A method for predicting fiber network resource failure

Abstract

This invention relates to the field of fault prediction technology, specifically to a method for predicting optical fiber network resource faults. The method includes the following steps: obtaining the arrival rate of optical cross-connection requests and the wavelength allocation signaling processing time of the optical network; substituting these into the controller system state space matrix to generate a block tridiagonal transfer rate array; iteratively solving the block tridiagonal transfer rate array to generate a signaling state steady-state probability vector. In this invention, the barcode length shortening rate under continuous time windows continues to be included in the paralysis risk quantification set along with the buffer over-limit probability assessment value. This preserves both the topology contraction trend and the signaling backlog, thus the output network resource fault early warning signal possesses both temporality, structural characteristics, and service stress resistance. It can distinguish between short-term disturbances and sustained instability before the fault manifests, improving the early warning lead time and risk location accuracy.

Description

Technical Field

[0001] This invention relates to the field of fault prediction technology, and in particular to a method for predicting faults in optical fiber network resources. Background Technology

[0002] Fault prediction technology involves continuous monitoring of system operating status, feature extraction, and trend analysis. By modeling the changing patterns of historical and real-time operating data, it identifies potential abnormal patterns and degradation characteristics, thereby providing early warning information before a fault actually occurs.

[0003] Existing technologies often rely on general feature analysis and trend comparison of monitoring results in practical operation. While they can detect state shifts, they lack sufficient characterization of the hierarchical transmission relationships during fault formation, easily treating anomalies from different sources as parallel signals, leading to coarse judgments. After collecting a large amount of operational data during the continuous monitoring phase, if only overall fluctuation amplitude, local degradation trends, or historical comparisons are considered, queue backlogs in the control plane may not yet be clearly manifested as anomalies, quality degradation in transmission links may still be localized, and connectivity contraction of redundant paths in the topology may be scattered across multiple time windows. A single observation is unlikely to accurately identify whether these three types of changes are accumulating in the same direction. In this type of operation, early warning criteria often focus on surface changes, lacking a continuous mapping from state distribution to risk aggregation, resulting in blurred boundaries between minor congestion and true instability precursors, and insufficient correlation between local link degradation and overall resource failure. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method for predicting optical fiber network resource failures.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for predicting optical fiber network resource faults, comprising the following steps: The arrival rate of optical cross-connection requests and the wavelength allocation signaling processing time of the optical network are obtained, and then substituted into the state space matrix of the controller system to generate a block tridiagonal transfer rate array. The block tridiagonal transfer rate array is iteratively solved to generate a signaling state steady-state probability vector. Based on the steady-state probability vector of the signaling state, calculate the queue length value of the fiber optic network controller, generate the buffer over-limit probability assessment value, extract the optical signal-to-noise ratio parameter of the fiber optic network link, take the reciprocal of the link optical signal-to-noise ratio parameter to generate the link degradation distance value, and merge the link degradation distance value with the buffer over-limit probability assessment value to generate the network resource state distribution tensor. Extract the link degradation distance value within the network resource state distribution tensor, construct the backbone loop connectivity feature sequence, and generate a redundant topology feature barcode sequence based on the backbone loop connectivity feature sequence. Extract the barcode length values ​​corresponding to continuous time windows within the redundant topological feature barcode sequence and calculate the length shortening rate. Combine this with the buffer over-limit probability assessment value to generate a paralysis risk quantification set. Generate a network resource fault early warning signal based on the size of the paralysis risk quantification set.

[0006] Preferably, the step of obtaining the signaling state steady-state probability vector is as follows: Acquire continuous records of optical cross-connection request arrival rate, arrange each continuous record value according to a fixed observation interval, count the change in the number of requests within adjacent observation intervals, mark the segments where the number of requests increases continuously and the segments where the number of requests appears in a concentrated manner, extract the burst parameter of optical cross-connection request arrival rate from the peak duration segment, peak interval segment, and unit observation interval deviation, count the extreme value of processing time, frequency of processing time interval, and cumulative percentage of processing time from the wavelength allocation signaling processing time record, extract the wavelength allocation signaling processing time distribution parameter, and form the controller state parameter combination; Based on the controller state parameter combination, read the queue level position, service stage position, and state transition position in the controller system state space matrix. Fill the arrival transition position according to the burst parameter of optical cross-connection request arrival rate and fill the processing transition position according to the wavelength allocation signaling processing time distribution parameter. Check the transition value relationship corresponding to the main diagonal position, upper diagonal position, and lower diagonal position line by line. Concatenate the transition values ​​of each position according to the state level order to generate a block tri-diagonal transfer rate array. Based on the block tri-diagonal transition rate array, the initial state probability value is written in the state hierarchy order. The transition values ​​in the main diagonal block, upper transition block, and lower transition block are read sequentially. The corresponding state transition element item is updated for each state position. The change of the state transition element item after each round of update is recorded. Duplicate state transition element items that meet the convergence requirement are removed. The state transition element items corresponding to all state positions are summarized to generate the signaling state steady-state probability vector.

