Optical fiber AI fault prediction method and system combined with big data analysis

Abstract

The invention provides an optical fiber AI fault prediction method and system combined with big data analysis, and the method comprises the steps: obtaining a continuous optical fiber operation data flow which is generated in real time in an optical fiber operation process and comprises optical power fluctuation data, signal transmission loss data and optical fiber link time delay data; performing fault feature time sequence evolution trajectory capture processing on the continuous optical fiber operation data flow to obtain a fault feature time sequence evolution trajectory, performing cross-dimension correlation tracing processing, generating a cross-dimension correlation tracing result, calling a preset optical fiber AI fault prediction model, inputting the cross-dimension correlation tracing result for dynamic modeling processing, and performing dynamic modeling processing on the cross-dimension correlation tracing result. According to the method, the optical fiber fault dynamic prediction sequence is generated, the optical fiber fault early warning operation is triggered according to the optical fiber fault dynamic prediction sequence, the optical fiber fault early warning instruction containing the complete fault early warning content is generated, operation and maintenance personnel can be timely and effectively notified to take corresponding measures, and the accuracy and timeliness of optical fiber fault prediction are remarkably improved. And stable operation of the optical fiber communication system is ensured.

Description

Technical Field

[0001] This invention relates to the field of big data technology, and more specifically, to a fiber optic AI fault prediction method and system that combines big data analysis. Background Technology

[0002] In the field of fiber optic communication, the stable operation of optical fibers is crucial for ensuring efficient and accurate data transmission. With the continuous expansion and increasing complexity of fiber optic networks, the frequency of fiber optic faults has increased, and the types of faults have become increasingly diverse. Timely and accurate prediction of fiber optic faults can effectively reduce data transmission interruptions caused by these faults, minimize the impact on communication services, and improve the reliability and stability of the communication system.

[0003] Currently, fiber optic fault prediction mainly relies on traditional monitoring methods and simple data analysis techniques. Traditional monitoring methods typically detect the operating status of fibers through periodic inspections or setting fixed thresholds, but these methods have significant limitations. Periodic inspections cannot capture subtle changes in the fiber's operation in real time, making it difficult to detect potential faults in a timely manner; while fixed threshold monitoring methods are too rigid and cannot adapt to the dynamic changes in the fiber's operating status, easily leading to misjudgments or missed detections. In addition, most existing data analysis techniques only perform simple processing on single-dimensional fiber operating data, failing to fully consider the correlation and temporal evolution characteristics between different dimensions of data, resulting in the accuracy and timeliness of fault prediction failing to meet practical needs. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a fiber optic AI fault prediction method incorporating big data analysis, the method comprising: Acquire a continuous optical fiber operation data stream generated in real time during optical fiber operation, the optical fiber operation data stream including optical power fluctuation data, signal transmission loss data, and optical fiber link delay data; The fault feature time-series evolution trajectory is captured and processed by performing fault feature time-series evolution trajectory on the optical fiber operation data stream. Based on the continuous change trend of optical power fluctuation data, signal transmission loss data, and optical fiber link delay data in the optical fiber operation data stream, and combined with the feature evolution rules corresponding to each fault type stored in the optical fiber historical fault feature database, the fault feature time-series evolution trajectory is obtained. Cross-dimensional correlation tracing processing is performed on the temporal evolution trajectory of fault characteristics. The correlation patterns of optical power fluctuation data, signal transmission loss data, and optical fiber link delay data in the optical fiber operation data stream are associated with multi-dimensional features before the fault in the optical fiber historical fault feature database, generating cross-dimensional correlation tracing results. Call the preset fiber optic AI fault prediction model, input the cross-dimensional correlation tracing results into the fiber optic AI fault prediction model to perform dynamic modeling processing, and generate a dynamic prediction sequence of fiber optic faults. Based on the fault occurrence status and corresponding fault type association information presented in the dynamic prediction sequence of optical fiber faults, the optical fiber fault early warning operation is triggered through the early warning triggering interface of the optical fiber operation and maintenance management system, generating an optical fiber fault early warning instruction containing complete fault early warning content.

[0005] In another aspect, embodiments of the present invention also provide a fiber optic AI fault prediction system that combines big data analysis, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0006] Based on the above, this embodiment of the invention acquires continuous fiber optic operation data streams generated in real time during fiber optic operation, comprehensively covering multi-dimensional key information such as optical power fluctuation data, signal transmission loss data, and fiber optic link delay data. It performs fault feature time-series evolution trajectory capture processing on the fiber optic operation data stream, combining the continuous change trend of the data with the feature evolution patterns in the historical fiber optic fault feature database to accurately obtain the fault feature time-series evolution trajectory, effectively capturing the dynamic changes before the fault occurs. Cross-dimensional correlation tracing processing is performed on the fault feature time-series evolution trajectory, associating multi-dimensional data changes with correlation patterns in the historical fault feature database to generate cross-dimensional correlation tracing results, deeply exploring the intrinsic connections between data of different dimensions. A preset fiber optic AI fault prediction model is invoked, and the cross-dimensional correlation tracing results are input for dynamic modeling processing to generate a dynamic fiber optic fault prediction sequence, achieving dynamic and accurate prediction of fiber optic faults. Based on the prediction sequence, a fiber optic fault early warning operation is triggered, generating a fiber optic fault early warning command containing complete fault warning content. This can promptly and effectively notify maintenance personnel to take corresponding measures, significantly improving the accuracy and timeliness of fiber optic fault prediction and ensuring the stable operation of the fiber optic communication system. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the execution flow of the fiber optic AI fault prediction method combined with big data analysis provided in an embodiment of the present invention.

[0008] Figure 2 This is a schematic diagram of exemplary hardware and software components of the fiber optic AI fault prediction system that combines big data analysis, provided in an embodiment of the present invention. Detailed Implementation

[0009] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1This is a flowchart illustrating a fiber optic AI fault prediction method combining big data analysis, provided in one embodiment of the present invention. The following is a detailed description of this fiber optic AI fault prediction method combining big data analysis.

[0010] Step S110: Obtain the continuous optical fiber operation data stream generated in real time during the operation of the optical fiber. The optical fiber operation data stream includes optical power fluctuation data, signal transmission loss data, and optical fiber link delay data.

[0011] In this embodiment, taking the operation and maintenance monitoring scenario of a fiber optic communication network in the core area of ​​a city as an example, the core area of ​​the city has a dense network of fiber optic lines, carrying a large number of government, commercial and residential daily communication services, and has extremely high requirements for network stability. In order to realize real-time monitoring of the fiber optic operation status in this area, dedicated data acquisition terminals are deployed at each fiber optic junction box, equipment room and important nodes in the area. The above terminals are connected to the interface of the fiber optic communication equipment and collect optical power fluctuation data, signal transmission loss data and fiber optic link delay data according to a preset collection cycle (such as every fixed time interval).

[0012] Among them, optical power fluctuation data is acquired through the optical power sensor built into the terminal, reflecting the real-time changes in the power of the optical signal during transmission; signal transmission loss data is calculated by comparing the difference in optical power between the transmitting and receiving ends, reflecting the energy attenuation of the optical signal in the transmission path; fiber optic link delay data is obtained by recording the time difference between the optical signal being sent from the transmitting end and received at the receiving end, characterizing the transmission duration of the optical signal on the link. All collected data are accompanied by precise timestamps to ensure the temporal correlation of the data. During the data acquisition process, for sensitive data such as specific fiber optic routing information, data anonymization technology is used to replace the specific routing identifier with a randomly generated anonymous code. Simultaneously, encrypted tunneling technology is used for data transmission between the acquisition terminal and the data processing center to ensure data security during transmission and prevent the leakage of sensitive information.

[0013] Step S120: Perform fault feature time-series evolution trajectory capture processing on the optical fiber operation data stream. Based on the continuous change trend of optical power fluctuation data, signal transmission loss data, and optical fiber link delay data in the optical fiber operation data stream, and combined with the feature evolution rules corresponding to each fault type stored in the optical fiber historical fault feature database, the fault feature time-series evolution trajectory is obtained.

[0014] In the aforementioned urban core area fiber optic communication network operation and maintenance monitoring scenario, after acquiring continuous fiber optic operation data streams, it is necessary to extract the temporal evolution trajectory that reflects the fault development process. This process requires in-depth analysis of the continuous changes in optical power fluctuation data, signal transmission loss data, and fiber optic link delay data, and comparison with the characteristic evolution patterns in historical fault cases, thereby constructing the temporal evolution trajectory of fault characteristics.

[0015] Step S121: Simultaneously expand the optical power fluctuation data, signal transmission loss data, and optical fiber link delay data in the optical fiber operation data stream according to the same time dimension to obtain continuous time series data of optical power fluctuation data, continuous time series data of signal transmission loss data, and continuous time series data of optical fiber link delay data.

[0016] In the fiber optic communication network scenario in the core urban area, all data in the fiber optic operation data stream carries timestamp information. In order to analyze the relationship between the changes of the three data items within a unified time frame, the optical power fluctuation data, signal transmission loss data, and fiber optic link delay data need to be arranged in chronological order according to their timestamps.

[0017] For example, all optical power fluctuation data collected within a given day can be arranged sequentially from morning to night according to their corresponding timestamps, forming a continuous curve with time on the horizontal axis and optical power value on the vertical axis, thus obtaining continuous time series data for optical power fluctuation. The same method can be used to process signal transmission loss data and fiber optic link delay data, yielding corresponding continuous time series data. This synchronous processing ensures that the three data sets are consistent across the time dimension, allowing for subsequent feature extraction and correlation analysis based on the same time interval.

[0018] Step S122: For the continuous time series data of optical power fluctuation data, extract the inflection points of the data value changes over time in the continuous time series data of optical power fluctuation data, record the time coordinate and data value change amplitude corresponding to each inflection point, and form a set of inflection points of optical power fluctuation data.

