment on the spliced ​​sequence, separate and label inconsistent sequences, and establish a cross-node anomaly event time series association table. S503: Based on the cross-node abnormal event time sequence association table, perform calculations on the state changes of adjacent event nodes, determine the transition relationship based on the state identifiers before and after the event, perform trajectory encoding and summarization on the determination results, and generate a safe state transition determination result.

Images