[0007] Preferably, the step of obtaining the buffer over-limit probability assessment value is as follows: The state position identifier corresponding to each element in the steady-state probability vector of the signaling state is analyzed, the queuing occupancy level corresponding to each state position identifier is extracted, the probability value is paired item by item according to the queuing occupancy level, the probability contribution corresponding to each queuing occupancy level is calculated, and the probability contribution corresponding to all queuing occupancy levels is accumulated to form the queuing queue length value of the fiber optic network controller. The queuing occupancy level range corresponding to the queuing queue length value of the fiber optic network controller is read, the capacity boundary position corresponding to the control plane memory overflow threshold is retrieved, and for each element in the signaling state steady-state probability vector, the difference between the queuing occupancy level and the capacity boundary position is calculated item by item. Element items with a difference greater than zero are filtered out, and the probability values ​​corresponding to the element items with a cumulative difference greater than zero are accumulated to generate a buffer over-limit probability assessment value.

[0008] Preferably, the step of obtaining the network resource state distribution tensor is as follows: Extract the link identifier, acquisition time sequence location, and parameter value corresponding to the optical signal-to-noise ratio parameter of the fiber optic network link. Perform a reciprocal transformation on the parameter value corresponding to each link identifier, and record the link degradation distance value corresponding to the reciprocal transformation result. Write the link degradation distance value item by item into the location unit corresponding to the buffer over-limit probability evaluation value according to the link identifier and acquisition time sequence location. Combine and arrange them according to the link dimension, time sequence dimension, and probability dimension to form a network resource state distribution tensor.

[0009] Preferably, the step of obtaining the backbone loop connectivity feature sequence is as follows: Extract the link degradation distance value corresponding to each link identifier in the network resource state distribution tensor, read the start node identifier and end node identifier corresponding to each link identifier, pair and register the node connection relationship according to the link degradation distance value, filter the node combination with direct connection relationship, record the sorting position and connection level position of the link degradation distance value corresponding to the node combination, and form the node simple complex parameter. The deduplication results of all link degradation distance values ​​within the simple complex parameters of the node are extracted and arranged in ascending order of value. A complex filtering threshold is set for each item. For each complex filtering threshold, the link degradation distance value corresponding to each node combination within the simple complex parameters of the node is compared item by item. Node combinations with link degradation distance values ​​not greater than the complex filtering threshold are retained. The number of connected node sets, the merging position of connected node sets, and the closing connection position under each complex filtering threshold are counted. The connectivity branch parameters corresponding to different complex filtering thresholds are extracted. The connectivity branch parameters are summarized according to the order of complex filtering thresholds to generate the backbone loop connectivity feature sequence.

[0010] Preferably, the step of obtaining the redundant topological feature barcode sequence is as follows: The connected component parameters corresponding to each complex filtering threshold in the main loop connected feature sequence are read item by item. The changes in the connected component parameters of adjacent positions are compared in the order of increasing complex filtering threshold. The feature positions where the connected component parameters change from the present state to the disappearance state are marked. The span of the complex filtering threshold corresponding to each feature position is calculated. The feature disappearance difference corresponding to each feature position is extracted. All feature disappearance differences are combined in the order of feature position and complex filtering threshold to generate a redundant topological feature barcode sequence.

[0011] Preferably, the steps for obtaining the paralysis risk quantification set are as follows: Extract the start time marker, end time marker, and barcode position marker corresponding to each continuous time window within the redundant topological feature barcode sequence. Read the barcode length value corresponding to each barcode position marker according to the arrangement order of the continuous time windows. Calculate the change in barcode length value between adjacent continuous time windows. Filter out the positions where the barcode length value is less than the barcode length value of the previous continuous time window. Divide the change in barcode length value by the corresponding barcode length value of the previous continuous time window. Record the shortening ratio value corresponding to each change position. Summarize the shortening ratio values ​​corresponding to all change positions to obtain the length shortening rate. Read the judgment boundary value, compare the size relationship between each shortening ratio value within the length shortening rate and the judgment boundary value, filter the continuous time window positions where the shortening ratio value is greater than the judgment boundary value, extract the buffer over-limit probability assessment value within the network resource state distribution tensor that is consistent with the continuous time window position, read the tolerance flag and tolerance value corresponding to the signaling storm tolerance value, perform calculations on each buffer over-limit probability assessment value and the corresponding tolerance value, record the calculation results corresponding to each continuous time window position, and obtain the paralysis risk quantification set.