[0019] In the aforementioned urban core area fiber optic communication network scenario, continuous time-series data of optical power fluctuations reflects the continuous change of optical power values ​​over time. Inflection points are points where the trend of optical power change changes significantly; these points often indicate abnormal changes in the fiber's operating status. Therefore, it is necessary to extract these inflection points from the continuous time-series data of optical power fluctuations.

[0020] Step S1221: Perform time-series segmentation on the continuous time series data of optical power fluctuation data. Each time-series segment contains a preset number of continuous data points, and there are preset lengths of overlapping data points between adjacent time-series segments.

[0021] In the context of the fiber optic communication network in the core urban area, to more accurately capture the changing trends of optical power fluctuation data, it is necessary to perform time-series segmentation processing on its continuous time series data. For example, each time series segment is set to contain a certain number of continuous data points, while a certain number of overlapping data points are retained between two adjacent time series segments. The purpose of this is to avoid the changing trend being interrupted at the segment boundaries due to data segmentation, and to ensure that the inflection points of change that may cross the segment boundaries can be fully captured.

[0022] Specifically, assuming that the continuous time series data of optical power fluctuation data has a total of several data points, it is divided into multiple continuous time series segments according to the preset segmentation rules. Each segment contains a number of continuous data points, and the latter part of the data points of the previous segment overlaps with the former part of the data points of the next segment.

[0023] Step S1222: Calculate the slope of the data value change in each time series segment by dividing the numerical difference of adjacent data points in each time series segment by the corresponding time difference.

[0024] For each time series segment, the difference in optical power between any two adjacent data points within that segment is calculated sequentially. This difference is then divided by the time difference between the two data points (i.e., the acquisition period) to obtain the slope of change between adjacent data points. Next, the slopes of change for all adjacent data point pairs within that time series segment are arithmetically averaged to obtain the average slope of that time series segment. This average slope reflects the overall trend of optical power fluctuations within that time series segment. A positive slope indicates an overall upward trend in optical power during that period, while a negative slope indicates an overall downward trend. The absolute value of the slope indicates the degree of drastic change.

[0025] Step S1223: Compare the change slopes of adjacent time series segments. If the difference in the change slopes of adjacent time series segments exceeds the preset slope difference standard, then mark the middle data point among the overlapping data points of the two adjacent time series segments as a candidate change inflection point.

[0026] The average slope of each time series segment is compared with the average slope of its adjacent time series segments, and the difference between the two (i.e., the slope difference) is calculated. A slope difference standard is preset, determined based on historical data and experience, to determine whether the trend has changed significantly. If the slope difference between two adjacent time series segments exceeds the preset standard, it indicates that the trend of optical power fluctuation data has changed significantly in the transition region between these two time series segments. In this case, the data point at the middle position of the overlapping data points of these two adjacent time series segments is marked as a candidate inflection point. For example, if there are 5 overlapping data points between two adjacent time series segments, the 3rd data point is marked as a candidate inflection point, which can better represent the position where the trend changes.

[0027] Step S1224: For each candidate change inflection point, calculate the average slope of the change of a preset number of data points before and after the candidate change inflection point, and obtain the average slope before the candidate change inflection point and the average slope after the candidate change inflection point respectively.

[0028] For each marked candidate inflection point, further verification is needed to confirm whether it is a genuine inflection point. Specifically, a predetermined number (e.g., several) of consecutive data points are selected before the candidate inflection point, and these data points are divided into a new temporary time series segment. The average slope of this temporary segment is calculated using the method in step S1222, and this average slope is used as the average slope before the candidate inflection point. Similarly, the same predetermined number of consecutive data points are selected after the candidate inflection point, and this segment is divided into another new temporary time series segment. Its average slope is calculated, and this average slope is used as the average slope after the candidate inflection point. By comparing the average slopes before and after the candidate inflection point, it is possible to more accurately determine whether the trend of change before and after the candidate point has truly changed significantly.

[0029] Step S1225: If the difference between the average slope before the candidate inflection point and the average slope after the candidate inflection point still exceeds the slope difference standard, then the candidate inflection point is confirmed as a formal inflection point.

[0030] The average slope before the candidate inflection point calculated in step S1224 is compared with the average slope after the candidate inflection point, and the difference between the two is calculated. If the difference still exceeds the preset slope difference standard, it indicates that the trend of change before and after the candidate inflection point has indeed undergone a significant and stable change, and therefore the candidate inflection point is confirmed as a formal inflection point. Conversely, if the difference does not exceed the preset standard, it indicates that the candidate point may just be a random fluctuation of the data and not a real inflection point, and it is removed from the candidate inflection point list.

[0031] Step S1226: Record the time coordinates corresponding to each formal change inflection point, and calculate the difference between the data value before the formal change inflection point and the data value after the formal change inflection point to determine the magnitude of the data value change.

[0032] For each confirmed inflection point, the timestamp of that data point is extracted from the continuous time series data of its corresponding optical power fluctuation, serving as the time coordinate of that inflection point. Simultaneously, the average optical power value of a predetermined number of data points before the inflection point (e.g., the same number as in step S1224) is selected as the pre-inflection point data value, and the average optical power value of the same predetermined number of data points after the inflection point is selected as the post-inflection point data value. The difference between the post-inflection point data value and the pre-inflection point data value is calculated; the absolute value of this difference represents the magnitude of the data value change at the inflection point. A positive difference indicates an upward trend in optical power after the inflection point, while a negative difference indicates a downward trend.

[0033] Step S1227: Organize the time coordinates and data value change amplitudes of all formal change inflection points in chronological order to form a set of optical power fluctuation data change inflection points.

[0034] All confirmed inflection points are sorted chronologically according to their time coordinates to form a set of inflection points for optical power fluctuation data. Each element in the set contains the time coordinate of an inflection point and the corresponding magnitude of the data value change. This set clearly shows at which time points the optical power fluctuation data changed significantly and the magnitude of the changes during the monitoring period.

[0035] Step S123: Using the same inflection point extraction method as the continuous time series data of optical power fluctuation data, perform inflection point extraction processing on the continuous time series data of signal transmission loss data and the continuous time series data of optical fiber link delay data respectively to obtain the inflection point set of signal transmission loss data and the inflection point set of optical fiber link delay data.

[0036] In the fiber optic communication network scenario of the city's core area, the continuous time series data of signal transmission loss and fiber optic link delay are processed using the same procedure as for extracting inflection points of optical power fluctuation data. Specifically, the continuous time series data of signal transmission loss is first segmented (corresponding to step S1221), the slope of each segment is calculated (corresponding to step S1222), the slope differences between adjacent segments are compared to mark candidate inflection points (corresponding to step S1223), the average slope before and after the candidate inflection point is calculated and compared with a standard to confirm the official inflection point (corresponding to steps S1224 and S1225), the time coordinates and calculated change amplitude of the official inflection point are recorded (corresponding to step S1226), and finally, the data is organized chronologically to form a set of inflection points for signal transmission loss. The same steps are used to process the continuous time series data of fiber optic link delay to obtain a set of inflection points for fiber optic link delay. This unified processing method ensures the comparability of inflection points for the three types of data, facilitating subsequent cross-dimensional correlation analysis.

[0037] Step S124: Retrieve the feature evolution patterns corresponding to each fault type from the optical fiber historical fault feature database. The feature evolution patterns include the order of occurrence and correlation patterns of change inflection points of optical power fluctuation data, signal transmission loss data, and optical fiber link delay data before the occurrence of each fault type.

[0038] The fiber optic historical fault feature database is a database that stores past fiber optic fault cases and their corresponding characteristic information. In the context of the fiber optic communication network in the core urban area, this database contains historical records of various common fault types, such as fiber breaks, loose connectors, fiber aging, and external construction interference. For each fault type, the feature evolution pattern details the typical order in which inflection points in optical power fluctuation data, signal transmission loss data, and fiber optic link delay data occur in the period preceding the fault, as well as the correlation patterns between the amplitudes of these inflection points. For example, for a loose fiber optic connector fault, the feature evolution pattern might be as follows: first, a small, frequent inflection point in optical power fluctuation data appears; then, a slowly increasing inflection point appears in signal transmission loss data; finally, irregular fluctuations occur in fiber optic link delay data, with a certain positive correlation between the amplitudes of optical power fluctuations and signal transmission loss fluctuations. These feature evolution patterns are retrieved from the fiber optic historical fault feature database as a reference for judging whether current fiber optic operation data shows signs of fault.

[0039] Step S125: Normalize the change amplitudes in the inflection point set of optical power fluctuation data, signal transmission loss data, and fiber optic link delay data respectively. Then, match the evolution trend of the normalized inflection point set of optical power fluctuation data, signal transmission loss data, and fiber optic link delay data with the change inflection point patterns in the characteristic evolution laws corresponding to each fault type.

[0040] Because optical power fluctuation data, signal transmission loss data, and fiber optic link delay data have different dimensions and ranges of variation, directly comparing their variation amplitudes cannot accurately reflect their indicative role in fault indication. Therefore, it is necessary to normalize the variation amplitudes of data values ​​in the inflection point sets of these three types of data. The normalization method is as follows: for each type of data (such as optical power fluctuation data), divide the absolute value of all variation amplitudes in the inflection point set of that type of data by the maximum value of the variation amplitude under the corresponding fault type in the historical fault feature database to obtain the normalized variation amplitude (the value range is between 0 and 1). After normalization, the elements in the normalized inflection point sets of the three types of data (including time coordinates and normalized variation amplitudes) are compared with the characteristic evolution patterns of each fault type in the historical fiber optic fault feature database. Specifically, this involves comparing whether the current order of occurrence of the inflection points of the three types of data is consistent with the order in the characteristic evolution pattern of a certain fault type, and whether the proportional relationship between the normalized variation amplitudes of each inflection point conforms to the association pattern of that fault type, i.e., performing evolution trend matching.