[0012] Preferably, the step of obtaining the network resource fault early warning signal is as follows: Read the calculation results corresponding to each continuous time window position within the paralysis risk quantification set, extract the risk level markers corresponding to different product results, write the warning category identifier, warning trigger position identifier, and warning output sequence identifier according to the risk level markers, verify the correspondence between the risk level markers and warning category identifiers for each continuous time window position, and generate a network resource failure warning signal.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, the arrival rate of optical cross-connection requests and the wavelength allocation signaling processing time are entered into the same state deduction process. First, a steady-state probability vector of the signaling state is formed. Then, the queue length value and buffer over-limit probability assessment value are further derived from the steady-state probability distribution. This transforms signaling load fluctuations from abstract anomalies into a risk quantity that can continue to be transmitted. The link optical signal-to-noise ratio parameter is further transcribed into a link degradation distance value and combined with the buffer over-limit probability assessment value to form a network resource state distribution tensor. Thus, control plane pressure and transmission plane degradation are given a unified characterization, avoiding the one-sidedness caused by judging a single parameter. Subsequently, the backbone loop connectivity feature sequence is extracted around the link degradation distance value, and a redundant topology feature barcode sequence is generated. This is equivalent to expanding the static topology relationship into a structural process that can evolve with a threshold, making the fragmentation symptoms hidden in the redundant path traceable. The barcode length shortening rate under continuous time windows continues to be included in the paralysis risk quantification set together with the buffer over-limit probability assessment value. This preserves both the topology shrinkage trend and the signaling backlog. As a result, the output network resource fault warning signal has both timeliness, structure and business pressure. It can distinguish between short-term disturbances and continuous instability before the fault manifests, thereby improving the early warning amount and the accuracy of risk location. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a simulation diagram of the steady-state probability of the signaling state. Figure 3 A simulation graph showing the arrival rate of optical cross-connect requests; Figure 4 A simulation diagram of data to buffer the probability assessment value of exceeding the limit. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0016] Please see Figure 1-4 This invention provides a technical solution, a method for predicting optical fiber network resource faults, comprising the following steps: The arrival rate of optical cross-connection requests and the wavelength allocation signaling processing time of the optical network are obtained, and then substituted into the state space matrix of the controller system to generate a block tri-diagonal transfer rate array. The block tri-diagonal transfer rate array is iteratively solved to generate a signaling state steady-state probability vector. Based on the steady-state probability vector of the signaling state, calculate the queue length value of the fiber optic network controller, generate the buffer over-limit probability assessment value, extract the optical signal-to-noise ratio parameter of the fiber optic network link, take the reciprocal of the link optical signal-to-noise ratio parameter to generate the link degradation distance value, and merge the link degradation distance value with the buffer over-limit probability assessment value to generate the network resource state distribution tensor. Extract the link degradation distance values ​​within the network resource state distribution tensor, construct the backbone loop connectivity feature sequence, and generate a redundant topology feature barcode sequence based on the backbone loop connectivity feature sequence. Extract the barcode length values ​​corresponding to continuous time windows within the redundant topological feature barcode sequence and calculate the length shortening rate. Combine this with the buffer over-limit probability assessment value to generate a paralysis risk quantification set. Generate a network resource fault early warning signal based on the size of the paralysis risk quantification set.

[0017] The steps for obtaining the steady-state probability vector of the signaling state are as follows: Acquire continuous records of optical cross-connection request arrival rate, arrange each continuous record value according to a fixed observation interval, count the change in the number of requests within adjacent observation intervals, mark the segments where the number of requests increases continuously and the segments where the number of requests appears in a concentrated manner, extract the burst parameter of optical cross-connection request arrival rate from the peak duration segment, peak interval segment, and unit observation interval deviation, count the extreme value of processing time, frequency of processing time interval, and cumulative percentage of processing time from the wavelength allocation signaling processing time record, extract the wavelength allocation signaling processing time distribution parameter, and form the controller state parameter combination; Based on the combination of controller state parameters, read the queue level position, service phase position, and state transition position in the controller system state space matrix. Fill the arrival transition position according to the burst parameter of optical cross-connection request arrival rate and fill the processing transition position according to the wavelength allocation signaling processing time distribution parameter. Check the transition value relationship corresponding to the main diagonal position, upper diagonal position, and lower diagonal position line by line. Piece together the transition values ​​of each position according to the state level order to generate a block tri-diagonal transfer rate array. Based on the block tri-diagonal transition rate array, the initial state probability value is written in the state hierarchy order. The transition values ​​in the main diagonal block, upper transition block, and lower transition block are read sequentially. The corresponding state transition element item is updated for each state position. The change of the state transition element item after each round of update is recorded. Duplicate state transition element items that meet the convergence requirement are removed. The state transition element items corresponding to all state positions are summarized to generate the signaling state steady-state probability vector.

[0018] Specifically, after obtaining continuous records of the optical cross-connection request arrival rate, each continuous record is first arranged at a fixed 1-second observation interval to form time series data. Then, the difference in the number of requests within adjacent observation intervals is calculated to statistically analyze the variation. For example, when the number of requests increases for three consecutive observation intervals, and the total increase exceeds 20% of the average number of requests over the previous 10 observation intervals, this segment is marked as a segment with continuously increasing request numbers. Simultaneously, a threshold for concentrated occurrence is set. This threshold is dynamically determined by calculating the sum of the average request arrival rate over the past 100 observation intervals and 1.5 times the standard deviation. For example, if the average is 1000 times / second and the standard deviation is 120 times / second, then the threshold is... The request arrival rate is calculated as follows: when the number of requests exceeds this threshold for more than two consecutive observation intervals, the segment is marked as a concentrated occurrence segment of requests. Based on these markings, peak duration parameters are extracted from the duration of the concentrated occurrence segment, and peak interval parameters are extracted from the start time interval of two consecutive concentrated occurrence segments. The deviation of the number of requests within each observation interval from the average of the past 100 observation intervals is calculated and used as the deviation parameter per unit observation interval. These three parameters together constitute the burst parameter of the optical cross-connection request arrival rate. Simultaneously, the maximum and minimum processing times are statistically recorded from the historical records of wavelength allocation signaling processing time as processing time extremes. The processing time range is divided into several equally wide intervals, for example, [0, 0.1ms), [0.1, 0.2ms), with a width of 0.1 milliseconds. ...and other multiple intervals, count the frequency of occurrence in each processing time interval, and calculate the proportion of each interval frequency to the total number of records. Then, accumulate them level by level to obtain the cumulative proportion of processing time. Extract the extreme value of processing time, the frequency of processing time interval, and the cumulative proportion of processing time together as wavelength allocation signaling processing time distribution parameters. Finally, combine the obtained optical cross-connection request arrival rate burst parameters with the wavelength allocation signaling processing time distribution parameters to form the controller state parameter combination.