[0041] Step S126: Based on the evolution trend matching results, construct a trajectory framework that includes the time coordinates and normalized change amplitude of the inflection point of optical power fluctuation data change, the time coordinates and normalized change amplitude of the inflection point of signal transmission loss data change, and the time coordinates and normalized change amplitude of the inflection point of fiber optic link delay data change, as well as the degree of correlation with the evolution law of fault type.

[0042] Based on the results of evolution trend matching, the feature evolution patterns of one or more fault types with the highest matching degree are selected. Using these feature evolution patterns as templates, a trajectory framework is constructed. This trajectory framework first includes the time coordinates and normalized change amplitudes of all inflection points in the sets of inflection points for optical power fluctuation data changes, signal transmission loss data changes, and fiber optic link delay data changes. Then, the positions of these inflection points are marked on the time axis and connected by lines or arrows to show the temporal relationship between the inflection points of the three types of data changes. Simultaneously, a correlation parameter is added to the trajectory framework to indicate the degree of similarity between the current extracted inflection point pattern and the feature evolution pattern of a certain fault type. This correlation parameter is determined based on the similarity score calculated during the matching process (such as a comprehensive score of sequence matching degree, amplitude ratio matching degree, etc.), and is used to represent the degree of similarity between the currently extracted change inflection point pattern and the feature evolution pattern of a certain fault type.

[0043] Step S127: Supplement the trajectory framework with continuous change curves of optical power fluctuation data, signal transmission loss data, and fiber optic link delay data between their respective inflection points to form a fault characteristic time-series evolution trajectory.

[0044] Based on the aforementioned trajectory framework, the continuous time series data of optical power fluctuation, signal transmission loss, and fiber optic link delay are connected, connecting the continuous data points corresponding to each inflection point to form three continuous change curves. These curves are superimposed on the time axis of the trajectory framework. These continuous change curves demonstrate the specific changes and details of each data point between inflection points, with the inflection points serving as key nodes on the curves. By combining the continuous change curves with inflection points, temporal relationships, and correlation parameters, a complete fault characteristic time-series evolution trajectory is ultimately formed. This trajectory visually reflects the similarity between the current trend of fiber optic operation data and the characteristic evolution patterns preceding a certain type of fault in the past.

[0045] Step S130: Perform cross-dimensional correlation tracing processing on the time-series evolution trajectory of fault characteristics, correlate the changes in optical power fluctuation data, signal transmission loss data, and optical fiber link delay data in the optical fiber operation data stream with the multi-dimensional feature correlation patterns before the fault occurred in the optical fiber historical fault feature database, and generate cross-dimensional correlation tracing results.

[0046] In the context of the fiber optic communication network in the core urban area, after obtaining the time-series evolution trajectory of fault characteristics, it is necessary to further analyze the intrinsic relationship between the three dimensions of optical power fluctuation data change, signal transmission loss data change, and fiber optic link delay data change, as well as the correspondence between these relationships and the characteristic patterns before the occurrence of historical faults, so as to trace the potential factors that may lead to the fault.

[0047] Step S131: Extract optical power fluctuation data change segments, signal transmission loss data change segments, and fiber optic link delay data change segments from the fault feature time-series evolution trajectory. Each optical power fluctuation data change segment, each signal transmission loss data change segment, and each fiber optic link delay data change segment contains information on the change of the corresponding data within the same time interval.

[0048] From the time-series evolution trajectory of fault characteristics, a continuous time interval containing multiple inflection points is selected as the analysis window. Within this time interval, corresponding portions of the continuous change curves for optical power fluctuation data, signal transmission loss data, and fiber optic link delay data are extracted to form three data change segments. These three data change segments have identical start and end points, ensuring that they contain change information within the same time interval. Each data change segment not only contains continuous data points within the time interval but also information on all inflection points within that interval (time coordinates and normalized change amplitudes).

[0049] Step S132: Retrieve the multi-dimensional feature association pattern before the fault occurred from the optical fiber historical fault feature database. The multi-dimensional feature association pattern describes the mutual influence relationship and timing coordination features between changes in optical power fluctuation data, changes in signal transmission loss data, and changes in optical fiber link delay data before the fault occurred.

[0050] The multi-dimensional feature correlation patterns stored in the fiber optic historical fault feature database summarize the interrelationships among optical power fluctuation data, signal transmission loss data, and fiber optic link delay data before various faults occurred in history. These patterns detail how, in a specific stage before a fault occurs, the other two dimensions typically change accordingly (mutual influence relationship) when one dimension of the data changes, and how these changes occur collaboratively over time (temporal coordination characteristics). For example, for fiber optic aging faults, the multi-dimensional feature correlation pattern might be described as follows: high-frequency, small-amplitude fluctuations in optical power fluctuation data appear before the slow increase in signal transmission loss data, and the greater the amplitude of the optical power fluctuation, the faster the subsequent increase in signal transmission loss. Meanwhile, fiber optic link delay data begins to show a significant increase after the signal transmission loss reaches a certain threshold.

[0051] Step S133: Assign correlation and tracing weights to the optical power fluctuation data change segment, the signal transmission loss data change segment, and the fiber optic link delay data change segment, respectively. The correlation and tracing weights are determined based on the degree of influence of each of the optical power fluctuation data change, signal transmission loss data change, and fiber optic link delay data change on the occurrence of the fault.

[0052] Based on historical failure case analysis and domain expert experience, the impact of changes in three dimensions—optical power fluctuation, signal transmission loss, and fiber optic link delay—on failure occurrence is determined. Dimensions with higher impact carry greater weight in the correlation and attribution analysis, i.e., they are assigned higher correlation and attribution weights. For example, in some failure types, changes in signal transmission loss may be the direct cause of the failure, thus its correlation and attribution weight might be the highest, followed by changes in optical power fluctuation, and then changes in fiber optic link delay. These weight values ​​need to be normalized so that the sum of the three weights is 1 for subsequent weighted calculations.

[0053] Step S134: Based on the assigned correlation and tracing weights, a multi-dimensional correlation and tracing model is constructed by setting the structural parameters of the input layer, feature extraction layer, correlation calculation layer, and output layer. The input of the multi-dimensional correlation and tracing model consists of optical power fluctuation data change segments, signal transmission loss data change segments, and fiber optic link delay data change segments. The output of the multi-dimensional correlation and tracing model is the matching degree parameter with the multi-dimensional feature correlation pattern.

[0054] Step S1341: Determine the input layer structure of the multi-dimensional correlation tracing model. The input layer of the multi-dimensional correlation tracing model contains three data input channels, which correspond to the input of optical power fluctuation data change segments, signal transmission loss data change segments, and optical fiber link delay data change segments, respectively.

[0055] The input layer of the multi-dimensional correlation tracing model is designed to simultaneously receive data change segments across three dimensions. Each data input channel corresponds to one dimension of data, and each channel contains several neurons. The number of neurons matches the number of data points contained in the data change segment. For example, if each data change segment contains M data points, then each input channel contains M neurons, each receiving one data point value from the corresponding data change segment. In this way, the input layer can completely receive and transmit the original information from the three-dimensional data change segments.

[0056] Step S1342: Set up a feature extraction submodule for each data input channel. Each feature extraction submodule is used to extract key change features in the corresponding data change segment and to standardize the key change features. The key change features include the data change period and the frequency of data peak occurrence.

[0057] Each data input channel is followed by an independent feature extraction submodule. This submodule consists of several layers of neural networks (such as convolutional and pooling layers). For each segment of input data, the convolutional layers slide across the data segment using kernels of different sizes to extract local variation features, such as the period of data variation (determined by identifying recurring patterns) and the frequency of peak occurrences (the number of peak points per unit time). The pooling layers then perform dimensionality reduction on the features extracted by the convolutional layers, preserving key feature information. The extracted key variation features (such as variation period values ​​and peak frequency values) need to be standardized, i.e., subtracting the feature value from the average value of historical data and then dividing by the standard deviation of the feature value in historical data, making different features comparable.

[0058] Step S1343: Construct an association calculation layer within the multi-dimensional association tracing model. The association calculation layer receives the standardized key change features output by the three feature extraction sub-modules, and, combined with the assigned association tracing weights, calculates the similarity between the key change features of the optical power fluctuation data change segment and the key change features of the optical power fluctuation data in the multi-dimensional feature association mode, the key change features of the signal transmission loss data change segment and the key change features of the signal transmission loss data in the multi-dimensional feature association mode, and the key change features of the fiber optic link delay data change segment and the key change features of the fiber optic link delay data in the multi-dimensional feature association mode.

[0059] The association calculation layer receives standardized key change feature vectors from three feature extraction submodules. Simultaneously, it retrieves the current multi-dimensional feature association pattern to be matched from the fiber optic historical fault feature database and extracts the standard key change feature vectors corresponding to the three dimensions of the pattern. For each dimension, the key change feature vector of the current data change segment is compared with the standard key change feature vector in the pattern, and the similarity between the two is calculated. Similarity can be calculated using methods such as cosine similarity or Euclidean distance; the closer the cosine similarity is to 1 or the smaller the Euclidean distance, the higher the similarity. During the calculation, the similarity result for each dimension is multiplied by the corresponding association source weight to reflect its importance in the association analysis.

[0060] Step S1344: Set up the output layer of the multi-dimensional correlation tracing model. The output layer of the multi-dimensional correlation tracing model will weight and integrate the similarity of key change features of optical power fluctuation data, key change features of signal transmission loss data, and key change features of optical fiber link delay data to obtain the overall matching degree parameter, and at the same time output the individual similarity parameters of each key change feature.

[0061] The output layer contains multiple neurons. One neuron outputs the overall matching degree parameter, while the other neurons output the similarity parameters of the key change features in each of the three dimensions. The overall matching degree parameter is calculated by summing the weighted similarities of the three dimensions output by the association calculation layer (since the weights have been normalized). The individual similarity parameters of each key change feature are directly taken from the calculation results of the association calculation layer without weighting. In this way, the model output includes both the overall matching situation and the specific matching details of each dimension.

[0062] Step S1345: Initialize the filtering parameters of the feature extraction submodule and the weight coefficients of the correlation calculation layer within the multi-dimensional correlation tracing model. The initialization parameters are obtained by training based on sample data in the optical fiber historical fault feature database.

[0063] In the initial stage of model construction, it is necessary to initialize the convolution kernel parameters (filtering parameters) in the feature extraction submodule and the weight coefficients in the association calculation layer (in addition to the preset association tracing weights, other coefficients used to adjust similarity calculations may also be included). These initialization parameters are not randomly set, but are obtained through pre-training using a large amount of sample data from the fiber optic historical fault feature database. Specifically, data change segments from historical fault cases are selected as input samples, and the matching results of the corresponding known multi-dimensional feature association patterns are used as the expected output. The internal parameters of the model are adjusted through the backpropagation algorithm to make the model output as close as possible to the expected output. After multiple rounds of iterative training, the final parameters are saved as initialization parameters.

[0064] Step S1346: By inputting sample data change segments from the optical fiber historical fault feature database, the multi-dimensional correlation tracing model is verified and adjusted so that the matching degree parameter output by the multi-dimensional correlation tracing model can accurately reflect the matching status between the data change segments and the multi-dimensional feature correlation pattern.

[0065] The constructed multi-dimensional correlation tracing model was validated using change segments of sample data from another portion of the fiber optic historical fault feature database that were not pre-trained. The aforementioned sample data was input into the model to obtain the matching degree parameter output by the model. This parameter was compared with the multi-dimensional feature correlation pattern of the actual fault type corresponding to the sample data, and the prediction error was calculated. If the error exceeded the preset range, the model's structural parameters (such as the number of layers and neurons in the feature extraction submodule, the calculation method of the correlation calculation layer, etc.) needed to be adjusted or the model parameters needed to be retrained until the matching degree parameter output by the model could accurately reflect the true matching between the data change segment and the multi-dimensional feature correlation pattern, and the validation error was controlled within an acceptable range.

[0066] Step S1347: After completing the construction of the multi-dimensional correlation tracing model, save the structure and parameters of the multi-dimensional correlation tracing model to form a multi-dimensional correlation tracing model that can be directly called.

[0067] After pre-training and validation adjustments, the structure (layer settings, number of neurons, etc.) and internal parameters (convolution kernel parameters, weight coefficients, etc.) of the multi-dimensional correlation tracing model are determined. The model structure is saved as a text file, and the model parameters are saved as a binary file, forming a complete model file that can be directly loaded and called. In subsequent cross-dimensional correlation tracing processing, this model file can be directly loaded to process the changing segments of the input data without the need to rebuild and retrain the model.

[0068] Step S135: Input the extracted optical power fluctuation data change segments, signal transmission loss data change segments, and fiber optic link delay data change segments into the multi-dimensional correlation tracing model. Through the feature collaborative comparison mechanism within the multi-dimensional correlation tracing model, calculate the degree of fit between the optical power fluctuation data change segments and the optical power fluctuation data change patterns in the multi-dimensional feature correlation model, the degree of fit between the signal transmission loss data change segments and the signal transmission loss data change patterns in the multi-dimensional feature correlation model, and the degree of fit between the fiber optic link delay data change segments and the fiber optic link delay data change patterns in the multi-dimensional feature correlation model.

[0069] The optical power fluctuation data segments, signal transmission loss data segments, and fiber optic link delay data segments extracted from the time-series evolution trajectory of fault characteristics are input into the three data input channels of the multi-dimensional correlation tracing model. Internally, the feature extraction submodule first extracts and standardizes key change features from each data segment. Then, in the correlation calculation layer, the model's internal feature collaborative comparison mechanism meticulously compares the extracted key change features of the current data with the standard key change features of the multi-dimensional feature correlation patterns stored in the model. The comparison process considers not only the similarity of individual feature values ​​but also the combination relationships between features and the consistency of change trends. Through the above comparison, the fit between the current optical power fluctuation data segment and the optical power fluctuation data change pattern in the model, the fit between the signal transmission loss data segment and the corresponding pattern, and the fit between the fiber optic link delay data segment and the corresponding pattern are calculated respectively.

[0070] Step S136: Calculate the overall correlation and tracing score based on the matching degree of optical power fluctuation data change segments, the matching degree of signal transmission loss data change segments, the matching degree of fiber optic link delay data change segments, and their respective correlation and tracing weights. The overall correlation and tracing score reflects the overall matching of the current data change with the multi-dimensional feature correlation pattern before the fault occurred.

[0071] The similarity scores (i.e., the fit scores) of the three dimensions calculated in step S135 are multiplied by their respective association source tracing weights. These three products are then summed to obtain the overall association source tracing score. Since the association source tracing weights reflect the degree of influence of each dimension on the occurrence of the fault, the overall association source tracing score can comprehensively evaluate the overall matching between the current data change segment and the multi-dimensional feature association pattern before the fault occurred. A higher score indicates that the current data change more closely matches the feature association pattern before the occurrence of this type of fault, and the greater the probability of the fault occurring.

[0072] Step S137: Integrate the overall correlation and tracing score, the fit of the optical power fluctuation data change segment, the fit of the signal transmission loss data change segment, the fit of the fiber optic link delay data change segment, and the corresponding multi-dimensional feature correlation pattern identifier to form a cross-dimensional correlation and tracing result.

[0073] The calculated overall correlation tracing score, along with the matching scores for each of the three dimensions, are integrated with the unique identifier (such as the fault type number) of this multi-dimensional feature correlation pattern in the optical fiber historical fault feature database to form a cross-dimensional correlation tracing result. This cross-dimensional correlation tracing result clearly reveals which historical fault mode is most similar to the current multi-dimensional changes in optical fiber operation data, as well as the matching details for each dimension.

[0074] Step S140: Call the preset fiber optic AI fault prediction model, input the cross-dimensional correlation tracing results into the fiber optic AI fault prediction model to perform dynamic modeling processing, and generate a dynamic prediction sequence of fiber optic faults.

[0075] In the context of the fiber optic communication network in the city's core area, cross-dimensional correlation tracing results revealed the degree of correlation between current data changes and historical failure modes. Next, a fiber optic AI failure prediction model needs to be used to dynamically extrapolate these correlations and predict the likelihood and development trend of failures in the future.

[0076] Step S141: Load the preset multi-dimensional feature collaborative inference parameters inside the fiber optic AI fault prediction model. The multi-dimensional feature collaborative inference parameters include feature inference step size and feature association update cycle.

[0077] Before performing dynamic modeling and prediction, the fiber optic AI fault prediction model needs to load internally preset multi-dimensional feature co-inference parameters. The feature inference step size refers to the time unit in which the model advances in one prediction inference. For example, if the inference step size is set to 1 hour, the model predicts the fault state for the next hour in each step. The feature association update cycle refers to how often the model updates the association parameters between various features based on the latest actual data during the inference process to ensure the accuracy of the inference. These parameters are preset based on historical prediction experience and model training results, stored in the model's configuration file, and loaded for use by the model during the inference process.

[0078] Step S142: Input the cross-dimensional correlation tracing results into the feature preprocessing subunit of the fiber optic AI fault prediction model, and convert the overall correlation tracing score, the fit of optical power fluctuation data change segments, the fit of signal transmission loss data change segments, and the fit of fiber optic link delay data change segments in the cross-dimensional correlation tracing results. Input the converted cross-dimensional correlation tracing results into the multi-dimensional feature collaborative inference subunit of the fiber optic AI fault prediction model, and perform time-series inference on the fit of optical power fluctuation data change segments, the fit of signal transmission loss data change segments, and the fit of fiber optic link delay data change segments according to the preset feature inference step size.

[0079] The scores and fit parameters in the cross-dimensional correlation tracing results need to be preprocessed to meet the input requirements of the fiber optic AI fault prediction model. The function of the feature preprocessing subunit is to standardize or normalize the above parameters so that their numerical range and distribution conform to the characteristics of the input data during model training. After the transformation, the preprocessed results are input into the multi-dimensional feature co-inference subunit of the model. This multi-dimensional feature co-inference subunit predicts and infers the changing trends of optical power fluctuation data fit, signal transmission loss data fit, and fiber optic link delay data fit for multiple future inference steps, starting from the current time, based on the preset feature inference step size. The inference process considers the mutual influence between the dimensions and the changing patterns in historical data.

[0080] Step S143: During the time series simulation, the correlation parameters between changes in optical power fluctuation data, signal transmission loss data, and fiber optic link delay data are updated periodically according to the feature correlation update cycle, so that the simulation process closely matches the real-time changes in data.

[0081] Step S1431: Record the start time of the time series deduction, and at the same time set the feature association update cycle timer. The feature association update cycle timer starts counting from the start time of the time series deduction.

[0082] At the start of the time series simulation, the model records the current start time (usually the current system time). At the same time, a feature association update cycle timer is started, which accumulates time from the start time to track the time elapsed since the last update of the association parameters.

[0083] Step S1432: During the time-series deduction process, the duration of the feature association update cycle timer is monitored in real time. When the duration of the feature association update cycle timer reaches the preset feature association update cycle, the association parameter update operation is triggered.

[0084] At each step of the model's time series simulation, the current duration of the feature association update cycle timer is checked. If the duration reaches the preset feature association update cycle (e.g., the preset cycle is 6 hours, and the timer shows that 6 hours have passed), the association parameter update operation is immediately triggered, and the current time series simulation process is paused.