[0019] Based on the controller state parameter combination, the controller system state is first defined as (n, p), where n represents the number of requests in the queue, and p represents the service stage of the wavelength allocation signaling processing. This state space is theoretically infinite, but in actual calculations, a sufficiently large queue capacity upper limit N is set, for example, N=1024. The number of service stages M is set according to the distribution characteristics of the wavelength allocation signaling processing time, for example, M=10. Next, a pre-constructed controller system state space matrix structure indexed by (n, p) is read. This structure defines the queue level position n, the service stage position p, and the state transition relationships between them. Then, the unit observation interval deviation in the aforementioned optical cross-connection request arrival rate burst parameter is used to fill in the transition positions representing request arrival. Specifically, the rate value of transitioning from state (n, p) to (n+1, p), i.e., the upper diagonal block matrix... The elements in the data will be set based on the deviation per unit observation interval, and combined with a baseline arrival rate. For example, the historical average arrival rate is used, and then the wavelength allocation signaling processing time distribution parameters are used to fill in the migration location representing the service processing, that is, the rate value of migration from state (n, p) to (n, p+1) or from (n, p) to (n-1, 1), i.e., the diagonal block matrix. and lower diagonal block matrix The elements in the data will be based on the service rate of the phase distribution fitted by the frequency and cumulative proportion of the processing time interval. To set up the process, after filling the matrix, check the numerical relationships of each position row by row to ensure that the value at each diagonal position is equal to the negative sum of the values ​​at all non-diagonal positions in that row. This represents the total rate of leaving that state, i.e., it satisfies the condition. ,in To determine the transition rate from state i to state j, finally, in ascending order of queue level n, the transition value blocks corresponding to each state level (i.e., ...) are... Together with the boundary matrix B, they are spliced ​​together to generate a block tridiagonal transfer rate array with a specific structure.

[0020] Based on the block three-diagonal transition rate array Q, the initial state probability vector is first written in the order of state hierarchy. Typically, the probability that the system is initially in an idle state is set to 1, i.e. The probabilities of all other states are 0. Then, the iterative solution process begins, sequentially reading the migration values ​​of the main diagonal block, upper migration block, and lower migration block in the tri-diagonal transition rate array Q. For each state position... (Representing a specific (n, p) combination), the steady-state probability is updated one by one using the Gauss-Seidel iterative method. The update rules are shown in the following formula: ; in, It is a state In the The probability value of each iteration. It is a state The total outflow rate (i.e., the main diagonal elements). From state Migration to state rate, Representative at the The latest probability value calculated in each iteration (for) Use the The value of the wheel, for Use the The value of the round), after each complete iteration (i.e., the probability of all state positions is updated once), records the state transition elements (i.e., the probability vector) updated in this round. (and the previous round) The change between the two vectors is measured by calculating the maximum absolute value of the difference between all corresponding elements of the two vectors, i.e. Then this change With a preset convergence accuracy requirement The accuracy requirement is compared. It is usually set to a very small value, for example ,like If the iteration converges, then subsequent repeated iterations are discarded, and the current probability vector is... As a final result, these converged probability values ​​corresponding to all state positions are aggregated to generate a signaling state steady-state probability vector.

[0021] The steps to obtain the buffer over-limit probability assessment value are as follows: The state position identifier corresponding to each element in the steady-state probability vector of the signaling state is analyzed, the queuing occupancy level corresponding to each state position identifier is extracted, the probability value is paired item by item according to the queuing occupancy level, the probability contribution corresponding to each queuing occupancy level is calculated, and the probability contribution corresponding to all queuing occupancy levels is accumulated to form the queuing queue length value of the fiber optic network controller. Read the queuing queue length value of the fiber optic network controller and the corresponding queuing occupancy level range. Retrieve the capacity boundary position corresponding to the control plane memory overflow threshold. For each element in the signaling state steady-state probability vector, calculate the difference between the queuing occupancy level and the capacity boundary position. Filter the elements with a difference greater than zero. Accumulate the probability values ​​corresponding to the elements with a difference greater than zero and generate a buffer over-limit probability assessment value.

[0022] Specifically, analyze the steady-state probability vector of the signaling state; each element of this vector... Each state contains a state location identifier (n, p) and a corresponding steady-state probability value, where n represents the queuing level, i.e., the number of requests waiting to be processed in the queue, and p represents the service stage. First, the entire signaling state steady-state probability vector is traversed, and the queuing level n in the state location identifier (n, p) of each element is extracted. Then, all probability values ​​are paired and grouped according to the queuing level n. For example, the probability values ​​corresponding to all states with queuing levels n=5 (5,1), (5,2), ..., (5,M) are grouped. Group them together, and then calculate the total probability contribution for each queuing level n. The calculation method is to add up the probability values ​​of all service stages under this level, that is... Where M is the total number of service stages, and finally, the desired queue length is obtained by summing the products of all queuing stages and their corresponding probability contributions, forming the queuing queue length value L of the fiber optic network controller. The calculation formula is as follows: Where L is the final queuing length of the fiber optic network controller, n is the number of queuing levels, N is the maximum number of queuing levels set in the model, p is the service stage index, and M is the total number of service stages. It is the steady-state probability value of the corresponding state (n, p) extracted from the steady-state probability vector of the signaling state.