[0085] Step S1433: In the correlation parameter update operation, extract the latest optical power fluctuation data change segment, the latest signal transmission loss data change segment, and the latest optical fiber link delay data change segment from the optical fiber operation data stream.

[0086] After the update operation is triggered, the model extracts the latest optical power fluctuation data, signal transmission loss data, and optical fiber link delay data from the real-time optical fiber operation data stream through the interface, from the last update (or the start of the simulation) to the current moment, and processes them into the latest data change segments in the same way as in step S131, ensuring that these segments contain the latest change information.

[0087] Step S1434: By calculating the influence coefficients of the latest extracted optical power fluctuation data change segment on the latest extracted signal transmission loss data change segment, the latest extracted optical power fluctuation data change segment on the latest extracted fiber optic link delay data change segment, and the latest extracted signal transmission loss data change segment on the latest extracted fiber optic link delay data change segment, the mutual influence coefficients among the three data change segments are obtained. The mutual influence coefficients reflect the degree to which the change of one data change segment affects the other two data change segments.

[0088] For the three most recent data change segments extracted, statistical analysis methods (such as correlation analysis and regression analysis) are used to calculate the mutual influence coefficients between them. For example, the correlation coefficient between the change in the optical power fluctuation data segment and the change in the signal transmission loss data segment is calculated as the influence coefficient of optical power on signal transmission loss; similarly, the influence coefficients of optical power on delay and signal transmission loss on delay are calculated. The magnitude of these coefficients reflects the degree to which the change in one dimension of data drives or is closely related to the change in another dimension of data.

[0089] Step S1435: Based on the calculated mutual influence coefficient, adjust the correlation parameters in the multi-dimensional feature collaborative inference sub-unit of the fiber optic AI fault prediction model so that the correlation parameter values ​​in the multi-dimensional feature collaborative inference sub-unit of the fiber optic AI fault prediction model are consistent with the latest mutual influence coefficient.

[0090] The multi-dimensional feature collaborative extrapolation subunit stores parameters describing the interrelationships between features of each dimension. After obtaining the latest interrelation coefficients, the fiber optic AI fault prediction model updates the values ​​of these parameters to the latest calculated interrelation coefficients, ensuring that the fiber optic AI fault prediction model uses parameters that reflect the latest correlations of the current data in subsequent extrapolations, thereby improving the accuracy of the extrapolations.

[0091] Step S1436: Reset the duration of the feature association update cycle timer to zero, so that the feature association update cycle timer starts counting again from the initial state and continues the timing deduction.

[0092] After updating the correlation parameters, the duration of the feature correlation update cycle timer is reset to zero, restarting the countdown from the current moment. Then, the fiber optic AI fault prediction model resumes the time-series extrapolation process, using the updated correlation parameters to continue predicting subsequent extrapolation steps.

[0093] Step S144: Through the multi-dimensional feature collaborative extrapolation subunit of the fiber optic AI fault prediction model, calculate the fault occurrence probability parameter and fault type attribution parameter corresponding to each extrapolation time node. The fault occurrence probability parameter reflects the probability of a fault occurring at that extrapolation time node, and the fault type attribution parameter reflects the possible fault types at that extrapolation time node.

[0094] At each time point in the time series simulation (i.e., at the end of each simulation step), the multi-dimensional feature collaborative simulation subunit calculates the probability of a failure occurring at that time point based on the changes in the fit of each dimension's features obtained from the current simulation, combined with the fault probability calculation algorithm within the model. This is the fault probability parameter (usually represented by a value between 0 and 1, with values ​​closer to 1 indicating a higher probability of failure). Simultaneously, based on the degree of matching between the fit of each dimension's features and different fault type patterns, the most likely fault type to occur at that time point is determined, i.e., the fault type attribution parameter (which can be a fault type number or name).

[0095] Step S145: Arrange the fault occurrence probability parameters and fault type attribution parameters corresponding to each simulation time node in chronological order to form an initial fault prediction sequence. Eliminate the abrupt changes in data between adjacent simulation time nodes in the initial fault prediction sequence to obtain a dynamic fiber optic fault prediction sequence after eliminating the abrupt changes in data between adjacent simulation time nodes.

[0096] The probability parameters and fault type attribution parameters for all projected time points are arranged chronologically to form an initial fault prediction sequence. Due to potential uncertainties during model projection, unreasonable abrupt changes may occur in the data of adjacent time points in the initial sequence (such as sudden and significant increases or decreases in probability parameters). Therefore, the initial fault prediction sequence needs to be smoothed, for example, using moving averages or exponential smoothing, to adjust the probability parameters of adjacent time points, eliminate abrupt changes, and make the sequence's trend more stable and reasonable. The smoothed sequence is the final dynamic fiber optic fault prediction sequence.

[0097] Step S150: Based on the fault occurrence status and corresponding fault type association information presented in the fiber optic fault dynamic prediction sequence, trigger the fiber optic fault early warning operation through the early warning trigger interface of the fiber optic operation and maintenance management system to generate a fiber optic fault early warning instruction containing complete fault early warning content.

[0098] In the context of the fiber optic communication network in the core urban area, the dynamic prediction sequence of fiber optic faults indicates the probability and type of future faults. When a high risk of fault occurrence is predicted, an early warning should be triggered promptly to notify maintenance personnel for handling.

[0099] Step S151: Analyze the dynamic prediction sequence of optical fiber faults, extract the fault occurrence probability parameter and fault type attribution parameter corresponding to each time node in the dynamic prediction sequence of optical fiber faults, and form a correspondence table between time nodes and fault information.

[0100] The data in the dynamic prediction sequence of fiber optic faults is analyzed line by line to extract the fault occurrence probability parameter and fault type attribution parameter corresponding to each prediction time node (e.g., 1 hour, 2 hours, etc.). The above information is organized into a tabular data structure (time node and fault information correspondence table), where each row corresponds to a time node and includes the timestamp of that prediction time node, the fault occurrence probability parameter value, and the fault type attribution parameter.

[0101] Step S152: Analyze the time node and fault information correspondence table, identify the time node intervals in the time node and fault information correspondence table where the fault occurrence probability parameter shows a continuous upward trend, and determine the start and end time nodes of the time node intervals where the fault occurrence probability parameter shows a continuous upward trend.

[0102] Step S1521: Extract all time nodes and their corresponding fault occurrence probability parameters from the time node and fault information correspondence table, and arrange them in chronological order of the time nodes to form a fault occurrence probability parameter sequence.

[0103] From the time node and fault information mapping table, all time nodes and their corresponding fault occurrence probability parameters are filtered out, ignoring the fault type attribution parameter. Then, the above time nodes are sorted according to their timestamp order, and the corresponding fault occurrence probability parameters are also arranged accordingly, forming a one-dimensional fault occurrence probability parameter sequence.

[0104] Step S1522: By comparing the failure probability parameter values ​​of two adjacent time nodes in the failure probability parameter sequence, determine the direction of change of the failure probability parameter for each pair of adjacent time nodes.

[0105] Compare the failure probability parameter values ​​at the i-th and (i+1)-th time points in the failure probability parameter sequence. If the (i+1)-th parameter value is greater than the i-th parameter value, the direction of change for that adjacent time point pair is upward; if the (i+1)-th parameter value is less than the i-th parameter value, the direction of change is downward; if they are equal, the direction of change is stable. Record the direction of change for all adjacent time point pairs.

[0106] Step S1523: Based on the continuity of the direction of change of the fault occurrence probability parameter, identify the time node sequence in which the direction of change of the fault occurrence probability parameter is continuously increasing. The time node sequence in which the direction of change of the fault occurrence probability parameter is continuously increasing indicates that the fault occurrence probability parameter shows a continuous upward trend.

[0107] Iterate through the records of adjacent time nodes and look for time node sequences in which multiple consecutive "increasing" directions of change occur. For example, if the direction of change is increasing from time node t1 to t2, from t2 to t3, and from t3 to t4, then time nodes t1, t2, t3, and t4 constitute a time node sequence in which the direction of change is continuously increasing. The probability parameter of the failure corresponding to this time node sequence shows a continuous upward trend.

[0108] Step S1524: For each time node sequence in which the direction of change of the probability parameter of fault occurrence is continuously increasing, determine the first time node of the time node sequence in which the direction of change of the probability parameter of fault occurrence is continuously increasing as the starting time node of the time node interval in which the probability parameter of fault occurrence shows a continuous upward trend, and determine the last time node of the time node sequence in which the direction of change of the probability parameter of fault occurrence is continuously increasing as the ending time node of the time node interval in which the probability parameter of fault occurrence shows a continuous upward trend.

[0109] For each identified sequence of continuously rising time nodes, the timestamp of the first time node in the sequence is determined as the starting time node of the continuously rising trend time node interval, and the timestamp of the last time node in the sequence is determined as the ending time node. In this way, the time when the probability of failure begins to rise continuously and the time when the current continuous rising trend ends (within the range of the prediction sequence) are clearly defined.

[0110] Step S1525: Check whether the time interval between the start time node of the time interval in which the probability parameter of fault occurrence shows a continuous upward trend and the end time node of the time interval in which the probability parameter of fault occurrence shows a continuous upward trend reaches a preset time interval standard. If the preset time interval standard is reached, the time interval in which the probability parameter of fault occurrence shows a continuous upward trend is retained. If the preset time interval standard is not reached, the time interval in which the probability parameter of fault occurrence shows a continuous upward trend is discarded.

[0111] A predefined time interval standard is used to determine whether a continuous upward trend is meaningful. If the time interval (i.e., duration) between the start and end points of a continuous upward trend reaches or exceeds the standard, the upward trend is considered significant and requires further attention; the time interval is retained. If the time interval does not reach the standard, it may only be a temporary fluctuation, and the time interval is discarded.