[0023] The system reads the queuing queue length value from the fiber optic network controller, corresponding to the queuing level range (e.g., 0 to 1024), and retrieves a pre-set control plane memory overflow threshold. This threshold is determined based on the physical capacity of the controller's hardware buffer and the network operation and maintenance strategy. For example, if the controller buffer's maximum capacity is 1024 requests, to maintain a certain safety margin, the capacity boundary for memory overflow alarms is set at 85% of the total capacity. Rounded down to 870, this 870 represents the capacity boundary position. Then, the process iterates through each element in the signaling state steady-state probability vector. For each element's state position identifier (n, p), the queuing occupancy level n is extracted, and the difference between the queuing occupancy level n and the capacity boundary position 870 is calculated. Next, all elements with a difference greater than zero are selected. These elements represent the system being in a state where the buffer occupancy exceeds the warning threshold. Finally, the probability values ​​corresponding to all selected elements (i.e., all states (n, p) satisfying n > 870) are calculated. The results are accumulated to obtain the total probability that the number of queued requests in the buffer exceeds a set threshold, thus generating a buffer overload probability assessment value. The calculation formula is as follows: ,in, This is the final generated buffer overrun probability assessment value. The capacity boundary position is set based on the memory overflow threshold (e.g., 870 in the previous example), N is the maximum number of queuing stages, and M is the total number of service stages. It is the steady-state probability of state (n, p).

[0024] The steps to obtain the network resource state distribution tensor are as follows: Extract the link identifier, acquisition time sequence location, and parameter value corresponding to the optical signal-to-noise ratio parameter of the fiber optic network link. Perform a reciprocal transformation on the parameter value corresponding to each link identifier, and record the link degradation distance value corresponding to the reciprocal transformation result. Write the link degradation distance value item by item into the location unit corresponding to the buffer over-limit probability evaluation value according to the link identifier and acquisition time sequence location. Combine and arrange them according to the link dimension, time sequence dimension, and probability dimension to form the network resource state distribution tensor.

[0025] Specifically, the optical signal-to-noise ratio (OSNR) parameter of the fiber optic network link is extracted from the optical network performance monitoring system. Each record contains a unique identifier for the link (e.g., "Link-AB"), the time position of the data point (e.g., timestamp "2023-10-27T10:30:00Z"), and the corresponding OSNR parameter value (e.g., 18 dB). For each link identifier and the OSNR parameter value collected at a specific time position, it is first converted from logarithmic units (dB) to a linear ratio using the following conversion formula: For example, an OSNR of 18 dB converted to a linear ratio of Then, the inverse of the linear ratio is performed, i.e. The reciprocal transformation result is recorded as the link degradation distance value at that time sequence position. Then, this calculated link degradation distance value is associated with the buffer over-limit probability assessment value calculated at the same acquisition time sequence position. Specifically, a data structure is constructed that pairs and stores each link degradation distance value with the corresponding (constant across all links) buffer over-limit probability assessment value according to the two dimensions of link identifier and acquisition time sequence position. Finally, these data are combined and arranged according to the three dimensions of link, time sequence, and feature to form a three-dimensional network resource state distribution tensor T. For example, an element in the tensor... This represents the k-th feature value of link l at time t, where It can store the link degradation distance value of link l at time t, and Then, the buffer overrun probability assessment value corresponding to storage time t.

[0026] The steps for obtaining the backbone loop connectivity feature sequence are as follows: Extract the link degradation distance value corresponding to each link identifier in the network resource state distribution tensor, read the start node identifier and end node identifier corresponding to each link identifier, pair and register the node connection relationship according to the link degradation distance value, filter the node combination with direct connection relationship, record the sorting position and connection level position of the link degradation distance value corresponding to the node combination, and form the node simplex complex parameter; The deduplication results of all link degradation distance values ​​within the simple complex parameters of a node are extracted and arranged in ascending order of value. A complex filtering threshold is set for each item. For each complex filtering threshold, the link degradation distance value corresponding to each node combination within the simple complex parameters of the node is compared item by item. Node combinations with link degradation distance values ​​not greater than the complex filtering threshold are retained. The number of connected node sets, the merging position of connected node sets, and the closing connection position under each complex filtering threshold are counted. The connectivity branch parameters corresponding to different complex filtering thresholds are extracted. The connectivity branch parameters are summarized according to the order of complex filtering thresholds to generate the backbone loop connectivity feature sequence.

[0027] Specifically, the link degradation distance values ​​corresponding to all link identifiers within a specific time window, such as the past 5 minutes, are extracted from the network resource state distribution tensor. Simultaneously, the start and end node identifiers corresponding to each link identifier are read from the network topology database. For example, link "L1" connects nodes "N1" and "N2," and its link degradation distance value is 0.0158. Then, this link information is treated as edges of a graph composed of node pairs and weights (link degradation distance values). All combinations of nodes with direct physical connections are filtered out, such as (N1, ...). Next, the link degradation distance values ​​of all these links are sorted, and the position of the link degradation distance value corresponding to each node combination (link) in the sorted list is recorded as its sort position. For example, if 0.0158 is the 10th smallest among all values, the sort position is 10. At the same time, since each link is a 1-dimensional simplex (edge), its connection level position is recorded as 1. For higher-dimensional structures formed by these links, such as triangles (3 nodes connected in pairs), its connection level position is 2. The identifier of each node combination, its corresponding link degradation distance value, sort position, and connection level position (mainly focusing on level 0 - node, level 1 - link) are integrated to form a set of node simplex parameters. This set describes the geometric and weighted structure of the network topology at a specific time.