[0112] Step S1526: Deduplicate the time intervals where the retained fault probability parameters show a continuous upward trend. If multiple time intervals where the fault probability parameters show a continuous upward trend overlap, merge the overlapping time intervals where the fault probability parameters show a continuous upward trend to form a new time interval where the fault probability parameters show a continuous upward trend.

[0113] In practical analysis, multiple consecutive rising time intervals may be identified, and these intervals may overlap in time. For example, one interval is t1-t5, and another is t3-t7, with the two overlapping in the t3-t5 period. In this case, it is necessary to merge these overlapping intervals into a new interval (such as t1-t7) from the earliest start time to the latest end time to avoid issuing duplicate warnings for the same time period of failure risk.

[0114] Step S1527: Record the start and end time nodes of the time interval in which the probability parameter of each fault occurrence shows a continuous upward trend after merging. At the same time, calculate the difference between the initial and final values ​​of the probability parameter of each fault occurrence within the time interval in which the probability parameter of each fault occurrence shows a continuous upward trend, obtain the total increase of the probability parameter of each fault occurrence within the time interval in which the probability parameter of each fault occurrence shows a continuous upward trend, and record it.

[0115] For each consecutively increasing time interval after merging, accurately record its start and end time nodes. Then, extract the fault occurrence probability parameter value corresponding to the start time node of the interval as the initial value, and extract the fault occurrence probability parameter value corresponding to the end time node as the final value. Calculate the difference between the final value and the initial value; this difference represents the total increase in the fault occurrence probability parameter within the interval. The larger the total increase, the more rapidly the fault risk increases.

[0116] Step S153: For the time intervals in which the identified fault probability parameters show a continuous upward trend, summarize the fault type attribution parameters corresponding to all time nodes in the time intervals in which the fault probability parameters show a continuous upward trend, count the frequency of each fault type attribution parameter in the time intervals in which the fault probability parameters show a continuous upward trend, and determine the fault type attribution parameter with the highest frequency as the main fault type information corresponding to the time intervals in which the fault probability parameters show a continuous upward trend.

[0117] For each merged consecutive rising time node interval, fault type attribution parameters corresponding to all time nodes within that interval are collected. Then, frequency statistics are performed on these parameters, i.e., the number of times each fault type attribution parameter appears within that interval is calculated. The fault type attribution parameter with the highest frequency is identified as the primary fault type information corresponding to that interval, because this type occurs most frequently within the prediction interval, indicating the highest probability of this type of fault occurring.

[0118] Step S154: Retrieve the preset fiber optic fault early warning rule base, which contains information on different fault types and the standard for generating early warning content under the changing trend of corresponding fault occurrence probability parameters.

[0119] The fiber optic fault early warning rule base is a database storing the specifications for generating early warning content. For different fault types and the varying trends in the probability parameters of a fault type (such as the rate of increase and the total magnitude of the increase), corresponding early warning content generation standards are pre-defined. These standards specify the elements, expression methods, and severity levels that the early warning information should include. For example, for the "loose fiber optic connector" fault type, if the probability parameter increases significantly in a short period, the early warning content generation standard may require emergency handling suggestions; while if the increase is small, only routine attention may be needed.

[0120] Step S155: Based on the time intervals in which the determined fault probability parameters show a continuous upward trend, the main fault type information corresponding to the time intervals in which the fault probability parameters show a continuous upward trend, and the warning content generation standard in the fiber optic fault warning rule base, construct fault warning content. The fault warning content includes the time interval in which the fault may occur, the main fault type, and a description of the changes in the fault probability.

[0121] For example, step S1551: retrieve the warning content generation standard corresponding to the main fault type information of the time node interval where the fault occurrence probability parameter shows a continuous upward trend from the fiber optic fault warning rule base. The warning content generation standard includes the time interval expression format of the possible fault, the description specification of the main fault type, and the description template of the change in the fault occurrence probability.

[0122] Based on the identified main fault types, the corresponding early warning content generation standard is retrieved from the fiber optic fault early warning rule base. This early warning content generation standard specifies in detail how to describe the time range in which a fault may occur (e.g., "a fault is expected to occur between X month X day X hour and X month X day X hour"), how to standardize the description of the main fault types (e.g., including fault name, common causes, hazards, etc.), and provides a template for describing changes in the probability of a fault occurring (e.g., "the probability of a fault occurring has increased from the initial Y% to the final Z%, showing a clear upward trend").

[0123] Step S1552: According to the time interval description format of the possible fault in the early warning content generation standard, the time node interval in which the determined probability parameter of the fault occurrence shows a continuous upward trend is converted into a specific time range description of the possible fault occurrence, and the start time and end time of the possible fault occurrence are clarified.

[0124] According to the time interval description format specified in the early warning content generation standard, the start and end time nodes of the merged continuous rising time node interval are converted into a time range described in natural language. For example, if the start time node is 8:00 on May 20, 2024, and the end time node is 12:00 on May 20, 2024, it may be described as "The fault is expected to occur between 08:00 on May 20, 2024 and 12:00 on May 20, 2024".

[0125] Step S1553: Based on the main fault type description specification in the early warning content generation standard, describe in detail the main fault type information corresponding to the time node interval where the fault occurrence probability parameter shows a continuous upward trend. The description includes the name of the fault type and the abnormal fiber optic operation phenomena that may occur when the fault occurs.

[0126] Based on the description specifications for major fault types in the early warning content generation standard, the identified major fault types are described in detail. First, the standard name of the fault type is clarified. Then, based on information from the rule base, the abnormal phenomena that typically occur in fiber optic operation when this fault type occurs are listed. For example, for the "loose fiber optic connector" fault, the description might be: "Fault type: loose fiber optic connector. When this fault occurs, possible abnormal phenomena in fiber optic operation include: frequent and increased fluctuations in optical power, unstable signal transmission rate, and intermittent interruptions of some services." Step S1554: Based on the changes in the fault occurrence probability parameter within the time interval during which the fault occurrence probability parameter shows a continuous upward trend, and in conjunction with the warning content, generate a fault occurrence probability change description template in the standard, and write a fault occurrence probability change description. This fault occurrence probability change description includes the initial state of the fault occurrence probability parameter, the final state of the fault occurrence probability parameter, and the key node states in the process of the fault occurrence probability parameter change.

[0127] Referring to the changes in the probability of failure parameters within a continuously rising time interval (initial value, final value, total increase, and values ​​at intermediate critical nodes), and combining this with the failure probability change description template provided in the early warning content generation standard, a specific change description is written. For example, the template might be: "The probability of failure increases from [initial value]% at the beginning of the interval to [final value]% at the end of the interval, with a total increase of [total increase]%. Around the [critical node time], the probability parameter experiences a significant jump, from [A]% to [B]%." The actual data is then filled into the template to form a complete description.

[0128] Step S1555: After describing the time range in which the fault may occur, connect it with a detailed description of the main fault types, and then connect it with a description of the changes in the probability of the fault occurring. Connect the contents of each part through preset logical connectors to finally form the fault warning content.

[0129] The description of the possible time range of the fault generated in step S1552, the detailed description of the main fault types generated in step S1553, and the description of the changes in the probability of the fault occurring generated in step S1554 are arranged in the above order. Pre-set logical connectors (such as "specifically manifested as:", "its probability changes as:", etc.) are used to connect the various descriptions naturally, making the entire warning content clear, logically coherent, and forming a complete text, i.e., the fault warning content.

[0130] Step S156: Add warning generation time information and data source information to the fault warning content. The data source information points to the fiber optic operation data stream used to generate the fault warning content.

[0131] At the beginning or end of the constructed fault warning text, add specific time information (accurate to the minute) indicating the timeliness of the warning generation so that operations and maintenance personnel can understand the timeliness of the warning. Simultaneously, add data source information, specifying the time period, region, or fiber optic link from which the warning was generated, such as "Data Source: Real-time operational data stream of fiber optic link A in the core urban area from May 19th to May 20th, 2024." This allows operations and maintenance personnel to trace the original data basis for the warning.

[0132] Step S157: According to the instruction format recognizable by the fiber optic operation and maintenance management system, encapsulate the fault warning content, which includes the warning generation time information and data source information, to generate a fiber optic fault warning instruction containing complete fault warning content.

[0133] Fiber optic maintenance and management systems typically have specific instruction format requirements to ensure the system can correctly parse and process early warning information. The text containing the early warning generation time, data source information, and fault warning content is encapsulated according to the system's required instruction format (such as XML, JSON, or a specific message format). The encapsulated instruction should include necessary fields such as instruction type identifier, early warning level (determined based on the probability and type of fault), and recipient identifier, ultimately forming a fiber optic fault early warning instruction. This instruction can be received and parsed by the fiber optic maintenance and management system, triggering subsequent early warning display, notification distribution, and other operations.

[0134] Figure 2 The illustration shows exemplary hardware and software components of a fiber optic AI fault prediction system 100 that incorporates big data analytics and can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 may be used in the fiber optic AI fault prediction system 100 that incorporates big data analytics and to perform the functions described in this application.

[0135] The fiber optic AI fault prediction system 100, which incorporates big data analytics, can be either a general-purpose server or a special-purpose server; both can be used to implement the fiber optic AI fault prediction method incorporating big data analytics of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0136] For example, the fiber optic AI fault prediction system 100 incorporating big data analytics may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the fiber optic AI fault prediction system 100 incorporating big data analytics may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The fiber optic AI fault prediction system 100 incorporating big data analytics also includes an I / O interface 150 between the computer and other input / output devices.

[0137] For ease of explanation, only one processor is described in the fiber optic AI fault prediction system 100 incorporating big data analytics. However, it should be noted that the fiber optic AI fault prediction system 100 incorporating big data analytics may also include multiple processors. Therefore, the steps performed by one processor as described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the fiber optic AI fault prediction system 100 incorporating big data analytics performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0138] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the fiber optic AI fault prediction method combined with big data analysis is implemented as described above.