[0028] First, extract all link degradation distance values ​​recorded within the node's simple complex parameters and perform deduplication to obtain a set without duplicate values, such as {0.01, 0.015, 0.02, ...}. Then, sort this deduplicated set in ascending order of values ​​to form a filtering sequence. Each value in this filtering sequence is used as the complex filtering threshold. For example, from Begin by applying a complex filtering threshold to each case. Iterate through the set of simple complex parameters of each node, compare the link degradation distance value corresponding to each node combination, and retain only those link degradation distance values ​​that are not greater than the current complex filtering threshold. The nodes (links) are combined, and a subgraph is constructed based on the retained links. Then, a graph traversal algorithm such as disjoint-set data structure or breadth-first search is used to count how many independent sets of connected nodes (i.e., the number of connected branches) exist in the subgraph, and the result is recorded at the current threshold. Under this condition, we record which originally independent sets of connected nodes have merged, the threshold position of the merge, and whether a new closed loop (e.g., three nodes forming a triangle) first appears under this threshold. We record the closed connection position and summarize the information such as the number of connected branches, merge events, and loop birth events under each threshold into the connected branch parameters corresponding to that threshold. Finally, according to the order of the complex filtering threshold from small to large, we arrange the connected branch parameters corresponding to each threshold in sequence to form the trunk loop connected feature sequence.

[0029] The steps for obtaining redundant topological feature barcode sequences are as follows: Read the connected component parameters corresponding to each complex filtering threshold in the main loop connected feature sequence one by one. Compare the changes of connected component parameters at adjacent positions in the order of increasing complex filtering threshold. Mark the feature positions where the connected component parameters change from the present state to the disappearance state. Calculate the complex filtering threshold span corresponding to each feature position. Extract the feature disappearance difference corresponding to each feature position. Combine all feature disappearance differences according to the feature position order and the complex filtering threshold order to generate a redundant topological feature barcode sequence.

[0030] Specifically, the backbone loop connectivity feature sequence is read item by item. This sequence records the changes in network topological connectivity under different complex filtering thresholds, arranged from smallest to largest complex filtering threshold. The order of the threshold positions (e.g.) is compared to the order of the threshold positions (e.g.) and The connected component parameters corresponding to the topological features (e.g., the H1 homology group, representing a loop) are of particular interest. When a certain topological feature (such as a specific loop) reaches a threshold... It exists at one time, and at the next threshold When the loop disappears (usually because it is filled by a higher-dimensional simplex), this position will be... An event marked as a feature vanishing is the range of complex filtering thresholds that this feature traverses from its creation to its disappearance. This is the duration of the feature, also known as the barcode length. The span of the complex filtering threshold corresponding to each marked feature disappearance position is calculated. This span value is the feature disappearance difference. The larger the disappearance difference of a feature, the more stable and important it is in the topology. All the calculated feature disappearance differences are combined and arranged according to the importance of their corresponding features in the topology (e.g., trunk loops take priority) and the order of the generated complex filtering thresholds to form a sequence of values ​​(barcode length), which is the redundant topological feature barcode sequence.

[0031] The steps for obtaining the paralysis risk quantification set are as follows: Extract the start time marker, end time marker, and barcode position marker corresponding to each continuous time window within the redundant topological feature barcode sequence. Read the barcode length value corresponding to each barcode position marker according to the arrangement order of the continuous time windows. Calculate the change in barcode length value between adjacent continuous time windows. Filter the positions where the barcode length value is less than the barcode length value of the previous continuous time window. Divide the change in barcode length value by the corresponding barcode length value of the previous continuous time window. Record the shortening ratio value corresponding to each change position. Summarize the shortening ratio values ​​corresponding to all change positions to obtain the length shortening rate. Read the judgment boundary value, compare the relationship between each shortening ratio value within the length shortening rate and the judgment boundary value, filter the continuous time window positions where the shortening ratio value is greater than the judgment boundary value, extract the buffer over-limit probability assessment value in the network resource state distribution tensor that is consistent with the continuous time window position, read the tolerance flag and tolerance value corresponding to the signaling storm tolerance value, perform calculations on each buffer over-limit probability assessment value and the corresponding tolerance value, record the calculation results corresponding to each continuous time window position, and obtain the paralysis risk quantification set.

[0032] Specifically, continuous time window data arranged in chronological order are extracted from the redundant topological feature barcode sequence. Each time window is set to 5 minutes and includes a start time marker and an end time marker. For example, the time window... The time window is [10:00, 10:05). The time window is set to [10:05, 10:10). Simultaneously, the most important topological features within each time window are extracted, such as barcodes representing the main loop, whose positions are specified by barcode position markers. Then, according to the time window arrangement order, the barcode length value corresponding to the specific barcode position marker within each time window is read sequentially. Next, the change in barcode length value between adjacent consecutive time windows is calculated. Filter out all changes that are negative (i.e. The locations of changes in the barcode length value indicate a decrease in topological redundancy. For each selected location of change, the shortening ratio is calculated by dividing the absolute value of the change in barcode length by the barcode length value of the previous consecutive time window. For example, if , The shortening ratio is then... Finally, all the shortening ratio values ​​recorded during all observation periods are summarized to form a set containing multiple shortening ratio values, thus obtaining the length shortening rate.

[0033] First, a preset judgment boundary value is read. This value is used to determine the severity of topology redundancy reduction. This value is set based on historical network operation data and expert experience. For example, by analyzing the distribution of barcode length reduction rates before historical failures, the judgment boundary value is set to the top 95th percentile value in historical data, such as 0.3. Then, each reduction ratio value in the length reduction rate set is compared with the judgment boundary value of 0.3. All continuous time windows with reduction ratio values ​​greater than 0.3 are selected. These positions are considered moments when the topology structure has potential risks. Next, buffer over-limit probability assessment values ​​consistent with these selected continuous time window positions are extracted from the network resource state distribution tensor. Simultaneously, a predefined signaling storm tolerance value is read. This tolerance value includes a tolerance flag (e.g., "high risk," "medium risk") and a corresponding tolerance value. For example, based on business importance, the tolerance value for "high risk" is set to 0.01, and the tolerance value for "medium risk" is set to 0.001. Then, the probability assessment value of each extracted buffer exceedance is calculated. Divide it by its corresponding signaling storm tolerance value. For example, if at a certain moment... If the value is 0.005, then the result of its calculation with the "medium risk" tolerance value is: Record the calculation result corresponding to each consecutive time window position, and all these calculation results together form a set of paralysis risk quantification.

[0034] The steps for obtaining network resource fault early warning signals are as follows: Read the calculation results corresponding to each continuous time window position within the paralysis risk quantification set, extract the risk level markers corresponding to different product results, write the risk level markers into the warning category identifier, warning trigger position identifier, and warning output order identifier, verify the correspondence between the risk level markers and warning category identifiers for each continuous time window position, and generate a network resource failure warning signal.

[0035] Specifically, the system reads the paralysis risk quantification set, which contains the calculation results at various risk time windows, such as {5.2, 8.1, 3.5, ...}. Then, according to a predefined set of risk level mapping rules, it extracts the risk level labels corresponding to different calculation result values. These mapping rules classify risk levels based on the magnitude of the calculation results; for example, calculation results in the range [1, 5) correspond to "low risk," and those in the range [5, 10) correspond to "medium risk." Corresponding to "high risk", the calculation results 5.2 are marked as "medium risk", 8.1 as "medium risk", and 3.5 as "low risk". Then, according to each risk level label, the corresponding warning category identifier (such as "L2-Warning" representing medium risk), warning trigger location identifier (i.e., the start and end time of the continuous time window in which the risk occurs), and warning output order identifier (determined according to the risk level and time sequence, such as high risk priority) are written from the warning strategy library. Then, the risk level label calculated for each continuous time window position is checked one by one to see if it matches the finally written warning category identifier. For example, it is confirmed that the time window with the calculation result of 8.1 is indeed assigned the warning category identifier of "L2-Warning". Through such verification, the consistency and accuracy of the warning information are ensured. Finally, all the verified warning information is integrated to generate a structured network resource failure warning signal, which contains information such as the warning level, specific time and priority.

[0036] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for predicting resource faults in an optical fiber network, characterized in that, Includes the following steps: The arrival rate of optical cross-connection requests and the wavelength allocation signaling processing time of the optical network are obtained, and then substituted into the state space matrix of the controller system to generate a block tridiagonal transfer rate array. The block tridiagonal transfer rate array is iteratively solved to generate a signaling state steady-state probability vector. Based on the steady-state probability vector of the signaling state, calculate the queue length value of the fiber optic network controller, generate the buffer over-limit probability assessment value, extract the optical signal-to-noise ratio parameter of the fiber optic network link, take the reciprocal of the link optical signal-to-noise ratio parameter to generate the link degradation distance value, and merge the link degradation distance value with the buffer over-limit probability assessment value to generate the network resource state distribution tensor. Extract the link degradation distance value within the network resource state distribution tensor, construct the backbone loop connectivity feature sequence, and generate a redundant topology feature barcode sequence based on the backbone loop connectivity feature sequence. Extract the barcode length values ​​corresponding to continuous time windows within the redundant topological feature barcode sequence and calculate the length shortening rate. Combine this with the buffer over-limit probability assessment value to generate a paralysis risk quantification set. Generate a network resource fault early warning signal based on the size of the paralysis risk quantification set.

2. The fiber optic network resource fault prediction method according to claim 1, characterized in that, The steps for obtaining the steady-state probability vector of the signaling state are as follows: Acquire continuous records of optical cross-connection request arrival rate, arrange each continuous record value according to a fixed observation interval, count the change in the number of requests within adjacent observation intervals, mark the segments where the number of requests increases continuously and the segments where the number of requests appears in a concentrated manner, extract the burst parameter of optical cross-connection request arrival rate from the peak duration segment, peak interval segment, and unit observation interval deviation, count the extreme value of processing time, frequency of processing time interval, and cumulative percentage of processing time from the wavelength allocation signaling processing time record, extract the wavelength allocation signaling processing time distribution parameter, and form the controller state parameter combination; Based on the controller state parameter combination, read the queue level position, service stage position, and state transition position in the controller system state space matrix. Fill the arrival transition position according to the burst parameter of optical cross-connection request arrival rate and fill the processing transition position according to the wavelength allocation signaling processing time distribution parameter. Check the transition value relationship corresponding to the main diagonal position, upper diagonal position, and lower diagonal position line by line. Concatenate the transition values ​​of each position according to the state level order to generate a block tri-diagonal transfer rate array. Based on the block tri-diagonal transition rate array, the initial state probability value is written in the state hierarchy order. The transition values ​​in the main diagonal block, upper transition block, and lower transition block are read sequentially. The corresponding state transition element item is updated for each state position. The change of the state transition element item after each round of update is recorded. Duplicate state transition element items that meet the convergence requirement are removed. The state transition element items corresponding to all state positions are summarized to generate the signaling state steady-state probability vector.