[0139] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A fiber optic AI fault prediction method combining big data analysis, characterized in that, The method includes: Acquire a continuous optical fiber operation data stream generated in real time during optical fiber operation, the optical fiber operation data stream including optical power fluctuation data, signal transmission loss data, and optical fiber link delay data; The fault feature time-series evolution trajectory is captured and processed by performing fault feature time-series evolution trajectory on the optical fiber operation data stream. Based on the continuous change trend of optical power fluctuation data, signal transmission loss data, and optical fiber link delay data in the optical fiber operation data stream, and combined with the feature evolution rules corresponding to each fault type stored in the optical fiber historical fault feature database, the fault feature time-series evolution trajectory is obtained. Cross-dimensional correlation tracing processing is performed on the temporal evolution trajectory of fault characteristics. The correlation patterns of optical power fluctuation data, signal transmission loss data, and optical fiber link delay data in the optical fiber operation data stream are associated with multi-dimensional features before the fault in the optical fiber historical fault feature database, generating cross-dimensional correlation tracing results. Call the preset fiber optic AI fault prediction model, input the cross-dimensional correlation tracing results into the fiber optic AI fault prediction model to perform dynamic modeling processing, and generate a dynamic prediction sequence of fiber optic faults. Based on the fault occurrence status and corresponding fault type association information presented in the dynamic prediction sequence of optical fiber faults, the optical fiber fault early warning operation is triggered through the early warning triggering interface of the optical fiber operation and maintenance management system, generating an optical fiber fault early warning instruction containing complete fault early warning content.

2. The fiber optic AI fault prediction method combining big data analysis according to claim 1, characterized in that, The process of capturing the time-series evolution trajectory of fault characteristics in the fiber optic operating data stream is based on the continuous changes in optical power fluctuation data, signal transmission loss data, and fiber optic link delay data in the fiber optic operating data stream. Combined with the characteristic evolution patterns corresponding to each fault type stored in the fiber optic historical fault characteristic database, the time-series evolution trajectory of fault characteristics is obtained, including: The optical power fluctuation data, signal transmission loss data, and optical fiber link delay data in the optical fiber operation data stream are synchronously expanded according to the same time dimension to obtain continuous time series data of optical power fluctuation data, continuous time series data of signal transmission loss data, and continuous time series data of optical fiber link delay data. For continuous time series data of optical power fluctuation data, extract the inflection points of data value changes over time in the continuous time series data of optical power fluctuation data, record the time coordinate and data value change amplitude corresponding to each inflection point, and form a set of inflection points of optical power fluctuation data. Using the same inflection point extraction method as for continuous time series data of optical power fluctuation data, inflection point extraction processing is performed on continuous time series data of signal transmission loss data and continuous time series data of optical fiber link delay data to obtain sets of inflection points for signal transmission loss data and sets of inflection points for optical fiber link delay data. The feature evolution patterns corresponding to each fault type are retrieved from the historical fault feature database of optical fibers. The feature evolution patterns include the order of occurrence and correlation pattern of change inflection points of optical power fluctuation data, signal transmission loss data, and optical fiber link delay data before the occurrence of each fault type. The amplitudes of changes in the inflection point sets of optical power fluctuation data, signal transmission loss data, and fiber optic link delay data are normalized respectively. The normalized inflection point sets of optical power fluctuation data, signal transmission loss data, and fiber optic link delay data are then matched with the inflection point patterns in the characteristic evolution laws corresponding to each fault type to determine their evolution trends. Based on the evolution trend matching results, a trajectory framework is constructed that includes the time coordinates and normalized change amplitude of the inflection point of optical power fluctuation data, the time coordinates and normalized change amplitude of the inflection point of signal transmission loss data, and the time coordinates and normalized change amplitude of the inflection point of fiber optic link delay data, as well as the degree of correlation with the evolution law of fault type. By supplementing the trajectory framework with continuous change curves of optical power fluctuation data, signal transmission loss data, and fiber optic link delay data between their respective inflection points, a fault characteristic time-series evolution trajectory is formed.

3. The fiber optic AI fault prediction method combining big data analysis according to claim 2, characterized in that, The continuous time series data of optical power fluctuation data is used to extract inflection points of data values ​​changing over time. The time coordinates and data value change amplitudes corresponding to each inflection point are recorded, forming a set of inflection points for optical power fluctuation data, including: The continuous time series data of optical power fluctuation data is segmented into time series segments. Each time series segment contains a preset number of continuous data points, and there are preset lengths of overlapping data points between adjacent time series segments. The slope of the data value change in each time series segment is calculated by dividing the numerical difference of adjacent data points in each time series segment by the corresponding time difference. Compare the change slopes of adjacent time series segments. If the difference in the change slopes of adjacent time series segments exceeds the preset slope difference standard, then mark the middle data point among the overlapping data points of the two adjacent time series segments as a candidate change inflection point. For each candidate inflection point, calculate the average slope of the changes of a preset number of data points before and after the candidate inflection point, and obtain the average slope before the candidate inflection point and the average slope after the candidate inflection point respectively. If the difference between the average slope before the candidate inflection point and the average slope after the candidate inflection point still exceeds the slope difference standard, then the candidate inflection point is confirmed as a formal inflection point. Record the time coordinates corresponding to each formal change inflection point, and calculate the difference between the data value before and after the formal change inflection point to determine the magnitude of the data value change. The time coordinates and data value change amplitudes of all formal change inflection points are organized in chronological order to form a set of inflection points for optical power fluctuation data changes.

4. The fiber optic AI fault prediction method combining big data analysis according to claim 1, characterized in that, The process involves performing cross-dimensional correlation tracing on the temporal evolution trajectory of fault characteristics. This involves correlating changes in optical power fluctuations, signal transmission loss, and fiber link delay in the fiber optic operation data stream with multi-dimensional feature correlation patterns from the historical fiber fault characteristic database before the fault occurred, generating cross-dimensional correlation tracing results, including: Extract optical power fluctuation data change segments, signal transmission loss data change segments, and fiber optic link delay data change segments from the fault characteristic time-series evolution trajectory. Each optical power fluctuation data change segment, each signal transmission loss data change segment, and each fiber optic link delay data change segment contains the change information of the corresponding data within the same time interval. Retrieve multi-dimensional feature association patterns before the fault occurs from the optical fiber historical fault feature database. The multi-dimensional feature association patterns describe the mutual influence and timing coordination characteristics between changes in optical power fluctuation data, changes in signal transmission loss data, and changes in optical fiber link delay data before the fault occurs. Correlation and tracing weights are assigned to the optical power fluctuation data change segment, the signal transmission loss data change segment, and the fiber optic link delay data change segment, respectively. The correlation and tracing weights are determined based on the degree of influence of each of the optical power fluctuation data change, signal transmission loss data change, and fiber optic link delay data change on the occurrence of the fault. Based on the assigned correlation and tracing weights, a multi-dimensional correlation and tracing model is constructed by setting the structural parameters of the input layer, feature extraction layer, correlation calculation layer, and output layer. The input of the multi-dimensional correlation and tracing model consists of optical power fluctuation data segments, signal transmission loss data segments, and fiber optic link delay data segments. The output of the multi-dimensional correlation and tracing model is the matching degree parameter with the multi-dimensional feature correlation pattern. The extracted optical power fluctuation data change segments, signal transmission loss data change segments, and fiber optic link delay data change segments are input into the multi-dimensional correlation tracing model. Through the feature collaborative comparison mechanism inside the multi-dimensional correlation tracing model, the degree of fit between the optical power fluctuation data change segments and the optical power fluctuation data change patterns in the multi-dimensional feature correlation pattern, the degree of fit between the signal transmission loss data change segments and the signal transmission loss data change patterns in the multi-dimensional feature correlation pattern, and the degree of fit between the fiber optic link delay data change segments and the fiber optic link delay data change patterns in the multi-dimensional feature correlation pattern are calculated. Based on the matching degree of optical power fluctuation data change segments, the matching degree of signal transmission loss data change segments, the matching degree of fiber optic link delay data change segments, and their respective correlation and tracing weights, the overall correlation and tracing score is calculated. The overall correlation and tracing score reflects the overall matching of the current data change with the multi-dimensional feature correlation pattern before the fault occurred. The overall correlation and tracing score, the fit of optical power fluctuation data segments, the fit of signal transmission loss data segments, the fit of fiber optic link delay data segments, and the corresponding multi-dimensional feature correlation pattern identifiers are integrated to form a cross-dimensional correlation and tracing result.

5. The fiber optic AI fault prediction method combining big data analysis according to claim 4, characterized in that, The assigned correlation tracing weights are used to construct a multi-dimensional correlation tracing model by setting structural parameters for the input layer, feature extraction layer, correlation calculation layer, and output layer. The inputs to the multi-dimensional correlation tracing model are data segments showing changes in optical power fluctuations, signal transmission loss, and fiber optic link delay. The output of the multi-dimensional correlation tracing model is a matching degree parameter with the multi-dimensional feature correlation pattern, including: The input layer structure of the multi-dimensional correlation tracing model is determined. The input layer of the multi-dimensional correlation tracing model contains three data input channels, which correspond to the input of optical power fluctuation data change segments, signal transmission loss data change segments, and fiber optic link delay data change segments, respectively. A feature extraction submodule is set up for each data input channel. Each feature extraction submodule is used to extract key change features in the corresponding data change segment and to standardize the key change features. The key change features include the data change period and the frequency of data peak occurrence. An association calculation layer is constructed within the multi-dimensional association tracing model. The association calculation layer receives the standardized key change features output by the three feature extraction sub-modules. Combined with the assigned association tracing weights, it calculates the similarity between the key change features of the optical power fluctuation data change segment and the key change features of the optical power fluctuation data in the multi-dimensional feature association mode, the key change features of the signal transmission loss data change segment and the key change features of the signal transmission loss data in the multi-dimensional feature association mode, and the key change features of the fiber optic link delay data change segment and the key change features of the fiber optic link delay data in the multi-dimensional feature association mode. The output layer of the multi-dimensional correlation tracing model is set up. The output layer of the multi-dimensional correlation tracing model integrates the similarity of key change features of optical power fluctuation data, signal transmission loss data, and fiber optic link delay data in a weighted manner to obtain the overall matching degree parameter, and outputs the individual similarity parameters of each key change feature. The filtering parameters of the feature extraction submodule and the weight coefficients of the correlation calculation layer within the multi-dimensional correlation tracing model are initialized. The initialization parameters are obtained by training based on sample data in the optical fiber historical fault feature database. By inputting sample data change segments from the optical fiber historical fault feature database, the multi-dimensional correlation tracing model is validated and adjusted so that the matching degree parameter output by the multi-dimensional correlation tracing model can accurately reflect the matching situation between the data change segments and the multi-dimensional feature correlation pattern. After completing the construction of the multi-dimensional correlation tracing model, save the structure and parameters of the multi-dimensional correlation tracing model to form a multi-dimensional correlation tracing model that can be directly called.

6. The fiber optic AI fault prediction method combining big data analysis according to claim 1, characterized in that, The process involves calling a preset fiber optic AI fault prediction model, inputting cross-dimensional correlation tracing results into the model for dynamic modeling, and generating a dynamic fiber optic fault prediction sequence, including: The fiber optic AI fault prediction model is loaded with preset multi-dimensional feature collaborative inference parameters, which include feature inference step size and feature association update cycle. The cross-dimensional correlation tracing results are input into the feature preprocessing subunit of the fiber optic AI fault prediction model. The overall correlation tracing score, the fit of optical power fluctuation data change segments, the fit of signal transmission loss data change segments, and the fit of fiber optic link delay data change segments in the cross-dimensional correlation tracing results are converted. The converted cross-dimensional correlation tracing results are input into the multi-dimensional feature collaborative inference subunit of the fiber optic AI fault prediction model. According to the preset feature inference step size, the fit of optical power fluctuation data change segments, the fit of signal transmission loss data change segments, and the fit of fiber optic link delay data change segments are time-series inferred. During the time series simulation, the correlation parameters between changes in optical power fluctuation data, signal transmission loss data, and fiber optic link delay data are updated periodically according to the feature correlation update cycle, so that the simulation process closely follows the real-time changes in data. Through the multi-dimensional feature collaborative extrapolation subunit of the fiber optic AI fault prediction model, the fault occurrence probability parameter and fault type attribution parameter corresponding to each extrapolation time node are calculated. The fault occurrence probability parameter reflects the probability of a fault occurring at that extrapolation time node, and the fault type attribution parameter reflects the possible fault type at that extrapolation time node. The fault occurrence probability parameters and fault type attribution parameters corresponding to each simulation time node are arranged in chronological order to form an initial fault prediction sequence. The abrupt changes in the data of adjacent simulation time nodes in the initial fault prediction sequence are eliminated to obtain the fiber optic fault dynamic prediction sequence after eliminating the abrupt changes in the data of adjacent simulation time nodes.

7. The fiber optic AI fault prediction method combining big data analysis according to claim 6, characterized in that, During the time-series simulation, the correlation parameters between changes in optical power fluctuation data, signal transmission loss data, and fiber optic link delay data are periodically updated according to the feature correlation update cycle, so that the simulation process closely reflects the real-time changes in data, including: Record the start time of the time series deduction and set a feature association update cycle timer. The feature association update cycle timer starts counting from the start time of the time series deduction. During the time-series simulation, the duration of the feature association update cycle timer is monitored in real time. When the duration of the feature association update cycle timer reaches the preset feature association update cycle, the association parameter update operation is triggered. In the correlation parameter update operation, the latest optical power fluctuation data change segment, the latest signal transmission loss data change segment, and the latest optical fiber link delay data change segment are extracted from the optical fiber operation data stream. By calculating the influence coefficients of the latest extracted optical power fluctuation data change segment on the latest extracted signal transmission loss data change segment, the latest extracted optical power fluctuation data change segment on the latest extracted fiber optic link delay data change segment, and the latest extracted signal transmission loss data change segment on the latest extracted fiber optic link delay data change segment, the mutual influence coefficients among the three data change segments are obtained. These mutual influence coefficients reflect the degree to which the change in one data change segment affects the other two data change segments. Based on the calculated mutual influence coefficient, adjust the correlation parameters in the multi-dimensional feature collaborative deduction sub-unit of the fiber optic AI fault prediction model to ensure that the correlation parameter values ​​in the multi-dimensional feature collaborative deduction sub-unit of the fiber optic AI fault prediction model are consistent with the latest mutual influence coefficient. The duration of the feature association update cycle timer is reset to zero, so that the feature association update cycle timer starts counting again from the initial state and continues the timing deduction.

8. The fiber optic AI fault prediction method combining big data analysis according to claim 1, characterized in that, Based on the fault occurrence status and corresponding fault type association information presented in the dynamic prediction sequence of fiber optic faults, the fiber optic fault early warning operation is triggered through the early warning triggering interface of the fiber optic operation and maintenance management system, generating a fiber optic fault early warning instruction containing complete fault early warning content, including: The optical fiber fault dynamic prediction sequence is analyzed, and the fault occurrence probability parameter and fault type attribution parameter corresponding to each time node in the optical fiber fault dynamic prediction sequence are extracted to form a time node and fault information correspondence table. Analyze the time node and fault information correspondence table, identify the time node intervals in the time node and fault information correspondence table where the fault occurrence probability parameter shows a continuous upward trend, and determine the start and end time nodes of the time node intervals where the fault occurrence probability parameter shows a continuous upward trend. For the time intervals in which the identified fault probability parameters show a continuous upward trend, the fault type attribution parameters corresponding to all time nodes in the time intervals in which the fault probability parameters show a continuous upward trend are summarized. The frequency of occurrence of each fault type attribution parameter in the time intervals in which the fault probability parameters show a continuous upward trend is counted. The fault type attribution parameter with the highest frequency is determined as the main fault type information corresponding to the time intervals in which the fault probability parameters show a continuous upward trend. The system retrieves a preset fiber optic fault early warning rule base, which contains information on different fault types and standards for generating early warning content under the changing trends of corresponding fault occurrence probability parameters. Based on the time intervals in which the probability of failure occurs continuously, the main fault types corresponding to the time intervals in which the probability of failure occurs continuously increase, and the warning content generation standards in the fiber optic fault warning rule base, a fault warning content is constructed. The fault warning content includes the time interval in which the fault may occur, the main fault types, and a description of the changes in the probability of failure. Add warning generation time information and data source information to the fault warning content, wherein the data source information points to the fiber optic operation data stream used to generate the fault warning content; According to the instruction format recognizable by the fiber optic operation and maintenance management system, the fault warning content, which includes the warning generation time information and data source information, is encapsulated to generate a fiber optic fault warning instruction containing complete fault warning content.

9. The fiber optic AI fault prediction method combining big data analysis according to claim 8, characterized in that, The analysis of the time node and fault information correspondence table identifies time node intervals in the table where the fault occurrence probability parameter shows a continuous upward trend, and determines the start and end time nodes of these time node intervals, including: Extract all time nodes and their corresponding fault occurrence probability parameters from the time node and fault information correspondence table, and arrange them in chronological order to form a fault occurrence probability parameter sequence. By comparing the failure probability parameter values ​​of two adjacent time nodes in the failure probability parameter sequence, the direction of change of the failure probability parameter for each pair of adjacent time nodes is determined. Based on the continuity of the direction of change of the probability parameter of failure occurrence, identify the time node sequence in which the direction of change of the probability parameter of failure occurrence is continuously increasing. The time node sequence in which the direction of change of the probability parameter of failure occurrence is continuously increasing indicates that the probability parameter of failure occurrence is showing a continuous upward trend. For each time node sequence in which the direction of change of the probability parameter of failure is continuously increasing, the first time node in the time node sequence in which the direction of change of the probability parameter of failure is continuously increasing is determined as the starting time node of the time node interval in which the probability parameter of failure is continuously increasing, and the last time node in the time node sequence in which the direction of change of the probability parameter of failure is continuously increasing is determined as the ending time node of the time node interval in which the probability parameter of failure is continuously increasing. Check whether the time interval between the start time of the time interval in which the probability parameter of the fault occurrence shows a continuous upward trend and the end time of the time interval in which the probability parameter of the fault occurrence shows a continuous upward trend reaches a preset time interval standard. If the preset time interval standard is reached, the time interval in which the probability parameter of the fault occurrence shows a continuous upward trend is retained; if the preset time interval standard is not reached, the time interval in which the probability parameter of the fault occurrence shows a continuous upward trend is discarded. The time intervals where the probability of failure occurs continuously are deduplicated. If there are multiple time intervals where the probability of failure occurs continuously are continuously increasing and there is overlap, the time intervals where the probability of failure occurs continuously are merged into a new time interval where the probability of failure occurs continuously is continuously increasing. Record the start and end times of the time interval in which the probability parameter of each fault occurrence shows a continuous upward trend after merging. At the same time, calculate the difference between the initial and final values ​​of the probability parameter of each fault occurrence within the time interval in which the probability parameter of each fault occurrence shows a continuous upward trend, obtain the total increase of the probability parameter of each fault occurrence within the time interval in which the probability parameter of each fault occurrence shows a continuous upward trend, and record it.

10. A fiber optic AI fault prediction system combining big data analytics, characterized in that, The fiber optic AI fault prediction system combining big data analysis includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the fiber optic AI fault prediction method combining big data analysis as described in any one of claims 1-9.

Images