3. The fiber optic network resource fault prediction method according to claim 1, characterized in that, The steps for obtaining the buffer over-limit probability assessment value are as follows: The state position identifier corresponding to each element in the steady-state probability vector of the signaling state is analyzed, the queuing occupancy level corresponding to each state position identifier is extracted, the probability value is paired item by item according to the queuing occupancy level, the probability contribution corresponding to each queuing occupancy level is calculated, and the probability contribution corresponding to all queuing occupancy levels is accumulated to form the queuing queue length value of the fiber optic network controller. The queuing occupancy level range corresponding to the queuing queue length value of the fiber optic network controller is read, the capacity boundary position corresponding to the control plane memory overflow threshold is retrieved, and for each element in the signaling state steady-state probability vector, the difference between the queuing occupancy level and the capacity boundary position is calculated item by item. Element items with a difference greater than zero are filtered out, and the probability values ​​corresponding to the element items with a cumulative difference greater than zero are accumulated to generate a buffer over-limit probability assessment value.

4. The fiber optic network resource fault prediction method according to claim 1, characterized in that, The steps for obtaining the network resource state distribution tensor are as follows: Extract the link identifier, acquisition time sequence location, and parameter value corresponding to the optical signal-to-noise ratio parameter of the fiber optic network link. Perform a reciprocal transformation on the parameter value corresponding to each link identifier, and record the link degradation distance value corresponding to the reciprocal transformation result. Write the link degradation distance value item by item into the location unit corresponding to the buffer over-limit probability evaluation value according to the link identifier and acquisition time sequence location. Combine and arrange them according to the link dimension, time sequence dimension, and probability dimension to form a network resource state distribution tensor.

5. The fiber optic network resource fault prediction method according to claim 1, characterized in that, The steps for obtaining the backbone loop connectivity feature sequence are as follows: Extract the link degradation distance value corresponding to each link identifier in the network resource state distribution tensor, read the start node identifier and end node identifier corresponding to each link identifier, pair and register the node connection relationship according to the link degradation distance value, filter the node combination with direct connection relationship, record the sorting position and connection level position of the link degradation distance value corresponding to the node combination, and form the node simple complex parameter. The deduplication results of all link degradation distance values ​​within the simple complex parameters of the node are extracted and arranged in ascending order of value. A complex filtering threshold is set for each item. For each complex filtering threshold, the link degradation distance value corresponding to each node combination within the simple complex parameters of the node is compared item by item. Node combinations with link degradation distance values ​​not greater than the complex filtering threshold are retained. The number of connected node sets, the merging position of connected node sets, and the closing connection position under each complex filtering threshold are counted. The connectivity branch parameters corresponding to different complex filtering thresholds are extracted. The connectivity branch parameters are summarized according to the order of complex filtering thresholds to generate the backbone loop connectivity feature sequence.

6. The optical fiber network resource fault prediction method according to claim 1, characterized in that, The steps for obtaining the redundant topological feature barcode sequence are as follows: The connected component parameters corresponding to each complex filtering threshold in the main loop connected feature sequence are read item by item. The changes in the connected component parameters of adjacent positions are compared in the order of increasing complex filtering threshold. The feature positions where the connected component parameters change from the present state to the disappearance state are marked. The span of the complex filtering threshold corresponding to each feature position is calculated. The feature disappearance difference corresponding to each feature position is extracted. All feature disappearance differences are combined in the order of feature position and complex filtering threshold to generate a redundant topological feature barcode sequence.

7. The fiber optic network resource fault prediction method according to claim 1, characterized in that, The steps for obtaining the paralysis risk quantification set are as follows: Extract the start time marker, end time marker, and barcode position marker corresponding to each continuous time window within the redundant topological feature barcode sequence. Read the barcode length value corresponding to each barcode position marker according to the arrangement order of the continuous time windows. Calculate the change in barcode length value between adjacent continuous time windows. Filter out the positions where the barcode length value is less than the barcode length value of the previous continuous time window. Divide the change in barcode length value by the corresponding barcode length value of the previous continuous time window. Record the shortening ratio value corresponding to each change position. Summarize the shortening ratio values ​​corresponding to all change positions to obtain the length shortening rate. Read the judgment boundary value, compare the size relationship between each shortening ratio value within the length shortening rate and the judgment boundary value, filter the continuous time window positions where the shortening ratio value is greater than the judgment boundary value, extract the buffer over-limit probability assessment value within the network resource state distribution tensor that is consistent with the continuous time window position, read the tolerance flag and tolerance value corresponding to the signaling storm tolerance value, perform calculations on each buffer over-limit probability assessment value and the corresponding tolerance value, record the calculation results corresponding to each continuous time window position, and obtain the paralysis risk quantification set.

8. The optical fiber network resource fault prediction method according to claim 1, characterized in that, The steps for obtaining the network resource fault early warning signal are as follows: Read the calculation results corresponding to each continuous time window position within the paralysis risk quantification set, extract the risk level markers corresponding to different product results, write the warning category identifier, warning trigger position identifier, and warning output sequence identifier according to the risk level markers, verify the correspondence between the risk level markers and warning category identifiers for each continuous time window position, and generate a network resource failure warning signal.

Images